Studies on Biomarker Development in Breast Cancer Ewan K a Millar

Studies on Biomarker Development

In Breast Cancer

Ewan K A Millar

Doctor of Medicine

The University of New South Wales

2011

Studies on Biomarker Development

In Breast Cancer

Ewan K A Millar

BSc (Hons) MB ChB (Glasg) FRCPath (UK) FRCPA

A report submitted for the degree of

Doctor of Medicine (MD) by Published Works

The University of New South Wales

August 2011

Originality Statement ‘I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project's design and conception or in style, presentation and linguistic expression is acknowledged.’

Ewan KA Millar

Date

Acknowledgements

I would like to thank several individuals, without whom the completion of the work in this thesis would not have been possible.

Firstly, I would like to thank my beautiful wife Julia, for all her love and support and who has always been so understanding and encouraging of my career and research work, which has at times, kept me from being quite as involved in family life as I should have been. To my rapidly growing and wonderful children David, William and

Georgia, thanks for all your love and support.

I would not be in the position of submitting this Thesis without the generous support and guidance offered to me by Professor Rob Sutherland. His mentorship, intellectual input and friendship over the past 6 years have been immense and have helped develop my research career immeasurably. His skills and expertise as a researcher have given me much to aspire to in the years ahead. Other members of the Translational Breast Cancer Research Group also stand out for a special mention: A/Professor Sandra O’Toole and Dr Catriona McNeil have always provided invaluable knowledge and support in all aspects of our joint research goals. To Alice

Boulghourjian and her predecessor Sarah Eggleton for tolerating, at times exhausting and demanding, requests for scientific support with their skills in immunohistochemistry. Associate Professor Peter Graham for his generous support and expert clinical input to our collaborative studies which will no doubt continue into the future. To Professor Soon Lee for long-standing support and encouragement, over the years, to pursue an academic career, with much useful advice.

To other colleagues at the Garvan Institute, Professor Liz Musgrove and Dr Alison

Butt who have continued to generously support my work and research career.

Dedication

For Julia, David, William and Georgia.

Contents

Page

Publications...... 8

Conference Abstracts...... 14

Abstract...... 17

Chapter 1. Background...... 18

Chapter 2. Immunophenotyping breast cancer using surrogate biomarker panels for molecular subtype ...... 35

Chapter 3. Endocrine Resistance in ER+ Breast Cancer...... 41

3.1 Apoptotic Pathways...... 43

3.2 Proliferative Pathways...... 46

Chapter 4. Signalling Pathways...... 51

Chapter 5. Tumour Hypoxia & HIF Pathway...... 63

Chapter 6. DNA Repair Pathways...... 68

Chapter 7. Ductal Carcinoma in-Situ...... 70

Chapter 8. Rare types of breast tumours: phyllodes tumours & non-Hodgkin’s lymphoma...... 72

Chapter 9. Authorship of papers for Thesis...... 75

Chapter 10. Significance of findings from published work...... 83

References...... 85

Publications arising from this work...... 102

Publications in peer reviewed journals

EM1. Murphy N, Millar E, Lee CS. Gene expression profiling in breast cancer: towards individualising patient management. Pathology. 2005; 37: 271-277.(IF 2.17).

EM2. O’Toole SA, Selinger T, Millar EKA , Lum T, Beith JM. Molecular assays in breast cancer pathology. Pathology. 2011; 43: 116-127. (IF 2.17).

EM3. Millar EKA, Graham PH, O’Toole SA, McNeil CM, Browne L, Morey AL,

Eggleton S, Beretov J, Theocharous C, Capp A, Nasser E, Kearsley JH, Papadatos

G, Delaney G, Fox C, Sutherland RL. Prediction of local recurrence, distant metastases and death following breast-conserving therapy in early-stage invasive breast cancer using a five biomarker panel. Journal of Clinical Oncology. 2009; 27:

4701-8. (IF 18.97)

EM4. Millar EKA, Graham PH, McNeil CM, Browne L, O’Toole SA, Boulghourjian A,

Papadatos G, Delaney G Nasser E, Kearsley JH, Fox C, Capp A, Sutherland RL.

Prediction of outcome in early ER+ breast cancer is improved using a biomarker panel which includes Ki-67 and p53. British Journal of Cancer. 2011. 105:272-80. (IF

4.83)

EM5. Graham P, Jagavkar R, Brown L, Millar E. Supraclavicular radiotherapy must be limited laterally by the coracoid to avoid significant adjuvant breast nodal radiotherapy lymphoedema risk. Australasian Radiology (Journal of Medical Imaging and Radiation Oncology). 2006. 50: 578-582. (IF 0.947)

EM6. Millar EKA, Anderson LR, McNeil CM, O’Toole SA, Pinese M, Crea P, Morey

AL, Biankin AV, Henshall SM, Musgrove EA, Sutherland RL, Butt AJ. BAG-1 predicts patient outcome and tamoxifen responsiveness in ER positive invasive ductal carcinoma of the breast. Br J Cancer.100; 122-133, 2009. (IF 4.83).

EM7. Roberts CG*, Millar EKA*, O’Toole SA, McNeil CM, Lehrbach GM, Pinese M,

Tobelmann P, McCloy RA, Musgrove EA, Sutherland RL and Butt AJ. Identification of PUMA as an estrogen target gene that mediates the apoptotic response to tamoxifen in human breast cancer cells and predicts patient outcome and tamoxifen responsiveness in breast cancer. Oncogene. 2011.30:3186-97. (IF 7.41)

*Both authors contributed equally to this work.

EM8. McNeil CM, Sergio CM, Anderson LR, Inman CK, Murphy NC, Millar EKA,

Crea P, Kench JG, Alles MC, Gardiner-Garden M, Ormandy CJ, Butt AJ, Henshall

SM, Musgrove EA, Sutherland RL. C-Myc overexpression and endocrine resistance in breast cancer. Journal of Steroid Biochemistry and Molecular Biology. 2006;

102:147-55. (IF 2.89).

EM9. Wang Y, Dean JL, Millar EKA, Tran TH, McNeil CM, Burd CJ, Henshall SM,

Utama FE, Witkiewicz A, Rui H, Sutherland RL, Knudsen KE, Knudsen ES. Cyclin

D1b is aberrantly regulated in response to therapeutic challenge and promotes resistance to estrogen antagonists. Cancer Research. 2008; 68:5628- 5638. (IF

8.23)

EM10. Millar EKA, Dean JL, McNeil CM, O’Toole SA, Henshall SM, Tran T, Lin J,

Quong A, Comstock CES, Witkiewicz A, Musgrove EA, Rui H, Le Marchand L,

Setiawan VW, Haiman CA, Knudsen KE, Sutherland RL, Knudsen ES. Cyclin D1b

Protein expression in breast cancer is independent of cyclin D1a and associated with poor disease outcome. Oncogene. 2009; 28:1812-20. (IF 7.41).

EM11. López-Knowles E, O’Toole SA, McNeil CM, Millar EKA, Qiu MR, Crea P,

Musgrove EA, Sutherland RL. PI3K pathway activation in breast cancer is associated with the basal-like phenotype and cancer-specific mortality. International

Journal of Cancer. 2010. 126:1121-31. (IF 4.93).

EM12. Fedele CG, Ooms LM, Ho M, Vieusseux J, O’Toole SA, Millar EKA, Lopez-

Knowles E, Sriratana A, Gurung R, Baglietto L, Giles GG, Bailey CG, Rasko JEJ,

Shields BJ, Price JT, Majerus PW, Sutherland RL, Tiganis T, McLean CA, Mitchell

CA. The PtdIns(3,4)P2 4-phosphatase, INPP4B, regulates ER-positive mammary cell proliferation and is lost in human basal-like breast carcinomas. Proceedings of the National Academy of Science (USA). 2010.107:22231-6. (IF 9.77).

EM13. O’Toole SA, Machalek D, Shearer R , Millar EKA, Nair R, McLeod D,

Cooper C, McFarland A, Ru Qiu M, McNeil CM, Rabinovich B, Martelotto L, Vu D,

Musgrove E, Sutherland RL, Watkins N, Swarbrick A. Hedgehog overexpression is associated with stromal interactions and predicts for poor outcome in breast cancer.

Cancer Research. 2011. 71:4002-4014. (IF 8.23).

EM14. Zardawi SJ, Zardawi I, McNeil CM, Millar EKA, McLeod D, Morey AL, Crea

P, Murphy NM, Lopez-Knowles E, Oakes SR, Ormandy CJ, Qiu MR, Hamilton A,

Spillane A, Lee CS, Sutherland RL, Musgrove EA, O’Toole SA. Notch1 expression is an early event in breast cancer development and is associated with the HER-2 molecular subtype. Histopathology. 2010; 56, 286–296. (IF 3.57).

EM15. López-Knowles E, Zardawi SJ, McNeil CM, Millar EKA, Crea P, Musgrove

EA, Sutherland RL,. O’Toole SA. Cytoplasmic localization of ß catenin is a marker of poor outcome in breast cancer patients. Cancer Epidemiology Biomarkers

Prevention. 2010;19:301-9. (IF 4.19).

EM16. Murphy NC*, Biankin AV*, Millar EKA*, McNeil CM, O’Toole SA, Segara D,

Crea P, Olayioye MA, Lee CS, Fox SB, Morey AL, Christie M, Musgrove EA, Daly

RJ, Lindeman GJ,. Henshall SM, Visvader JE, Sutherland RL. Loss of STARD10 expression identifies a group of poor prognosis breast cancers independent of

HER2/Neu and triple negative status. International Journal of Cancer. 2010.

126:1445-53. (IF 4.93).

*These authors contributed equally to this work.

EM17. Fleuren EDG, O’Toole SA, Millar EKA, McNeil CM, Lopez-Knowles E,

Brummer T, Sutherland RL, Daly RJ. Overexpression of the oncogenic signal transducer Gab2 occurs early in breast cancer development. International Journal of

Cancer. 2010. 127: 1486–1492. (IF 4.93).

EM18. Tan EY, Yan M, Campo L, Han C, Takano E, Turley H, Pezzella F, Gatter KC,

Millar EKA, O’Toole SA, McNeil CM, Crea P, Segara D, Sutherland RL, Harris AL,

Fox SB. The key hypoxia regulated gene CAIX is upregulated in basal-like breast tumors and is associated with resistance to chemotherapy. British Journal of Cancer.

2009. 100: 405-411. (IF 4.83).

EM19. Yan M, Jene N, Byrne D, Millar EKA, O’Toole SA, McNeil CM, Sutherland

RL, Fox SB. Recruitment of regulatory T cells is driven by hypoxia induced CXCR4 expression, and is associated with poor prognosis in basal-like breast cancers.

Breast Cancer Research. 2011. 13:R47. (IF 5.79).

EM20. Chan P, Moeller A, Liu MCP, Sceneay JE, Wong CSF, Wadell N, Huang K,

Dobrovic A, Millar EKA, O’Toole SA, McNeil CM, Sutherland RL, Bowtell D, Fox SB.

The expression of the ubiquitin ligase SIAH (seven in absentia homolog) 2 is mediated through gene copy number in breast cancer and is associated with a basal-like phenotype and p53 expression. Breast Cancer Research. 2011. 13:R19.

(IF 5.79).

EM21. Xu H, Yan M, Patra J, Yan Y, Swagemakers S, Thomasewski J, Verschoor S,

Millar EKA, van der Spek P, Reis-Filho JS, Ramsay R, O’Toole SA, McNeil CM,

Sutherland RL, McKay MJ, Fox SB. Enhanced Rad21 cohesin expression confers poor prognosis and resistance to chemotherapy in high-grade luminal, basal and

HER2 breast cancer. Breast Cancer Research. 2011;13:R9. (IF 5.79).

EM22. Millar EKA and Leong A S-Y. Significance and assessment of margins of excision in ductal carcinoma in-situ of the breast. Advances in Anatomical Pathology.

2001. 8:338-344. (IF 3.3).

EM23. Millar EKA, Tran K, Marr P, Graham PH. p27kip-1, cyclin A and cyclin D1 protein expression in ductal carcinoma in-situ of the breast: p27kip-1 correlates with hormone receptor status but not with local recurrence. Pathology International.

2007. 57: 183-9. (IF 1.48).

EM24. Millar EKA, Beretov J, Marr P, Sarris M, Clarke RA, Kearsley JM and Lee

CS. Malignant phyllodes tumours of the breast display increased stromal p53 protein expression. Histopathology. 1999. 34:491-496. (IF 3.57).

EM25. Yilmaz MH, Millar EKA, Theocharous C, Graham PH. Metachronous bilateral primary low-grade mucosa-associated lymphoid tissue (MALT) non-Hodgkin’s lymphoma of the breast. Asia-Pacific Journal of Clinical Oncology. 2009. 5:154-158.

(IF 0.296).

Conference Abstracts

1. Millar EKA, Graham PH, McNeil CM, Browne L, O’Toole SA, Boulghourjian A,

Papadatos G, Delaney G Nasser E, Kearsley JH, Fox C, Capp A, Sutherland

RL. Prediction of outcome in early ER+ breast cancer is improved using a

biomarker panel which includes Ki-67 and p53. i) International Academy of

Pathology, Annual Scientific Meeting, Sydney 4-6th June, 2011 (Commended);

ii) 13th Milan International Breast Cancer Conference, Milan, Italy 23-25th June

2011.

2. O’Toole SA, Machalek D, Shearer R, Millar EKA, Nair R, McLeod D, Cooper

C, Ru Qui M, Sutherland RL, Watkins DN, Swarbrick A. Hedgehog over-

expression predicts poor outcome in breast cancer and is a potential

therapeutic target for metastatic breast cancer. i) San Antonio Breast Cancer

Symposium, San Antonio, USA Dec 2010; ii) Breakthrough Breast Cancer

Triple Negative Breast Cancer Meeting, London March 2011; iii) International

Academy of Pathology Annual Scientific Meeting Sydney 4-6th June 2011.

(Best poster prize winner).

3. Millar EKA, Graham PH, O’Toole SA, McNeil CM, Browne L, Morey AL,

Eggleton S, Beretov J, Theocharous C, Capp A, Nasser E, Kearsley JH,

Papadatos G, Delaney G, Fox C, Sutherland RL. Prediction of local

recurrence, distant metastases and death following breast-conserving therapy

in early-stage invasive breast cancer using a five biomarker panel. Oral

presentation Australasian Society of Breast Diseases, Gold Coast, Australia,

2-4th Oct 2009.

4. Lopez-Knowles E, O’Toole SA, McNeil CM, Millar EKA, Qui MR, Crea P, Daly

RJ, Musgrove EA, Sutherland RL. PI3K pathway activation in breast cancer is

associated with the basal-like phenotype and cancer-specific mortality. San

Antonio Breast Cancer Symposium, Texas, USA, 9-13 Dec 2009.

5. Lopez-Knowles E, Zardawi SJ, McNeil CM, Millar EKA, Crea P, Musgrove EA,

Sutherland RL, O’Toole SA. Cytoplasmic localisation of B-catenin is a marker

of poor outcome in breast cancer patients. AACR conference: Advances in

breast cancer research, San Diego, USA, 13-16 Oct 2009.

6. McNeil CM, O’Toole SA, Millar EKA, Kench JG, Segara D, Morey AL, Lopez-

Knowles E, Crea P, Henshall, Musgrove EA, Sutherland RL. Differences in cell

cycle and apoptotic biomarker expression between molecular subtypes of

invasive ductal breast carcinoma. AACR conference: Advances in breast

cancer research, San Diego, USA, 13-16 Oct 2009.

7. O’Toole S, Swarbrick A, Millar E, McLeod D, McNeil C, Qiu MR, Lopez-

Knowles E, Caldon E, Oakes S, Ormandy C, Morey A, Musgrove E, Henshall

S, Sutherland R. Aberrant Hedgehog signaling is an early event in breast

cancer development. San Antonio Breast Cancer Symposium, San Antonio,

USA Dec 2008.

8. Millar EKA, Anderson LR, McNeil CM, O’Toole SA, Pinese M, Crea P, Morey

AL, Biankin AV, Henshall SM, Musgrove EA, Sutherland RL, Butt AJ. BAG-1

predicts patient outcome and tamoxifen responsiveness in ER positive

invasive ductal carcinoma of the breast. i) International Academy of Pathology,

Annual Scientific Meeting Sydney, June 2008 (Commended); ii) Sydney

Cancer Conference, Sydney University July 2008; iii) Leura VI International

Breast Cancer Conference, Sydney, Sept 2008. 15

9. Millar EKA, Murphy NC, McNeil CM, O’Toole SA, Segara D, Crea P, Olayioye

MA, Lee CS, Fox SB, Morey AL, Christie M, Musgrove EA, Daly RJ, Lindeman

GJ,. Henshall SM, Visvader JE, Sutherland RL. Loss of StarD10 expression

identifies a poor prognosis group of primary breast cancers independent of

HER2/Neu status. Australian Breast Cancer Conference, Melbourne, Nov

2007.

10. McNeil CM, Kench J, Millar EKA , O’Toole SA, Crea P, Alles MC, Gardiner-

Garden M, Ormandy CJ, Butt AJ, Henshall SM, Musgrove EA, Sutherland RL.

Alteration in the subcellular localisation of c-myc is associated with poor

prognosis and resistance to anti-estrogen therapy in invasive breast cancer.

AACR meeting San Diego, Oct 2007.

11. Murphy N, Eggleton SA, Millar E, Biankin S, Crea P, Lindeman GJ, Visvader

JE, Henshall SM, Sutherland RL. Expression of STARD10 and its Association

with Disease Outcome in Breast Cancer. St Vincent’s Symposium Sept 2006 &

3rd PacRim Breast and Prostate Cancer Meeting, Fraser Island, Nov 2006.

12. Biankin S, Eggleton SA, Crea P, Millar E, Murphy N, Henshall S, Sutherland

R. Overexpression of sonic hedgehog may be an early event in the

development and progression of breast cancer. St Vincent’s Symposium Sept

2006 & 3rd PacRim Breast and Prostate Cancer Meeting, Fraser Island, Nov

2006.

Abstract

Aim: To identify new prognostic and predictive biomarkers for clinical breast cancer, thereby improving patient selection for currently available therapies.

Methods: Data derived from gene expression profiling of human breast cancer or human breast cancer cell lines, were interrogated to identify putative biomarkers in

ER positive and ER negative disease. These findings were validated using immunohistochemistry on tissue microarrays constructed from the development of two independent clinical breast cancer cohorts (n=292 and n=498).

Results: Prognosis in ER+ disease can be predicted by expression of BAG-1, PUMA, c-Myc and an improved biomarker signature for Luminal A and B cancer which includes Ki67 and p53. Studies identifying abnormalities in signalling pathways (PI3- kinase, Hedgehog, STARD10), HIF-1Į (CAIX, FOXP3/CXCR4, SIAH2), proliferation

(cyclin D1b) and DNA repair pathways (Rad21) have also identified potential biomarkers and therapeutic targets for ER negative and basal-like breast cancer.

Conclusions: New biomarkers for clinical breast cancer have been identified which have the potential to improve patient selection and therapeutic decision making.

Validation studies are underway in independent international randomised clinical trials to confirm these findings. The use of immunohistochemistry allows the potential rapid translation of these findings into routine Hospital Pathology clinical practice.

Chapter 1. Background to research area

1.1 Current treatment strategies and dilemmas: who to treat and why?

The Role of current biomarkers

Breast cancer is the most common cancer in women in Australia affecting one in every nine women, with approximately 12,600 new cases and 2,600 deaths reported in 2006 (Australian Institute of Health, 2006). Although survival has increased by

15% over the last ten years, largely as the result of increased early detection through breast screening and improvements in treatment, 20-30% of breast cancers sufferers will die of their disease.

The most recent recommendations from the St Gallen consensus meeting for the treatment of early breast cancer have highlighted the importance of the development of biomarkers of prognosis and response to treatment, to better select patients for currently available therapies (Goldhirsch et al 2009). Thus there is a great need to identify new biomarkers for existing therapies, which will better serve the needs of personalised breast cancer therapy in the 21st century. Currently however, there are only three established biomarkers to assist in clinical decision making which have remained unchanged for over a decade: estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2). These biomarkers are however very powerful predictors of prognosis and therapeutic responsiveness. The aim of current breast cancer treatment is to individualise management with targeted adjuvant systemic therapy wherever possible. Those patients who are ER+ derive most benefit from drugs which target estrogen synthesis (aromatase inhibitors) or block ER (tamoxifen). Although endocrine

therapy can reduce recurrences and death at 15 years by 46% and 31% respectively, 30% of these patients will have intrinsic or acquire resistance to endocrine therapy during the course of their treatment (Early Breast Cancer Trialist’s

Collaborative Group [EBCTCG] 2005). Those patients who over-express HER2 have a poor prognosis, are resistant to non-anthracycline based chemotherapeutic regimes and are usually also resistant to anti-estrogen drugs (Wilken and Maihle

2010). The HER2 targeted drug trastuzumab has however revolutionised treatment for this group of patients which accounts for approximately 16% of all patients with breast cancer. Treatment with trastuzumab results in a 50% increase in recurrence- free survival and overall survival. However only 30-50% of patients with HER2 amplification will derive benefit from treatment. For some ER- tumours such as the biologically aggressive basal-like breast cancer, no specific targeted therapy exists and much research is required to find one (eg PARP (poly-ADP-ribose polymerase) inhibitors).

The success of radiotherapy in controlling local relapse has driven the trend for conservative surgery, which when combined with adjuvant endocrine or chemotherapy can lead to long term survival with low rates of recurrence.

Mastectomy, which had previously been the main-stay of surgical treatment for all breast cancer, is now reserved for larger tumours or more extensive disease where attempts at conservative excision have failed. Whole breast radiotherapy when combined with local surgical excision can reduce local recurrence from 30% to under

10%. The benefits of good local control also translate into improved survival where local recurrence within the breast (ipsilateral breast tumour recurrence, IBTR) is an adverse prognostic indicator. A large meta-analysis performed by the EBCTCG

found that one life would be saved for every four local recurrences prevented

(EBCTCG 2005). However, as successful has radiotherapy is, no established biomarker to predict local failure has been identified to date. Thus the development of a tumour bank from the St George Breast Radiotherapy Boost study cohort is a valuable asset in the development of biomarkers of local failure following radiotherapy, an area of immediate need for improved patient care.

1.2 Biomarker Development

A biomarker is an objective measure of a biological process which may be normal or pathological and associated with disease or responses to a therapeutic intervention.

In terms of cancer management, biomarkers are utilised for risk assessment and predisposition, screening, diagnosis, prognosis or as predictive markers of therapeutic responsiveness. The National Cancer Institute (NCI) published guidelines for the recommended development of a biomarker, which follow a sequential pathway or “pipeline” from discovery to full clinical implementation culminating in a change in clinical practice (Sullivan Pepe et al 2001, Figure 1).

Figure 1. The biomarker development “pipeline” from discovery to full clinical application, as defined by the National Cancer Institute. The uptake of genetic and molecular markers, which optimises stratification of therapy, into routine cancer management has resulted in significant gains in cancer 20

outcomes, such that optimal treatment is given without delay and unnecessary side effects of ineffective therapies are minimised. Significant international effort is currently focussed on characterising and refining phenotypic subgroups in particular cancers. The heterogeneity which exists in all human cancers and which was highlighted within the early gene expression profiling (GEP) studies in breast cancer has driven the consensus view that the effective management of cancer requires a personalised medicine approach based on the use of therapies guided by biomarkers that can be measured in the patient or the tumour prior to therapy.

Biomarker research embraces a number of contemporary technologies including: genomics/epigenomics, transcriptomics, proteomics and metabolomics. Biomarkers may be assessed in tissue, blood and other bodily fluids using a number of technologies that assess mutations, epigenetic modifications and gene expression at the level of mRNA and protein. The work in this thesis has largely taken an approach of developing biomarkers that can be applied to the existing technology available within diagnostic pathology laboratories employing immunohistochemistry (IHC).

This enables an easier and more direct translation of discovery into current infrastructure and therefore potentially to patient care. However, much supportive data to generate or test hypotheses was derived from GEP data generated by members of our research group or by interrogation of data in the public domain. In the future biomarker discovery and detection is expected to become increasingly accompanied by more sophisticated approaches using deep sequencing, imaging and other emerging technologies.

1.3 The role of Tissue Microarray (TMA) as a tool in breast cancer research

TMAs are now widely utilised as a vehicle for high-throughput analysis of biomarker expression using IHC. They were initially described in 1987 (Wan et al 1987) but it was not until 1996 that a manual tissue arrayer was developed (Kononen et al 1998).

Using this technique, cores of tumour derived from a patient’s formalin fixed paraffin embedded (FFPE) tumour block are sampled and inserted into a recipient block which can contain tumour cores for up to 30-40 patients per block (Figure 2). Cores of normal breast tissue, kidney or colon are also inserted into the recipient block to enable accurate orientation according to a block map as well as to act as internal controls for staining specificity.

Figure 2. Construction of tissue microarrays (taken from: Dolled-Filhart and Rimm)

This technique has several important advantages over whole section IHC. Most importantly tissue is conserved. By sampling small areas of a tumour it is possible to

preserve much of the precious original tumour block for future analysis. Once the recipient block has been constructed all of the cores on the slide can be assessed in a more uniform manner with less variation between batches of slides or even between individual slides, as can happen with whole sections. This introduces standardisation of staining which provides a greater degree of confidence in the data produced. Additional savings in time and volume of reagents are also made as a whole cohort of 400-500 patients can be assessed in 18-20 slides for each target antigen. One possible limitation of interpretative problems with heterogeneity in staining for any given antigen, is a result of the fact that TMAs sample only a small proportion of the original tumour. Several studies have examined the optimal number of TMA cores and compared their concordance with whole section IHC. Many of these studies have concluded that 1 or 2, 0.6mm cores is sufficient to adequately sample a tumour (Camp et al 2008, Anagnostou et al 2010). Indeed several key studies have employed single core (Cheang et al 2009, Neilsen et al 2004) or double core sampling (Abd El-Rehim et al 2005). Two clinical TMA cohorts of breast cancer patients were constructed by our group using three 1mm cores as the sample: The

Garvan St Vincent’s Breast Cancer Cohort (GSVBCC, n=292) and St George Breast

Boost study cohort (n=498). The GSVBCC comprises a series of invasive ductal carcinomas of no special type, all treated by a single surgeon with collated clinicopathological data and clinical follow-up and the St George Breast Boost study cohort which comprises invasive breast cancers derived from a radiotherapy trial randomised to assess the effectiveness of a cavity “boost”. The eligibility and exclusion criteria for the St George “boost” trial are described in table 1, the

CONSORT diagram summarizing the flow of patients and tissue in Figure 1 and the clinico-pathological details in table 2. The St George cohort contains predominantly 23

good prognosis, lower-grade, ER+ breast cancers, all treated with breast conservation (ie wide local excision and radiotherapy) which is mirrored by the low numbers of events observed. This is most notable for local recurrences in the breast

– the key end-point for this clinical trial of a local radiotherapy “boost”. Unlike other

European trials, this study reduces the whole breast dose in the boost arm to avoid cosmetic degradation.

Table 1: Eligibility & Exclusion criteria for the St George “Boost” Trial. Eligibility criteria: 1. Histologically proven carcinoma of the breast, Tis (ductal carcinoma in-situ) - 2 (0-5cm) N0- 3, M0. 2. Pure ductal carcinoma in situ accepted if completely excised. 3. Any receptor status. 4. Extensive ductal carcinoma in-situ accepted if completely excised.

Exclusion Criteria: 1. Unable to consent 2. Vascular/collagen disorder 3. Prior malignancy except minor skin squamous or basal carcinoma or cervix in-situ. 4. Gross multifocal disease (microscopic single quadrant multifocality is not an exclusion) 5. Involvement of margins. Focal (< 2mm) superficial or deep margins confirmed by surgeon to be at skin or pectoralis fascia respectively considered technically clear. 6. Bilateral breast cancer

Table 2. St George Breast Boost Cohort: patient tumour characteristics, treatments and outcomes.

Characteristic No of patients (%) Median Range Length of follow-up (months) 498 84 1-134 Age (years) 61 24-84

Tumor size (mm) 16 1-60 T1a (1- 5) 4 (0.8) T1b (6 -10) 77(16.3) T1c (11-20) 270(54.2) T2 (21-50) 136(27.3) T3 (>50) 1(0.2)

Tumor Grade 1 167 (33.5) 2 185 (37.1) 3 145 (29.1)

Lymph node metastases 146(29.3) N0 339(69.9) N1 (1-3) 128 (25.7) N2 (4-10) 17(3.5) N3 (>10) 2 (0.4) LN unsampled 12 (2.4)

ER+ 393 (78.9) PR+ 334( 68.3) HER-2 amplified (FISH) 36 (7.2)

Intrinsic subtype Luminal A 394 (79.1) Modified Luminal A 321 (64.5%) Luminal B 23 (4.6) Modified Luminal B 96 (19.3%) Basal-like 52 (10.4) HER-2 enriched 13 (2.6) Unclassified 16 (3.2) Triple negative 68 (13.6)

Margin + 17 (3.4) Cavity boost positive 247 (49.5) Cavity boost negative 251 (50.5) Endocrine therapy 223 (44.7) Chemotherapy 117 (23.4) Endocrine & chemotherapy 48 (9.6)

Patients with IBTR 24 (4.8) Patients with LRR 35 (7) Patients with distant metastases 47 (9.4) Breast cancer specific deaths 37 (7.4)

5 year IBTR free survival 97.4% 5 year LRR free survival 95.6% 5 year DDFS 92.9% 5 year breast cancer-specific survival 96.3%

Comparatively the GCVBCC contains a greater percentage of high grade cancers, with more events than the St George cohort, all treated by a single surgeon (Dr Paul

Crea) with either local excision or mastectomy, but with no standardised adjuvant therapies. The clinico-pathological details of the GSVBCC are presented in table 3.

This is an excellent cohort on which to generate hypotheses regarding biomarker expression and prognosis.

Table 3. Clinico-pathological details of the Garvan St Vincent’s Breast Cancer Cohort (n=292)

Characteristic n (%) Age >50 184 (63) Grade 3 132 (45) Size >20mm 117 (40) Node positive 125 (43) HER-2 positive (FISH) 51 (18) ER+ 192 (68) PR+ 161 (57) Triple Negative 48 (17)

Treatment and survival details

Median length of follow-up: 64 months (range 0-152) Endocrine therapy 144/292(49.3) Chemotherapy 111/292(38.0) Endocrine & chemotherapy 71/292(24.3) All Recurrences 75/292(25.7) Distant metastases 68/292(23.3) Deaths 67/292(22.9) Breast cancer specific deaths 52/292(17.8) 5 year disease free survival 74% 5 year metastasis free survival 76.8% 5 year breast cancer specific survival 86.0%

Enrollment Assessed for eligibility* (n=NK)

Excluded (n=NK )

Randomized (n=688)

Allocation Allocated to Boost (n=346) Allocated to No Boost (n=342) iReceived Boost (n=338) iReceived No Boost (n=336) iDid not receive Boost (n=4) iReceivedBoost (n=6)

FollowǦUp Lost to follow-up (n=1) Lost to follow-up (n=3 )

Tissue not available for TMA analysis n=98 Tissue not available for TMA analysis n=92

Analysis

iTissueavailableforTMAanalysisn=247 i TissueavailableforTMAanalysis n=251

Figure 3. CONSORT flow diagram for the St George Breast Boost study (Taken from Millar et al 2011). *The trial recruited from three main centres (St George, Wollongong & Liverpool Hospitals). Whilst the total number of patients assessed for eligibility and excluded for all centres is not known, this data is available for the main recruiting centre at St George Hospital, which contributed the majority of patients participating in the trial n=546 (number assessed n=2046, excluded n=1500:not meeting inclusion criteria n=943, declined to participate n=235, other reasons n=322; patients randomized in trial n= 546).

1.4 Optimisation of antibodies and assessment of staining using immunohistochemistry

For all antibodies utilised, prior to assessment on our TMA cohorts, a process of optimisation was employed to determine the best conditions for ideal staining and subsequent scoring of antigen expression. This process was carried out by a specialised IHC research scientist. This labour intensive task starts with the manufacturer’s guidelines which infrequently work. Positive and negative human tissue controls are identified from tissue datasets such as Genecard

(http://www.genecards.org/index.shtml) or Human Protein Atlas

(http://www.proteinatlas.org/). Additionally, sections from genetically modified cell line pellets or xenografts (ie gene knockout or gene amplified) can also be used if available, although we have often found them uninformative to use, due to problems with heat-induced epitope retrieval (HIER) and subsequent high non-specific background staining. Using serial titrations and varied antigen retrieval conditions from “gentle” (eg water bath with enzyme digestion) to “harsh” (low or high pH microwaving for up to 2 minutes) we have been able to define the best staining for each antibody. Subsequently we would assess staining on an atlas of normal body tissue and correlate the expression levels with known expression datasets. Using a preliminary test array of breast cancers (n=30) we would next proceed to assessing expression in this small group of cases to determine patterns of over-expression

(including sub-cellular localization) or loss (according to the putative role of the target antigen). If there was satisfactory staining and sufficient data to suggest a role in human breast cancer, including assessment of publically available mRNA expression profile data (GEO http://www.ncbi.nlm.nih.gov/geo/ , oncomine

https://www.oncomine.org/content/unsecured/bookmarkRedirect.html, and/or on-line kaplan-meier plotter http://kmplot.com/breast/ ) we would then proceed to stain one of the TMA cohorts.

Assessment of target antigen expression was performed using an Olympus microscope usually at x200 magnification. Depending on the localisation of the antigen (nuclear, cytoplasmic or cell membrane) I would record the intensity of the staining as 0,1, 2 or 3 corresponding to absent, weak, moderate or strong and also assess the percentage of cells staining for each compartment assessed. All data were recorded and entered into “Cansto”, a data handling software used at the

Garvan, which stores and matches all biomarker data with patient clinicopathological and outcome data. Once scoring and data entry were complete, Cansto generated a spreadsheet which would include minimum, maximum, median and mean scores for target antigen expression across the 3 tumour cores present, along with matched patient follow-up data. Subsequently we would also calculate a simplified “H” (histo) score for the target antigen by multiplying the intensity by the percentage of positively staining cells, generating an index from 0-300.

Statistical analyses were performed using Statview 5.0. (Abacus systems, Berkeley,

CA). Initial analyses of raw data would include basic parameters of distribution, mean, median with bar charts to observe the spread of the data. The optimal cut- point was then determined using several different methods. If the analysed data distribution showed an obvious split between two subgroups this would be examined.

If there was no obvious cut point the data could be dichotomized at the mean or median, or if for more commonly employed antigen such Ki67 or p53 a commonly used standard cut-point from the literature of eg >10% would be employed. For those

antigens where none of these methods revealed any biological significance we would then employ serial cut-point determination using Kaplan-Meier analysis to determine the point at which the obtained p value from the Logrank test was most significant. In one of our studies we also determined the optimal cut-point using the

ROC (receiver-operator curve) analysis. Subsequent Kaplan-Meier and Cox analyses were performed to assess univariate and multivariate models and identify possible prognostic significance.

1.5 The impact of gene expression profiling on breast cancer classification and management

The last decade has seen an unparalleled explosion of data resulting from the use of

GEP studies which have highlighted the diversity which exists in human cancer.

Using this approach, mRNA expression in cohorts of breast cancer has been interrogated using supervised and unsupervised cluster analysis, which have provided information relating to “intrinsic subtype” and those identifying good and poor prognosis signatures. Two seminal papers by the same group of authors described so-called intrinsic molecular subtypes which spanned all types of breast cancer (Perou et al 2000, Sorlie et al 2001) and were associated with distinct clinical outcomes. These intrinsic subtypes correlated to different constituent cellular components of breast epithelium and divided ER+ and ER- tumours as follows

(Figure 4): ER+: Luminal A, luminal B, ER-: basal-like, HER2-enriched and unclassified.

Figure 4. Gene expression profiling “heat map” and intrinsic molecular phenotype of breast cancer

(taken from Sorlie et al 2001).

Although originally performed on relatively small numbers of tumours, these studies have subsequently been performed by several groups of investigators and have been reproducible across different platforms and patient cohorts (Sorlie et al 2003,

Hu et al 2006). However the classifiers used to assign any particular tumour to a specific group do not appear to be stable ie a luminal A tumour in one study may not be classified as luminal A in another (Weigelt et al 2010). Whilst there are problems inherent in using such classifications, which are dependent on the platform used, the probe sets analysed and the cut-points used in the classifiers, it does however provide invaluable information regarding the genetic profile and the heterogeneity which exists in breast cancer. The molecular classification of breast cancer is now in common clinical usage. The other key area of GEP analyses lies within the development of prognostic signatures, many of which require fresh tissue. The

problems with reproducibility, current cost and the information technology support required for these tests means that they will not be available for testing in routine pathology laboratories for some time. Those that have achieved some degree of uptake include the Oncotype Dx assay (Paik et al 2004; Genomic Health, RT-PCR on a paraffin block) and the 70 gene (van’t Veer et al 2002) mammaprint NKI signature (Agendia, requires fresh tumour tissue). On average the Oncotype Dx costs A$4000 but is not widely used in Australia. Many observers feel that most GEP prognostic signatures are not yet ready for full clinical application with evidence that they are no better than tumour morphology (ie histological type and grade) with routine IHC for ER, PR and HER2 (Weigelt and Reis-Filho 2010). The role of such

GEP analysis and its contribution to breast cancer management in the evolution of the era of personalised medicine were the main points of consideration in two reviews of GEP in breast cancer (Murphy et al 2005 [EM1], O’Toole et al 2011

[EM2], summarised in Figure 5).

Figure 5. The contribution of gene expression profiling studies to our understanding of the molecular basis of breast cancer. (Taken from Sotiriou and Pusztai 2005).

As described above the identification and application of any new biomarker follows a well developed “pipeline” from discovery, to pilot studies on exploratory cohorts of patients, to validation in randomised clinical trials and subsequent acceptance by the medical and scientific community as a test of value. Examples of current molecular biomarkers in clinical breast cancer which highlight this pathway are HER2 and

Topoisomerase IIĮ (TOPO2Į). Their roles in the molecular pathology and contribution to personalised breast cancer medicine by in-situ hybridisation, are further discussed at length in our recent review (O’Toole et al 2011 [EM2]). As with any current biomarker, the importance of rigorous validation, interpretative and quality control are significant aspects of the routine evaluation of such molecular predictive biomarkers.

With the identification of any new potential biomarker, its mechanistic role within the functioning of the cancer cell is of interest as it may offer new insight into our understanding of a highly complex process involving cross-talk between many divergent internal and external (microenvironmental) factors, reviewed in detail elsewhere (Hanrahan and Weinberg 2011, Figure 6). Additionally new data may offer a potential new avenue of therapeutic attack.

Figure 6. The complex interaction of signalling networks within cancer cells comprising pathways regulating proliferation, differentiation, apoptosis and motility (taken from Hanrahan and Weinburg 2011).

Publications arising from this work: EM1. Murphy N, Millar E, Lee CS. Gene Expression Profiling in Breast Cancer:

Towards Individualising Patient Management. Pathology. 2005; 37: 271-277.

EM2. O’Toole SA, Selinger T, Millar EKA , Lum T, Beith JM.Molecular Assays in

Breast Cancer Pathology. Pathology. 2011; 43: 116-127.

Chapter 2. Immunophenotyping breast cancer using surrogate biomarker panels for molecular subtype: predicting outcome in conservatively treated early breast cancer.

The drive towards tailoring specific treatments to individual patients and providing as much prognostic information as possible, early in the disease work-up, will continue to expand the field of biomarker development. As most patients will not have access to affordable genetic profiling of their tumours for some time, we and others have explored the approach of translating the GEP intrinsic subtypes of breast cancer into a simplified panel of IHC markers suitable for routine use to determine its clinical relevance. This approach was first described by Nielsen using an IHC panel of ER,

HER2, CK5/6 and EGFR (Neilsen et al 2004) and was recommended for application to clinical practice at the most recent St Gallen conference (Goldhirsch et al 2011 in press). We performed an analysis of a five biomarker panel as a surrogate for intrinsic molecular phenotype using ER, PR, HER2 (using fluorescent in-situ hybridisation), CK5/6 and EGFR (Millar et al 2009 [EM3]) and compared this classification against existing clinicopathological prognostic parameters, something which some earlier studies had failed to do. We demonstrated that IHC intrinsic subtype, so defined, was of inferior predictive power to more traditional prognostic indices such as tumour grade, lymph node status and lymphatic vascular invasion.

This 5 biomarker classification was however capable of making a distinction between the five subtypes, with luminal A (LA) having the best prognosis and luminal B (LB), basal, HER2 enriched and unclassified all having a worse prognosis (Figure 7). The subtyping information so provided, could be used to augment rather than replace established prognostic parameters and help further refine treatment decisions.

The St George cohort, on which this study was performed is composed predominantly of ER+ luminal breast cancers, representing the most common type of breast cancer, over 70% of all breast cancers diagnosed. This provides the opportunity to investigate a better definition of luminal A and B tumours. This would be of significant clinical value in determining those luminal A tumours with an excellent outcome which could be treated with endocrine therapy only and avoid chemotherapy and also identify those poorer prognosis luminal B cancers that would benefit from the addition of chemotherapy and/or trastuzumab. GEP studies defined the luminal B group on the basis of high expression of HER2 and proliferation related genes such as HER2, GRB7, MKI67, MYBL2 to comprise 11-20% of most cohorts

(Sorlie et al 2004, Hu et al 2006). However the luminal B population only constitutes

5-10% of most cohorts using ER, PR and HER-2 by IHC (Nguyen et al 2008,

Cheang et al 2008, Millar et al 2009 [EM3], Blows et al 2010) indicating this group of tumours is likely under represented using these biomarkers, probably reflecting the fact that only 30% of luminal B tumours are HER2 amplified by GEP (Carey et al

2006). A previous study using IHC had shown that the addition of the proliferation marker Ki67 to the definition of ER+ and/or PR+, HER2+ improved the predictive power of this definition of luminal B (Cheang et al 2009). This group of patients was the target for the development of the Oncotype Dx assay (Paik et al 2004) and more recently the Mammostrat IHC assay (Ring et al 2006) to predict risk of recurrence in

ER+, lymph node negative patients. Our further analysis of the St George Boost cohort showed that there was a significant difference in expression of Ki67 and p53 between the LA and LB groups from our initial 5 biomarker data and that these two markers better discriminate between these two groups of patients (Millar et al 2011

[EM4]. Additionally, prior analysis of the GSVBCC as part of Dr Catriona McNeil’s

PhD thesis (McNeil 2008) had shown a similar association of Ki-67 and p53 overexpression within the luminal B subtype. Therefore we modified the definition of LA to include: ER and/or PR+ and HER negative, p53 negative (<10%) and Ki67 low

(<10%) and LB to include: ER and/or PR+ and HER2+ and/or p53+ and /or Ki67 high. Using this definition, our detection of poor prognosis luminal B tumours was increased by over 4 fold (compared to using ER, PR HER2 alone), increasing the size of the LB group from 4.6 to 19.7% of the cohort, which better reflected GEP estimates of the size of this group and improved the separation between all intrinsic subtypes of breast cancer (Figure 8). More significantly this definition of LB was an independent predictor of outcome in multivariate analysis with a hazard ratio 3-4 37

times that of LA for locoregional recurrence (HR 3.612, 95% CI 1.555-8.340, p=0.003), distant metastases (HR 3.023 95% CI 1.501-6.087, p=0.002) and breast cancer specific death (HR 3.617, 95% CI 1.629-8.031, p=0.002). Additional exploratory multivariate analyses for patients treated with tamoxifen alone (n=169,

10 events) showed that LB retained independent prognostic significance in the final resolved model for breast cancer specific death (HR 5.361, 95% CI 1.418-20.25, p=0.013).

This finding suggests that this refined definition of LB identifies a subgroup of ER+ patients with five times the risk of death when compared to LA patients treated with endocrine therapy. The predictive value of this classification requires further independent validation within the setting of a randomized trial of endocrine therapy and such studies have been initiated. The simplicity of this IHC panel means that it could be transferred into clinical practice almost immediately, costing a fraction of gene testing kits. Recently the validity of this approach was confirmed by another group of investigators using a four marker panel for IHC (“IHC4”: ER,PR, HER2 and

Ki67) which was found to be at least equivalent to Oncotype Dx in its predictive value in a head to head comparison using data from the ATAC trial (Arimidex Tamoxifen

Alone or in Combination; Cuzick et al 2009).

In terms of radiotherapy planning, this signature independently predicts locoregional recurrence (LRR: breast, axilla, chest wall and supraclavicular fossa) in conservatively managed breast cancer. This is a significant finding for the identification of those ER+ patients at highest risk of LRR. Previously only pathological factors such as positive margins and grade were significant predictors of local failure. Following axillary surgery, lymphoedema occurs in approximately 10% of patients. This problem can be predicted in those patients who have extra-nodal extension, whilst the total number of involved nodes does not (Graham et al 2006

EM3]). This highlights the continuing importance of the role of routine pathology in predicting patient outcomes.

Conclusions:

Immunohistochemistry using a 5 biomarker panel can provide meaningful information regarding intrinsic subtype of breast cancer, but the information provided is inferior to established pathological indices. The addition of Ki67 and p53 significantly improves discriminatory power in identifying poor prognosis ER+ luminal

B breast cancer using a new definition, enabling their early identification for more aggressive therapy.

Publications arising from this work:

EM3. Millar EKA, Graham PH, O’Toole SA, McNeil CM, Browne L, Morey AL,

Eggleton S, Beretov J, Theocharous C, Capp A, Nasser E, Kearsley JH, Papadatos

EM4. Millar EKA, Graham PH, McNeil CM, Browne L, O’Toole SA, Boulghourjian A,

Papadatos G, Delaney G Nasser E, Kearsley JH, Fox C, Capp A, Sutherland RL.

Prediction of outcome in early luminal breast cancer is improved using a biomarker panel which includes Ki-67 and p53. British Journal of Cancer. 2011.105:272-80.

Chapter 3. Endocrine resistance in ER+ breast cancer

Given that luminal ER+ breast cancer is the most prevalent subtype of breast cancer, any improvement in prognosis would be of benefit to the greatest number of patients.

The luminal B group comprises those ER+ cancers with a relatively poor prognosis which probably corresponds to tumours which have intrinsic or acquired resistance to endocrine therapy, the mainstay of treatment for ER+ disease. The mechanisms underlying endocrine resistance are complex but are a key interest of our group. The molecular details of this area have been recently reviewed by members of our group

(Musgrove and Sutherland 2009, Figure 9). In summary there are three distinct pathways of estrogen activation of gene expression: ligand bound ER binds to specific DNA sequences, estrogen responsive elements (ERE), via complexes which are formed with co-activators (CoA), histone acetyl transferases (HAT) or through protein-protein interactions with transcription factors eg AP1, Sp1 which cooperate to bind serum response elements (SREs) to initiate transcription. Ligand independent activation of ER can also occur via receptor tyrosine kinase signalling eg activation of EGFR, HER2 or IGFR results in activation of the PI3K-AKT or the ERK pathways which phosphorylate ER. Lastly ER which is present in the cytoplasm or bound to the cell membrane can also be activated by non-genomic mechanisms involving formation of signalling complexes involving PI3K, Src and FAK. Major ER target genes include Myc, cyclins D1, E1 and E2 which drive proliferation and cell growth.

Figure9.Oestrogensignallinginbreastcancer(TakenfromMusgrove&Sutherland2009).

An in-vitro model system of anti-oestrogen growth arrested MCF-7 breast cancer cells, salvaged by subsequent administration of exogenous oestrogen was developed by members of our group to examine the role of estrogen regulated genes in overcoming growth arrest with Tamoxifen. mRNA GEP data derived from these experiments following oestrogenic rescue (Musgrove et al 2009) demonstrated the rapid upregulation of several genes including those regulating proliferation: c-Myc, cyclins D1 and E2, and apoptosis: BAG-1 (BCL2 athanogene 1), and downregulation of another apoptosis regulating gene PUMA (p53 upregulated mediator of apoptosis).

3.1 Apoptosis in Endocrine Resistance

In two parallel studies, analysis of publically available mRNA GEP datasets and protein expression using IHC, showed that high expression of BAG-1 and PUMA correlated with improved prognosis and low expression with an adverse outcome

(Figures 10-13).

A.BAGͲ1mRNA B.BAGͲ1protein

1 1

.8 .8 High High

.6 .6 Low .4 Low .4 Cum. Survival Cum. Cum. Survival Cum.

.2 .2

0 0

0 25 50 75 100 125 150 175 200 225 0 20 40 60 80 100 120 140 160 180 Time Time

Figure 10. Kaplan-Meier analysis for BAG-1 expression and breast cancer specific death in the Wound/NKI (van de Vijver et al 2002) cohort using mRNA expression (A) and protein expression in the GSVBCC cohort using IHC (B).

Detailed IHC analyses (Figure 11) validated the mRNA findings in 290 breast cancer patients and showed that BAG-1 correlated with a luminal A phenotype and was an independent predictor of favourable outcome in ER+ patients in multivariate analysis

(Millar et al 2009 [EM6], HR 0.302, 95% CI 0.122 – 0.744, p=0.0093). In vitro studies further demonstrated that over-expression of BAG-1 in MCF-7 cells, augmented the growth arresting potential of Tamoxifen. As a result of these findings and the independent prognostic power of BAG-1 IHC in ER+ breast cancer, its expression will be assessed in a large international trial of endocrine therapy in ER+ breast cancer through an on-going collaboration with the ANZBCTG (Australian New

Zealand Breast Cancer Trials Group) and the IBCSG (International Breast Cancer

Study Group). This exciting development again highlights the “pipeline” for development of new biomarkers in clinical oncology in that it will provide independent validation in the context of a prospective randomized clinical trial ie NCI Phase II.

Figure 11. BAG-1 IHC: strong (3+) nuclear and weak (1+) cytoplasmic staining in invasive ductal carcinoma; weak nuclear staining in normal duct (arrow), (x400, taken from Millar et al 2009,).

High expression of PUMA, a BCL-2 homology 3 (BH3)-only, pro-apoptotic regulator was also found to associate with improved prognosis at the mRNA and protein level

(Figures 12, 13). We identified that PUMA (BBC3) is an estrogen target gene that is acutely downregulated in response to estrogen in breast cancer cell lines, independently of their p53 status (Roberts et al 2011 [EM7]). PUMA is transcriptionally upregulated following treatment with tamoxifen and knock down of

PUMA expression in these cells attenuates the apoptotic response to tamoxifen. Low

PUMA expression in breast carcinomas is significantly associated with breast cancer-specific death (p = 0.0014 and p= 0.0115, for mRNA and protein, respectively) and with a worse outcome in tamoxifen-treated patients for mRNA,

(p=1.49 x10-05). These findings suggest that the dysregulation of apoptotic signaling

pathways such as those executed via PUMA, can significantly impact on both the progression and therapeutic responsiveness of ER+ breast cancer and its identification suggests a new avenue for therapeutic attack.

Figure 12. PUMA immunohistochemistry in invasive ductal carcinoma surrounding a normal duct (arrow). Cytoplasmic staining was present at variable intensity that ranged from 0-3+ (3+ staining present here, with 1+ staining within a normal duct (arrow), taken from Roberts et al 2011).

A B

Figure 13. Kaplan-Meier analysis (logrank test) of breast cancer specific death A: PUMA mRNA (BBC3) in the Wound/NKI cohort (van de Vijver et al 2002) and B: PUMA protein by IHC in the GSVBCC (taken from Roberts et al 2011).

3.2 Proliferation and Endocrine Resistance

As described above, previous work examining the role of over-expression the protooncogene c-Myc had demonstrated its ability to overcome tamoxifen induced growth arrest in MCF-7 cells (Prall et al 1998). Its potential role as a driver of anti-estrogen resistance and as a biomarker formed the basis for the PhD Thesis of Dr Catriona

McNeil (McNeil 2008). This body of work outlines the identification of a high-risk phenotype of c-Myc expression characterised by high cytoplasmic expression and low nuclear expression which is associated with poor prognosis in ER+ cancers and is associated with a basal phenotype in ER- cancers (McNeil et al 2006 [EM8],

Figure 14).

Figure 14. Patterns of C-Myc staining using IHC in invasive ductal carcinoma x400 (taken from McNeil et al 2006).

Cyclin D1 is a key estrogen regulated gene and important mediator of cell cycle progression that is aberrantly regulated in multiple cancers, especially in breast cancer (Buckley et al 1993, Arnold and Papanikolaou 2005, Musgrove et al 2011). A number of studies have indicated that a polymorphism in a splice donor site in the cyclin D1 gene is associated with alternate splicing and the production of the alternative cyclin D1b transcript (Figure 15). Furthermore, this polymorphism may be selectively associated with disease outcomes (Knudsen et al 2006a, Knudsen

2006b).

Figure 15. Alternate splicing of the cyclin D1 transcript occurs at the exon4/intron 4 junction which normally results in the usual cyclin D1 product (bottom left), which contains discrete motifs that regulate cell cycle control, subcellular localization, and transcriptional regulation. Failure to splice at the exon4/intron 4 boundary results in the cyclin D1b protein product, which harbors a divergent C- terminus (hatched). The G/A870 polymorphism (yellow circle) is thought to influence the splicing event (taken from Knudsen 2006b).

However, relatively little is known regarding the protein product of the alternatively spliced message, cyclin D1b. We demonstrated that this protein is readily detectable in a number of cancer cell lines and primary breast cancers (Wang et al 2008 [EM9]

Figure 16). Whereas cyclin D1b interacts with cyclin-dependent kinase 4 (CDK4), it is relatively inefficient at mediating RB phosphorylation and cell cycle progression in model systems due to the lack of exon 5 of cyclin D1–encoded sequences. However, cyclin D1b protein levels are not significantly attenuated by DNA damage or 47

antiestrogen treatment, indicating that the protein may have significant effect on the response to such therapeutic modalities. Whereas enforced expression of cyclin D1b was not sufficient to abrogate DNA damage checkpoint responses, it did efficiently overcome cell cycle arrest mediated by antiestrogen therapeutics. This action of cyclin D1b was not associated with effects on estrogen receptor activity, but was rather dependent on functional association with CDK4. These studies indicate that the cyclin D1b protein is aberrantly regulated and could contribute to therapeutic failure in the context of ER+ breast cancer.

Figure 16. Cyclin D1b IHC. 3+ nuclear staining in invasive ductal carcinoma.

A single nucleotide polymorphism (SNP) at nucleotide 870 in the CCND1 gene, rs603965, influences the relative production of the cyclins D1(D1a) and D1b and can impart increased risk for tumour development. Our study included data from a large multiethnic case–control study where the G/A870 polymorphism conferred no significant risk for breast cancer overall or by stage or estrogen receptor (ER) status

(Millar et al 2009 [EM10]). However, cyclin D1b protein was upregulated in breast cancer, independent of cyclin D1 levels, and exhibited heterogeneous levels in breast cancer specimens. High cyclin D1 expression inversely correlated with the

Ki67 proliferation marker and was not associated with clinical outcome. In contrast, elevated cyclin D1b expression was independently associated with adverse outcomes, including recurrence, distant metastasis and decreased survival.

Interestingly, cyclin D1b was particularly associated with poor outcome in the context of ER-negative breast cancer. Thus, specific cyclin D1 isoforms are associated with discrete forms of breast cancer and high cyclin D1b protein levels hold prognostic potential.

Conclusions:

Endocrine resistance in breast cancer is a significant clinical and mechanistic problem. We have identified c-Myc, BAG-1 and PUMA as potential biomarkers of endocrine responsiveness/resistance which may help better identify patients with poor prognosis ER+ breast cancer for optimal therapy. These markers will be assessed in a large randomised clinical trial of endocrine therapy.

Publications arising from this work:

EM6. Millar EKA, Anderson LR, McNeil CM, O’Toole SA, Pinese M, Crea P,

Morey A, Biankin AV, Henshall SM, Musgrove EA, Sutherland RL, Butt AJ. BAG-

1 predicts patient outcome and tamoxifen responsiveness in ER positive invasive

ductal carcinoma of the breast. Br J Cancer 2009.100; 122-133.

EM7. Roberts CG*, Millar EKA*, O’Toole SA, McNeil CM, Lehrbach GM, Pinese

M, Tobelmann P, McCloy RA, Musgrove EA, Sutherland RL and Butt AJ.

Identification of PUMA as an estrogen target gene that mediates the apoptotic

response to tamoxifen in human breast cancer cells and predicts patient outcome

and Tamoxifen responsiveness in breast cancer. Oncogene. 2011. 30:3186-97.

*Both authors contributed equally to this work.

EM8. McNeil CM, Sergio CM, Anderson LR, Inman CK, Murphy NC, Millar EKA,

Crea P, Kench JG, Alles MC, Gardiner-Garden M, Ormandy CJ, Butt AJ,

Henshall SM, Musgrove EA, Sutherland RL. C-Myc overexpression and

endocrine resistance in breast cancer. J Ster Biochem Mol Biol. 2006; 102:147-

55.

EM9. Wang Y, Dean JL, Millar EKA, Tran TH, McNeil CM, Burd CJ, Henshall

SM, Utama FE, Witkiewicz A, Rui H, Sutherland RL, Knudsen KE, Knudsen ES.

Cyclin D1b is aberrantly regulated in response to therapeutic challenge and

promotes resistance to estrogen antagonists. Cancer Research 2008; 68:5628-

5638.

EM10. Millar EKA, Dean JL, McNeil CM, O’Toole SA, Henshall SM, Tran T, Lin

J, Quong A, Comstock CES, Witkiewicz A, Musgrove EA, Rui H, Le Marchand L,

Setiawan VW, Haiman CA, Knudsen KE, Sutherland RL, Knudsen ES. Cyclin

D1b Protein expression in breast cancer is independent of cyclin D1a and

associated with poor disease outcome. Oncogene. 2009. 28:1812-20.

Chapter 4. Signalling Pathways in Breast Cancer

Basal-like breast cancer comprises about 10-15% of all cases and has a poor prognosis with predilection for younger patients, an association with particular ethnic groups such as African-Americans and is associated with BRCA1 mutations. It is often of high histological grade with frequent mitotic activity has a variable lymphocytic infiltrate and may have broad areas of central necrosis (Figure 17), or sclerosis (reviewed in detail elsewhere: Rakha et al 2008).

Figure 17. Basal-like invasive breast cancer demonstrating prominent tumour necrosis.

Basal-like cancer is triple negative (ie ER, PR and HER2 negative) and unlike other forms of breast cancer such ER+ or HER2-enriched, no specific targeted therapy exists. Thus there is an urgent need to further develop better markers of diagnosis and treatment of this aggressive cancer which has a predilection for haematogenous spread especially to brain. Most recurrences and death occur within 5 years of diagnosis but those that survive beyond this period, appear to have a prognosis similar to other types of breast cancer. As a step towards investigating this aggressive form of disease we planned to examine several key signalling pathways involved in cellular proliferation in the GSVBCC. The first of these was the

phosphoinositol-3-kinase (PI3K) pathway (Lopez-Knowles et al 2010 [EM11].

Amplification and activating mutations of the PIK3CA gene (3 hotspots in exons 9 and 20) were assessed by PCR. Loss of PTEN expression and over-expression of its downstream target phospho-AKT were assessed by IHC. All previous publications in breast cancer examining this pathway have described each individual component on its own and not together as a complete pathway. Thus we performed the first integrative analysis to assess all components which would therefore lead to an active pathway ie mutation or amplification of PIK3CA, PTEN loss or overexpression of pAKT. This analysis demonstrated that PIK3CA mutations and PTEN loss were mutually exclusive events and that at least one component of the pathway was abnormal and over-active in most breast cancers (71%) and in over 90% of basal- like breast cancer. These findings suggested that this may therefore represent a potential therapeutic approach to the basal group of tumours.

PI3K signalling is inhibited by inositol polyphosphate 4-phosphatase-II (4-ptase-II,

INPP4B) and is implicated as a tumour suppressor in epithelial carcinomas. INPP4B loss of heterozygosity (LOH) is detected in some human breast cancers, however, the expression of 4-ptase-II protein in breast cancer subtypes and the normal breast is unknown. In collaboration with colleagues at Monash University, we identified that

4-ptase-II is expressed in luminal, non-proliferative ER+ cells in the normal breast, and in ER+, but not ER-ve breast cancer cell lines (Fedele et al 2010 [EM12]).

Ectopic 4-ptase-II expression in ER-ve breast cancer cells reduced Akt activation and cell proliferation. Conversely, 4-ptase-II knock-down in ER+ve MCF-7 cells increased AKT activation, cell proliferation and xenograft tumour growth. Analysis of primary human breast cancers from two independent patient datasets revealed

frequent loss of 4-ptase-II protein expression was associated with high clinical grade and tumour size. 4-ptase-II expression correlated with ER and PR and was lost most frequently in aggressive basal-like breast carcinomas. Interestingly, 4-ptase-II protein loss in human breast cancers occurred mutually exclusive of PIK3CA mutation, but was frequently observed in PTEN-null tumours, revealing concomitant loss of two tumour suppressors that regulate PI3K signaling. Therefore 4-ptase-II protein is a significant regulator of ER+ mammary cell proliferation and its loss of expression is a marker of aggressive basal-like breast carcinomas.

Stem cell & Developmental Signalling Pathways in Breast Cancer

The role of the hedgehog (Hh), Wnt (Wingless-type) and Notch signalling pathways which are important in embryogenesis and in the development and self renewal of stem cells have recently emerged in the pathogenesis, prognosis and potential treatment of breast cancer and are reviewed in detail elsewhere (Kakarala and

Wicha 2008, Zardawi et al 2009, O’Toole et al 2009, Figure 18).

Fig 18. Key pathways involved in stem cell signalling, growth and renewal. Dysregulation of stem cell renewal in sporadic cancers may involve loss of PTEN activity, HER2 amplification or activation of Hedgehog or Notch pathways acting via the polycomb protein BMI-1. BRCA1 loss is important in hereditary cancers. Subsequently, there is expansion of the stem cell compartment which may undergo further carcinogenic events. (Taken from Karakal & Wicha 2008).

Our investigation of these pathways yielded new data highlighting their associations with pathogenesis, intrinsic subtype and outcome. The role of Hedgehog (Hh) and its association with a stem cell phenotype and with the basal type of breast cancer was investigated in the PhD thesis of collaborator Dr Sandra O’Toole (O’Toole 2008).

Over-expression of Hh by IHC in invasive ductal carcinoma is present in 34% of tumours and is associated with a poor prognosis (death HR 2.3 p=0.0002), basal-like subtype (p=0.004) as well as resistance to endocrine therapy in ER+ cancers and those patients treated with chemotherapy (O’Toole et al 2011 [EM13]). A mouse model over-expressing Hh, based on the M6 cell line derived from the C3/SV40T transgenic mouse, demonstrated a role for Hh in driving tumour growth and metastasis, which could be inhibited by a novel monoclonal antibody, 5E1. This work identifies the key role of this important pathway as well as suggesting a potential

treatment strategy for metastatic breast cancer and basal breast cancer, for which there is currently no targeted therapy.

Notch signalling is critical in mammalian embryonic development, particularly in neurogenesis, angiogenesis and in development of the breast, heart and lymphoid systems (Artavanis-Tsakonas et al 1999). More recently a key role for Notch signalling in the regulation of mammary progenitor cells was described (Bouras et al

2008). Activation of Notch signalling results in hyperplasia and tumorigenesis in murine mammary epithelium but there is little information regarding the expression of

Notch1 in premalignant lesions and early breast cancer. We observed enhanced expression of Notch1 protein as an early event in both murine and human breast cancer development with progressive increases in expression with the development of hyperplasia and malignancy (Zardawi et al 2010 [EM14]). High Notch1 was not prognostic in the outcome cohort but there was a highly significant association of high Notch1 protein with the HER-2 molecular subtype of breast cancer. This signalling pathway may be a potential therapeutic target worthy of further investigation.

ȕ-catenin, a pivotal downstream mediator of Wnt signalling, is involved in cell adhesion through catenin-cadherin complexes and as a transcriptional regulator in the Wnt signaling pathway. Its deregulation is important in the genesis of a number of human malignancies, particularly colorectal cancer (Polakis 2000). In breast cancer, there are conflicting associations reported for ȕ-catenin expression, clinicopathological variables, and outcome. Whilst we were unable to demonstrate any association with breast cancer–specific death for cytoplasmic or membrane 55

expression alone a novel continuous score representing both locations (membrane minus cytoplasmic expression: MTC score, Figure 19) was associated with a worse outcome in univariate analysis, and approached significance in multivariate analysis

(Lopez-Knowles et al 2010 [EM15]). An association was identified between high cytoplasmic expression (low MTC score), and high tumour grade, positive Ki67, negative ER, positive HER2 over-expression and an active PI3K pathway. Low cytoplasmic expression (high MTC score) was associated with the luminal A subtype. A low ȕ-catenin MTC score is associated with an adverse outcome in breast cancer and may be of mechanistic significance in the disease process.

Figure 19. ȕ-catenin expression in human tissues. A. Normal colonic crypts show a membranous pattern of expression. B. Invasive colon adenocarcinoma showing predominantly nuclear expression. C. Invasive lobular breast carcinoma negative control lacks expression of ȕ-catenin. D. Normal breast duct with membranous expression in the basal cell layer. E. Weak predominantly membranous expression in invasive ductal carcinoma. F. Strong predominantly membranous expression in invasive ductal carcinoma. G. Predominantly cytoplasmic expression in invasive ductal carcinoma. H. Strong membranous and cytoplasmic expression in invasive ductal carcinoma. All images, ×400 magnification (taken from Lopez-Knowles et al 2010).

The phospholipid transfer protein STARD10 is expressed in normal breast and during lactation. It cooperates with c-erbB signalling (Olayioye et al 2004, Alpy and

Tomasetto 2005) and its expression in breast cancer may indicate preservation of a differentiated luminal phenotype.

A B

Fig 20. STARD10 immunohistochemistry. A: over-expressing breast carcinoma (3+); B: loss of expression.

We interrogated mRNA gene expression profiling data from the Wound/NKI cohort

(van de Vijver 2002) and found that patients with low STARD10 or high HER2 tumour mRNA levels formed discrete groups each associated with a poor disease- specific survival (Murphy et al 2010 [EM16], p = 0.0001 and p = 0.0058, respectively). These findings were validated using IHC in the GSVBCC (Figure 20).

In multivariate analyses, low STARD10 expression was an independent predictor of death from breast cancer (HR 2.56 95% CI 1.27-5.18, p=0.0086) independent of

HER2 status and triple negative phenotype, suggesting a potential role as a biomarker identifying a subgroup of patients with a particularly adverse prognosis.

Gab2, a docking-type signalling protein with demonstrated oncogenic potential, is overexpressed in breast cancer, but its prognostic significance and role in disease evolution remain unclear. Our analysis demonstrated that while Gab2 expression

(Figure 21) was positively correlated with increased tumour grade, it did not correlate with disease recurrence or breast cancer-related death in the total cohort or in patients stratified according to lymph node, estrogen receptor (ER) or HER2 status

(Fleuren et al 2010 [EM17]).

Figure 21. Gab2 IHC in invasive ductal carcinoma x400 (taken from Fleuren et al 2010).

Interestingly, analysis of a ‘‘progression series’’ that included premalignant and preinvasive breast lesions as well as samples of metastatic disease revealed that

Gab2 expression was significantly enhanced in the earliest lesion examined, usual ductal hyperplasia, with a further increase detected in ductal carcinoma in situ

(DCIS). Furthermore, expression was less in invasive cancers and lymph node metastases than in DCIS, but still higher than in normal breast. These findings indicate that while Gab2 expression is not prognostic in breast cancer, its role in 59

early disease evolution warrants further analysis, as Gab2 and its effectors may provide targets for novel strategies aimed at preventing breast cancer development.

Conclusions:

Our analysis of signalling pathways has identified several new potentially useful biomarkers and avenues of therapeutic attack most notably for basal-like breast cancer (Hedgehog, PI3K, PPInd4). This may help improve the prognosis for this group of tumours for which there is currently no specific targeted therapy.

Publications arsing from this work:

EM11. López-Knowles E, O’Toole SA, McNeil CM, Millar EKA, Qiu MR, Crea P,

Musgrove EA, Robert L. Sutherland. PI3K pathway activation in breast cancer is

associated with the basal-like phenotype and cancer-specific mortality. Int J

Cancer. 2010. 126:1121-31.

EM12. Fedele CG, Ooms LM, Ho M, Vieusseux J, O’Toole SA, Millar EKA,

Lopez-Knowles E, Sriratana A, Gurung R, Baglietto L, Giles GG, Bailey CG,

Rasko JEJ, Shields BJ, Price JT, Majerus PW, Sutherland RL, Tiganis T,

McLean CA, Mitchell CA. The PtdIns(3,4)P2 4-phosphatase, INPP4B, regulates

ER-positive mammary cell proliferation and is lost in human basal-like breast

carcinomas. Proc Natl Acad Sci. 2010. 107: 22231-6.

EM13. O’Toole SA, Machalek D, Shearer R , Millar EKA, Nair R, McLeod D,

Cooper C, McFarland A, Ru Qiu M, McNeil CM, Rabinovich B, Martelotto L, Vu D,

Musgrove E, Sutherland RL, Watkins N, Swarbrick A. Hedgehog overexpression

is associated with stromal interactions and predicts for poor outcome in breast

cancer. Cancer Research. 2011. 71:4002-4014.

EM14. Zardawi SJ, Zardawi I, McNeil CM, Millar EKA, McLeod D, Morey AL,

Crea P, Murphy NM, Lopez-Knowles E, Oakes SR, Ormandy CJ, Qiu MR,

Hamilton A, Spillan A, Lee CS, Sutherland RL, Musgrove EA, O’Toole SA.

Notch1 expression is an early event in breast cancer development and is

associated with the HER-2 molecular subtype. Histopathology. 2010; 56, 286–

296.

EM15. López-Knowles E, Zardawi SJ, McNeil CM, Millar EKA, Crea P,

Musgrove EA, Sutherland RL,. O’Toole SA. Cytoplasmic localization of ß catenin

is a marker of poor outcome in breast cancer patients. Cancer Epidemiol

Biomarkers Prev. 2010;19:301-9.

EM16. Murphy NC*, Biankin AV*, Millar EKA*, McNeil CM, O’Toole SA, Segara

D, Crea P, Olayioye MA, Lee CS, Fox SB, Morey AL, Christie M, Musgrove EA,

Daly RJ, Lindeman GJ,. Henshall SM, Visvader JE, Sutherland RL. Loss of

STARD10 expression identifies a group of poor prognosis breast cancers

independent of HER2/Neu and triple negative status. Int J Cancer.

2010;126:1445-53.

*These authors contributed equally to this work. 61

EM17.Fleuren EDG, O’Toole SA, Millar EKA, McNeil CM, Lopez-Knowles E,

Brummer T, Sutherland RL, Daly RJ. Overexpression of the oncogenic signal

transducer Gab2 occurs early in breast cancer development. Int J Cancer.

2010;127: 1486–1492.

Chapter 5. Tumour Hypoxia in Breast Cancer

Basal breast cancers can often exhibit prominent areas of central necrosis (Fig 17) suggesting they may have an abnormal hypoxic response with aberrant Hypoxia

Inducible Factor (HIF) pathway activity. In collaboration with colleagues at the Peter

MacCallum Cancer Centre we undertook a series of studies addressing the roles of several key downstream targets in the HIF pathway: CAIX, CXCR4/FOXP3 interactions and SIAH2, summarized below (Figure 22).

Figure 22. The HIF pathway is activated by hypoxic inhibition of prolylhydroxylases (PHDs) and resultant transcription of hypoxia responsive elements (HREs) involving activation of diverse downstream pathways and targets involved in angiogenesis, autophagy/cell survival, metastasis, pH homeostasis and metabolism.(Taken from Brahimi et al 2007).

CAIX (carbonic anhydrase IX), a key downstream target of HIF-1Į, is a transmembrane protein involved in maintaining a low pericellular pH through its reversible conversion of carbon dioxide and water into carbonic acid. IHC analysis showed that CAIX was nine times more likely to be expressed in basal breast cancers compared to the other subtypes (Tan et al 2009 [EM18], Figure 23).

Figure 23. A: Positive CK 5/6 staining in a basal-like carcinoma, B positive nuclear staining for HIF-1Į, C: strong membranous staining for CAIX in a basal-like carcinoma, D: negative CAIX staining in a luminal tumour. (taken from Tan et al 2009).

Additionally those breast cancers which over expressed CAIX were resistant to chemotherapy, possibly the result of trapping of chemotherapeutic agents outside the cell where the extracellular environment is acidic. These findings suggested another potential avenue of therapeutic attack for this group of breast cancers.

Despite the presence of a host lymphocytic infiltrate (Figure 24) within most basal- like cancers they are associated with a poor prognosis. Recently it has become evident that a subpopulation of T cells, T regulatory cells (Tregs), may play an increasing role in maintaining tolerance to self-antigens. Tregs identified by the marker transcription factor FoxP3 (Forkhead box P3) appear to be important mediators of peripheral tolerance, suppress undesirable immune responses and are associated with a poor prognosis (Bates et al 2006). We hypothesised that hypoxia-induced secretion of CXCL12 (C-X-C motif chemokine 12) within tumours recruits Tregs via binding to its receptor CXCR4 and may drive immune suppression and help explain

the increased aggressiveness of basal-like breast cancers. This may be of therapeutic significance as Treg treatments are now in clinical trials.

Figure 24. Basal-like breast cancer with peri-tumoral lymphocytic infiltrate and tumour infiltrating lymphocytes (arrows).

Tumour CXCL12 positivity correlated with Treg FOXP3 expression and basal phenotype and CXCL12 positivity correlated with improved survival (p=0.005; Yan et al 2011 [EM19]). High Treg correlated with shorter survival for all breast cancers, luminal cancers and basal-like cancers that was confirmed in a multivariate analysis

(p=0.042). Despite the association between high Treg with CXCL12 and basal phenotype, there was no difference in CXCL12 expression between luminal and basal-like cancers. Up-regulation of CXCR4 in Treg correlated with basal-like phenotype and tumour hypoxia, as indicated by CAIX expression. Although CXCL12 expression is associated with a good prognosis, in the setting of hypoxia and CXCR4 up regulation in Treg, this may have the negative consequence of enhancing Treg recruitment while suppressing the anti-tumour immune response.

During tissue hypoxia, the ubiquitin ligase SIAH2 (seven in absentia homologous 2) plays a significant role by targeting the prolyl hydroxylases that regulate HIF-1Į for proteasomal degradation, indicating a potential important role in basal-like breast

cancer which was previously unknown. We observed a significant increase in nuclear SIAH2 expression from normal breast tissues through to DCIS and progression to invasive cancers (p<0.0001) with an inverse correlation between

SIAH2 and ER, PR and a positive association with high tumour grade, HER2 , p53 and intrinsic basal-like subtype (Chan et al 2011 [EM20]). No SIAH2 promoter methylation was identified but there was a significant correlation between SIAH2 mRNA and gene copy number. There was also a significant association between

SIAH2 positive tumours and a shorter relapse-free survival in a univariate but not multivariate analysis. SIAH2 expression was upregulated in basal-like breast cancers via copy number changes and/or transcriptional activation by p53 and may be partly responsible for the enhanced hypoxic drive through abrogation of the prolyl hydroxylases. This suggests that anti-SIAH therapeutics may be a potential site of therapeutic attack for these chemoresistant tumours.

Conclusions:

Basal-like cancers are associated with over-expression of HIF pathway components which play a pivotal role in driving pathways critical to tumour survival, aggressiveness and resistance to treatment. Some of these, such as CAIX, have prognostic and predictive potential for this group of poor prognosis tumours.

Publications arising from this work:

EM18. Tan EY, Yan M, Campo L, Han C, Takano E, Turley H, Pezzella F, Gatter KC,

Millar EKA, O’Toole SA, McNeil CM, Crea P, Segara D, Sutherland RL, Harris AL,

Fox SB. The key hypoxia regulated gene CAIX is upregulated in basal-like breast tumours and is associated with resistance to chemotherapy. Br J Cancer. 2009; 100:

405-411. 66

EM19. Yan M, Jene N, Byrne D, Millar EKA, O’Toole SA, McNeil CM, Sutherland

RL, Fox SB. Recruitment of regulatory T cells is driven by hypoxia induced CXCR4 expression, and is associated with poor prognosis in basal-like breast cancers.

Breast Cancer Research. 2011.13:R47.

EM20. Chan P, Moeller A, Liu MCP, Sceneay JE, Wong CSF, Wadell N, Huang K,

Dobrovic A, Millar EKA, O’Toole SA, McNeil CM, Sutherland RL, Bowtell D, Fox SB.

Chapter 6. DNA Repair Pathways

Basal breast cancers have an association with BRCA1 mutations and harbour potential dysfunctional DNA repair processes. Rad21/SCC1, a component of the cohesin complex, is essential for sister chromatid cohesions which are important during mitosis and error-free DNA repair. Two cohesin core subunits, Smc1 and

Smc3, are members of the “structural maintenance of chromosomes” (SMC) family whose members are large ATPases with polypeptide chains which fold back on themselves around a central “hinge” domain (Figure 25) forming a ring-like structure

(reviewed in Peters et al 2008). Within the cohesin complex, the hinge domains of

Smc1 and Smc3 bind tightly to each other, whereas the ATPase heads of both proteins are physically connected by the Rad21/Scc1 subunit.

Figure25.ThecohesincomplexisformedbyamultiͲproteinstructurecontainingRad21/Scc1,Smc1 andSmc3formingaringͲlikestructurearoundchromatin(takenfromPetersetal2008). Aberrant expression of Rad21 has been described in cancer and it was identified as one of sixty-nine signature genes in undifferentiated cancers that had aggressive in vitro or clinical courses and poor patient outcomes (Rhodes et al 2004). We observed its expression to correlate with early relapse in all breast cancer patients largely due to the effect of grade 3 tumours, in which Rad21 expression correlated

with early relapse in all subtypes of breast cancer (Xu et al 2011 [EM21]). With chemotherapy, Rad21 expression associated with a poorer prognosis. Knockdown of

Rad21 mRNA in the MDA-MB-231 basal breast cancer cell line, significantly enhanced sensitivity to cyclophosphamide, 5-fluorouracil and etoposide, supporting the clinical findings.

Conclusions:

Rad21 expression may be a novel therapeutic target and predictive biomarker as it appears to confer a poor prognosis and resistance to chemotherapy in all high grade tumours (luminal, basal or HER2). This is particularly relevant for those basal-like cancers for whom no specific therapy currently exists.

Publication arising from this work:

EM21. Xu H, Yan M, Patra J, Yan Y, Swagemakers S, Thomasewski J, Verschoor S,

Millar EKA, van der Spek P, Reis-Filho JS, Ramsay R, O’Toole SA, McNeil CM,

Sutherland RL, McKay MJ, Fox SB. Enhanced Rad21 cohesin expression confers poor prognosis and resistance to chemotherapy in high-grade luminal, basal and

HER2 breast cancer. Breast Cancer Research. 2011.13:R9.

Chapter 7. Predictive factors in ductal carcinoma in-situ (DCIS)

Despite the on-going drive towards identifying biomarkers which predict disease outcome, unlike invasive carcinoma, for DCIS, no established biomarkers exist which predict recurrence or progression to invasive disease (Polyak 2010). As a first step to investigating cell cycle marker expression and outcome in DCIS we analysed

60 cases of pure DCIS for cyclin A, cyclin D1 and p27 expression. We were unable to identify any association with outcome, although there was a positive association of p27 with ER/PR positivity, correlating with low nuclear grade (Millar et al 2007

[EM22]). Thus, given the absence any predictive biomarker, the role of traditional pathological factors still assume a vital role in guiding optimal management of DCIS.

This latter aspect is of particular importance in conservatively managed breast cancer where local excision with clear margins is important. We reviewed the data discussing the relevance of margin involvement and local recurrence in DCIS (Millar and Leong 2001 [EM23]). Although local recurrence is less than 10% with optimal surgical excision and radiotherapy, prevention is important since up to 50% of recurrences may be as invasive carcinoma. The practical considerations in assessing margin status are of key importance in obtaining a firm final opinion to guide subsequent radiotherapy to aid in local control.

Conclusions:

No established predictive biomarkers exist for DCIS and as a result, key pathological indicators such as margin status, grade and size continue to guide therapy. Further studies are needed to identify biomarkers which may predict outcome.

Publications arising from this work.

EM22. Millar EKA, Tran K, Marr P, Graham PH. p27kip-1, cyclin A and cyclin D1 protein expression in ductal carcinoma in-situ of the breast: p27kip-1 correlates with hormone receptor status but not with local recurrence. Pathology International. 2007.

57: 183-9.

EM23. Millar EKA and Leong A S-Y. Significance and assessment of margins of excision in ductal carcinoma in-situ of the breast. Advances in Anatomic Pathology.

2001. 8:338-344.

Chapter 8. Biomarkers in rare types of breast tumours.

8.1 Phyllodes tumours

Phyllodes tumours of the breast are rare tumours characterised by increasing stromal overgrowth, cytological atypia and malignancy and are currently classified as benign, “borderline malignant” or malignant using histological criteria. They do however suffer from some difficulty in terms of definitive classification and therefore accurate prediction of outcome. As a first step towards identifying a prognostic biomarker for this disease, we made the first description of p53 protein expression in

15 phyllodes tumours (six malignant, nine benign) and 20 fibroadenomas (Millar et al

1999, [EM24]). Five of the six malignant phyllodes tumours showed moderate or strong p53 positivity at sites of periepithelial stromal condensation and atypia (Figure

26). All 20 fibroadenomas, nine benign phyllodes tumours and one malignant phyllodes tumour showed either negativity or focal weak nuclear positivity of scattered stromal cells. The observed p53 staining in malignant phyllodes was thought at that time to be a distinctive pattern, confirmatory of a malignant phenotype and potentially of prognostic importance. However subsequent studies by other investigators have shown p53 expression to be of no prognostic value (Tan et al

2005). However, abnormalities in the p53 pathway may be important in the progression of benign to malignant phyllodes tumours.

Figure 26. p53 IHC in a malignant phyllodes tumour of the breast (taken from Millar et al 2001). 72

8.2 Low-grade mucosa associated lymphoid tissue (MALT) non-Hodgkin’s lymphoma.

Primary non-Hodgkin’s lymphoma (NHL) of the breast comprises 0.04–0.53% of primary malignant breast tumours. Variable frequencies of primary mucosa associated lymphoid tissue (MALT) lymphoma of the breast are reported and comprise 4% of all extranodal MALT lymphomas. Other more common non- gastrointestinal sites of extranodal MALT lymphomas include the head and neck, lung, ocular adnexa, and salivary glands. Bilateral, primary MALT lymphoma of the breast is exceptionally rare, with only eight cases reported in the published work.

The optimal treatment is not well established. We described the occurrence of bilateral low-grade MALT NHL of the breast in a 73 year-old female, occurring seven years apart. As no predictive biomarkers exist for this disease we reviewed the current literature for potential prognostic and predictive targets. Multiple cytogenetic abnormalities exist, some correlating with the site of disease. Two translocations of the MALT1 gene on 18q21 are known: t(11; 18)(q21; q21) with API2 found mainly in gastric and pulmonary tumours and t(14; 18)(q32; q21) with IGH, which is found predominantly in ocular/adnexal and salivary lesions. The relatively rare but well known translocation involving the BCL10 gene t(1; 14)(p22; q32) with IGH is associated with lung and gastric disease and appears to be specific for MALT lymphoma. Both MALT1 and BCL10 translocations are associated with activation of the NFkB pathway and cell survival. t(3; 14)(p13; q32) is a newly identified chromosomal translocation in MALT lymphoma, which deregulates the expression of the FOXP1 gene and is described in thyroid, ocular, skin and lung disease. The genetic features of MALT breast lymphoma are not yet well established. Talwalkar et

al (2006) showed no evidence of t(11; 18) and t(14; 18) translocations in MALT lymphoma of the breast. There are no established markers of outcome in this disease, although a recent study indicated that gastric MALT lymphoma with t (1; 14) or strong BCL10 nuclear expression are unlikely respond to H. Pylori eradication.

Further investigation in the field is therefore required.

Conclusions:

No predictive or prognostic biomarkers exist for phyllodes tumours or MALT lymphoma of the breast. Further investigation is needed to better identify those patients at risk of recurrence.

Publications arising from this work:

EM24. Millar EKA, Beretov J, Marr P, Sarris M, Clarke RA, Kearsley JM and Lee

CS. Malignant phyllodes tumours of the breast display increased stromal p53 protein expression. Histopathology. 1999. 34:491-496.

EM25. Yilmaz MH, Millar EKA, Theocharous C, Graham PH. Metachronous bilateral primary low-grade mucosa-associated lymphoid tissue (MALT) non-Hodgkins lymphoma of the breast. Asia-Pacific Journal of Clinical Oncology. 2009. 5:154-158.

Chapter 9. Authorship of Publications for Thesis

Contemporary major advances in biomarkers of breast cancer phenotype and therapeutic responsiveness require multidisciplinary/multi-institutional teams. The

Cancer Program at the Garvan Institute, developed over a decade, has several multidisciplinary and investigational teams. Breast pathology is a significant component of this program and has a central pivotal role upon which observations about underlying biology and prognosis can be made. The subsequent discussion outlines in more specific detail, my contribution to publications included in this

Thesis. The particular position that one occupies on the list of authors demonstrates that for those publications where I was first author, I was principally responsible for driving the study, analysing and interpreting the data and drafting the manuscript.

This is also the case for two further publications where I am second or third author but have been credited with an equal level of contribution to the publication as the first author. However, as outlined above, a specialist breast Pathologist was pivotal to all studies, since the appropriate diagnosis of IHC data derived from human tissue underpinned the quality and clinical relevance of all the data reported in the publications contained within this Thesis.

I always sought to be involved in research during my specialist training in Edinburgh

(1993-1998). When I moved permanently to Australia in 1998, working initially in

Newcastle (1998-2001), then St George Hospital, Sydney (from November 2001), by late 2004, I had carried out some small studies and written some papers on breast

(p53 expression in Phyllodes tumours 1999; DCIS margins invited review 2001) but soon realised that attempting to perform any truly internationally competitive research whilst working as a full time Specialist in Anatomical Pathology was

impossible. Subsequently in March 2005, I was invited by Professor Rob Sutherland,

Director of the Cancer Research Program, Garvan Institute of Medical Research,

Sydney to join the Breast Cancer Translational Research Group which had just been formed. I committed to giving up one day per week of my clinical work to join this multi-disciplinary group of investigators which includes senior laboratory-based

Scientists, breast surgeons, medical and radiation Oncologists and Pathologists. The majority of the work in this Thesis was carried out as a member of this group, which is, by necessity, collaborative in nature, requiring the expert contribution of several key individuals from a number of complementary disciplines. Weekly meetings are held by our group at which there is discussion on all on-going studies with input into their direction and drafting of manuscripts by all members of the team. My primary role was to provide key specialist, pathological input to underpin and drive the design, implementation and analysis of studies to be undertaken by the group.

During this time I also finalised some other outstanding work with A/Prof Peter

Graham (cell cycle markers in DCIS, mucosa associated lymphoid tissue (MALT) lymphoma) to which I contributed by driving the projects, assessing all pathology and drafting the manuscripts. An additional study examining factors associated with supraclavicular fossa radiotherapy associated lymphoedema, I reviewed all the pathology and contributed to the writing of the manuscript. Concurrently Professor

Soon Lee (Royal Prince Alfred Hospital, University of Sydney) invited me to collaborate to the writing and provide specialist pathological input to a review on gene expression profiling in breast cancer with Dr Niamh Murphy, his Post-Doctoral research scientist who was also just about to begin work at the Garvan. This timely project helped me gain further insights into this rapidly evolving area at an early stage. More recently I was invited to contribute to another review of molecular testing 76

in breast cancer, by Associate Professor Sandra O’Toole a Visiting Scientist at the

Garvan(Royal Prince Alfred Hospital and University of Sydney). This is an area of expanding importance in research and routine reporting of all breast cancer cases with the advent of chromogenic in-situ hybridisation (CISH). As the supervising

Pathologist for the CISH service at SEALS, St George Hospital, I was responsible for setting up this service (2006) which provides testing for all breast cancers from St

George and Prince of Wales Hospitals (approximately 500 cases per annum, approximately 12% of all breast cancers in NSW).

This published work examines the role of biomarkers in breast cancer: their role in pathogenesis, subtype classification, prediction of tumour behaviour and in some studies, the response to therapy. This work has predominantly utilised immunohistochemistry (IHC) to examine protein biomarker expression in TMAs constructed from two locally derived breast cancer clinical cohorts. In both of these cohorts (The Garvan St Vincent’s Breast Cancer Cohort, GSVBCC, (n=292) and the

St George Breast Boost Study, (n=498), I oversaw the development of these critical resources covering all aspects from case reviewing and marking up of the tissue blocks for TMA construction to optimisation of antibodies, scoring of IHC, data analysis and drafting of manuscripts. IHC was performed for baseline parameters such as ER, PR, HER2, CK5/6 and EGFR. For many of the studies outlined below I was responsible for directing the optimisation of IHC staining protocols for multiple monoclonal antibodies used, as well as assessment of staining/scoring, which was performed in association with Associate Professor Sandra O’Toole. Clinical follow-up data were obtained for the GSVBCC from patient files made available by Dr Paul

Crea, breast Surgeon at St Vincent’s Hospital, largely by Dr Catriona McNeil, a

medical Oncologist. Clinical outcome data for the St George Boost cohort was provided by a collaborator and colleague at St George Hospital, Associate Professor

Peter Graham, Radiation Oncologist, Cancer Care Centre with statistical support from Ms Lois Browne, senior statistician St George Hospital. Prior to publication, data from all studies was discussed by the group and suggestions made as how best to present the study for publication with intellectual input from all authors.

An early project I was involved in was the assessment of c-Myc staining in a

“progression series” of cases which included pre-malignant lesions, ductal carcinoma in-situ and invasive carcinomas with Dr Catriona McNeil, who was doing her PhD in endocrine resistance with Professors Sutherland and Musgrove. This early work formed the basis, along with in-vitro work performed by Catriona and Professor Liz

Musgrove for the study on the role of c-Myc in driving endocrine resistance in breast cancer, a key interest of our group, to which I contributed along with the group to the writing of the manuscript. Subsequently further data derived from Prof Liz Musgrove,

Prof Sutherland and Dr Alison Butt identified the expression of estrogen target genes in MCF-7 breast cancer cells. These data lead to two publications on BAG1 and

PUMA both of which are key regulators of apoptosis. I coordinated the BAG1 paper and performed all of the IHC analyses, data analyses and drafted the manuscript. mRNA expression datasets were extracted by Mark Pinese and statistically analysed by myself. Luke Anderson, Dr Alison Butt’s PhD student at that time performed the in-vitro work which demonstrated the mechanistic role of BAG1 over-expression in

MCF7 cell lines. Similarly for PUMA I performed the IHC assessment and data analysis of the IHC results and GEP mRNA datasets, while Caroline Roberts carried

out the cell line work and western blots and Dr Alison Butt drafted the initial manuscript with further detailed input from all authors.

During this phase of work, the expertise of myself and other members of the group was actively sought by several other collaborative research groups within Australia and overseas. STARD10 was a novel protein which had been identified by

Professors Geoff Lindeman and Jane Visvader, Walter and Eliza Hall Institute,

Melbourne. They had produced a monoclonal antibody to STARD10, a novel lipid transport molecule, and had approached me to assess its staining on our breast cancer cohorts. Dr Niamh Murphy and I optimised the staining and assessed its expression by IHC. Niamh Murphy drafted the original manuscript which went through several substantial revisions, for which Professor Andrew Biankin and I were responsible for finalising prior to publication, which did not occur for three years after the study was completed.

During this time the GSVBCC was expanded in size by the addition of extra patients.

The publication of IHC surrogates of intrinsic subtype by Neilson in 2004 had prompted us and others to subtype our own cohorts using a simplified biomarker panel of ER, PR, HER2, EGFR and CK5/6. HER2 was assessed by IHC and fluorescent in-situ hybridisation (FISH) which was performed by Associate Professor

Adrienne Morey, SydPath, St Vincent’s Hospital Sydney. Additional markers of proliferation, Ki67 and the tumour suppressor p53 were later added, along with cell cycle proteins cyclin D1, cyclin E1 & E2 (the latter in support of a UNSW Medical student Honours project for Edwina Wing-Lum), p27, p21 to further assess differing classifications of intrinsic subtype. This large panel of markers on the GSVBCC was included in the PhD Thesis of Dr Catriona McNeil (UNSW 2008) and has been

prepared for publication with detailed assessment of staining by myself, Dr Catriona

McNeil and Associate Professor Sandra O’Toole. A similar IHC panel approach was subsequently taken to assess the St George Boost cohort for which I was responsible for designing, coordinating, assessing all IHC staining and performing the statistical analyses. This study lead to two large publications both of which I was responsible for drafting (18 months apart), the first of which was published in the highly prestigious Journal of Clinical Oncology, (impact factor 18.97, one of the top ranking international Oncology journals) which was a major achievement. The second paper was recently published in The British Journal of Cancer. Other ongoing collaborations at that time included work for the BSc Thesis of Sarah Zardawi, examining the role of Notch signalling in breast cancer. Shortly after this Dr Elena

Lopez-Knowles took over in the lab as Post-Doctoral scientist after the departure of

Niamh Murphy. Elena was responsible for the PI3K mutation analysis and in optimising IHC for downstream pathway components phospho AKT, FHKR and

PTEN, the IHC analysis of which I performed with Sandra O’Toole. Elena drafted the initial manuscript and also worked up the project on ȕ-catenin with Sandra O’Toole with further input from all authors. Whilst these projects were on-going Professor Eric

Knudsen, Kimmel Cancer Centre, Philadelphia, USA sought our expertise to assess a new antibody to the cyclin D1b isoform. Cyclin D1 is a major interest of our group, given its role in breast cancer. I was responsible for directing the optimisation of the antibody, assessing the scoring and statistical analysis of the data in the GSVBCC.

The observations I made indicated its potential importance and this highly successful work lead to two publications in the highly reknowned cancer journals Cancer

Research and Oncogene, (the latter also incorporating some further SNP data of

Prof Knudsen’s from a large case control study assessing risk of breast cancer in 80

CCND1 polymorphism carriers). At this time Sandra O’Toole was also completing her thesis on stem cell signalling in breast cancer which focussed principally on the role of hedgehog (Hh) signalling. This work involved a large amount of IHC performed to which I contributed with analysis of Hh in a progression cohort of early precursor lesions (normal, hyperplasia, atypical hyperplasia, ductal carcinoma in situ and invasive carcinomas). This work resulted in a large manuscript recently published in Cancer Research. Another collaborative study with Professor Roger

Daly, Garvan Institute, on the role of the signal transducer docking protein Gab2 was also ongoing at this time, which was a project driven by Prof Daly’s research assistant Gaby Fleuren, who drafted the manuscript.

During this time a further close working collaboration was established with Professor

Stephen Fox and Dr Max Yan, Peter MaCallum Cancer Centre, Melbourne, following a breast cancer workshop I presented at for the National Breast Cancer Foundation in 2007. A mutual interest in basal-like cancers and an interest in tumour hypoxia lead us to complete a number of studies examining CAIX, FOXP3, SIAH2,

RAD21/cohesin, which were all performed on TMAs from the GSVBCC. Data generated by Prof Fox and Dr Yan was correlated with data on our existing biomarkers/parameters and discussed with our team, with input into the drafting of the manuscripts by all members of the group.

Most recently our group was again approached to collaborate with Professor

Christine Mitchell’s group at Monash University, Melbourne. This novel work addressed the role a phosphatase inhibitor (PPInd4B) in the PI3K pathway which they had identified and resulted in a significant publication in the highly prestigious journal Proceedings of the National Academy of Science, another major

achievement. Using their data on the in-vitro role of this molecule we optimised and performed an IHC and data analysis in our GSVBCC cohort and correlated its expression with our previously published and detailed PI3K pathway data to generate significant and novel observations about this protein. The paper was drafted by Prof Mitchell’s group with significant input from all of our team.

Chapter 10. Significance of findings from published work

The work presented in this thesis contains 25 publications, the significance and impact of which are reflected in the quality of the journal, its associated impact factor and the number of citations. This volume of work has resulted in my Scopus H Index increasing from 5 (2005) to 11 (July 2011). This is however, expected to rise significantly and rapidly. Citations have been sourced from Scopus and Google

Scholar and are summarised below. Recent publications from 2011 have no citations as yet. One of these publications [EM20] has been cited as one of “high accessed” relative to age by Breast Cancer Research with 1901 downloads to date. Similarly

EM4 ranks as the fourth most highly accessed recent publication in the British

Journal of Cancer at time of writing.

Publication Journal Impact Factor Citations Year

EM3 Journal of Clinical Oncology 18.97 26 2009

EM12 Proceedings of the National Academy of Science (USA) 9.77 5 2010

EM9 Cancer Research 8.23 24 2008 EM13 Cancer Research 8.23 0 2011

EM7 Oncogene 7.41 0 2011 EM10 Oncogene 7.41 18 2009

EM19, 20, 21 Breast Cancer Research 5.79 0 2011

EM11 International Journal of Cancer 4.93 19 2010 EM16 International Journal of Cancer 4.93 3 2010 EM17 International Journal of Cancer 4.93 5 2010

EM 4 British Journal of Cancer 4.83 0 2011 EM6 British Journal of Cancer 4.83 12 2009 EM18 British Journal of Cancer 4.83 24 2009

EM15 Cancer Epidemiology Biomarkers & Prevention 4.19 11 2005

EM14 Histopathology 3.57 3 2010 EM24 Histopathology 3.57 42 1999

EM22 Advances in Anatomic 3.3 4 2001 Pathology

EM8 Journal of Steroid Biochemistry & Molecular Biology 2.89 17 2006

EM1 Pathology 2.168 33 2005 EM 2 Pathology 2.168 1 2011

EM23 Pathology International 1.48 5 2007

EM5 Australasian Radiology (Journal of Medical Imaging & Radiation Oncology) 0.947 10 2006

EM25 Asia-Pacific Journal of Clinical Oncology 0.296 0 2007

References

Abd El-Rehim DM, Ball G, Pinder SE, Rakha E, Paish C, Robertson JF,

Macmillan D, Blamey RW, Ellis IO. High-throughput protein expression

analysis using tissue microarray technology of a large well-characterised

series identifies biologically distinct classes of breast cancer confirming recent

cDNA expression analyses.Int J Cancer. 2005. 116:340-50.

Alpy F, Tomasetto C. Give lipids a START: the StAR-related lipid transfer

(START) domain in mammals. J Cell Sci. 2005;118:2791–801.

Anagnostou VK, Lowery FJ, Syrigos KN, Cagle PT, Rimm DL. Quantitative

Evaluation of Protein Expression as a Function of Tissue Microarray Core

Diameter. Is a Large (1.5 mm) Core Better Than a Small (0.6 mm) Core?

Arch Pathol Lab Med. 2010;134:613–619.

Arnold A, Papanikolaou A. Cyclin D1 in breast cancer pathogenesis. J Clin

Oncol. 2005; 23:4215–24.

Artavanis-Tsakonas S, Rand MD, Lake RJ. Notch signaling: cell fate control

and signal integration in development. Science.1999; 284;770–776.

Australian Institute of Health and Welfare & National Breast and Ovarian

Cancer Centre. Breast cancer in Australia: an overview. Cancer series no. 50.

Cat. no. CAN 46. Canberra: AIHW. Breast cancer incidence, 2009.

Bates GJ, Fox SB, Han C, Leek RD, Garcia JF, Harris AL, Banham AH.

Quantification of regulatory T cells enables the identification of high-risk

breast cancer patients and those at risk of late relapse. J Clin Oncol. 2006;

1;24:5373-80.

Blows FM, Driver KE, Schmidt MK, Broeks A, van Leeuwen FE, Wesseling J,

Cheang MC, Gelmon K, Nielsen TO, Blomqvist C, Heikkilä P, Heikkinen T,

Nevanlinna H, Akslen LA, Bégin LR, Foulkes WD, Couch FJ, Wang X,

Cafourek V, Olson JE, Baglietto L, Giles GG, Severi G, McLean CA, Southey

MC, Rakha E, Green AR, Ellis IO, Sherman ME, Lissowska J, Anderson WF,

Cox A, Cross SS, Reed MW, Provenzano E, Dawson SJ, Dunning AM,

Humphreys M, Easton DF, García-Closas M, Caldas C, Pharoah PD,

Huntsman D. Subtyping of breast cancer by immunohistochemistry to

investigate a relationship between subtype and short and long term survival: a

collaborative analysis of data for 10,159 cases from 12 studies. PLoS

Medicine 2010. 7 e1000279.

Bouras T, Pal B, Vaillant F, Harburg G, Asselin-Labat ML, Oakes SR,

Lindeman GJ, Visvader JE. Notch signaling regulates mammary stem cell

function and luminal cell-fate commitment. Cell Stem Cell. 2008; 3; 429–441.

Brahimi-Horn MC, Chiche J, Pouysségur J. Hypoxia and cancer. J Mol Med.

2007; 85:1301-1307.

Buckley MF, Sweeney KJ, Hamilton JA, Sini RL, Manning DL, Nicholson RI,

deFazio A, Watts CK, Musgrove EA, Sutherland RL. Expression and

amplification of cyclin genes in human breast cancer. Oncogene 1993;

8:2127–33.

Carey LA, Perou CM, Livasy CA, Dressler LG, Cowan D, Conway K, Karaca

G, Troester MA, Tse CK, Edmiston S, Deming SL, Geradts J, Cheang MC,

Nielsen TO, Moorman PG, Earp HS, Millikan RC. Race, breast cancer

subtypes and survival in the Carolina breast cancer study. JAMA. 2006; 295:

2492-2502.

Camp RL, Neumeiste V and Rimm DL. A Decade of Tissue Microarrays:

Progress in the Discovery and Validation of Cancer Biomarkers. J Clin Oncol

2008 26:5630-5637.

Chan P, Moeller A, Liu MCP, Sceneay JE, Wong CSF, Wadell N, Huang K,

Dobrovic A, Millar EKA, O'Toole SA, McNeil CM, Sutherland RL, Bowtell D,

Fox SB. The expression of the ubiquitin ligase SIAH (seven in absentia

homolog) 2 is associated with a basal-like breast cancer that is mediated

through gene copy number and/or p53. Breast Cancer Research. 2011.

13:R19.

Cheang MC, Voduc D, Bajdik C, Leung S, McKinney S, Chia SK, Perou CM,

Nielsen TO. Basal like breast cancer defined by five biomarkers has superior

prognostic value than triple-negative phenotype. Clin Cancer Res.

2008.14:1368-1376.

Cheang MC, Chia SK, Voduc D, Gao D, Leung S, Snider J, Watson M, Davies

S, Bernard PS, Parker JS, Perou CM, Ellis MJ, Nielsen TO. Ki67 index, HER2

status, and prognosis of patients with luminal B breast cancer. J Natl Cancer

Inst 2009. 101:736-750.

Cuzick J, Dowsett M, Wale C, Salter J, Quinn E, Zabaglo L, Howell A, Buzdar

A, Forbes JF: Prognostic Value of a Combined ER, PgR, Ki67, HER2

Immunohistochemical (IHC4) Score and Comparison with the GHI

Recurrence Score - Results from TransATAC. Cancer Res.2009; 69:503S-

503S.

Dolled-Filhart M and Rimm DL.

http://www.cap.org/apps/docs/futurescape/course_material/Sat-

5_Rimm_1045-1130.pdf Construction of TMAs diagram.

Early Breast Cancer Trialists’ Collaborative Group (EBCTCG). Effects of

chemotherapy and hormonal therapy for early breast cancer on recurrence

and 15-year survival: an overview of the randomised trials. Lancet. 2005.

365:1687-1717.

Early Breast Cancer Trialists’ Collaborative Group (EBCTCG): Effects of

radiotherapy and of differences in the extent of surgery for early breast cancer

on local recurrence and 15-year survival: An overview of the randomised

trials. Lancet. 2005. 366:2087-2106.

Fedele CG, Ooms LM, Ho M, Vieusseux J, O’Toole SA, Millar EKA, Lopez-

Knowles E, Sriratana A, Gurung R, Baglietto L, Giles GG, Bailey CG, Rasko

JEJ, Shields BJ, Price JT, Majerus PW, Sutherland RL, Tiganis T, McLean

CA, Mitchell CA. The PtdIns(3,4)P2 4-phosphatase, INPP4B, regulates ER-

positive mammary cell proliferation and is lost in human basal-like breast

carcinomas. Proc Natl Acad Sci. 2010.107: 22231–22236.

Fleuren EDG, O’Toole SA, Millar EKA, McNeil CM, Lopez-Knowles E,

Brummer T, Sutherland RL, Daly RJ. Overexpression of the oncogenic signal

transducer Gab2 occurs early in breast cancer development. Int J Cancer.

2010; 127: 1486–1492.

Goldhirsch A, Ingle JN, Gelber RD, Coates AS, Thürlimann B, Senn HJ.

Thresholds for therapy: Highlights of the international expert consensus on the

primary therapy of early breast cancer. Ann Oncol. 2009. 20: 1319-1329.

Goldhirsch A, Wood WC, Coates AS, Gelber RD, Thürlimann B, Senn H-J.

Strategies for subtypes - dealing with the diversity of breast cancer: highlights

of the St.Gallen International Expert Consensus on the Primary Therapy of

Early Breast Cancer 2011. Annals of Oncology 2011. In press. 89

Hanrahan D and Weinburg RA. Hallmarks of cancer: The next generation.

Cell. 2011; 144:646-674.

Hu Z, Fan C, Oh DS, Marron JS, He X, Qaqish BF, Livasy C, Carey LA,

Reynolds E, Dressler L, Nobel A, Parker J, Ewend MG, Sawyer LR, Wu J, Liu

Y, Nanda R, Tretiakova M, Ruiz Orrico A, Dreher D, Palazzo JP, Perreard L,

Nelson E, Mone M, Hansen H, Mullins M, Quackenbush JF, Ellis MJ, Olopade

OI, Bernard PS, Perou CM. The molecular portraits of breast tumors are

conserved across microarray platforms. BMC Genomics. 2006. 7:96.

Kakarala M and Wicha M. Implications of the Cancer Stem-Cell Hypothesis for

Breast Cancer Prevention and Therapy. J Clin Oncol. 2008; 26:2813-2820.

Knudsen KE, Diehl JA, Haiman CA, Knudsen ES. Cyclin D1: polymorphism,

aberrant splicing and cancer risk. Oncogene. 2006a; 25:1620–1628.

Knudsen K. The CD1b splice variant: an old dog learns new tricks. Cell

Division. 2006b; 1:15.

Kononen J, Bubendorf L, Kallioniemi A, Bärlund M, Schraml P, Leighton S,

Torhorst J, Mihatsch MJ, Sauter G, Kallioniemi OP. Tissue microarrays for

high-throughput molecular profiling of tumor specimens. Nat Med 1998. 4:844-

847.

López-Knowles E, O’Toole SA, McNeil CM, Millar EKA, Qiu MR, Crea P,

Musgrove EA, Robert L. Sutherland. PI3K pathway activation in breast cancer

is associated with the basal-like phenotype and cancer-specific mortality. Int J

Cancer. 2010; 126:1121-31.

López-Knowles E, Zardawi SJ, McNeil CM, Millar EKA, Crea P, Musgrove

EA, Sutherland RL,. O’Toole SA. Cytoplasmic localization of ß catenin is a

marker of poor outcome in breast cancer patients. Cancer Epidemiol

Biomarkers Prev. 2010. 19:301-9.

McNeil CM, Sergio CM, Anderson LR, Inman CK, Murphy NC, Millar EKA,

Crea P, Kench JG, Alles MC, Gardiner-Garden M, Ormandy CJ, Butt AJ,

Henshall SM, Musgrove EA, Sutherland RL. C-Myc overexpression and

endocrine resistance in breast cancer. J Ster Biochem Mol Biol. 2006;

102:147-55.

McNeil CM. Downstream targets of the oestrogen receptor and endocrine

resistance. PhD Thesis. UNSW 2008.

Musgrove EA and Sutherland RL. Biological determinants of endocrine

resistance in breast cancer. Nat Rev Cancer. 2009; 9:631-643.

Musgrove EA, Caldon CE, Barraclough J, Stone A, Sutherland RL. Cyclin D 1

as a therapeutic target in cancer. Nat Rev Cancer. 2011.11:558-72.

Millar EKA, Beretov J, Marr P, Sarris M, Clarke RA, Kearsley JM and Lee CS.

Malignant phyllodes tumours of the breast display increased stromal p53

protein expression. Histopathology. 1999. 34:491-496.

Millar EKA and Leong A S-Y. Significance and assessment of margins of

excision in ductal carcinoma in-situ of the breast. Advances in Anatomical

Pathology. 2001. 8:338-344.

Millar EKA, Tran K, Marr P, Graham PH. p27KIP-1 cyclin A and cyclin D1

protein expression in ductal carcinoma in situ of the breast: p27KIP-1 correlates

with hormone receptor status but not with local recurrence. Pathology

International. 2007. 57:183-189.

Millar EKA, Anderson LR, McNeil CM, O’Toole SA, Pinese M, Crea P, Morey

AL, Biankin AV, Henshall SM, Musgrove EA, Sutherland RL, Butt AJ. BAG-1

predicts patient outcome and tamoxifen responsiveness in ER positive

invasive ductal carcinoma of the breast. Br J Cancer.100; 122-133, 2009.

Millar EKA, Graham PH, O’Toole SA, McNeil CM, Browne L, Morey AL,

Eggleton S, Beretov J, Theocharous C, Capp A, Nasser E, Kearsley JH,

Papadatos G, Delaney G, Fox C, Sutherland RL. Prediction of local

recurrence, distant metastases, and death after breast-conserving therapy in

early-stage invasive breast cancer using a five-biomarker panel. J Clin Oncol.

2009; 27: 4701–08.

Millar EKA, Dean JL, McNeil CM, O'Toole SA, Henshall SM, Tran T, Lin J,

Quong A, Comstock CES, Witkiewicz A, Musgrove EA, Rui H, Le Marchand L,

Setiawan VW, Haiman CA, Knudsen KE, Sutherland RL, Knudsen ES. Cyclin

D1b Protein expression in breast cancer is independent of cyclin D1a and

associated with poor disease outcome. Oncogene. 2009. 28:1812-20.

Millar EKA, Graham PH, McNeil CM, Browne L, O’Toole SA, Boulghourjian A,

Delaney G, Papadatos G, Nasser E, Capp A, Fox C, Sutherland RL.

Prediction of outcome of early luminal breast cancer is improved using a

biomarker panel which includes Ki-67 and p53. British Journal of Cancer.

2011. 105: 272-280.

Murphy N, Millar E, Lee CS. Gene Expression Profiling in Breast Cancer:

Towards Individualising patient management. Pathology. 2005. 37: 271-277.

Murphy NC, Biankin AV Millar EKA, McNeil CM, O'Toole SA, Segara D, Crea

P, Olayioye MA, Lee CS, Fox SB, Morey AL, Christie M, Musgrove EA, Daly

RJ, Lindeman GJ,. Henshall SM, Visvader JE, Sutherland RL. Loss of

STARD10 expression identifies a group of poor prognosis breast cancers

independent of HER2/Neu and triple negative status. Int J Cancer.

2010.126:1445-53

Musgrove EA, Sergio CM, Loi S, Inman CK, Anderson LR, Alles MC, Pinese

M, Caldon CE, Schütte J, Gardiner-Garden M, Ormandy CJ, McArthur G, Butt

AJ, Sutherland RL. Identification of functional networks of estrogen- and c-

Myc-responsive genes and their relationship to response to tamoxifen therapy

in breast cancer. PLoS One. 2008 Aug 20;3(8):e2987.

Nielsen TO, Hsu FD, Jensen K, Cheang M, Karaca G, Hernandez-Boussard

T, Livasy C, Cowan D, Dressler L, Akslen LA, Ragaz J, Gown AM, Gilks CB,

van de Rijn M, Perou CM. Immunohistochemical and clinical characterization

of the basal-like subtype of invasive breast cancer. Clin Cancer Res. 2004.

10:5367-5374.

Nguyen PL, Taghian AG, Katz MS, Niemierko A, Abi Raad RF, Boon WL,

Bellon JR, Wong JS, Smith BL, Harris JR. Breast cancer subtype

approximated by estrogen receptor, progesterone receptor, and HER-2 is

associated with local and distant recurrence after breast conserving therapy. J

Clin Oncol. 2008; 26:2373-2378.

Olayioye MA, Hoffmann P, Pomorski T, Armes J, Simpson RJ, Kemp BE,

Lindeman GJ, Visvader JE. The phosphoprotein STARD10 is overexpressed

in breast cancer and cooperates with ErbB receptors in cellular

transformation. Cancer Res. 2004; 64:3538–44.

O’Toole SA, Selinger T, Millar EKA , Lum T, Beith JM. Molecular assays in

breast cancer pathology. Pathology. 2011; 41:116-127.

O’Toole SA, Swarbrick A, Sutherland RL. The Hedgehog signalling pathway

as a therapeutic target in early breast cancer development. Expert Opin. Ther.

Targets. 2009; 13:1095-1103.

O’Toole SA. Studies in stem cell biology and developmental pathway

regulation in the pancreas and breast. PhD Thesis UNSW 2008.

O’Toole SA, Machalek D, Shearer R , Millar EKA, Nair R, McLeod D,

Cooper C, McFarland A, Ru Qiu M, McNeil CM, Rabinovich B, Martelotto L,

Vu D, Musgrove E, Sutherland RL, Watkins N, Swarbrick A. Hedgehog

overexpression is associated with stromal interactions and predicts for poor

outcome in breast cancer. Cancer Res. 2011.71: 4002-4014.

Paik S, Shak S, Tang G, et al: A multigene assay to predict recurrence of

tamoxifen-treated, node-negative breast cancer. New England J Medicine

2004. 351:2817-2826.

Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack

JR, Ross DT, Johnsen H, Akslen FA, Fluge O, Pergamenschikov A, Williams

C, Zhu SX, Lonning PE, Borreson-Dale AL, Brown PO, Botstein D. Molecular

portraits of human breast tumours. Nature. 2000.406:747-752.

Peters JM, Tedeschi A, Schmitz J. The cohesin complex and its roles in

chromosome biology. Genes Dev. 2008 22: 3089-3114.

Polakis P. Wnt signaling and cancer. Genes Dev 2000;14:1837–51.

Polyak K. Molecular markers for the diagnosis and management of ductal

carcinoma in situ. J Natl Cancer Inst Monogr. 2010;2010:210-3.

Prall OW, Rogan EM, Musgrove EA, Watts CK, Sutherland RL. c-Myc or

cyclin D1 mimics estrogen effects on cyclin E-Cdk2 activation and cell cycle

reentry. Mol Cell Biol. 1998. 18:4499-508.

Rakha EA, Reis-Filho JS, Ellis IO: Basal-like breast cancer: A critical review. J

Clin Oncol. 2008; 26:2568-2581.

Rhodes DR, Yu J, Shanker K, Deshpande N, Varambally R, Ghosh D,

Barrette T, Pandey A, Chinnaiyan AM. Large-scale meta-analysis of cancer

microarray data identifies common transcriptional profiles of neoplastic

transformation and progression. Proc Natl Acad Sci U S A. 2004. 101:9309-

14.

Ring BZ, Seitz RS, Beck R, Shasteen WJ, Tarr SM, Cheang MC, Yoder BJ,

Budd GT, Nielsen TO, Hicks DG, Estopinal NC, Ross DT. Novel prognostic

immunohistochemical biomarker panel for estrogen receptor-positive breast

cancer. J Clin Oncol. 2006; 24:3039-3047.

Roberts CG, Millar EKA, O’Toole SA, McNeil CM, Lehrbach GM, Pinese M,

Tobelmann P, McCloy RA, Musgrove EA, Sutherland RL and Butt AJ.

Identification of PUMA as an estrogen target gene that mediates the apoptotic

response to tamoxifen in human breast cancer cells and predicts patient

outcome and Tamoxifen responsiveness in breast cancer. Oncogene. 2011.

30:3186-97.

Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie T,

Eisen MB, van de Rijn M, Jeffrey SS, Thorsen T, Quist H, Matese JC, Brown

PO, Botstein D, Eystein Lonning P, Borreson-Dale AL. Gene expression

patterns of breast carcinomas distinguish tumor subclasses with clinical

implications. Proc Natl Acad Sci USA 2001.98:10869-10874.

Sorlie T, Tibshirani R, Parker J, Hastie T, Marron JS, Nobel A, Deng S,

Johnsen H, Pesich R, Geisler S, Demeter S, Perou CM, Lonning PE, Brown

PO, Borreson-Dale AL, Botstein D. Repeated observation of breast tumor

subtypes in independent gene expression data sets. Proc Natl Acad Sci U S

A. 2003.100:8418-8423.

Sotiriou C, Pusztai L. Gene-expression signatures in breast cancer. N Engl J

Med. 2009; 360: 790–800.

Sullivan Pepe M, Etzioni R, Feng Z, Potter JD, Thompson ML, Thornquist M,

Winget M, Yasui Y. Phases of Biomarker Development for Early Detection of

Cancer. Journal of the National Cancer Institute, 14:1054-61, 2001. 97

Talwalkar SS, Valbuena JR, Abruzzo LV, Admirand JH, Konoplev SN, Bueso-

Ramos CE, Medeiros LJ. MALT1 gene rearrangements and NF-KB activation

involving p65 and p50 are absent or rare in primary MALT lymphomas of the

breast. Mod Pathol. 2006. 19: 1402–8.

Tan PH, Jayabaskar T, Yip G, Tan Y, Hilmy M, Selvarajan S, Bay BH. p53 and

c-kit (CD117) protein expression as prognostic indicators in breast phyllodes

tumors: a tissue microarray study.Modern Pathology. 2005. 18: 1527–1534.

Tan EY, Yan M, Campo L, Han C, Takano E, Turley H, Pezzella F, Gatter KC,

Millar EKA, O’Toole SA, McNeil CM, Crea P, Segara D, Sutherland RL, Harris

AL, Fox SB. The key hypoxia regulated gene CAIX is upregulated in basal-like

breast tumors and is associated with resistance to chemotherapy. Br J

Cancer. 2009; 100: 405-411.

van’t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AAM, Mao M, Peterse

HL, van der Kooy K, Marton MJ, Witteveen AT, Schreiber GJ, Kerkhoven RM,

et al. Gene expression profiling predicts clinical outcome of breast cancer.

Nature 2002;415:530–6.

van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Voskuil DW,

Schreiber GJ, Peterse JL, Roberts C, Marton MJ, Parrish M, Atsma D,

Witteveen A, Glas A, Delahaye L, van der Velde T, Bartelink H, Rodenhuis S,

Rutgers ET, Friend SH, Bernards R. A gene expression signature as a

predictor of survival in breast cancer. N Engl J Med. 2002; 347: 1999–2009.

Wan WH, Fortuna MB, Furmanski P: A rapid and efficient method for testing

immunohistochemical reactivity of monoclonal antibodies against multiple

tissue samples simultaneously. J Immunol Methods.1987.103:121-129.

Wang Y, Dean JL, Millar EKA, Tran TH, McNeil CM, Burd CJ, Henshall SM,

Utama FE, Witkiewicz A, Rui H, Sutherland RL, Knudsen KE, Knudsen ES.

Cyclin D1b is aberrantly regulated in response to therapeutic challenge and

promotes resistance to estrogen antagonists. Cancer Research. 2008;

68:5628- 5638.

Weigelt B, Mackay A, A’Hern R, Natrajan R, Tan DS, Dowsett M, Ashworth A,

Reis-Filho JS. Breast cancer molecular profiling with single sample predictors:

a retrospective analysis. Lancet Oncology. 2010. 11: 339–49.

Weigelt B and Reis-Filho JS. Molecular profiling currently offers no more than

tumour morphology and basic immunohistochemistry. Breast Cancer Res.

2010; 12 (suppl 4), S5.

Wilken JA, Maihle NJ. Primary trastuzumab resistance: new tricks for an old

drug. Ann N Y Acad Sci. 2010.1210:53-65.

Xu H, Yan M, Patra J, Yan Y, Swagemakers S, Thomasewski J, Verschoor S,

Millar EKA, van der Spek P, Reis-Filho JS, Ramsay R, O’Toole SA, McNeil

CM, Sutherland RL, McKay MJ, Fox SB. Enhanced Rad21 cohesin

expression confers poor prognosis and resistance to chemotherapy in high-

grade luminal, basal and HER2 breast cancer. Breast Cancer Research.

2011;13:R9.

Yan M, Jene N, Byrne D, Millar EKA, O'Toole SA, McNeil CM, Sutherland RL,

Fox SB. Recruitment of regulatory T cells is driven by hypoxia induced

CXCR4 expression, and is associated with poor prognosis in basal-like breast

cancers. Breast Cancer Research 2011. 13:R47.

Yilmaz MH, Millar EKA, Theocharous C, Graham PH. Metachronous bilateral

primary low-grade mucosa-associated lymphoid tissue (MALT) non-Hodgkin’s

lymphoma of the breast. Asia-Pacific J Clin Oncol. 2009. 5:154-158.

Zardawi SJ, O’Toole, Sutherland RL, Musgrove EA. Dysregulation of

Hedgehog, Wnt and Notch signalling pathways in breast cancer. Histol

Histopathol. 2009; 24:385-98.

100

Zardawi SJ, Zardawi I, McNeil CM, Millar EKA, McLeod D, Morey AL, Crea P,

Murphy NM, Lopez-Knowles E, Oakes SR, Ormandy CJ, Qiu MR, Hamilton A,

Spillane A, Lee CS, Sutherland RL, Musgrove EA, O’Toole SA. Notch1

expression is an early event in breast cancer development and is associated

with the HER-2 molecular subtype. Histopathology. 2010; 56: 286–296.

101

Pathology (August 2005) 37(4), pp. 271–277

REVIEW

Gene expression profiling in breast cancer: towards individualising patient management

N. MURPHY*{,E.MILLAR{§ AND C. S. LEE{d *Sydney Breast Cancer Institute, Sydney Cancer Centre; and {Department of Anatomical Pathology, Royal Prince Alfred Hospital; {Division of Anatomical Pathology, St George Hospital, South Eastern Area Laboratory Services, and §University of New South Wales; and dDepartment of Pathology, University of Sydney, Sydney, New South Wales, Australia

Summary has largely been attributed to the implementation of Breast cancer is a complex and clinically heterogeneous widespread mammography screening and advances in disease. The increase in knowledge of breast cancer biology adjuvant treatment of early-stage disease.2 has led to a number of clinical advances in the treatment of Breast cancer is a complex and heterogeneous disease, breast cancer, most notably the implementation of wide- encompassing a wide range of pathological entities and spread mammography screening and advances in adjuvant molecular profiles. The clinical behaviour and response of treatment of early-stage disease. In the last 20 years, arrays these tumours to treatment are equally disparate.3 of potential prognostic and/or predictive markers of breast Adjuvant hormonal therapy and polychemotherapy cancer have been analysed. However, relatively few have reduces the risk of both recurrence and death from breast proven to be clinically useful. To date, the only widely cancer; however, adjuvant therapy does have associated accepted markers for routine use in breast cancer are the risk. Defining a patient’s risk of developing metastatic oestrogen receptor (ER), progesterone receptor (PR) and disease at diagnosis (prognostic assessment) is crucial for human epidermal growth factor receptor, HER-2 (c-erbB2/ clinicians in deciding which patients should receive neu). Expression microarray technology and laser capture expensive and toxic adjuvant therapy. Following this, the microdissection have now been employed to further our clinician must decide which combination of treatment is the understanding of the molecular pathogenesis of breast most suitable for each individual patient (predictive cancer. Recently reported advances in array technology assessment).4 and RNA amplification methods are having a considerable Axillary lymph node status, tumour size and grade, impact in this field, allowing the analysis of pre-malignant and patient age and lymphovascular invasion are among the pre-invasive lesions. A number of studies have identified prognostic indicators currently employed to predict an prognostic and predictive gene ‘signatures’, whose prediction individual’s risk of metastasis.5 Axillary nodal status is of disease outcome and response to treatment is superior to currently the best prognostic marker available; however, it conventional prognostic indicators. Despite major technolo- is an imperfect indicator. Approximately 25% of node- gical advances, a number of confounding issues remain negative patients harbour micrometastases and are destined concerning the potential clinical utility of gene expression to develop recurrence. Conversely, up to 50% of node- profiling, including differences in study design, patient positive patients will not experience recurrence even with- selection, array technology, chemistry, and methods of out adjuvant treatment and after many years of follow-up.6 analysis. It seems likely, however, that following careful Considering only a minority of node-negative patients will ‘hypothesis driven’ validation studies and clinical trials, develop a recurrence, there is a crucial need to identify expression profiling will be applied in the future to identify accurate prognostic markers that can identify patients with patient-specific disease profiles and provide rationale for extremely low risks of recurrence to avoid over-treatment individualised treatment. This review focuses on the current of these patients. use and future potential of microarray profiling in breast The ability to accurately select which particular adjuvant cancer. regimes are most likely to benefit individual patients remains a dilemma. Despite years of research, the only recommended Key words: Prognostic/predictive markers, gene expression profiling. predictive markers in breast cancer are the estrogen receptor Received 1 March, revised 2 May, accepted 9 May 2005 (ER) and the progesterone receptor (PR) for selecting endocrine-sensitive breast cancers7,8 and the human epidermal growth factor receptor, HER-2 (c-erbB2/neu) for identifying breast cancer patients with metastatic disease INTRODUCTION who may benefit from trastuzumab.9 HER-2 also indi- Breast cancer is a major cause of cancer-related morbidity cates an enhanced sensitivity to high-dose anthracycline- and mortality among females worldwide and remains a based regimens, while high concentrations of HER-2 major global health burden.1 In recent years, mortality correlate with a lower probability of response to hormone from breast cancer in the Western world has declined. This therapy in both early and advanced breast cancer.8,10,11

ISSN 0031-3025 printed/ISSN 1465-3931 # 2005 Royal College of Pathologists of Australasia DOI: 10.1080/00313020500169586 272 MURPHY et al. Pathology (2005), 37(4), August

The use of these predictive markers is less then ideal, amplified and labelled with a fluorescent dye. The labelled however, and can result in the unnecessary application of RNA is then hybridised with the complementary nucleic treatment or the application of ineffective treatment. The acid probes on the chip. ability to accurately select patients who will benefit from Two methods of target labelling for microarray analysis particular adjuvant regimes would be a major advance in exist: single-colour and two-colour. Single-colour labelling the clinical management of breast cancer. is used for platforms such as Affymetrix’s GeneChips During the last 20 years, improved molecular and genetic (Affymetrix, USA) in which one RNA target is hybridised research has provided a greater understanding of the to each chip. In this approach, differentially expressed molecular events underlying normal breast development genes are detected by comparing the intensity of hybridisa- and the pathogenesis of breast cancer.12 Increasing knowl- tion signals between different chips. Two-colour labelling is edge of deregulated processes involved in tumour progres- used for platforms such as Agilent’s cDNA and oligonu- sion, including cell cycle regulation, angiogenesis and cleotide microarrays (Agilent, USA). In this approach, two apoptosis, have facilitated the identification of a number RNA targets are labelled with different dyes, for example of new ‘potential’ prognostic and/or predictive markers in cyanine-3 or cyanine-5. The targets are then combined and breast cancer.13 Despite the evaluation of hundreds of hybridised to the same microarray. After hybridisation and potential markers, relatively few have demonstrated real washing, the microarray is scanned in two wavelengths and clinical application. In recent years, major advances have the resulting data are combined to provide a red/green been made in the area of expression microarray technology. colour image in which differentially expressed genes will be Methods have been developed which allow the simulta- either green or red, and equally expressed genes will show neous examination of thousands of genes. Microarray gene up as yellow. Values for each gene are extracted as a ratio expression profiling is having an increasing impact in all of one colour to the other. Computer algorithms are then areas of breast cancer classification, prognosis and predic- employed to cluster specimens based on similarity of gene tion of response to therapy. Many believe expression expression pattern. microarray analysis might hold the future for patient- specific disease profiling and individualised treatment for breast cancer patients. This review aims to describe the PREDICTION OF PATIENT OUTCOME current use of DNA microarray-based profiling in breast cancer and its future potential in breast cancer prognos- A number of cDNA microarray expression platforms have tication and patient management. been used in recent studies examining their prognostic utility in breast cancers (Table 1). In a pioneering study by Perou et al., DNA microarray analysis was used to classify breast cancers according to EXPRESSION PROFILING IN BREAST CANCER gene expression profiles.14 This study demonstrated that by Individual breast tumours demonstrate multiple mutations employing an intrinsic gene set of 456 cDNAs, breast and epigenetic changes that influence the expression of tumours can be phenotypically classified into four major thousands of cancer driving genes. Considering this, it is subgroups based on their gene expression profiles: a not surprising that the analysis of one gene at a time is luminal epithelial/ER-positive group; a basal epithelial-like unlikely to demonstrate the same prognostic significance as group negative for ER expression; a HER-2-positive group; the simultaneous examination of thousands of genes. and a group exhibiting a ‘normal-like’ expression signa- cDNA microarray technology employs nucleic acid poly- ture.14 It was also demonstrated in a subsequent extension mers/probes immobilised on a solid surface such as a glass of this study15 that patients whose tumours were classified slide called a ‘chip’. Thousands of nucleic acid polymers as luminal epithelial/ER-positive could be divided into at each representing a specific gene can be spotted on a single least two subgroups, each with a distinctive expression chip. RNA is then extracted from the sample of interest, profile referred to as luminal A and B. This study showed

TABLE 1 Prognostic studies

Study Chip platform Specimen number (n) Tissue

Perou et al., 200014 Customised in-house array 40 breast carcinomas Fresh frozen 8102 cDNA clones 3 normal breast (Research Genetics, USA) Sorlie et al., 200115 Customised in-house array 78 breast carcinomas Fresh frozen 8102 cDNA clones 3 fibroadenomas (Research Genetics, USA) 4 normal breast (3/4 pooled normals) van’t Veer et al., 200216 Inkjet deposited Oligo Arrays 117 breast carcinomas Fresh frozen (Agilent Technologies, USA) van de Vijver et al., 200217 Inkjet deposited Oligo Arrays 295 breast carcinomas Fresh frozen (Agilent Technologies, USA) Piccart et al., 200420 Inkjet deposited Oligo Arrays 301 breast carcinomas Fresh frozen (Agilent Technologies, USA) Wang et al., 200521 U133A GeneChips Training set: Fresh frozen (Affymetrix Human, USA) 115 breast carcinomas Validation set: 171 breast carcinomas GENE EXPRESSION PROFILING IN BREAST CANCER 273 that the basal-like and HER-2-positive subtypes had a therapy, thereby preventing over-treatment in a consider- poorer prognosis, while the luminal-cell-like subtype was able number of patients. associated with a higher survival rate. It was also Importantly, comparison of the recent Wang et al., study demonstrated that ER expression was not a significant with previous tumour classification studies by van’t Veer et prognostic marker on its own and that there is a significant al. highlights a number of important and confounding difference in outcome for the two ER-positive groups.15 issues regarding the application of expression microarrays On the other hand, an alternative expression array in breast cancer classification. Both studies identified a platform16 identified a 70-gene ‘prognostic signature’ in a signature which could predict metastasis-free survival in cohort of 78 young breast cancer patients with axillary patients with lymph node-negative breast cancer. The most lymph node negative disease, that was found to comprise striking observation is the lack of overlap of genes in both genes involved in cell cycle regulation, invasion, angiogen- gene signature lists. An overlap of only three genes esis and metastasis development, and was indicative of was identified, namely cyclin E2, origin recognition poor prognosis. This prognostic signature correctly pre- complex (ORC) and the tumour necrosis factor (TNF) dicted disease outcome in 83% of patients, predicting poor superfamily protein. However, there was considerable outcome with a sensitivity of 85% and good outcome with a overlap in common gene pathways involved in tumour 16,21 sensitivity of 81%.17 This 70-gene prognostic signature was progression. validated on 295 patients with node-negative and node- Much of the disparity between gene signatures can be positive breast disease. All patients had Stage 1–2 tumours attributed to fundamental differences in patient selection, and were less than 53 years of age. The gene expression array technology and chemistry, and method of analysis. profile had a strong independent value on multivariate As stated, the Wang et al., study was carried out using the analysis and was more powerful then the St Gallen18 Affymetrix U133A arrays, while van’t Veer et al. used criteria and the National Institute of Health (NIH) Agilent arrays. Differences in microarray technology consensus criteria.19 Last year a multicentre external (including probe type, labelling and hybridisation chemis- validation study of the 70-gene signature was reported by tries) and the selection of genes on each array are likely to Piccart et al., on behalf of the TRANSBIG Consortium contribute to the lack of coherence observed between (TRANSBIG is an abbreviation for ‘translating molecular signature gene sets. Other confounding issues are likely to knowledge into early breast cancer management: building be patient selection criteria. The van’t Veer et al. study , on the Breast International Group network for improved selected patients 53 years of age who had not received treatment tailoring’). Tissue from 301 patients who were adjuvant therapy, whereas the Wang et al. study included followed for at least 10 years was examined. Again the 70- patients who had not received adjuvant therapy but gene profile signature was strongly prognostic, outperform- selection was irrespective of age, menopausal status and ing classic prognostic criteria including the St Gallen tumour size. Methods of data analysis are likely to be the most important confounding factor in expression analysis consensus panel. However, the magnitude of effect was 22 less than that previously reported.20 The European studies. Variation between these studies and others Organisation for Research and Treatment of Cancer indicate that in the future much larger validation studies such as the MINDACT study must be carried out before (EORTC) and the Breast International Group (BIG) are 23 now designing a large, prospective clinical trial to validate any consensus on predictive signatures can be reached. the utility of this assay in patients with newly diagnosed breast cancer, the ‘MINDACT’ (microarray in node negative disease may avoid chemotherapy) study. GENE PROFILES IN RESPONSIVENESS TO A gene expression-based algorithm, which provides TREATMENT: TAILORING TREATMENT FOR quantitative predictions on disease outcome in patients INDIVIDUAL PATIENTS with lymph node-negative breast cancer, together with gene Breast cancer adjuvant treatment can significantly decrease expression profiling using Affymetrix Human U133A risk of disease relapse and death. Currently, estrogen GeneChips, found a 76-gene signature comprising 60 genes receptor (ER) and progesterone receptor (PR) status are from ER-positive patients and 16 genes from ER-negative used to select patients with breast cancer likely to respond 21 patients. This gene signature was subsequently validated to hormone therapy. For over 25 years, the gold standard on a test set of 171 lymph node-negative patients and for the endocrine treatment of all stages of ER-positive showed a sensitivity of 93% and a specificity of 48%. When breast cancer has been tamoxifen. However, in the applied, this algorithm accurately identified patients at high advanced setting, approximately half of patients with ER- risk of distant tumour recurrence within 5 years. This gene positive breast tumours will not respond to endocrine signature could be applied to any lymph node-negative treatment.24 Therefore, additional markers are required patients independently of age, tumour size and grade and that could identify patients who will not respond and select ER status. In particular, this gene signature accurately patients for tailored treatment. predicted risk of recurrence in patients with tumours 10– Gene expression profiling has recently been used to 20 mm in size for whom prediction of risk is difficult. identify markers that could predict response to hormonal Importantly, the 76-gene algorithm recommended systema- and neoadjuvant therapy (Table 2). In one study, gene tic adjuvant therapy to only 52% of low-risk patients, expression profiling was performed on a cohort of compared with 90% and 89% by the St Gallen18 and NIH oestrogen receptor-positive primary breast carcinomas guidelines.19 The ability of this algorithm to predict from patients with advanced disease and clearly defined favourable outcome in patients may provide a valuable types of response to tamoxifen treatment. Response was tool to identify patients who do not require adjuvant assessed based on the effect of treatment on tumour size 274 MURPHY et al. Pathology (2005), 37(4), August

TABLE 2 Predictive studies

Study Chip platform Specimen number (n) Tissue

Chang et al., 200325 HgU95-Av2 GeneChips 24 breast carcinomas Fresh frozen (Affymetrix, USA) (biopsied before neoadjuvant docetaxel treatment) (core biopsy samples) Chang et al., 200526 HgU95-Av2 GeneChips 24 breast carcinomas Fresh frozen (Affymetrix, USA) (biopsied before and after neoadjuvant docetaxel treatment) (core biopsy samples) LCM Ayers et al., 200427 30, 721 cDNA clones Training set: Fine needle aspiration (Millennium Pharmaceuticals, USA) 24 breast carcinomas Snap frozen/RNAlater Validation set: 18 breast carcinomas (samples taken before and after T/FAC neoadjuvant therapy) Jansen et al., 200524 Customized in-house array Training set: Fresh frozen 19,200 cDNA clones 46 breast carcinomas (Research Genetics, USA) Validation set: 66 breast carcinomas

LCM, laser capture microdissection. and time until tumour progression (progression-free gene expression profiling of cellular material collected by survival [PFS]). A training set of 81 differentially expressed fine needle aspiration prior to neoadjuvant chemotherapy genes was identified. These genes were involved in may identify those women least likely to achieve oestrogen action, immune response, apoptosis and extra- complete pathological response to T/FAC neoadjuvant cellular matrix formation. From the 81 genes, a predictive chemotherapy.27 set of 44 genes was then selected and validated. This While these findings may have a significant impact in predictive signature allowed discrimination between breast future selection of breast cancer patients for both cancer patients with progressive disease and objective neoadjuvant chemotherapy and tamoxifen therapy, further response to tamoxifen.24 validation studies will be required in the future before these Docetaxel is one of the most effective drugs available in gene signatures will become sufficiently validated for the treatment of breast cancer; however, nearly half of the patient management. treated patients do not respond and suffer from a range of side effects. Recently, gene profiling of core biopsies of breast cancer has been successfully employed to accurately UNDERSTANDING THE PATHOGENESIS OF 25 predict response to neoadjuvant docetaxel. Differential HEREDITARY BREAST CANCER patterns of expression of 92 genes were shown to correlate Germline mutations in BRCA1 and BRCA2 account for 5– with docetaxel response. Responsive tumours had higher 10% of all sporadic breast cancers. Women carrying germ- expression of genes involved in cell cycle, cytoskeleton, line mutations in one BRCA1 or BRCA2 allele have an adhesion, protein transport, protein modification, tran- extremely high lifetime risk of developing breast cancer. A scription and apoptosis. Resistant tumours showed number of studies have utilised expression microarray increased expression of a number of genes involved in analysis to study inherited breast cancer. In agreement with transcription and signal transduction. The 92 ‘predictor’ accumulating histological and historical evidence, expres- gene set had positive and negative predictive values of 92% sion array analysis revealed that tumours arising from and 83%, respectively. A subsequent study26 examined core mutations in either BRCA1 or BRCA2 are biologically biopsies taken prior to neoadjuvant docetaxel and surgical distinct. BRCA1 breast cancers were typically negative for specimens obtained at 3 months following neoadjuvant ER and PR expression while BRCA2 breast tumours were chemotherapy (four cycles, 100 mg/m2 daily for 3 weeks). likely to express both receptors.28 In addition, sporadic This study again found that the initial core biopsies showed differential expression patterns of the 92-gene set, correlat- breast cancers are similarly biologically distinct from ing with response to docetaxel. Importantly, the expression tumours arising from mutations in either BRCA1 or profiles of the residual tumours after 3 months of docetaxel BRCA2. Even after removal of ER/PR-related genes as a treatment were strikingly similar and independent of initial potential confounding issue from analysis, the gene expression profiles of the two inherited tumour groups sensitivity or resistance, suggesting selection of a resistant 29 subpopulation of cells. Notably, resistance to docetaxel was are discernible. Studies in this area to date have been associated with the expression of genes involved in G2M limited, particularly with respect to specimen number. phase cell cycle arrest and survival pathways involving the Future work in this area will significantly increase our mammalian target of rapamycin. These genes may prove to understanding of underlying pathogenesis of inherited be useful therapeutic targets.26 breast cancer. In a separate neoadjuvant study, expression analysis was employed to develop a multigene predictor gene set of response to paclitaxel, fluorouracil, doxorubicin and UNRAVELLING THE MOLECULAR EVOLUTION cyclophosphamide (T/FAC) neoadjuvant therapy. A 75- OF BREAST CANCER gene predictor set was identified that demonstrated a Traditionally, the development of breast cancer was viewed predictive accuracy of 78%. This study demonstrates that as a multistep process involving progressive changes from GENE EXPRESSION PROFILING IN BREAST CANCER 275 normal to atypical ductal hyperplasia (ADH), carcinoma isolated from the same specimen, suggesting that the three in situ (DCIS), invasive carcinoma (IDC) and eventually distinct stages of breast cancer are highly similar to each culminating in metastatic disease.30 In the last 15 years, other at the transcriptome level.37 However, there were immunohistochemistry and molecular genetics have chan- distinct gene expression differences between high-grade and ged the way the breast cancer multistep model is perceived. low-grade tumours. Expression profiles of Grade 2 Breast cancer progression is no longer viewed as a single tumours were mixed, showing either low-grade or high- pathway but rather a complex series of genetic events grade gene expression changes. These findings corroborate leading to distinct and divergent pathways culminating in previous studies, which suggest that well-differentiated invasive cancer.3 The molecular mechanisms involved in DCIS progresses to well-differentiated IDC, and poorly this process are still not completely defined. Gene expres- differentiated DCIS progresses to poorly differentiated sion analysis is being applied to help unravel the complex IDC.38,39 In addition, expression of a small subset of genes interplay of genes and pathways involved in breast cancer involved in cellular proliferation and cell cycle regulation progression. In the past, expression profiling of precursor correlated with advanced tumour grade and with the lesions was hampered by a number of technical factors transition from DCIS to IDC, suggesting a link between including difficulty in sampling pure populations, lack of tumour stage and tumour grade. When ADH was frozen material and the inability to extract sufficient compared with patient-matched normal epithelium, a quantities of high quality RNA from in vivo samples. significant global change in gene expression was observed. However, the ability of laser-capture microdissection These alterations are maintained in the later stages of DCIS (LCM) techniques to sample pure cell populations and and IDC. Together these data suggest a clonal relationship advances in microarray technologies has enabled research between the distinct pathological entities of breast cancer in this area to advance considerably. Standard microarray and that the expression profile of early stage disease may protocols require 5–10 mg of total RNA as starting material predict the clinical behaviour of later stages of disease. In to allow accurate detection and quantitation of relative the future, expression profiles of other benign proliferative RNA expression levels. In vivo samples, and in particular breast lesions such as adenosis, papillomas and tubular LCM samples, typically generate only nanogram (ng) adenomas may shed further light on the underlying genetics quantities of RNA. As such, it is necessary to amplify of breast cell changes. RNA to obtain sufficient quantities for expression profiling. Until recently, it was generally perceived that additional amplification steps resulted in both a 39 bias CLINICAL IMPACT OF EXPRESSION PROFILING and poor reproducibility for low abundance transcripts. AND FUTURE PERSPECTIVES However, more recently, technical advances in RNA The use of gene expression profiling in breast cancer amplification methods and array technologies have become prognostication and prediction has not yet reached the available. An increasing number of studies demonstrate stage where it can be implemented clinically. An array of that reliable and reproducible gene expression measure- confounding issues remains, including differences in patient ments can be obtained from amplified mRNA.31–34 selection, array technology and chemistry, and methods of One of the first studies in this area compared gene analysis. Validation of new makers must be performed in expression within a cohort of DCIS cases with and without the context of prospective clinical trials in which the necrosis.35 A set of 69 genes was identified that were prognostic or predictive questions can be answered. It is consistently differentially expressed. It was found that hoped that large-scale multicentre trials such as the DCIS with necrosis could be distinguished from DCIS MINDACT study will help to reconcile many of these issues. without necrosis by the pattern of gene expression, and that A major difficulty in gene expression profiling studies has up-regulation of angio-associated migratory cell protein been the necessity for fresh frozen material. In addition, the (AAMP), a gene associated with angiogenesis and tumour current expense and degree of expertise necessary to carry out progression, was also associated with high nuclear grade expression array profiling precludes its use in a routine morphology and necrosis in DCIS. clinical setting. However, recent technical advances promise A more recent study that analysed gene expression to address some of these issues. profiles of DCIS and IDC cases employed laser capture Until recently, gene expression array profiling required microdissection to yield 98–99% pure populations of breast frozen tissue. In the majority of studies, tissue was obtained cancer cells.36 Microarray analysis and subsequent unsu- from small tissue banks where tissue collection took place pervised hierarchical clustering distinguished two distinct without regard to planned patient treatment and other groups mainly in terms of ER status. A set of 325 genes was confounding issues. It was believed that expression profil- found to be differentially expressed in DCIS and IDC and ing of degraded nucleic acids extracted from formalin these genes might play important roles in malignant fixation and paraffin-embedded (FFPE) tissue samples transformation of breast ductal cells. In addition, it was would be impossible. However, Affymetrix in conjunction found that there was a subset of genes whose expression with Arcturus reagents (Arcturus, USA) have recently was altered in IDC but not in DCIS. Furthermore, there released the Gene ChipXP3 that enables extraction and was another set of 34 genes that was differentially expressed amplification of RNA from FFPE samples. This array in tumours from patients with lymph node metastasis as contains probe sets selected from the 300 bases at the most opposed to those without metastases.36 39 prime end of transcripts, thus maximising detection of A study that employed laser capture microdissection to shorter degraded RNAs extracted from FFPE material. sample pure epithelial cell populations from normal breast The ability to carry out accurate and reproducible lobules, ADH, DCIS and IDC found no consistent major molecular analysis of archival tissue combined with long- transcriptional differences between ADH, DCIS and IDC term clinical follow-up could provide breast cancer 276 MURPHY et al. Pathology (2005), 37(4), August researchers with a tremendous and previously inaccessible trial that will be performed by the Breast Intergroup resource. In particular, the ability to analyse FFPE tissue (TBIG) of North America. from completed and ongoing clinical trials will provide an A similar study involved the reduction of a complex unrivalled opportunity to validate previous findings and to microarray signature to a two-gene expression ratio, which ask important ‘hypothesis’ driven questions regarding can be detected in FFPE material using quantitative RT- patient outcome and response to treatment. PCR. The expression ratio of these two genes, namely the Tissue microarrays (TMAs) are composed of hundreds homeobox gene (HOXB13) and the interleukin 17B of tissue specimens from multiple patients on a single receptor (IL17BR), accurately predicts tumour recurrence microscope slide.40 TMAs represent a major advancement in tamoxifen-treated patients with early stage ER-positive in molecular pathology over traditional methods, as they breast cancer.44 Before RT-PCR ‘predictor’ assays such as provide a high throughput method to rapidly examine these can become routine, a number of technical considera- molecular targets in hundreds and potentially thousands of tions must be addressed. Issues regarding the routine use of tissue specimens. TMAs are increasingly being employed in RT-PCR in FFPE blocks, such as fixation and storage of the area of breast cancer research to examine the expression blocks, must be resolved. However, in the future, quanti- of novel prognostic/predictive makers.41 Like conventional tative RT-PCR may provide a robust, high throughput and FFPE material, tissue microarrays are amenable to a wide economical method of gene expression profiling which is range of techniques including histochemical stains, immu- amenable to routine clinical use. nological stains with either chromogenic or fluorescent visualisation, in situ hybridisation (including both mRNA ISH and FISH), and tissue microdissection techniques. The CONCLUSION analysis of archival material from large-scale clinical and research trials using TMAs, combined with long-term The advent of reliable and reproducible expression profil- clinical follow-up, is allowing the rapid validation of ing technology is having an increasing impact on our markers originally identified through gene expression understanding of breast cancer biology. Laser-capture profiling. microdissection techniques combined with reliable RNA Real-time reverse transcriptase polymerase chain reac- amplification methods now enable researchers to accurately tion (RT-PCR) assays have been developed which allow the and reproducibly examine small cell populations such as simultaneous analysis of several hundred genes. those in ADH and DCIS. In addition, the potential to carry Importantly, these assays can examine gene expression out gene expression profiling on FFPE tissues is a using limited amounts of RNA extracted from FFPE significant technological leap forward. Expression profiling sections. RT-PCR is a high throughput method which can of archival material combined with long-term clinical be employed to validate ‘expression signatures’, originally follow-up will be a tremendous research resource. identified through expression microarrays analysis. Using Considering the complexity and clinical heterogeneity of this technique, Paik et al. have developed a 21-gene RT- breast cancer, it seems inevitable that single gene markers PCR assay (Oncotype DX assay; Genomic Health, USA), will be replaced with gene profiling ‘signatures’. However, which can quantify the likelihood of distant recurrence in before expression profiling can be implemented clinically, a tamoxifen-treated patients with node-negative, ER-positive number of issues must be resolved. As discussed, the breast cancer. An initial set of 250 candidate genes was selection of ‘gene signatures’ depends on a constellation of identified from published literature, genomic databases, confounding factors such as microarray technology and and in-house gene profiling experiments performed on chemistry, patient selection criteria and methods of data frozen tissue. The relationship between quantitative expres- analysis. As such, the importance of experimental design sion of these 250 candidate genes and breast cancer cannot be overstressed. Large-scale clinical validation recurrence was then accessed on archival material obtained studies must be carried out before a consensus on from three independent clinical studies on breast cancer ‘predictive gene’ signatures and the future role of expres- including the B20 National Surgical Adjuvant Bowel and sion profiling in clinical breast cancer decision-making can Breast Project (NSABP) trial, in which node-negative be reached. patients with ER-positive tumours were randomly assigned ACKNOWLEDGEMENTS The authors would like to to tamoxifen alone or with CMF chemotherapy. The sincerely thank the Sydney Breast Cancer Foundation for tamoxifen alone cohort was included in the initial evalua- its generous support and Drs Anne Hamilton, Sue tion. From these data a panel of 16 cancer-related genes Pendlebury, Andrew Spillane and Hugh Camalt for and five reference genes were selected to develop a critically reviewing the manuscript. recurrence algorithm which can estimate the odds of recurrence over 10 years from diagnosis. Following this, Address for correspondence: Dr N. Murphy, Department of Anatomical Pathology, Royal Prince Alfred Hospital, Missenden Road, Camperdown, archival tissue samples were examined from the NSABP NSW 2050, Australia. E-mail: [email protected] B14 trial, in which patients with node-negative, ER-positive breast cancer were randomly assigned to tamoxifen versus placebo. Again, it was demonstrated that the 21-gene profile is predictive for benefit from tamoxifen.42 An References additional study utilising chemotherapy-treated patients 1. Jemal A, Thomas A, Murray T, et al. Cancer Statistics. CA from the B20 cohort demonstrated that the 21-gene assay is Cancer J Clin 2002; 52: 23. 43 2. Bray F, McCarron P, Parkin DM. The changing global patterns of a strong predictor of benefit from chemotherapy. These female breast cancer incidence and mortality. Breast Cancer Res 2004; results are the basis of a planned prospective randomised 6: 229–39. GENE EXPRESSION PROFILING IN BREAST CANCER 277

3. Simpson PT, Reis-Filho JS, Gale T, et al. Molecular evolution of 25. Chang JC, Wooten EC, Tsimelzon A, et al. Gene expression profiling breast cancer. J Pathol 2005; 205: 248–54. for the prediction of therapeutic response to docetaxel in patients with 4. van’t Veer LJ, Paik S, Hayes DF. Gene expression profiling of breast breast cancer. Lancet 2003; 362: 362–9. cancer: a new tumor marker. J Clin Oncol 2005; 23: 1631–5. 26. Chang JC, Wooten EC, Tsimelzon A, et al. Patterns of resistance and 5. Primary Therapy of Early Breast Cancer, 9th International Conference. incomplete response to docetaxel by gene expression profiling in breast January 26–29 2005, St Gallen, Switzerland. Breast 2005; 14 (Suppl 1): cancer patients. J Clin Oncol 2005; 23: 1169–77. S1–56. 27. Ayers M, Symmans WF, Stec J, et al. Gene expression profiles predict 6. Early Breast Cancer Trialists’ Collaborative Group (EBCTCG). complete pathologic response to neoadjuvant paclitaxel and fluorour- Tamoxifen for early breast cancer: an overview of the randomized acil, doxorubicin, and cyclophosphamide chemotherapy in breast trials. The Lancet 1998; 351: 1451–67. cancer. J Clin Oncol 2004; 22: 2284–93. 7. Duffy MJ. Predictive markers in breast and other cancers: a review. 28. Verhoog LC, Brekelmans CT, Seynaeve C, et al. Survival in hereditary Clin Chem 2005; 51: 494–503. breast cancer associated with germline mutations of BRCA2. J Clin 8. Diaz LK, Sneige N. Estrogen receptor analysis for breast cancer: Oncol 1999; 11: 3396–402. current issues and keys to increasing testing accuracy. Adv Anat Pathol 29. Hedenfalk I, Duggan D, Chen Y, et al. Gene-expression profiles in 2005; 12: 10–9. hereditary breast cancer. N Engl J Med 2001; 344: 539–48. 9. Vogel CL, Cobleigh MA, Tripathy D, et al. Efficacy and safety of 30. Shackney SE, Silverman JF. Molecular evolutionary patterns in breast trastuzumab as a single agent in first-line treatment of HER2- cancer. Adv Anat Pathol 2003; 10: 278–90. overexpressing metastatic breast cancer. J Clin Oncol 2002; 20: 719–26. 31. King C, Guo N, Frampton GM, et al. Reliability and reproducibility 10. Lewis F, Jackson P, Lane S, et al. Testing for HER2 in breast cancer. of gene expression measurements using amplified RNA from laser- Histopathology 2004; 45: 207–17. microdissected primary breast tissue with oligonucleotide arrays. J Mol 11. Zhou BP, Li Y, Hung MC. HER-2/Neu signaling and therapeutic Diagn 2005; 7: 57–64. approaches in breast cancer. Breast Dis 2002; 15: 13–24. 32. Man YG, Zhang Y, Shen T, et al. cDNA expression profiling reveals 12. Rosen JM. Hormone receptor patterning plays a critical role in normal elevated gene expression in cell clusters overlying focally disrupted lobuloalveolar development and breast cancer progression. Breast Dis myoepithelial cell layers: implications for breast tumour invasion. 2003; 18: 3–9. Breast Cancer Res Treat 2005; 89: 199–208. 33. Schlingemann J, Thuerigen O, Ittrich C, et al. Effective transcriptome 13. Lerwill MF. Current practical applications of diagnostic immunohis- amplification for expression profiling on sense-oriented oligonucleotide tochemistry in breast pathology. Am J Surg Pathol 2004; 28: 1076–91. microarrays. Nucleic Acids Res 2005; 33: 29. 14. Perou CM, Sorlie T, Eisen MB, et al. Molecular portraits of human 34. Singh R, Maganti RJ, Jabba SV, et al. Microarray based comparison breast tumours. Nature 2000; 406: 747–52. of three amplification methods for nanogram amounts of total RNA. 15. Sorlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of Am J Physiol Cell Physiol 2004; 288: 1179–89. breast carcinomas distinguish tumour subclasses with clinical implica- 35. Adeyinka A, Emberley E, Niu Y, et al. Analysis of gene expression in tions. Proc Natl Acad Sci USA 2001; 98: 10869–74. ductal carcinoma in situ of the breast. Clin Cancer Res 2002; 12: 16. van’t Veer LJ, Dai H, van de Vijver MJ, et al. Gene expressing 3788–95. profiling predicts clinical outcome of breast cancer. Nature 2002; 415: 36. Seth A, Kitching R, Landberg G, et al. Gene expression profiling of 530–6. ductal carcinomas in situ and invasive breast tumours. Anticancer Res 17. van de Vijver MJ, He YD, van’t Veer LJ, et al. A gene-expression 2003; 23: 2043–51. signature as a predictor of survival in breast cancer. N Engl J Med 37. Ma XJ, Salunga R, Tuggle JT, et al. Gene expression profiles of human 2002; 25: 1999–2009. breast cancer progression. Proc Natl Acad Sci USA 2003; 100: 5974–9. 18. Goldhirsch A, Wood WC, Gelber RD, et al. Meeting highlights: 38. Porter DA, Krop IE, Nasser S, et al. A SAGE (serial analysis of gene updated international expert consensus on the primary therapy of early expression) view of breast tumor progression. Cancer Res 2001; 61: breast cancer. J Clin Oncol 2003; 21: 3357–65. 5697–702. 19. Eifel P, Axelson JA, Costa J, et al. National Institutes of Health 39. Warnberg F, Nordgren H, Bergkvist L, et al. Tumour markers in Consensus Development Conference Statement: adjuvant therapy for breast carcinoma correlate with grade rather than with invasiveness. breast cancer, November 1–3, 2000. J Natl Cancer Inst 2001; 93: Br J Cancer 2001; 85: 869–74. 979–89. 40. Kononen J, Kallioniemi A, Barlund M, et al. Tissue microarrays for 20. Piccart MJ, Loi S, van’t Veer LJ, et al. Multi-centre external validation high-throughput molecular profiling of tumor specimens. Nat Med study of the Amsterdam 70-gene prognostic signature in node negative 1998; 4: 844–7. untreated breast cancer: are resuts still outperforming the clinical 41. Kumar B, Venter DJ, Armes JE, et al. Tissue microarrays: a practical pathological criteria? Breast Cancer Res Treat 2004; 88: 17 (Abstr 38). guide. Pathology 2004; 37: 295–300. 21. Wang Y, Klijn JG, Zhang Y, et al. Gene-expression profiles to predict 42. Paik S, Shak S, Tang G. A multigene assay to predict recurrence of distant metastasis of lymph-node-negative primary breast cancer. tamoxifen-treated, node-negative breast cancer. N Engl J Med 2004; Lancet 2005; 365: 671–9. 351: 2817–26. 22. Eifel P, Axelson JA, Costa J, et al. Outcome signature genes in breast 43. Paik S, Shak S, Tang G, et al. Expression of 21 genes in the recurrence cancer: is there a unique set? Bioinformatics 2005; 21: 171–8. score assay and prediction of clinical benefit from tamoxifen in NSABP 23. Jenssen TK, Hovig E. Gene-expression profiling in breast cancer. study B-14 and chemotherapy in NSABP B-20. Breast Cancer Res Lancet 2005; 365: 634–5. Treat 2004; 88: 15 (Abstr 24). 24. Jansen MP, Foekens JA, van Staveren IL, et al. Molecular classifica- 44. Ma XJ, Wang Z, Ryan PD, et al. A two-gene expression ratio predicts tion of tamoxifen-resistant breast carcinomas by gene expression clinical outcome in breast cancer patients treated with tamoxifen. profiling. Clin Oncol 2005; 23: 732–40. Cancer Cell 2004; 6: 607–16. Pathology (February 2011) 43(2), pp. 116–127

INVITED REVIEW

Molecular assays in breast cancer pathology

SANDRA A. O’TOOLE*z§jj,CHRISTINA I. SELINGER*, EWAN K. A. MILLARzô**{{, TRINA LUM* AND JANE M. BEITH{z§**

*Department of Tissue Pathology and Diagnostic Oncology, Royal Prince Alfred Hospital, {Department of Medical Oncology, Sydney Cancer Centre, Royal Prince Alfred Hospital, Camperdown, zGarvan Institute of Medical Research, Darlinghurst, §Sydney Medical School, University of Sydney, jjSt Vincent’s Clinical School, ôFaculty of Medicine, University of New South Wales, Randwick, **Department of Anatomical Pathology, South Eastern Area Laboratory Service, St George Hospital, Kogarah, and {{School of Medicine and Health Sciences, University of Western Sydney, Campbelltown, New South Wales, Australia

Summary It has been known for some time that breast cancer is a Recent advances in understanding the molecular pathology heterogeneous disease, first recognised many years ago with the 3 of breast cancer offer significant potential to identify discovery of hormone receptor positive breast cancer and later 4,5 patients who may benefit from adjuvant therapies. To date, HER2 amplified disease. More recently, gene expression few of these advances are utilised in a routine setting. We microarray (GEM) studies have refined our understanding that review molecular assays that are currently in use or are in based on the genetic profile of a breast cancer, the biology and 6 the advanced stages of development, which may be used clinical behaviour varies significantly. Although molecular as predictive or prognostic biomarkers in breast cancer. biology techniques have significant potential to improve the The only widely used breast cancer molecular assay is selection of optimal breast cancer therapy for individuals, only in situ hybridisation (ISH) for human epidermal growth one molecular biomarker, human epidermal growth factor factor receptor 2 (HER2) gene amplification and we highlight receptor 2 (HER2) gene amplification, is in routine use. The key issues with the interpretation of this assay, with particu- aim of this review is to highlight recent developments and lar attention to the difficulties of the equivocal category. controversies in HER2 molecular testing and to discuss the New molecular assays such as ISH for the topoisomerase most promising molecular markers in breast cancer pathology II alpha (TOP2A) gene and for the aberrations in the that are currently in practice or closest to introduction in a copy number of the centromeric region of chromosome routine setting, specifically topoisomerase II alpha (TOP2A), 17 are readily performed in a standard histopathology labora- and two multigene assays: Oncotype DX and Mammaprint. tory, but to date there are insufficient data to support their These assays will also be compared to routine immunohisto- routine use. We also review the current data on two commer- chemical markers for their predictive potential. cially available multigene expression assays, Oncotype DX and MammaPrint and discuss their potential use. CURRENT ISSUES IN TESTING FOR HER2 Overall, while new molecular assays have significant GENE AMPLIFICATION potential to improve patient selection for therapy, well-per- The HER2 gene is located on 17q12–q21 and encodes a 185 formed histopathology with reliable interpretation of stan- kDa protein that is part of the epidermal growth factor family. dard hormone and HER2 assays provides the most The HER2 protein is a transmembrane tyrosine kinase receptor important predictive and prognostic information in early that forms either homodimers or heterodimers with other breast cancer. members of the HER family (EGFR, HER3 and HER4). Activation of HER2 results in activation of the RAS-MAPK Key words: Breast cancer, molecular assays, prognostic biomarkers, in situ pathway stimulating cell proliferation, while interaction with hybridisation, HER2, TOP2A, chromosome 17, Oncotype DX, MammaPrint. the phosphatidylinositol 3’-kinase (PI3K) pathway inhibits cell death (reviewed by Barros et al.7). The net effect of this is Received 7 November, revised 24 November, accepted 24 November 2010 promotion of an aggressive tumour phenotype, reflected in the association of HER2 amplification with larger, higher grade INTRODUCTION tumours and a poor outcome.8 There have been significant improvements in outcome from HER2 gene amplification can be detected using in situ breast cancer over the past two decades1 due to earlier diagnosis hybridisation (ISH) relying on the hybridisation of a comp- and the use of targeted therapies, especially hormonal therapy lementary labelled DNA probe to the HER2 gene as shown in for oestrogen receptor expressing breast cancer.2 Despite these Fig. 1. Reports of the incidence of HER2 gene amplification advances, there are still women with breast cancer who have a vary widely, with earlier studies suggesting as many as 30% of poor outcome and a key research and clinical question is how to breast cancers were HER2 amplified.4,5 More recent studies select the right treatment for the right patient. This requires the suggest that around 15% of newly diagnosed invasive breast use of biomarkers, candidate genes in a breast cancer patient cancers are HER2 positive, although higher grade and node that can predict outcome (prognostic biomarkers) or response to positive tumours which are more likely to receive adjuvant therapy (predictive biomarkers). chemotherapy have a higher incidence of HER2 positivity of

Print ISSN 0031-3025/Online ISSN 1465-3931 # 2011 Royal College of Pathologists of Australasia DOI: 10.1097/PAT.0b013e3283430926

need to accurately identify those patients who may benefit from these targeted therapies. There are a number of modalities to assess HER2 status. One of the first techniques used to assess HER2 status was immu- G T A nohistochemistry (IHC); early trastuzumab metastatic breast C A A GGT þ þ C T C C cancer trials enrolled patients with 2 or 3 expression of Her2 protein. Subsequent retrospective analyses showed that only Her2 patients with 3þ IHC or gene amplification by FISH benefited.18–21 Subsequently, a common approach has been to triage C A T patient eligibility for trastuzumab via IHC; those with no or C G C G weak staining are termed negative and no further testing is 21 þ G performed. Her2 protein positive patients (3 uniform strong Chr 17 C G C membranous staining in >30% of tumour cells) may receive C G trastuzumab in many studies, while patients with equivocal staining on IHC (2þ) are referred for FISH testing. This approach is endorsed by the most recent American Society of Clinical Oncology (ASCO)/College of American Pathol- Fig. 1 Fluorescent in situ hybridisation (FISH). Within the cell the HER2 ogists (CAP) guidelines on HER2 ISH testing, as shown in ﬂuorescent DNA probe (red) hybridises to the HER2 gene and the chromosome Table 1 with representative examples shown in Fig. 2. Briefly, a 17 centromeric enumeration DNA probe (CEP17, green) hybridises with FISH result of more than six HER2 copies per nucleus or a ratio chromosome 17 DNA. of the number of copies of HER2 to the centromeric probe for chromosome 17 (CEP17) of >2.2 is reported as a positive, amplified result (Fig. 2A). Cases with a mean HER2 copy number per nucleus of <4 or a HER2/CEP17 ratio of <1.8 are around 25%.9 Our own figures from Royal Prince Alfred negative, and cases with copy number between 4 and 6 or a ratio Hospital’s recently commenced HER2 ISH testing programme of 1.8–2.2 are considered equivocal (Fig. 2B) and require show around 17% of 325 early invasive breast cancers are further investigation to determine their status. amplified, while St George Hospital has reported a very similar In Australia, Pharmaceutical Benefit Scheme (PBS) funded proportion of HER2 amplified cases of 16.2% in 1708 patients use of trastuzumab requires confirmation of gene amplification tested to date (unpublished data). Farshid et al. reported an via ISH. This testing of all early breast cancers is supported overall mean of 15.42% for newly diagnosed breast cancers through the Roche Australian ISH Testing Program for Breast across the 20 Australian ISH testing centres in 2009.10 Cancer. All breast cancers irrelevant of their IHC status may be HER2 targeted treatments are making an impact in this tested through this program. This approach ensures that only otherwise poor prognosis breast cancer. The first HER2 specific patients with HER2 gene amplification are eligible to receive therapy was a monoclonal antibody, trastuzumab, directed HER2 targeted therapy. Initially FISH was the only modality against the juxtamembrane portion of the extracellular domain used for this, but newer bright field modalities of ISH are now of the HER2 receptor.11 A number of trials suggest that being widely employed. The two most utilised bright field trastuzumab improved the disease free survival (DFS) and techniques are chromogenic in situ hybridisation (CISH) and overall survival (OS) of women with early stage HER2 positive silver in situ hybridisation (SISH).22 While these new tech- breast cancers by as much as 50%.12–14 A recent meta-analysis niques are not specifically addressed in the ASCO guidelines, of randomised control trials of trastuzumab in early breast they state that any new assay should show >95% concordance cancer has confirmed a highly significant reduction in breast with an established assay and several studies have confirmed cancer deaths, recurrence and metastasis (all p < 0.00001).15 the utility and accuracy of these techniques.22–25 Advantages of Trastuzumab also improves survival in metastatic breast can- these techniques are the durability of the signal which does not cer, with a recent study showing a 44% reduction in the risk of fade appreciably with time and the ability to be interpreted with death compared to non-HER2 metastatic breast cancer.16 More a standard light microscope with easier interpretation of tissue recently, a dual tyrosine kinase small molecule inhibitor (with morphology rather than requiring an expensive fluorescence activity against EGFR and HER2) lapatinib is also proving to microscope. SISH is an automated system (Ventana, Roche be an effective therapy in metastatic breast cancer in combi- Diagnostics, USA), while CISH is a two day manual procedure nation with capecitabine,17 with a 51% reduction in the risk of utilising a kit (SPotLight; Invitrogen, USA). Disadvantages of disease progression. Taken together, these data emphasise the these bright field methods, in our experience, are a lower

Table 1 Interpretation of HER2 ISH testing in breast cancer from ASCO/CAP guidelines on HER2 testing21

Result Single probe (e.g., CISH or SISH) Dual probe (e.g., FISH or C/SISH with CEP17 probe)

Negative Mean HER2 copy number <4 signals per tumour cell nucleus HER2/ CEP17 ratio <1.8 Positive Mean HER2 copy number >6 signals per tumour cell nucleus HER2/ CEP17 ratio >2.2 Equivocal Mean HER2 copy number 4–6 signals per tumour cell nucleus HER2/CEP17 ratio 1.8–2.2

ASCO/CAP, American Society of Clinical Oncology/College of American Pathologists; CISH, chromogenic in situ hybridisation; FISH, ﬂuorescence in situ hybridisation; SISH, silver in situ hybridisation.

Fig. 2 (A) Fluorescent in situ hybridisation example of HER2 amplification and (B) an example of equivocal HER2 copy number. (C) A case with clonal amplification (red circle, region of HER2 amplification; blue circle, region of diploid Her2 copy number) by Her2 immunohistochemistry and (D) HER2 chromogenic in situ hybridisation. (E) Chromogenic in situ hybridisation showing topoisomerase II alpha amplification and (F) chromosome 17 ‘polysomy’.

sensitivity compared to FISH and reduced efficacy on sub- might be to select another tumour block if available for testing. optimally fixed tissues or those that have undergone decalci- The ASCO/CAP guidelines suggest that in such ISH equivocal fication (e.g., biopsy of a bony metastatic site). At Royal cases, consideration of the IHC result may help resolve this Prince Alfred and St George Hospitals, New South Wales, dilemma.21 Australia, we have also found that the CISH signal tends to However, Dowsett and colleagues27 report that even in be weaker in archival blocks of an age >5years.Theyare experienced laboratories, borderline FISH cases can be difficult generally used as single probe tests for HER2; those cases that to interpret.27 Twenty breast cancer cases were FISH tested by have <4 (negative) or those with >6 (positive) copies of five large reference laboratories in this concordance study HER2 per nucleus require no further testing. Cases in the which reported HER2/CEP17 ratios in the range of 1.7 (i.e., equivocal range (between 4 and 6 copies of HER2 per negative) to 2.3 (i.e., positive), with an overall discordance rate nucleus) require a second probe applied to a parallel section of 20%. There is a deficiency in the literature regarding the fromthesametissueareaforCEP17,whichenablescalcu- clinical significance of cases that fall in the equivocal range and lation of a HER2/CEP17 ratio. As for FISH, cases with a ratio further studies are required to clarify this issue.21 of >2.2 are positive and <1.8 are negative. Around 2% of The use of a chromosome 17 centromeric probe may also cases fall within the equivocal range (1.8–2.2 copies)26 and contribute to the difficulties involved in assessing equivocal are usually subject to FISH.21 cases. Chromosome 17 polysomy has been reported to occur in There is also some debate about the utility and clinical around 2–9% of breast cancers.26 However, recent studies significance of the equivocal category of HER2/CEP17 1.8– based on comparative genomic hybridisation (CGH) arrays, 2.2.19 Some argue that it is unnecessary and creates diagnostic which assess the copy number of multiple genes along the and therapeutic dilemmas.26 Instead, in cases with a HER2/ entire chromosome suggest that some cases of so called CEP17 ratio in the ‘equivocal’ range, the authors argue that an chromosome 17 ‘polysomy’ are not true increases in the additional 20 nuclei should be scored by the primary scorer number of copies of the whole chromosome 17, but in fact while a second independent scorer counts a minimum of 40 reflect co-amplification of the centromeric region.28 This could nuclei. When these two ratios are in agreement, this result is result in cases where HER2 is truly amplified, but because reported. If there is no agreement, the entire assay should be there is co-amplification of the centromeric region reflected in repeated and the specimen rescored.26 An alternative approach an increased chromosome 17 probe count, the ratio may

incorrectly be reported at <2,29 not reflecting the true amplified While there are well recognised problems with HER2 IHC, status of the tumour.30 the technique is still valuable in assessing HER2 status of breast Bartlett and colleagues conversely argue that cases where cancers. We find it useful as an additional internal quality chromosome 17 is not used may falsely underestimate the control measure. It is reassuring that 3þ IHC cases are in the incidence of HER2 amplification.31 They comment that due large majority (>80%) amplified, and if the ISH signal is weak to nuclear transection in thin cut sections and incomplete in a 3þ IHC, CISH negative case, we will often repeat the assay hybridisation of DNA probes, the observed mean chromosomal with increased pretreatment or go on to perform FISH to ensure copy number in reality falls significantly below the theoretical we are not missing an amplified case due to technical problems. threshold of 2 copies per cell. A previous study in normal breast IHC is also very valuable for detecting heterogeneity of HER2 by this group identified a range for disomy of chromosome 17 amplification. While this phenomenon is not widely recognised of 1.3–1.85 copies per nucleus, arguing that polysomy is better Morey et al. report that this occurs in around 0.4% (33/9035) of defined as >1.85 copies per nucleus.32 One consequence diagnostic HER2 ISH cases.35 This clonality is reflected by the of these calculations is that theoretically a tumour with a immunohistochemistry and is easier to detect as it is readily HER2 copy number of 3 and monosomy for chromosome 17 apparent at low power, in comparison to ISH where the signal is (<1.3 copies) would have a ratio of >2 and be amplified. Using only easily seen at high power magnification, increasing the dual colour FISH for HER2 and chromosome 17, they assessed risk of missing a small amplified clone (an example of a 1711 cases of breast cancer for HER2, including 593 cases with ‘clonal’ case is shown in Fig. 2C,D). Many of their reported 2þ Her2 IHC enriching for this borderline group. The authors cases showed background polysomy and a merging of ampli- found that using a dual probe (HER2 and chromosome 17), fied and non-amplified components. The majority also showed theoretically 16.4% of cases with a HER2 copy number of 3–4 a mixture of amplified and non-amplified DCIS.35 Interest- were amplified for HER2. They also observed that 3.28% of ingly, although forming a small minority of a largely non- cases with an observed HER2 copy number of between 2 and 3 amplified tumour, the nodal metastasis contained amplified were also ‘amplified’. Therefore, the authors argue that chrom- tumour cells. This issue of heterogeneity also reinforces the some 17 signal should be assessed in all breast cancer cases caution that is needed when interpreting HER2 assays (whether with a HER2 copy number of >2 and that the current guidelines IHC or ISH) on core biopsies. result in underdiagnosis of HER2 ‘amplification’. However, A critical issue for HER2 testing, whichever method is whether these technically amplified cases represent true ampli- selected, is the need for strict quality control and quality fication in terms of their biology and response to trastuzumab assurance of HER2 testing, with >95% concordance with was not determined by this study, and it is clear that such another validated test. A recent study from the North Central technically amplified ratios are mostly generated through loss Cancer Treatment Group (NCCTG) Intergroup trial N9831, a of chromosome 17. Further study is required to determine the randomised phase III clinical trial evaluating trastuzumab as outcome and response to treatment of this group of ‘techni- adjuvant therapy for patients with HER2 positive early breast cally’ amplified low HER2 copy number cancers. Recent cancer, has highlighted the need for regulated testing. A advice from the CAP Quality Assurance Program (QAP; preliminary protocol specific review of the first 119 patients ISH-A Participant Summary 2010) suggests caution in classi- showed only 67% of samples classified as HER2 positive by fying any case as HER2 ‘amplified’ if the mean HER2 count is FISH performed by the local laboratory were confirmed as <4 and CEP17 is monosomic (regardless of the ratio). FISH HER2 amplified at the central laboratory. Criteria for the The Australian approach of requiring ISH confirmation of trial were subsequently altered to require central re-testing for HER2 gene amplification has been supported by a recent HER2 and concordance was only 88.1% for FISH and 81.6% review addressing issues raised by the 2007 ASCO/CAP for immunohistochemistry. Interestingly, most of the local- HER2 testing guidelines,26 in which the authors argue strongly central discordant cases were re-tested at a reference labora- for primary FISH testing of breast cancers. Immunohistochem- tory, and there was good concordance between the central and istry for HER2 has a number of well recognised problems; the reference laboratory (95.2%), within the suggested ASCO/CAP assay is significantly affected by tissue fixation, edge and crush guidelines. These data emphasise that HER2 testing is best artefacts which are a particular problem in core biopsies, and performed in relatively high volume laboratories. there is no internal positive control. Although FISH may be While HER2 is best recognised as a predictive marker for affected by fixation, DNA in formalin fixed, paraffin embedded response to trastuzumab therapy, there is accumulating evi- (FFPE) material is relatively stable and there is an endogenous dence that it may also predict response to a number of other internal control in the nucleus of every cell, which should have breast cancer therapies. It has been reported to contribute to up to two copies of the HER2 gene. Furthermore IHC is endocrine therapy resistance,36–39 possibly taxane response,40 subjective, requiring interpretation of intensity of membranous and a number of clinical trials suggest that patients with HER2 expression. In contrast, FISH is semi-quantitative, relying on amplified tumours may derive benefit from anthracyclines,41,42 counting signals within tumour nuclei. The greater reliability also seen in a recent meta-analysis.43 and reproducibility of FISH as a HER2 assay is supported by Anthracyclines such as doxorubicin and epirubicin are data showing much greater concordance in external quality widely used as chemotherapeutic agents in breast cancer, but assurance programs (United Kingdom National External Qual- are also associated with a variety of serious adverse effects, ity Assurance Scheme, UKNEQAS Immunocytochemistry particularly cardiotoxicity, which is probably under-reported journal; http://www.ukmeqasicc.ucl.ac.uk/neqasicc.shtml),33 but is becoming more apparent with longer term survivors and compared to IHC, where up to 20% of HER2 assays performed in older patients.44 While a clear benefit is derived from in routine laboratories are incorrect.21 CAP also published anthracycline chemotherapy in the adjuvant setting,2,45,46 the findings from its proficiency testing program and found that effects overall are quite modest, which when coupled with the 100% of participating laboratories correctly classified higher risk of adverse effects and toxicity highlights the need to unknown samples for HER2 status by FISH.34 accurately identify those patients with the greatest potential

benefit. A number of studies have suggested a link between An interim report of the Breast Cancer International HER2 and anthracycline benefit, (reviewed by Munro et al.47) Research Group (BCIRG) randomised phase III Trial 006 in but this link has no known biological basis.48 In contrast, 4943 patients comparing three chemotherapy regimens [(1) TOP2A, which is closely located to the HER2 gene on chromo- doxorubicin and cyclophosphamide followed by docetaxel some 17 and frequently co-amplified with it, is a direct (AC/ET), with (2) doxorubicin and cyclophosphamide molecular target of anthracycline chemotherapy. followed by docetaxel and trastuzumab (AC/ETH), and (3) docetaxel, carboplatin and trastuzumab (TCH) in HER2 positive early breast cancer patients] also suggested that TOPOISOMERASE II ALPHA (TOP2A) deletion or amplification of TOP2A was indicative of a TOP2A is located on chromosome 17q21-q22 and encodes a poor outcome and predicted a greater benefit of regimens 170 kDa enzyme, topoisomerase II alpha. Located close to and containing anthracycline.66 frequently co-amplified with HER2 gene, TOP2A plays a key Most recently, retrospective analysis of TOP2A gene ampli- role in fundamental nuclear processes including DNA replica- fication by CISH in 391 patients of Trial 9401 from the tion, transcription, recombination, chromosome structure, con- Scandanavian Breast Group of anthracycline-based chemother- densation and segregation.49 The prevalence of TOP2A aberra- apy dose escalation demonstrated that HER2 did not predict tions differs widely in the literature ranging from 9% in specific response to therapy, but found that TOP2A amplifica- unselected breast cancer to 46–90% of HER2 amplified breast tion was associated with a better relapse free survival in patients tumours,50,51 whilst it is only rarely detected in HER2 non- treated with a tailored and dose escalated epirubicin containing amplified cancers.52,53 Our own unpublished data in 69 HER2 regimen (FEC).56 amplified patients (manuscript in preparation) found a high However, as outlined in two recent reviews,40,47 other trials frequency of TOP2A aberrations; TOP2A was amplified in 21 have not identified such a clear-cut role for TOP2A. The Cancer cases (34%) and deleted in nine cases (15%). Other studies have and Leukaemia Group B (CALGB) Trial 8541–150013 retro- reported TOP2A deletion rates in HER2 amplified tumours of spectively evaluated whether TOP2A amplification could pre- 16–43% (reviewed by Glynn et al.54). There are a number of dict benefit from intensive dose cyclophopshamide, doxorubi- likely reasons for this variability, including differences in study cin and fluorouracil in 687 cases of HER2 amplified early populations as well as inconsistent definitions of what consti- breast cancer using a triple FISH probe for TOP2A, HER2 and tutes amplification or deletion. TOP2A may be assessed via chromosome 17, but found no association with outcome.63 FISH55 or bright field techniques such as CISH, with an Another large study using FISH to retrospectively assess example of a TOP2A amplified tumour shown in Fig. 2E.56 HER2 and TOP2A status in 2123 patients with early stage Cut-points for TOP2A gene aberrations have also varied con- breast cancer treated with doxorubicin based adjuvant che- siderably in the literature with studies using gene copy numbers motherapy found no association with outcome for TOP2A, or TOP2A/chromosome 17 ratios. For those who used copy although high level HER2 amplification was a prognostic numbers, amplification ranged from >5 signals per nucleus in marker in anthracycline treated patients.55 more than 50% of cells57 to 6 or more gene copies when Thus, the data on the predictive benefit of TOP2A amplifi- detected in at least 20% of screened malignant cells.58 In cation are conflicting and there are a number of reasons for this, contrast, TOP2A was also considered amplified when the including differing methods of assessment of TOP2A status. In TOP2A/chromosome 17 ratios were 1.5, 2.0 or 2.1 and particular, all of these trials relied on retrospective analysis of deleted when the TOP2A/chromosome 17 ratio was less than TOP2A and HER2 genomic status and were statistically under- 0.67, 0.7, 0.8 or 1.0.50,59–64 Therefore, it is important to powered to reliably assess their capacity as a predictive bio- standardise the methodology, particularly the scoring criteria marker. Furthermore, many utilised pre-trastuzumab regimens, used to define amplification and deletion. This would help to and so the role of anthracyclines in trastuzumab treated patients eliminate inconsistencies in results and make reporting is not yet clear. The role of TOP2A deletion is even more more uniform. unclear, with studies showing conflicting associations with One of the major mechanisms of anthracycline action is via sensitivity or resistance to anthracycline therapy.54,61,65 The inhibition of the TOP2A enzyme,52 by impairing DNA replica- issue is further complicated by a small study (n ¼ 81) showing tion and repair58 via p53 DNA damage sensors and caspase that, dissimilar to HER2, there is no association between mechanisms, thereby promoting apoptosis.44 In view of its TOP2A amplification by FISH and expression of the protein direct interaction with anthracycline chemotherapy, TOP2A by immunohistochemistry.67 This finding is supported by a has been proposed as a likely candidate biomarker for the recent study showing no association between TOP2A deletion beneficial effect of anthracycline therapy and this is supported and loss of protein expression.68 by a number of studies. Knoop et al.65 retrospectively analysed The issues have all contributed to suggest that assessment of 805 tumours for HER2 and TOP2A gene aberrations from the TOP2A gene aberrations is not yet ready for the clinic,69 with a Danish Breast Cancer Cooperative Group trial 89D comparing need to design prospective trials that are adequately powered to a CMF regimen (cyclophosphamide, methotrexate and fluor- address the predictive potential of this gene for anthracycline ouracil) to CEF (cyclophosphamide, epirubicin and fluorour- therapy response with rational and uniform criteria for defining acil). They found that while no predictive value for anthracy- gene aberrations. cline (epirubicin) benefit was seen for HER2 amplification, TOP2A amplification (TOP2A/chromosome 17 ratio of >2.0) or deletion (ratio <0.8) was associated with increased recur- CHROMOSOME 17 ABERRATIONS AS A rence-free survival and overall survival (hazard ratio of 0.57 for MARKER OF ANTHRACYCLINE BENEFIT TOP2A amplification and 0.63 for TOP2A deletion). In con- In view of this uncertainty regarding TOP2A, an intensive trast, patients who had a normal TOP2A genotype had a similar search is underway to identify and validate alternative markers outcome with both regimens. in this region of chromosome 17 that may explain the overall

association of HER2 amplified tumours with greater benefit guidelines recommend chemotherapy be considered for the from anthracycline chemotherapy. There has been speculation majority of women, even including tumours that are hormone that polysomy of chromosome 17 (example shown in Fig. 2F), receptor positive, HER2 negative, if they are larger than 1 cm.72 rather than the specific genes HER2 or TOP2A, may in fact be a However, these prognostic and predictive factors are relatively marker of an unknown gene that predicts anthracycline crude measures and many patients are over-treated or under- response. Chromosome 17 is the second most dense human treated as a result. In the last decade there has been considerable chromosome in terms of genes, containing many genes import- interest in developing assays that may help select patients for ant in cancer such as HER2, TP53, and BRCA1 as well as adjuvant therapies, both endocrine and chemotherapy. With the TOP2A.47 development of new technologies which allow for screening of To date there are relatively few published data regarding the the relative abundance of messenger RNA transcripts in the utility of chromosome 17 in this setting. Reinholz and col- cancer tissue, representing the entire genome, there has been leagues presented in abstract form at the San Antonio Breast much research directed at developing assays to address this key Cancer Symposium in 200770 their study examining whether issue in breast cancer management. chromosome 17 could predict outcome in 1888 patients in the Perou et al.6 used cDNA microarrays representing 8102 HER2 positive N9831 intergroup adjuvant trastuzumab trial. human genes to characterise gene expression patterns in a Chromosome 17 was not associated with tratuzumab response set of 65 surgical specimens of human breast tumours from but the data indicated that patients on standard chemotherapy 42 different individuals. A subset of 456 genes, termed the who did not receive trastuzumab with chromosome 17 polys- ‘intrinsic’ gene subset, consisted of genes with significantly omy benefited more than those with a normal chromosome 17 greater expression variation between different tumours than count. Bartlett and colleagues have identified in the UK paired samples from the same tumour. Using this subset, the National Epirubicin Adjuvant Trial (NEAT) that in 1762 authors were then able to identify different molecular subtypes patients who were assessed for HER2, TOP2A and chromo- of breast cancer: luminal A, luminal B, HER2 enriched, basal- some 17 aberrations using a triple FISH probe, the most like and normal breast-like. These five molecular subtypes have powerful predictor of anthracycline benefit was seen with been confirmed to show distinct differences in behaviour in a CEP17 (the chromosome 17 centromeric enumeration probe) number of independent data sets;73–75 Sorlie et al.74 examined duplication.59 As discussed earlier, these workers argue that a subset of 49 patients with locally advanced breast cancer who counting of signals in thin tissue sections is likely to result in were treated with doxorubicin and found that the recurrence relative under-counting of signals,32 thus they define CEP17 free survival (RFS) and OS differed significantly among the duplication as greater than 1.86 observed signals per cell (in breast cancer subtypes, with the luminal A having the longest contrast to the standard definition of ‘polysomy’ as >3 signals survival times, the basal-like and HER2 positive subtypes the per nucleus30). Although HER2 and TOP2A were predictive of shortest survival times, and the luminal B tumours having an relapse free and overall survival in this cohort, there was no intermediate survival time. Importantly, these gene expression interaction with anthracycline benefit. Interestingly, around subtypes appear stable between primary and subsequent meta- two-thirds of patients with CEP17 duplication were not static lesions occurring years later.76 While gene expression HER2 amplified, suggesting that anthracycline benefit may array studies provide a large amount of useful prognostic and not be confined to HER2 amplified patients as described in predictive data it is clearly not practical or possible to perform some studies.50,51 The authors conclude that assessment of these studies on all patients with breast cancer. Consequently, CEP17 duplication is the most powerful predictor described there is an ongoing search for reliable immunohistochemical to date of anthracycline chemotherapy benefit and suggest that surrogate markers of these subtypes for application to routine validation in a larger meta-analysis would be helpful in leading diagnostic pathology laboratories, particularly to identify basal- to introduction of this predictive biomarker into routine prac- like cancers and the high risk, hormone receptor positive tice. Clearly, further investigation into candidate genes for this luminal B subgroup. Current biomarker panels use a combi- effect in the centromeric region of chromosome 17 is required. nation of ER, PR, HER2, cytokeratins 5 and 6 (CK5/6) and the The observed changes in CEP17 copy number may reflect epidermal growth factor receptor (EGFR), although debate still unbalanced translocations, subchromosomal amplification or exists as to which is the best combination of markers, with deletion or whole chromosomal duplication (which as dis- recent publications (Carey et al. 2006;77 Cheang et al. 2008,78 cussed above is a rare event in breast cancer28). Hugh et al. 2009,79 Livasy et al. 2007,80 Rakha et al. 200981) all proposing different methods of defining the basal-like and luminal B subtypes in particular. MOLECULAR CLASSIFICATION OF BREAST Hugh and colleagues79 discriminate luminal A and B patients CANCER on the basis of Ki-67 expression in tumours (luminal A, Traditionally, pathological determinants of tumour size, lymph hormone receptor positive, HER2 negative and Ki-67 node status, endocrine receptor status, grade, lymphovascular 13%; luminal B, same pattern but Ki-67 >13%). They studied invasion and HER2 status have driven prognostic predications tumours from more than 1300 patients participating in the and, ultimately, adjuvant therapy recommendations for women Breast Cancer International Research Group (BCIRG) 001 trial with early breast cancer. A large meta-analysis of adjuvant comparing FAC (5-fluorouracil, doxorubicin and cyclopho- chemotherapy has shown an improvement of 24% in disease sphamide) to TAC (docetaxel, doxorubicin cyclophospha- free survival (DFS) and 15% in overall survival (OS) in women mide). In this study TAC improved relapse free and overall receiving adjuvant chemotherapy.2 This analysis did not survival compared to FAC among patients with luminal B class, include taxane based regimens which showed an even greater HER2 class and triple negative tumours, but not for tumours benefit, providing up to 30% improvement in both DFS and OS belonging to the luminal A class (receptor positive, HER2 in hormone receptor negative tumours, although their role in negative and Ki-67 13%). The hazard ratio for a relapse hormone receptor positive tumours is still not clear.71 Current among patients treated with the TAC versus FAC regimen was

0.50 for triple negative patients, 0.46 for HER positive and 0.66 had a low recurrence rate of 15% over 5 years, meaning 85% of for patients with luminal B tumours. Thus, the addition of a ER positive patients did not require adjuvant chemotherapy.84 simple proliferation index resulted in a highly effective sep- The problem is how to identify those low risk patients, who can aration of ER-positive patients into two intrinsically different safely be spared chemotherapy and are currently not reliably luminal A and luminal B populations underscoring the import- identified using standard clinicopathological factors. ance of proliferation and suggesting that incorporation of a There is considerable interest in developing assays that may proliferation score into therapy decisions may complement better help select patients for adjuvant therapies, both endocrine histological grade. and cytotoxic chemotherapy. Traditional biomarker research Our own study investigating panels of immunhistochemical/ has relied on assessing the expression of single or small in situ hybridisation surrogate markers for intrinsic molecular numbers of genes at a time mostly using IHC and ISH as subtypes revealed differences in recurrence and breast cancer diagnostic tools (see Fig. 3). Analysis of single genes has been specific death between subtypes.82 We defined five different so far successful for a handful of markers (such as ER, PR, subgroups: luminal A ¼ ERþ and/or PRþ, HER-2; luminal HER2 and more recently Ki-67), with their widespread adop- B ¼ ERþ and/or PRþ, HER-2þ; HER-2 ¼ ER and PR, tion into routine practice. However, cancer often involves HER-2þ; basal-like ¼ ER,PR, HER-2, CK 5/6 þ and/ aberrations in many genes and multiple pathways can be or EGFRþ; unclassified ¼ negative for all five markers using defective. Gene expression profiling using RNA microarrays criteria similar to those recently described by Cheang et al.78 or polymerase chain reaction (PCR) technology is an efficient but using FISH to determine HER-2 status.83 Using these surro- way of taking a snapshot of the gene expression signature of gate panels we found there was a markedly shorter recurrence tumours. With the development of new technologies which time for the more aggressive basal-like, HER2 and unclassified allow for relatively affordable screening of the relative abun- subtypes. Critically, however, while these surrogates were able to dance of messenger RNA transcripts in cancer tissue, repre- provide useful information regarding recurrence, they were not senting the entire genome, there has been much research as powerful predictive markers as standard clinicopathological directed at developing assays to address this key issue in breast variables such as tumour size, lymph node status, lymphatic cancer management. invasion, histological tumour grade and hormone receptor An increasing number of diagnostic tools/tests that make use expression suggesting that well performed histopathological of gene expression signatures are now available to assess examination of breast cancer is still the gold standard for patient risk and survival as well as the benefit of adjuvant providing prognostic and predictive data. therapy (reviewed in Ross et al.85). These tools promise improved identification of patients who will benefit from treatment and those patients who could be spared unnecessary GENE EXPRESSION PROFILING ASSAYS AS treatment. Many large microarray studies have controversially PREDICTIVE BIOMARKERS differed in the relative abundance of top genes involved in The benefit of adjuvant chemotherapy has been demonstrated breast cancer with relatively little overlap between them, in a number of clinical trials, reducing overall risk of recurrence however this is thought to be due to differences in array by up to 25%, however the absolute benefit for individual platforms and the complexities and differing methodologies patients is small (1–5%).84 The NSABP trials B14 and B20 of data analysis. A way to resolve this is to use standardised showed that women with lymph node negative, hormone methods, a feature that commercialised diagnostic testing can receptor positive tumours treated with endocrine therapy alone provide. Two assays in particular have been validated with

Fig. 3 Molecular methods for clinical diagnostics can assess changes in DNA, mRNA and protein levels and include examples such as: HER2 FISH, PathVysion; qPCR, Oncotype DX; microarray, MammaPrint; and HER2 IHC. Each method has speciﬁc tissue requirements and different levels of throughput.

Table 2 Comparison of multi-gene assays: Oncotype DX and MammaPrint

Assay Oncotype DX MammaPrint

Method qPCR Microarray Genes tested 21 70 Material required FFPE or fresh tissue Fresh/frozen tissue Processing limitations May require microdissection May impact routine surgical procedures Current indication Node negative, ERþ Node negative Validated in retrospective studies Yes Yes Prospective clinical trials in progress TAILORx MINDACT

ER, oestrogen receptor; FFPE, formalin ﬁxed, parafﬁn embedded.

clinical trials and are in current clinical use to varying degrees: intermediate recurrence score group was not clear, but currently Oncotype DX and MammaPrint (Table 2). there appears to be no clear cut benefit in this group.90 The Oncotype DX assay has had fairly wide uptake, at least within North America, with over 6000 physicians requesting Oncotype DX the test for over 40 000 patients. There is accumulating evi- Oncotype DX is a diagnostic assay that employs quantitative dence that the test is altering patient management,92 primarily reverse transcriptase PCR (RT-PCR) using FFPE breast cancer in the low recurrence score group, increasing the number of specimens, to measure the expression of a panel of 21 genes patients who receive hormonal therapy only.93 The test is primarily in ER positive, node negative patients. These genes generally ordered for those patients who ultimately prove to comprise groups related to ER (ER, PR, Bcl2 and Scube2), have low (48%) and intermediate (37%) risk scores, with a proliferation (Ki-67, STK15, Survivin, Cyclin B1 and MYBL2), lower number of high risk score patients (15% of assays invasion (Stromelysin3, Cathepsin L2), HER2 (HER2 and performed) who are usually identified using standard clinico- GRB7), the macrophage marker CD68, the anti-apoptosis gene pathological variables and are less likely to be referred for BAG183 and GSTM1, as well as five reference ‘housekeeping’ testing as decisions regarding chemotherapy are usually more genes. They are given relative weighting by a scoring algorithm straightforward. The testing process consists of pathologist- (with the heaviest weighting for ER and proliferation related guided selection of a representative block of tumour. Fifteen genes) to develop a recurrence score. There are three categories 5 mm sections are cut, with recommendations to minimise based on this score originally defined as low risk (<18), inter- contamination (utilising a new section of the microtome blade mediate risk (18–30) and high risk (31). or a new blade between cases, cleaning the water bath between This 21 gene signature was tested prospectively in the cases and wearing clean gloves for cutting and mounting National Surgical Adjuvant Breast and Bowel Trial (NSABP) process). These sections are ultimately sent to Genomic Health B-14, comprising 2644 patients with ER positive, histological in the USA for in-house performance of the assay in which a node negative tumours.86 The randomly allocated groups were report outlining and explaining the recurrence score is pro- tamoxifen only or placebo, with the trial showing that tamox- vided. The current cost to Australian patients for whom there is ifen reduced recurrence over 15 of years follow-up. In a subset no rebate is just under $4000 with turnaround time of around of 668 patients for whom paraffin tissue blocks were available, 2 weeks. the 21 gene signature revealed a 5 year distant recurrence rate of Despite the data showing the prognostic and predictive 22.1% for patients with a high recurrence score, compared to potential of the Oncotype DX assay, there is emerging evidence 2.1% for the low recurrence score, and 30.5% and 6.8% at that routinely, well performed immunohistochemical markers 10 years, respectively. Furthermore, the majority of patients may provide just as much information to aid therapeutic with high or intermediate recurrence scores relapsed within decision making. Preliminary data from the translational arm 5 years, compared to around one-third of recurring patients of the arimidex, tamoxifen, alone or in combination (Trans- with low recurrence scores. In multivariate analysis of distant ATAC) trial94 presented at the 2009 San Antonio Breast Cancer recurrence, the recurrence score was independent of age and Symposium compared the prognostic power of Oncotype DX tumour size. Further analysis revealed it performed better than recurrence score, with a formula utilising four standard immu- Adjuvant! (www.adjuvantonline.com) which uses standard nohistochemical markers (‘IHC4’: combined ER, PR, Ki67, clinicopathological variables in predicting recurrence.87 HER2). Quantitative IHC scores were obtained for ER, PR and Further studies confirmed the utility of this 21 gene assay, Ki-67 and HER2 in 1125 women on the TransATAC trial with now called OncotypeDX, including a retrospective case control Oncotype DX results and for whom FFPE sections were also study88 which identified its role as a predictive biomarker for available. The IHC4 score showed reasonable correlation with hormonal therapy in the NSABP B14 trial as well as for the recurrence score (Pearson coefficient 0.7) and provided a chemotherapy in NSABP trial B-20.89 The trial in which the similar amount of prognostic information as the recurrence 21 gene assay was performed in a subset of 651 patients, [227 score. These results suggest that four standard IHC assays who received tamoxifen only and 424 who also received performed in a high quality laboratory can provide similar chemotherapy, either methotrexate and fluorouracil (MF) or prognostic information for endocrine treated ER positive breast cyclophosphamide, methotrexate and fluorouracil (CMF)] cancer patients as the OncotypeDX recurrence score. Measure- showed a large benefit of chemotherapy for patients with a ment of ER and PR has been performed by IHC rather than by high recurrence score, and minimal benefit for those with a low ligand binding assay (LBA) since the early 1990s. However, score.90 A benefit for node positive patients with a high there is a well recognised problem with reliability and repro- recurrence score has also been shown.91 The data for the ducibility of testing. There can be a large discordance in

measurement of these key receptors between laboratories, with MammaPrint the Royal College of Pathologists of Australasia Quality The Netherlands Cancer Institute (NKI) developed the ‘70 Assurance Program (RCPA-QAP) reporting ER positive rates gene’ signature using gene expression microarrays on tissue ranging from 26 to 100% of breast cancers in a multi-laboratory from a retrospective series of node negative breast cancer audit.95 It is possible that since a significant component of the patients who were under 55 years of age, with tumours smaller 21 gene signature relies on expression of ER associated genes than 5 cm and who were treated with loco-regional therapies that any accurate measure of ER status may provide data just as only.98 This initial study found 213 genes that could identify useful to aid in therapeutic clinical decision making. patients with a high risk of developing distant metastases. This Despite yielding potentially informative assessments of risk signature which was subsequently refined to 70 genes was in patients considered indeterminate by routine clinical vari- validated on a set of 295 patients where it was able to ables, OncotypeDX still returns 40–66% of cases as intermedi- distinguish patients at high and low risk of distant metastasis ate risk96 with no clear data to suggest a benefit of chemother- based on 10 year survival figures. This 70 gene signature was apy. A large clinical trial called TAILORx (Trial Assigning developed into an FDA approved diagnostic test named Individualised Options for Treatment Rx) conducted by the MammaPrint using the Agilent microarray platform, and is North American Breast Cancer Intergroup is currently under- recommended for node negative patients under 61 years of age, way to prospectively determine whether ER positive, node with stage I or II disease with tumour size <5 cm. It can also be negative patients with an intermediate score may benefit from used for patients with 1–3 positive nodes, although currently chemotherapy, and aims to recruit at least 10 000 patients. The only outside the USA as it has not been FDA approved yet for cut-offs for each category have been altered from their initial this indication. This assay requires either fresh frozen or tissue descriptions (see above), to low (<11), intermediate (11–25) collected at room temperature into a RNA preserving solution and high (>25) risk categories to minimise potential for under- and is currently not suitable for FFPE tissue. MammaPrint uses treatment in the high risk and intermediate group. Patients with the 70 gene signature to discriminate patients with high or low a low recurrence score receive hormonal therapy only, while risk of recurrence and encompasses genes associated with high risk patients receive standard chemotherapy. The inter- proliferation, metastases, stromal invasion, and angiogenesis. mediate recurrence score group are randomised to receive MammaPrint does not directly assess ER, PR or HER2 mRNA, either hormonal therapy or hormonal therapy and chemother- although a modified assay TargetPrint does. The MammaPrint apy. The outcome of this trial will not be known until at least assay dichotomises patients into low or high risk groups, with 2013; until then, although the assay has significant benefit in no intermediate group, in comparison to the Oncotype DX identifying low risk ER positive node negative patients who can assay which generates a continuous score and unlike the be spared chemotherapy, it offers little benefit for intermediate Oncotype DX assay, ER negative patients can be assessed. risk patients, who often also have equivocal clinicopathological Compared to the St Gallen and NIH consensus criteria, the 70 features. gene signature is equally as effective at predicting patients who There are preliminary data to suggest that addition of stan- would benefit from adjuvant treatment98 and is able to identify dard clinicopathological variables to the Oncotype DX recur- patients with a higher risk of developing distant metastases than rence score can help reduce the number of cases that fall into by traditional methods. However, with 70–80% of breast the intermediate risk group. At the 2010 ASCO meeting, Tang cancer patients receiving unnecessary treatment (EBCTCG et al.97 examined both pathological and clinical factors such as 98a, EBCTCG 98b), the greatest value of MammaPrint is in tumour size, grade, and patient age, in combination with the its ability to identify patients who could be spared unnecessary recurrence score, to assess whether the recurrence score may adjuvant therapy in the ‘low risk’ group who show a greater achieve more prognostic power. All patients in the NSABP trial than 90% chance of being disease free for a minimum of 5 B-14 and the ATAC study with ER positive tumour specimens years. At the moment, the MammaPrint assay is largely a and a successful Oncotype DX recurrence score assay were prognostic, rather than predictive assay, although a large pro- included. The meta-analysis included 647 B-14 patients and spective trial to assess its predictive capability is underway 1088 ATAC patients; B-14 patients were node negative and called the MINDACT trial (microarray in node negative disease were treated with tamoxifen while the ATAC patients were may avoid chemotherapy). Node negative patients, and more node positive or node negative and were treated with tamoxifen recently some lymph node positive patients, are eligible for the or anastrozole. Meta-analysis assessed the risk of distant recur- trial whether ER positive or negative. Patients classified as high rence combining the individual study multivariate risk assess- risk using standard clinicopathological factors as assessed by ments using recurrence score and pathologic and clinical Adjuvant!Online and via MammaPrint receive chemotherapy, (RSPC) information. RSPC prognosis combining clinical and while patients identified as low risk by both methods receive pathology information with recurrence score was significantly hormonal therapy as appropriate. However, any discord more powerful than using recurrence score alone. Furthermore, between standard criteria and the MammaPrint assays results compared with the Oncotype DX recurrence score alone, fewer in randomisation to receive either adjuvant chemotherapy or patients were classified as intermediate risk using the RSPC hormonal therapy as clinically appropriate. index (18% versus 26%; p ¼ 0.001), and 72% of patients with A validation cohort was analysed using MammaPrint, which intermediate recurrence score 18–30 were pushed into either included both node negative and node positive patients and also high or low risk categories. The RSPC index combining patients who received systemic adjuvant treatment. The 70 recurrence score with pathology and clinical information with gene signature was found to be the strongest predictor for recurrence score supplied more powerful prognosis for early distant metastasis free survival, independent of adjuvant treat- stage breast cancer patients than recurrence score alone and it ment, tumour size, lymph node status, histological grade and was estimated that its use would reduce the number of patients age.99 In addition, the prognosis signature significantly with intermediate risk by 30% and enhance individualised improved identification of patients at high risk and low risk, treatment decisions. reducing potential clinical under-treatment or over-treatment of

these patients. Another independent validation cohort using 307 References node negative breast cancer patients who did not receive 1. Peto R, Boreham J, Clarke M, Davies C, Beral V. UK and USA breast cancer deaths down 25% in year 2000 at ages 20–69 years. Lancet 2000; systemic adjuvant treatment also confirmed significant benefit . 100 355: 1822 of the prognostic categories identified by MammaPrint. 2. Early Breast Cancer Trialists’ Collaborative Group (EBCTCG). Effects of Although these assays seem to show superior performance to chemotherapy and hormonal therapy for early breast cancer on recurrence aid clinical decision making than standard clinicopathological and 15-year survival: an overview of the randomised trials. Lancet 2005; 365: 1687–717. variables in particular groups of patients, there is significant 3. Jensen E, DeSombre E, Jungblut P. Estrogen receptors in hormone criticism about the overall utility of ‘gene signatures’, especi- responsive tissues and tumours. Chicago: University of Chicago Press; ally in regards to the selection of genes within them. There are 1969. 4. Slamon DJ, Clark GM, Wong SG, Levin WJ, Ullrich A, McGuire WL. very few genes in common between the many gene signatures Human breast cancer: correlation of relapse and survival with amplifica- published; for example, there is only a single gene in common tion of the HER-2/neu oncogene. Science 1987; 235: 177–82. between the 21 genes of the Oncotype DX assay and the 70 5. Slamon DJ. Proto-oncogenes and human cancers. N Engl J Med 1987; genes of the MammaPrint assay (SCUBE2 which is an oestro- 317: 955–7. 6. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, et al. gen regulated gene). A major contributor to this problem is the Molecular portraits of human breast tumours. Nature 2000; 406: 747–52. varying different composition of patients within the discovery 7. Barros FF, Powe DG, Ellis IO, Green AR. Understanding the HER family patient cohorts.101 Michiels and colleagues repeatedly (500 in breast cancer: interaction with ligands, dimerization and treatments. Histopathology 2010; 56: 560–72. times) generated signatures of the top 50 prognostic genes from 8. Slamon DJ, Godolphin W, Jones LA, et al. Studies of the HER-2/neu the Van’t Veer dataset (on which the MammaPrint assay is proto-oncogene in human breast and ovarian cancer. Science 1989; 244: based) and found that by manipulating patient selection, 707–12. et al entirely different signatures were generated. After 500 repeats, 9. Bartlett JM, Ellis IO, Dowsett M, . Human epidermal growth factor receptor 2 status correlates with lymph node involvement in patients with only 20% of genes in the original published 70 gene signature estrogen receptor (ER) negative, but with grade in those with ER-positive were seen in more than half of the new signatures generated by early-stage breast cancer suitable for cytotoxic chemotherapy. J Clin changes in patient selection and an additional 10 genes were Oncol 2007; 25: 4423–30. 10. Farshid G, Armes JE, Bell R, et al. Establishment of the Australian in situ frequently identified that were not present in the 70 gene hybridization program for the assessment of HER2 amplification in breast 102 signature. These data suggest that these signatures are not cancer: A model for the introduction of new biomarkers into clinical stable and will likely vary significantly within different groups practice. Diagn Mol Pathol 2010; 19: 187–93. 11. Hudis CA. Trastuzumab–mechanism of action and use in clinical practice. of patients. This is reflected in the studies of Fan and col- . 103 N Engl J Med 2007; 357: 39–51 leagues who compared the predictions of five published 12. Romond EH, Perez EA, Bryant J, et al. Trastuzumab plus adjuvant breast cancer gene signatures on the same dataset. In regards to chemotherapy for operable HER2-positive breast cancer. N Engl J Med the 21 (OncotypeDX) and 70 gene (MammaPrint) signatures, 2005; 353: 1673–84. 13. Piccart-Gebhart MJ, Procter M, Leyland-Jones B, et al. Trastuzumab after outcome prediction agreement was only 80% and the analysis adjuvant chemotherapy in HER2-positive breast cancer. N Engl J Med also revealed that 50% of patients with an intermediate risk 2005; 353: 1659–72. score by the Oncotype DX assay were classified as high risk by 14. Joensuu H, Kellokumpu-Lehtinen PL, Bono P, et al. Adjuvant docetaxel or 102 vinorelbine with or without trastuzumab for breast cancer. N Engl J Med the 70 gene MammaPrint signature. 2006; 354: 809–20. 15. Viani GA, Afonso SL, Stefano EJ, De Fendi LI, Soares FV. Adjuvant trastuzumab in the treatment of her-2-positive early breast cancer: a meta- CONCLUSIONS analysis of published randomized trials. BMC Cancer 2007; 7: 153. The recent advances in molecular biology are yet to have a 16. Dawood S, Broglio K, Buzdar AU, Hortobagyi GN, Giordano SH. Prognosis of women with metastatic breast cancer by HER2 status and significant impact on the routine diagnosis and management of trastuzumab treatment: an institutional-based review. J Clin Oncol 2010; breast cancer, with the exception of HER2. This review has 28: 92–8. described a number of promising tests that may have a role to 17. Geyer CE, Forster J, Lindquist D, et al. Lapatinib plus capecitabine for play in specific subsets of breast cancer (summarised in Fig. 3). HER2-positive advanced breast cancer. N Engl J Med 2006; 355: 2733– 43. While it seems that the ISH assays for TOP2A or chromosome 18. Cobleigh MA, Vogel CL, Tripathy D, et al. Multinational study of the 17 may be readily adapted to a routine pathology laboratory efficacy and safety of humanized anti-HER2 monoclonal antibody in setting, there are currently insufficient clinical data to support women who have HER2-overexpressing metastatic breast cancer that has progressed after chemotherapy for metastatic disease. J Clin Oncol 1999; their use. Current information on the multigene assays such as 17: 2639–48. Oncotype DX or MammaPrint in our opinion also does not yet 19. Mass RD, Press MF, Anderson S, et al. Evaluation of clinical outcomes support their routine use, at least until the outcome of the according to HER2 detection by fluorescence in situ hybridization in TAILORx and MINDACT trials are known in view of the cost women with metastatic breast cancer treated with trastuzumab. Clin Breast Cancer 2005; 6: 240–6. and limitations of the assays. Finally, our review of the 20. Vogel CL, Cobleigh MA, Tripathy D, et al. Efficacy and safety of literature still highlights the importance of well performed trastuzumab as a single agent in first-line treatment of HER2-overexpres- routine histopathology and accurate assessment of hormone sing metastatic breast cancer. J Clin Oncol 2002; 20: 719–26. 21. Wolff AC, Hammond ME, Schwartz JN, et al. American Society of receptor and HER2 assays in leading to optimal patient Clinical Oncology/College of American Pathologists guideline recom- outcomes. mendations for human epidermal growth factor receptor 2 testing in breast cancer. J Clin Oncol 2007; 25: 118–45. 22. Penault-Llorca F, Bilous M, Dowsett M, et al. Emerging technologies for Acknowledgements: The authors wish to thank the Sydney assessing HER2 amplification. Am J Clin Pathol 2009; 132: 539–48. Breast Cancer Foundation for their support. Sandra O’Toole 23. van de Vijver M, Bilous M, Hanna W, et al. Chromogenic in situ hybridisation for the assessment of HER2 status in breast cancer: an also receives support from the NHMRC and the Cancer international validation ring study. Breast Cancer Res 2007; 9: R68. Institute NSW. 24. Bilous M, Morey A, Armes J, Cummings M, Francis G. Chromogenic in situ hybridisation testing for HER2 gene amplification in breast cancer produces highly reproducible results concordant with fluorescence in situ Address for correspondence: Dr S. O’Toole, Department of Tissue Pathology hybridisation and immunohistochemistry. Pathology 2006; 38: 120–4. and Diagnostic Oncology, Royal Prince Alfred Hospital, Missenden Rd, 25. Papouchado BG, Myles J, Lloyd RV, et al. Silver in situ hybridization Camperdown, NSW 2050, Australia. E-mail: [email protected] (SISH) for determination of HER2 gene status in breast carcinoma:

comparison with FISH and assessment of interobserver reproducibility. 49. Nielsen KV, Ejlertsen B, Moller S, et al. The value of TOP2A gene copy Am J Surg Pathol 2010; 34: 767–76. number variation as a biomarker in breast cancer: Update of DBCG trial 26. Sauter G, Lee J, Bartlett JM, Slamon DJ, Press MF. Guidelines for human 89D. Acta Oncol 2008; 47: 725–34. epidermal growth factor receptor 2 testing: biologic and methodologic 50. Konecny GE, Pauletti G, Untch M, et al. Association between HER2, considerations. J Clin Oncol 2009; 27: 1323–33. TOP2A, and response to anthracycline-based preoperative chemotherapy 27. Dowsett M, Hanna WM, Kockx M, et al. Standardization of HER2 testing: in high-risk primary breast cancer. Breast Cancer Res Treat 2010; 120: results of an international proficiency-testing ring study. Mod Pathol 2007; 481–9. 20: 584–91. 51. Bouchalova K, Cizkova M, Cwiertka K, Trojanec R, Hajduch M. Triple 28. Marchio C, Lambros MB, Gugliotta P, et al. Does chromosome 17 negative breast cancer–current status and prospective targeted treatment centromere copy number predict polysomy in breast cancer? A fluores- based on HER1 (EGFR), TOP2A and C-MYC gene assessment. Biomed cence in situ hybridization and microarray-based CGH analysis. J Pathol Pap Med Fac Univ Palacky Olomouc Czech Repub 2009; 153: 13–7. 2009; 219: 16–24. 52. Oakman C, Moretti E, Galardi F, Santarpia L, Di Leo A. The role of 29. Troxell ML, Bangs CD, Lawce HJ, et al. Evaluation of Her-2/neu status in topoisomerase IIalpha and HER-2 in predicting sensitivity to anthracy- carcinomas with amplified chromosome 17 centromere locus. Am J Clin clines in breast cancer patients. Cancer Treat Rev 2009; 35: 662–7. Pathol 2006; 126: 709–16. 53. Jarvinen TA, Tanner M, Barlund M, Borg A, Isola J. Characterization of 30. Viale G. Be precise! The need to consider the mechanisms for CEP17 copy topoisomerase II alpha gene amplification and deletion in breast cancer. number changes in breast cancer. J Pathol 2009; 219: 1–2. Genes Chromosomes Cancer 1999; 26: 142–50. 31. Bartlett JM, Campbell FM, Mallon EA. Determination of HER2 ampli- 54. Glynn RW, Miller N, Whelan MC, Kerin MJ. Topoisomerase 2 alpha and fication by in situ hybridization: when should chromosome 17 also be the case for individualized breast cancer therapy. Ann Surg Oncol 2010; determined? Am J Clin Pathol 2008; 130: 920–6. 17: 1392–7. 32. Watters AD, Going JJ, Cooke TG, Bartlett JM. Chromosome 17 aneusomy 55. Tubbs R, Barlow WE, Budd GT, et al. Outcome of patients with early- is associated with poor prognostic factors in invasive breast carcinoma. stage breast cancer treated with doxorubicin-based adjuvant chemother- Breast Cancer Res Treat 2003; 77: 109–14. apy as a function of HER2 and TOP2A status. J Clin Oncol 2009; 27: 33. Bartlett JM, Campbell FM, Ibrahim M, et al. A UK NEQAS ICC and ISH 3881–6. multicentre study using the Kreatech Poseidon HER2 FISH probe: inter- 56. Tanner M, Isola J, Wiklund T, et al. Topoisomerase IIalpha gene ampli- site variation can be rigorously controlled using FISH. Histopathology fication predicts favorable treatment response to tailored and dose-esca- 2010; 56: 297–304. lated anthracycline-based adjuvant chemotherapy in HER-2/neu- 34. Pathologists CMaCCCoA. Clinical laboratory assays for HER-2/neu amplified breast cancer: Scandinavian Breast Group Trial 9401. J Clin amplification and overexpression: Quality assurance, standardisation, Oncol 2006; 24: 2428–36. and proficiency testing. Arch Pathol Lab Med 2002; 126: 803–8. 57. Arriola E, Rodriguez-Pinilla SM, Lambros MB, et al. Topoisomerase II 35. Morey AL, Buckland M, O’Toole S, et al. Clonal heterogeneity of HER alpha amplification may predict benefit from adjuvant anthracyclines in amplification: Nodal metastasis from a minority clone. RCPA Updates HER2 positive early breast cancer. Breast Cancer Res Treat 2007; 106: Sydney, 2009. 181–9. 36. Hurtado A, Holmes KA, Geistlinger TR, et al. Regulation of ERBB2 by 58. Villman K, Sjostrom J, Heikkila R, et al. TOP2A and HER2 gene oestrogen receptor-PAX2 determines response to tamoxifen. Nature 2008; amplification as predictors of response to anthracycline treatment in 456: 663–6. breast cancer. Acta Oncol 2006; 45: 590–6. 37. Knoop AS, Bentzen SM, Nielsen MM, Rasmussen BB, Rose C. Value of 59. Bartlett JM, Munro AF, Dunn JA, et al. Predictive markers of anthracy- epidermal growth factor receptor, HER2, p53, and steroid receptors in cline benefit: a prospectively planned analysis of the UK National predicting the efficacy of tamoxifen in high-risk postmenopausal breast Epirubicin Adjuvant Trial (NEAT/BR9601). Lancet Oncol 2010; 11: cancer patients. J Clin Oncol 2001; 19: 3376–84. 266–74. 38. Dowsett M, Houghton J, Iden C, et al. Benefit from adjuvant tamoxifen 60. Schindlbeck C, Mayr D, Olivier C, et al. Topoisomerase IIalpha expres- therapy in primary breast cancer patients according oestrogen receptor, sion rather than gene amplification predicts responsiveness of adjuvant progesterone receptor, EGF receptor and HER2 status. Ann Oncol 2006; anthracycline-based chemotherapy in women with primary breast cancer. 17: 818–26. J Cancer Res Clin Oncol 2010; 136: 1029–37. 39. Ryden L, Jirstrom K, Bendahl PO, et al. Tumor-specific expression of 61. O’Malley FP, Chia S, Tu D, et al. Topoisomerase II alpha and respon- vascular endothelial growth factor receptor 2 but not vascular endothelial siveness of breast cancer to adjuvant chemotherapy. J Natl Cancer Inst growth factor or human epidermal growth factor receptor 2 is associated 2009; 101: 644–50. with impaired response to adjuvant tamoxifen in premenopausal breast 62. Kawachi K, Sasaki T, Murakami A, et al. The topoisomerase II alpha gene cancer. J Clin Oncol 2005; 23: 4695–704. status in primary breast cancer is a predictive marker of the response to 40. Pritchard KI, Messersmith H, Elavathil L, Trudeau M, O’Malley F, Dhesy- anthracycline-based neoadjuvant chemotherapy. Pathol Res Pract 2010; Thind B. HER-2 and topoisomerase II as predictors of response to 206: 156–62. chemotherapy. J Clin Oncol 2008; 26: 736–44. 63. Harris LN, Broadwater G, Abu-Khalaf M, et al. Topoisomerase II{alpha} 41. Pritchard KI, Shepherd LE, O’Malley FP, et al. HER2 and responsiveness amplification does not predict benefit from dose-intense cyclophospha- of breast cancer to adjuvant chemotherapy. N Engl J Med 2006; 354: mide, doxorubicin, and fluorouracil therapy in HER2-amplified early 2103–11. breast cancer: results of CALGB 8541/150013. J Clin Oncol 2009; 27: 42. Paik S, Bryant J, Park C, et al. erbB-2 and response to doxorubicin in 3430–6. patients with axillary lymph node-positive, hormone receptor-negative 64. Hagen AI, Bofin AM, Ytterhus B, et al. Amplification of TOP2A and breast cancer. J Natl Cancer Inst 1998; 90: 1361–70. HER-2 genes in breast cancers occurring in patients harbouring BRCA1 43. Dhesy-Thind B, Pritchard KI, Messersmith H, O’Malley F, Elavathil L, germline mutations. Acta Oncol 2007; 46: 199–203. Trudeau M. HER2/neu in systemic therapy for women with breast 65. Knoop AS, Knudsen H, Balslev E, et al. retrospective analysis of cancer: a systematic review. Breast Cancer Res Treat 2008; 109: topoisomerase IIa amplifications and deletions as predictive markers in 209–29. primary breast cancer patients randomly assigned to cyclophosphamide, 44. Minotti G, Menna P, Salvatorelli E, Cairo G, Gianni L. Anthracyclines: methotrexate, and fluorouracil or cyclophosphamide, epirubicin, and molecular advances and pharmacologic developments in antitumor activ- fluorouracil: Danish Breast Cancer Cooperative Group. J Clin Oncol ity and cardiotoxicity. Pharmacol Rev 2004; 56: 185–229. 2005; 23: 7483–90. 45. Levine MN, Pritchard KI, Bramwell VH, Shepherd LE, Tu D, Paul N. 66. Slamon D, Eiermann W, Robert N, et al. Phase III trial comparing AC-T Randomized trial comparing cyclophosphamide, epirubicin, and fluor- with AC-TH and with TCH in the adjuvant treatment of HER2 positive ouracil with cyclophosphamide, methotrexate, and fluorouracil in pre- early breast cancer patients: second interim efficacy analysis. San Antonio menopausal women with node-positive breast cancer: update of National Breast Cancer Symposium, Dec 14–17, 2006, Abstract 52. Cancer Institute of Canada Clinical Trials Group Trial MA5. J Clin Oncol 67. Mueller RE, Parkes RK, Andrulis I, O’Malley FP. Amplification of the 2005; 23: 5166–70. TOP2A gene does not predict high levels of topoisomerase II alpha protein 46. Poole CJ, Earl HM, Hiller L, et al. Epirubicin and cyclophosphamide, in human breast tumor samples. Genes Chromosomes Cancer 2004; 39: methotrexate, and fluorouracil as adjuvant therapy for early breast cancer. 288–97. N Engl J Med 2006; 355: 1851–62. 68. Usha L, Tabesh B, Morrison LE, Rao RD, Jacobson K, Zhu A, et al. 47. Munro AF, Cameron DA, Bartlett JM. Targeting anthracyclines in early Topoisomerase II alpha gene copy loss has adverse prognostic significance breast cancer: new candidate predictive biomarkers emerge. Oncogene in ERBB2-amplified breast cancer: a retrospective study of paraffin- 2010; 29: 5231–40. embedded tumor specimens and medical charts. J Hematol Oncol 48. Pegram MD, Finn RS, Arzoo K, Beryt M, Pietras RJ, Slamon DJ. The 2008; 1: 12. effect of HER-2/neu overexpression on chemotherapeutic drug sensi- 69. Pritchard KI. Are HER2 and TOP2A useful as prognostic or predictive tivity in human breast and ovarian cancer cells. Oncogene 1997; 15: 537– biomarkers for anthracycline-based adjuvant chemotherapy for breast 47. cancer? J Clin Oncol 2009; 27: 3875–6.

70. Reinholz MM, Jenkins RB, Hillman D, et al. The clinical significance of 88. Habel LA, Shak S, Jacobs MK, et al. A population-based study of tumor polysomy 17 in the HER2þ N9831 intergroup adjuvant trastuzumab trial. gene expression and risk of breast cancer death among lymph node- San Antonio Breast Cancer Symposium, Dec 13–16, 2007, Abstract 36. negative patients. Breast Cancer Res 2006; 8: R25. 71. Martin M, Mackey J, Vogel C. Benefit from adjuvant taxanes and 89. Fisher B, Dignam J, Wolmark N, et al. Tamoxifen and chemotherapy for endocrine responsiveness in breast cancer. Breast 2007; 16 (Suppl 2): lymph node-negative, estrogen receptor-positive breast cancer. J Natl S127–31. Cancer Inst 1997; 89: 1673–82. 72. Carlson RW, Brown E, Burstein HJ, et al. NCCN Task Force Report: 90. Paik S, Tang G, Shak S, et al. Gene expression and benefit of chemother- Adjuvant therapy for breast cancer. J Natl Compr Canc Netw 2006; 4 apy in women with node-negative, estrogen receptor-positive breast (Suppl 1): S1–26. cancer. J Clin Oncol 2006; 24: 3726–34. 73. Yu K, Lee CH, Tan PH, Tan P. Conservation of breast cancer molecular 91. Albain KS, Barlow WE, Shak S, et al. Prognostic and predictive value of subtypes and transcriptional patterns of tumor progression across distinct the 21-gene recurrence score assay in postmenopausal women with node- ethnic populations. Clin Cancer Res 2004; 10: 5508–17. positive, oestrogen-receptor-positive breast cancer on chemotherapy: a 74. Sorlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast retrospective analysis of a randomised trial. Lancet Oncol 2010; 11: 55– carcinomas distinguish tumor subclasses with clinical implications. Proc 65. Natl Acad Sci USA 2001; 98: 10869–74. 92. Asad J, Jacobson AF, Estabrook A, et al. Does oncotype DX recurrence 75. Sorlie T, Tibshirani R, Parker J, et al. Repeated observation of breast score affect the management of patients with early-stage breast cancer? tumor subtypes in independent gene expression data sets. Proc Natl Acad Am J Surg 2008; 196: 527–9. Sci USA 2003; 100: 8418–23. 93. Lyman GH, Cosler LE, Kuderer NM, Hornberger J. Impact of a 21-gene 76. Weigelt B, Hu Z, He X, et al. Molecular portraits and 70-gene prognosis RT-PCR assay on treatment decisions in early-stage breast cancer: an signature are preserved throughout the metastatic process of breast cancer. economic analysis based on prognostic and predictive validation studies. Cancer Res 2005; 65: 9155–8. Cancer 2007; 109: 1011–8. 77. Carey LA, Perou CM, Livasy CA, et al. Race, breast cancer subtypes, and 94. Cuzick J, Dowsett M, Wale C, et al. Prognostic value of a combined ER, survival in the Carolina Breast Cancer Study. JAMA 2006; 295: 2492–502. PgR, Ki67, HER2 immunohistochemical (IHC4) score and comparison 78. Cheang MC, Voduc D, Bajdik C, et al. Basal-like breast cancer defined by with the GHI recurrence score – results from TransATAC. San Antonio five biomarkers has superior prognostic value than triple-negative phe- Breast Cancer Symposium, Dec 10–13, 2009, Abstract 74. notype. Clin Cancer Res 2008; 14: 1368–76. 95. Francis GD, Dimech M, Giles L, Hopkins A. Frequency and reliability of 79. Hugh J, Hanson J, Cheang MC, et al. Breast cancer subtypes and response oestrogen receptor, progesterone receptor and HER2 in breast carcinoma to docetaxel in node-positive breast cancer: use of an immunohistochem- determined by immunohistochemistry in Australasia: results of the RCPA ical definition in the BCIRG 001 trial. J Clin Oncol 2009; 27: 1168–76. Quality Assurance Program. J Clin Pathol 2007; 60: 1277–83. 80. Livasy CA, Perou CM, Karaca G, et al. Identification of a basal-like 96. Kelly CM, Krishnamurthy S, Bianchini G, et al. Utility of oncotype DX subtype of breast ductal carcinoma in situ. Hum Pathol 2007; 38: 197– risk estimates in clinically intermediate risk hormone receptor-positive, 204. HER2-normal, grade II, lymph node-negative breast cancers. Cancer 81. Rakha EA, Elsheikh SE, Aleskandarany MA, et al. Triple-negative breast 2010; 116: 5161–7. cancer: distinguishing between basal and nonbasal subtypes. Clin Cancer 97. Tang G, Cuzick J, Wale C, et al. Recurrence risk of node-negative and ER- Res 2009; 15: 2302–10. positive early-stage breast cancer patients by combining recurrence score, 82. Millar EK, Graham PH, O’Toole SA, et al. Prediction of local recurrence, pathologic, and clinical information: A meta-analysis approach. J Clin distant metastases, and death after breast-conserving therapy in early- Oncol 2010; 28 (Suppl): abstr 509. stage invasive breast cancer using a five-biomarker panel. J Clin Oncol 98. van’t Veer LJ, Dai H, van de Vijver MJ, et al.Geneexpression 2009; 27: 4701–8. profiling predicts clinical outcome of breast cancer. Nature 2002; 83. Millar EK, Anderson LR, McNeil CM, et al. BAG-1 predicts patient 415: 530–6. outcome and tamoxifen responsiveness in ER-positive invasive ductal 99. van de Vijver MJ, He YD, van’t Veer LJ, et al. A gene-expression carcinoma of the breast. Br J Cancer 2009; 100: 123–33. signature as a predictor of survival in breast cancer. N Engl J Med 84. Fisher B, Jeong J, Dignam J, et al. Findings from recent National Surgical 2002; 347: 1999–2009. Adjuvant Breast and Bowel Project adjuvant studies in stage I breast 100. Buyse M, Loi S, van’t Veer L, et al. Validation and clinical utility of a cancer. J Natl Cancer Inst Monogr 2001; 30: 62–6. 70-gene prognostic signature for women with node-negative breast cancer. 85. Ross JS, Hatzis C, Symmans WF, Pusztai L, Hortobagyi GN. Commer- J Natl Cancer Inst 2006; 98: 1183–92. cialized multigene predictors of clinical outcome for breast cancer. 101. Michiels S, Koscielny S, Hill C. Prediction of cancer outcome with Oncologist 2008; 13: 477–93. microarrays: a multiple random validation strategy. Lancet 2005; 365: 86. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, et al. A multigene 488–92. assay to predict recurrence of tamoxifen-treated, node-negative breast 102. Koscielny S. Critical review of microarray-based prognostic tests and cancer. N Engl J Med 2004; 351: 2817–26. trials in breast cancer. Curr Opin Obstet Gynecol 2008; 20: 47–50. 87. Bryant J. Oncotype DX correlates more closely with prognosis than 103. Fan C, Oh DS, Wessels L, et al. Concordance among gene- Adjuvant Online. 9th International Conference on Primary Therapy of expression-based predictors for breast cancer. N Engl J Med 2006; Early Breast Cancer, St Gallen, Switzerland, Jan 26-29, 2005. 355: 560–9.

JOURNAL OF CLINICAL ONCOLOGY ORIGINAL REPORT

Prediction of Local Recurrence, Distant Metastases, and Death After Breast-Conserving Therapy in Early-Stage Invasive Breast Cancer Using a Five-Biomarker Panel Ewan K.A. Millar, Peter H. Graham, Sandra A. O’Toole, Catriona M. McNeil, Lois Browne, Adrienne L. Morey, Sarah Eggleton, Julia Beretov, Constantine Theocharous, Anne Capp, Elias Nasser, John H. Kearsley, Geoff Delaney, George Papadatos, Christopher Fox, and Robert L. Sutherland

From the Cancer Research Program, Garvan Institute of Medical Research, ABSTRACT and Department of Pathology (SydPath), St Vincent’s Hospital, Darlin- Purpose ghurst; Department of Anatomical To determine the clinical utility of intrinsic molecular phenotype after breast-conserving therapy Pathology, South Eastern Area Labora- (BCT) with lumpectomy and whole-breast irradiation with or without a cavity boost. tory Service, and Department of Radia- tion Oncology, Cancer Care Centre, St Patients and Methods George Hospital, Kogarah; Department Four hundred ninety-eight patients with invasive breast cancer were enrolled into a randomized of Medical Oncology, University of trial of BCT with or without a tumor bed radiation boost. Tumors were classified by intrinsic Sydney, Westmead Hospital, West- molecular phenotype as luminal A or B, HER-2, basal-like, or unclassified using a five-biomarker mead; Department of Radiation Oncol- panel: estrogen receptor, progesterone receptor, HER-2, CK5/6, and epidermal growth factor ogy, Liverpool Hospital, Liverpool; receptor. Kaplan-Meier and Cox proportional hazards methodology were used to ascertain Department of Radiation Oncology, relationships to ipsilateral breast tumor recurrence (IBTR), locoregional recurrence (LRR), distant Wollongong Hospital, Wollongong; St Vincent’s Clinical School, University of disease-free survival (DDFS), and death from breast cancer. New South Wales, and University of Results Western Sydney, Sydney; and Depart- Median follow-up was 84 months. Three hundred ninety-four patients were classified as luminal ment of Anatomical Pathology, Royal Prince Alfred Hospital, Camperdown, A, 23 were luminal B, 52 were basal, 13 were HER-2, and 16 were unclassified. There were 24 Australia. IBTR (4.8%), 35 LRR (7%), 47 distant metastases (9.4%), and 37 breast cancer deaths (7.4%). The overall 5-year disease-free rates for the whole cohort were: IBTR 97.4%, LRR 95.6%, DDFS Submitted December 23, 2008; accepted April 22, 2009; published 92.9%, and breast cancer–specific death 96.3%. A significant difference was observed for survival online ahead of print at www.jco.org on between subtypes for LRR (P ϭ .012), DDFS (P ϭ .0035), and breast cancer–specific death August 31, 2009. (P ϭ .0482), but not for IBTR (P ϭ .346). Supported by grants from the National Conclusion Health and Medical Research Council of The 5-year and 10-year survival rates varied according to molecular subtype. Although this Australia, the Cancer Institute New approach provides additional information to predict time to IBTR, LRR, DDFS, and death from South Wales, the Petre Foundation, and the R.T. Hall Trust. breast cancer, its predictive power is less than that of traditional pathologic indices. This information may be useful in discussing outcomes and planning management with patients Authors’ disclosures of potential con- after BCT. flicts of interest and author contributions are found at the end of this article. J Clin Oncol 27:4701-4708. © 2009 by American Society of Clinical Oncology

Corresponding author: Ewan K.A. Millar, MD, Cancer Research Program, Garvan the whole breast with 40 to 50 Gy with or without a Institute of Medical Research, 384 INTRODUCTION Victoria St, Darlinghurst, New South cavity boost of 16 Gy. The importance of optimal Wales 2010, Australia; e-mail: e.millar@ Breast conserving therapy (BCT) is the preferred local control is highlighted by meta-analysis which garvan.org.au. option in the management of early breast cancer shows that overall mortality is reduced with im- The Acknowledgment is included in (EBC). Surgical excision to clear margins combined proved local control, which equates to one fewer the full-text version of this article, with local radiotherapy can successfully lower the death for every four local recurrences prevented available online at www.jco.org. 2 It is not included in the PDF version rate of local recurrence to approximately 5% at 5 after 5 years. IBTR has a significant impact on (via Adobe® Reader®). years.1 However, prediction of outcome for indi- overall survival with 5-year survival rate of ap- 1 © 2009 by American Society of Clinical vidual patients is uncertain and the development proximately 60%. Oncology of new biomarkers to guide clinical decision mak- Gene expression profiling has identified breast 0732-183X/09/2728-4701/$20.00 ing is needed. Specifically, no established biomark- cancer subtypes with five main gene expression pro- DOI: 10.1200/JCO.2008.21.7075 ers that predict ipsilateral breast tumor recurrence files,3,4 which divides patients into groups with dis- (IBTR) after BCT have been validated. Current ra- tinct tumor phenotypes and outcomes.5 Recent diotherapy regimens typically include treatment of immunohistochemical validation of these intrinsic

© 2009 by American Society of Clinical Oncology 4701 Information downloaded from jco.ascopubs.org and provided by University of New South Wales on October 6, 2009 from 129.94.224.120. Copyright © 2009 by the American Society of Clinical Oncology. All rights reserved. Millar et al molecular phenotypes have suggested that a five-marker panel of estrogen receptor (ER), progesterone receptor (PR), HER-2, CK5/6, Table 1. Patient Baseline Characteristics, Treatments, and Outcomes and epidermal growth factor receptor (EGFR) can predict distant Patients failure and death.6,7 A recent study of conservatively treated breast Characteristic No. % Median Range cancer confirmed the value of this approach in predicting IBTR, using Length of follow-up, months 498 84 1-134 a simplified panel of ER, PR, and HER-2.8 Further refining this classi- Age, years 61 24-84 fication and translating these features into a useful panel for routine Tumor size, mm 16 1-60 pathology is a priority. However, it is now believed that the triple- T1a (1-5) 4 0.8 negative phenotype (TNP; ie, ERϪ,PRϪ, and HER2Ϫ) is a heteroge- T1b (6-10) 77 16.3 T1c (11-20) 270 54.2 neous group comprising basal-like, normal-like breast cancer, and T2 (21-50) 136 27.3 other unclassified breast cancers and does not equate with a basal-like T3 (Ͼ 50) 1 0.2 9 phenotype as originally suggested. Positive expression of basal mark- Tumor grade ers are needed to better define this group of tumors, as evidenced by a 1 167 33.5 recent study showing that the poor prognosis of TNP tumors was 2 185 37.1 conferred by those tumors expressing basal markers CK5/6 and/or 3 145 29.1 EGFR with a specificity of 100% and a sensitivity of 76%.7 Lymph node metastases 146 29.3 N0 339 69.9 Among women with EBC, management decisions regarding lo- N1 (1-3) 128 25.7 cal therapy are generally made without regard to breast cancer subtype N2 (4-10) 17 3.5 and more refined data regarding risk of local and distant failure would N3 (Ͼ 10) 2 0.4 allow for better patient-specific tailoring of therapy. However, identi- LN unsampled 12 2.4 fying risk factors for failure in BCT is problematic as current local ERϩ 393 78.9 ϩ recurrence rates in early-stage breast cancer are low and therefore large PR 334 68.3 HER-2 amplified (FISH) 36 7.2 numbers of patients are required for sufficient statistical power to Subtype detect a significant difference. Luminal A 394 79.1 In this study, we compared the clinical utility of intrinsic molec- Luminal B 23 4.6 ular phenotype as assessed by the five-biomarker panel, ER, PR, Basal-like 52 10.4 HER-2, CK 5/6, and EGFR with traditional pathologic indices, in HER-2 13 2.6 predicting local or distant failure and death in conservatively Unclassified 16 3.2 treated EBC. Triple negative 68 13.6 Treatment and outcome Marginϩ 17 3.4 PATIENTS AND METHODS Cavity boost positive 247 49.5 Cavity boost negative 251 50.5 Endocrine therapy 223 44.7 Patient Selection Chemotherapy 117 23.4 All patients were enrolled into a randomized clinical trial that compared Endocrine and chemotherapy 48 9.6 the benefit of the addition of a local cavity boost of radiotherapy to BCT. The Patients with IBTR 24 4.8 complete study cohort included 688 women with breast cancer, 629 of whom Patients with LRR 35 7 had invasive carcinoma, with an additional 59 patients with ductal carcinoma Patients with distant metastases 47 9.4 in situ (Clinical Trials Registry NCT00138814). Formalin-fixed paraffin- Breast cancer–specific deaths 37 7.4 embedded tissue blocks were available for 498 patients with invasive carci- 5-year survival noma. The clinicopathologic characteristics of this cohort are summarized in IBTR free 97.4 Table 1. HER-2 status was unknown at the time of treatment. Seventy-three LRR free 95.6 percent of patients were postmenopausal, 20% were premenopausal, and 7% DDFS 92.9 were perimenopausal. This study was approved by the Human Research Ethics Breast cancer–specific 96.3 Committee of St George Hospital, Sydney, Australia (ref No: 96/84). Abbreviations: ER, estrogen receptor; PR, progesterone receptor; FISH, Treatment fluorescent in situ hybridization; IBTR, ipsilateral breast tumor recurrence; All patients with invasive carcinoma received local excision and axil- LRR, locoregional recurrence; DDFS, distant disease free survival. lary sentinel node biopsy or axillary clearance. Adjuvant chemotherapy (adriamycin and cyclophosphamide, or cyclophosphamide, methotrexate, and 5-fluorouracil) was given to 23.7% of patients and 44.9% received adjuvant tamoxifen. No patients received adjuvant trastuzumab. Patients were randomly assigned to whole-breast radiotherapy of 50 Gy in 25 fractions or of the first event (see End Points) or to the last known confirmed date of whole-breast radiotherapy of 45 Gy in 25 fractions plus a tumor bed boost of breast cancer disease-free status. Median follow-up was 84 months (range, 16 Gy in 8 fractions. Supraclavicular fields were not added unless there were 1 to 134 months). four or more nodes positive. Seventeen patients had positive margins (clear- End Points ance ϭ 0 mm), 65 had clearance of smaller than 1 mm, and an additional 86 The primary end point was time to ipsilateral breast tumor recurrence had smaller than 2 mm clearance, the remainder being well clear. (IBTR). This included any ipsilateral in-breast recurrence (invasive or nonin- Follow-Up vasive). The secondary end points were locoregional recurrence (LRR, which Patients were assessed 6 weeks after radiation therapy, every 6 months included patients with IBTR and regional recurrences in the axilla, chest wall, for 2 years, then annually thereafter with annual breast imaging. Follow-up internal mammary, or supraclavicular fossa lymph nodes) and time to distant time was calculated from the date of the first surgical procedure to the date metastases and death.

4702 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY Information downloaded from jco.ascopubs.org and provided by University of New South Wales on October 6, 2009 from 129.94.224.120. Copyright © 2009 by the American Society of Clinical Oncology. All rights reserved. Predicting Failure in Early Breast Cancer

Table 2. Patient Tumor Characteristics and Event Rates Classiﬁed by Intrinsic Molecular Phenotype Whole Cohort Luminal A Luminal B Basal HER-2 Unclassiﬁed Patient Tumor Characteristics No. % No. % No. % No. % No. % No. % No. of patients 498 394 79.1 23 4.6 52 10.4 13 2.6 16 3.2 Size Ͻ 20 mm 357 70.3 289 73.4 17 73.9 31 59.6 7 53.8 7 43.7 LVIϩ 79 15.8 62 15.7 4 17.4 9 17.3 2 15.3 2 12.5 LNϩ 146 29.0 117 29.6 5 21.7 13 25 8 61.5 4 25 Grade 3 145 29.1 65 16.5 16 69.5 47 90.3 9 69.2 8 50 EICϩ 45 9.0 29 7.4 5 21.7 6 11.5 3 23.1 1 6.3 Median age, years 61 62 57 54 53 50 Events Median follow-up 84 83.5 71 85 83 72.5 IBTR 24 4.8 15 3.8 2 8.7 5 9.6 1 7.6 1 6.3 LRR 35 7 20 5.1 2 8.7 9 17.3 2 15.4 2 12.5 DDFS 47 9.4 30 7.6 2 8.7 8 15.3 2 15.4 5 31 Breast cancer death 37 7.4 23 5.8 2 8.7 7 13.5 2 15.4 3 18.8

Abbreviations: LVI, lymphatic/vascular invasion; LNϩ, lymph node positive; EIC, extensive intraduct carcinoma; IBTR, ipsilateral breast tumor recurrence; LRR, locoregional recurrence; DDFS, distant disease–free survival.

Tissue Microarray Construction Immunohistochemistry and rates at 5 and 10 years in Table 4. At a median follow-up of 84 months Fluorescent In Situ Hybridization IBTR was observed in 24 patients. The 5-year recurrence-free rate was Immunohistochemistry for ER, PR, CK 5/6, and EGFR was per- 97.4% for the whole cohort, 98.8% for luminal A, 95.5% for luminal B, formed on tissue microarrays (TMAs), assessed by one breast pathologist (E.K.A.M.) blinded to clinical outcome. Tumors were considered HER-2 90% for basal, 92.3% for HER-2, and 92.9% for unclassified. Consis- 11 positive if amplified on fluorescent in situ hybridization (FISH) using a tent with the overall randomized 6-year analysis, in this biomarker HER-2: chromosome 17 ratio higher than 2.2. study cohort, no reduction in IBTR was observed in patients treated Classification of Intrinsic Molecular Phenotype with a radiotherapy boost either in the whole cohort (P ϭ .214 or Patients were categorized based on the status of their primary tumor as between subtypes). The luminal A phenotype was associated with a previously described10: luminal A (ERϩ and/or PRϩ and HER2–), luminal B lower rate of IBTR, compared to all other groups at 5 and 10 years ϩ ϩ ϩ Ϫ Ϫ ϩ (ER and/or PR and HER-2 ), HER-2 (ER and PR and HER-2 ), and (Table 4; hazard ratio [HR], 0.433; 95% CI, 0.186 to 1.005; P ϭ .051; basal (ERϪ and PRϪ, HER-2Ϫ,CK5/6ϩ, and/or EGFRϩ), unclassified Table 5). Kaplan-Meier analysis comparing survival of all five molec- (negative for all five markers). The TNP was assigned on the basis of ERϪ, ϭ PRϪ, HER-2Ϫ. ular subtypes was not statistically significant (P .346; Fig 1A). The median times to event (Table 3) was significantly shorter for basal, Statistical Analysis Kaplan-Meier analyses for IBTR, LRR, distant disease-free survival unclassified, and HER-2 compared to luminal A and B. The only (DDFS), and breast cancer–specific death were estimated for each subtype and variables that predicted IBTR in univariate and multivariate analysis compared using the log-rank test. Crude rates of survival by subtype for each were grade 3 (HR, 3.372; 95% CI, 1.488 to 7.642; P ϭ .004) and end point were also calculated at 5 and 10 years. We used Cox proportional positive margins (HR, 5.838; 95% CI, 1.690 to 20.172; Pϭ.005, Tables hazards univariate analysis to analyze the association between prognostic 5 and 6). Of the 24 patients with IBTR, eight recurrences were located variables and molecular subtype with IBTR, LRR, metastases, and breast cancer–specific death. Those variables significant in univariate analysis in the same quandrant as the primary tumor (four luminal A, three were used in multivariate analysis to construct models identifying variables basal, one HER-2) and 16 were elsewhere (11 luminal A, two luminal which were independently prognostic and not the result of confounding fac- B, two basal, one unclassified). All 17 patients with positive margins tors. Subsequently step-wise removal of redundant variables was employed were luminal A. until resolution. Further analyses characterized how IBTR influenced DDFS and mortality using Kaplan-Meier analysis where survival times were reported using the times from the IBTR event until distant disease or death. Patients who developed distant metastases within 3 months of IBTR were excluded from the analysis. Five-year results were reported for these end points. All analyses were performed using Statview 5.0 (Abacus Systems, Berkeley, CA). Table 3. Median Time to Event, in Months, According to Molecular Subtype Competing risks proportional hazards models (Fine and Gray) were con- Whole structed using ACCoRD software (http://boffinsoftware.com). Event Cohort Luminal A Luminal B Basal HER2 Unclassified IBTR 60 80.5 78 20 23 30 RESULTS LRR 49 72 78 26 25 28.5 DDFS 33 44 59 23 25 27 Breast cancer IBTR death 61 66 94.5 23 33 18

The clinicopathologic characteristics, number of events, crude Abbreviations: IBTR, ipsilateral breast tumor recurrence; LRR, locoregional rates, and median follow-up within each subtype of invasive cancer are recurrence; DDFS, distant disease–free survival. summarized in Table 2, median times to event in Table 3, and crude www.jco.org © 2009 by American Society of Clinical Oncology 4703 Information downloaded from jco.ascopubs.org and provided by University of New South Wales on October 6, 2009 from 129.94.224.120. Copyright © 2009 by the American Society of Clinical Oncology. All rights reserved. Millar et al

Table 4. 5- and 10-Year Event Rates According to Molecular Phenotype IBTR LRR DDFS Breast Cancer–Speciﬁc Death 5 Year 10 Year 5 Year 10 Year 5 Year 10 Year 5 Year 10 Year Parameter No. % No. % No. % No. % No. % No. % No. % No. % No. of patients 24 35 47 37 Whole cohort (n ϭ 498) 12/498 2.4 23/498 4.6 21/498 4.2 35/498 6.8 34/498 6.8 47/498 9.4 18/498 3.6 37/498 7.4 12/24 50 23/24 95.8 21/35 60 35/35 100 34/47 72.3 47/47 100 18/37 48.6 37/37 100 Luminal A (n ϭ 394) 4/394 1 14/394 3.6 8/394 2 19/394 4.8 19/394 4.8 30/394 7.6 7/394 1.8 23/394 5.8 4/15 26.6 14/15 93.3 8/20 40 19/20 95 19/30 63.3 30/30 100 7/23 30.4 23/23 100 Luminal B (n ϭ 23) 1/23 4.3 2/23 8.7 1/23 4.3 2/23 8.6 2/23 8.6 2/23 8.6 1/23 4.3 2/23 8.6 1/2 50 2/2 100 1/2 50 2/2 100 2/2 100 2/2 100 1/2 50 2/2 100 Basal (n ϭ 52) 5/52 9.6 5/52 9.6 8/52 14.8 9/52 17.3 7/52 13.5 8/52 14.8 6/52 11.5 7/52 13.5 5/5 100 5/5 100 8/9 88.8 9/9 100 7/8 87.5 8/8 100 6/7 85.7 7/7 100 HER-2 (n ϭ 13) 1/13 7.7 1/13 7.7 2/13 15.3 2/13 15.3 2/13 15.3 2/13 15.3 2/13 15.3 2/13 15.3 1/1 100 1/1 100 2/2 100 2/2 100 2/2 100 2/2 100 2/2 100 2/2 100 Unclassiﬁed (n ϭ 16) 1/16 6.3 1/16 6.3 2/16 12.6 2/16 12.6 4/16 25 5/16 31.3 2/16 12.6 3/16 18.8 1/1 100 1/1 100 2/2 100 2/2 100 4/5 80 5/5 100 2/3 66.7 3/3 100

NOTE. Rates are presented relating to the whole cohort and molecular subtype and as the relative rate of all events for each subtype. Abbreviations: IBTR, ipsilateral breast tumor recurrence; LRR, locoregional recurrence; DDFS, distant disease–free survival.

LRR than half that of basal (13.5%) and less than one third of HER-2 Thirty-five patients developed LRR with a 5-year disease-free (15.4%) and unclassified (18.8%), with a statistically significant differ- rate of 95.6% for the whole cohort, luminal A of 99%, basal of ence in survival between subtypes (P ϭ .048; Fig 1D). PR (HR, 0.369; 93.7%, HER-2 of 84.6%, and unclassified of 92.9% with a signifi- 95% CI, 0.189 to 0.722; P ϭ .004), size larger than 20 mm (HR, 2.178; cant difference in survival between subtypes (Fig 1B). Crude recur- 95% CI, 1.050 to 4.521; P ϭ .037), lymph node involvement (HR, rence rates of luminal A were less than one third of those of basal 3.984; 95% CI, 1.850 to 8.851; P ϭ .001), lymphatic invasion (HR, (5% v 17.3%; Table 5). In a resolved multivariate model, grade 3 2.858; 95% CI, 1.336 to 6.113; P ϭ .007), and IBTR (HR, 3.608; 95% (HR, 3.365; 95% CI, 1.848 to 7.151; P ϭ .001), lymph node posi- CI, 1.341 to 9.706; P ϭ .011) were significant in a resolved multivariate tivity (HR, 1.986; 95% CI, 1.01 to 3.906; P ϭ .047) and extensive analysis (Table 8). intraduct carcinoma (HR, 3.212; 95% CI, 1.382 to 7.463; P ϭ .007) were independently predictive of LRR. Competitive Risks Modeling Further analyses showed no significant alteration in the final DDFS resolved models presented above (Tables 6-8) for IBTR, DDFS and Forty-seven patients developed distant metastases with a 5-year breast cancer–specific death. For LRR the final resolved model DDFS rate for the whole cohort of 92.9%, luminal A of 95%, luminal contained grade 3 (HR 3.423, 95% CI 1.746-6.709, P Ͻ .001) and B of 90%, basal of 86.3%, HER-2 of 84.6%, and unclassified of 75%. extensive intraduct carcinoma (HR 2.950, 1.250-6.963, P ϭ .014) Luminal A had a crude event rate of less than half that of basal, HER-2, with lymph nodal status no longer significant (HR 1.748, 0.890- and unclassified tumors (Table 4), with a statistically significant dif- 3.433, P ϭ .105). ference in survival between subtypes (P ϭ .0035, Fig 1C). Lymph node positivity (HR, 3.558; 95% CI, 1.937 to 6.536; P Ͻ .001), lymphatic DDFS and Overall Survival After IBTR and Effect of invasion (HR, 1.977; 95% CI, 1.054 to 3.710; P ϭ .034), grade 3 (HR, Subtype, Endocrine Therapy, and Chemotherapy 1.912; 95% CI, 1.046 to 3.495; P ϭ .035), and PR (HR, 0.523; 95% CI, on Outcome 0.287 to 0.952; P ϭ .034) were independently significant in a resolved Two of 24 patients developed distant metastases within 3 months multivariate analysis (Table 7). of IBTR and were not included in this analysis. After IBTR, the 5-year DDFS rate was 81% and breast cancer–specific survival rate was Breast Cancer–Specific Death 77.3%. IBTR did not predict distant metastases but was significant for There were 37 deaths attributable to breast cancer with a 5-year breast cancer–specific death in univariate and multivariate analysis breast cancer–specific survival for the whole cohort of 96.3%, luminal (HR, 3.608; 95% CI, 1.341 to 9.706; P ϭ .011; Table 8). A of 98.2%, luminal B of 95.7%, basal of 88.3%, HER-2 of 84.6%, and No association between subtype, treatment, and outcome unclassified of 87.5%. Luminal A (5.8%) had a crude death rate less was observed.

4704 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY Information downloaded from jco.ascopubs.org and provided by University of New South Wales on October 6, 2009 from 129.94.224.120. Copyright © 2009 by the American Society of Clinical Oncology. All rights reserved. Predicting Failure in Early Breast Cancer

Table 5. Univariate Crude Rates and Cox Analysis for IBTR, LRR, DDFS, and Breast Cancer–Speciﬁc Death Breast Cancer–Speciﬁc Death IBTR (n ϭ 24) LRR (n ϭ 35) DDFS (n ϭ 47) (n ϭ 37) Crude Rate Crude Rate Crude Rate Crude Rate Parameter No. %HR P No. %HR P No. %HR P No. % HR P ERϩ 16/393 4.1 0.564 .195 21/393 5.3 0.401 .009 29/393 7.4 0.427 .005 22/393 5.6 0.398 .006 PRϩ 15/334 4.5 0.810 .622 19/334 5.7 0.559 .089 23/334 6.9 0.450 .006 17/334 5.1 0.395 .005 G3 13/145 8.9 3.025 .007 20/145 13.8 3.441 .001 24/145 16.6 2.632 .001 21/145 14.5 3.270 .001 LNϩ 11/146 7.5 2.041 .082 15/146 10.3 1.791 .089 29/146 19.9 4.046 < .001 24/146 16.4 5.068 < .001 LVI 4/79 5 1.208 .731 8/79 10.1 1.752 .166 16/79 20.2 3.026 .001 15/79 18.9 3.966 < .001 Size 10/147 6.8 1.906 .125 13/147 8.8 1.531 .228 24/147 16.3 2.614 .001 23/147 15.6 4.011 < .001 HER2ϩ 3/36 8.3 1.661 .421 4/36 11.1 1.544 .421 4/36 11.1 1.218 .706 4/36 11.1 1.611 .368 Age Ͻ 50 years 8/102 7.8 1.998 .114 12/102 11.8 2.085 .041 14/102 13.7 1.635 .123 13/102 12.7 1.959 .051 EICϩ 4/45 8.8 1.887 .249 7/45 15.6 2.474 .033 ———— ———— Margϩ 3/17 17.6 4.437 .016 4/38 10.5 1.744 .282 — — — — — — — — Boostϩ 15/247 6.1 1.674 .226 19/247 7.7 1.195 .599 — — — — — — — — Endo 7/223 3.1 0.545 .181 11/223 4.9 0.587 .146 18/223 8.1 0.774 .393 15/223 6.7 0.876 .692 Chemo 9/117 7.7 1.921 .128 13/117 11.1 1.919 .066 19/117 16.2 2.379 .004 18/117 15.4 3.194 .001 IBTR — — — — — — — — 4/21 19 2.281 .116 5/22 22.7 3.602 .008 LA 15/394 3.8 0.433 .051 20/394 5 0.333 .002 30/394 7.6 0.446 .008 23/394 5.8 0.414 .009 LB 2/23 8.7 2.132 .307 2/23 8.7 1.365 .669 2/23 8.7 0.963 .958 2/23 8.7 1.258 .753 Basal 5/52 9.6 2.182 .126 9/52 17.3 3.025 .005 8/52 15.4 1.745 .151 7/52 13.5 1.662 .108 HER2 1/13 7.7 1.056 .959 2/13 15.4 1.725 .475 2/13 15.4 1.612 .509 2/13 15.4 2.103 .307 Uncls 1/16 6.2 1.905 .530 2/16 12.5 2.621 .188 5/16 31.2 4.533 .001 3/16 18.8 3.419 .042 TNP 6/68 8.8 2.227 .093 11/68 16.2 3.194 .002 13/68 19.1 2.538 .004 10/68 14.7 2.446 .016

Abbreviations: IBTR, ipsilateral breast tumor recurrence; LRR, locoregional recurrence; DDFS, distant disease-free survival; HR, hazard ratio; ER, estrogen receptor; PR, progesterone receptor; G3, grade 3; LNϩ, lymph node positive; LVI, lymphatic vascular invasion; Size, tumor size Ͼ 20 mm; EIC, extensive intraduct component; Margϩ, resection margin positive; Boostϩ, cavity boost of 16 Gy given; endo, endocrine therapy; chemo, chemotherapy; LA, luminal A; LB, luminal B; uncls, unclassiﬁed; TNP, triple negative phenotype. Bold indicates signiﬁcance.

now important as triple-negative status alone is not synonymous with DISCUSSION this group of tumors.9 We applied the five-biomarker panel to assess its predictive and The importance of achieving optimal local control in BCT is high- prognostic value in locally treated EBC in a clinical trial setting. We lighted by its association with improved overall survival; conversely, found very low rates of IBTR with 97.4% recurrence-free survival at 5 IBTR is a poor prognostic indicator for subsequent distant failure and years for the whole cohort, which is predominantly comprised of death.12-14 The recent trend of progressive decline in local recurrence luminal A cancers (79.1%) with small components of luminal B rates to approximately 5% is likely the result of several factors, includ- (4.6%), basal-like (10.4%), HER-2 (2.6%), and unclassified (3.2%). ing improved preoperative breast imaging, greater emphasis on Our Australian cohort is similar to a population-based study from pathological margin assessment, achieving clear surgical margins, and 7 more frequent use of adjuvant systemic therapies. Local failure has North America (64.8% luminal A, 5.5% luminal B, 6.4% HER-2, 9% been associated with young age (Ͻ 50), tumor size (Ͼ T2), negative basal, 8% unclassified), although we have lower rates of HER-2, lumi- hormone receptor status, and lymph node involvement, although nal B and unclassified cancers, which may reflect the selection of cases these vary between studies,15-18 and an algorithm to define risk of in this trial setting. We found that molecular subtype was associated IBTR has been described.19 However, there is a need to improve with differences in IBTR, LRR, DDFS, and breast cancer–specific predictive and prognostic information to better tailor discussion re- death, although this was not significant for IBTR, which is likely the garding recurrence risk and hence treatments to individual patients.20 result of insufficient numbers of events. Our 5-year subtype IBTR-free The subclassification of breast cancer into five main intrinsic rates (98.8% luminal A, 95.5% luminal B, 92.3% HER-2, 90% basal, subtypes correlates with outcome but there are limited data available 92.9% unclassified) are similar to those described in a recent study regarding its predictive value. Specific subtypes, such as basal-like which utilized this approach with a simplified triple-marker assess- 8 cancers, have no specific targeted therapy, unlike ERϩ and HER-2ϩ ment of ER, PR, and HER-2. This latter cohort of 793 patients was of disease, and their identification is important for therapeutic decision similar composition to ours, but their use of the triple assessment making. The recent validation of the intrinsic molecular signature method for classification did not require positive basal marker expres- using a panel of five markers, ER, PR, HER-2, CK 5/6, and EGFR, sion and did not include an unclassified group. Two other studies that demonstrated its superiority over ER, PR, and HER-2 (TNP) alone, as also examined the predictive utility of the TNP and IBTR in BCT did it identifies basal-like tumors with a specificity of 100% and sensitivity not find an association, compared with non–triple-negative can- of 75%, compared with classification by gene expression profiling.7 cers21,22 although the mean time to local recurrence was shortened Specifically, the inclusion of the latter two antigens as basal markers is (2.8 v 4.2 years).22 Using the TNP as a classifier, our cohort contained www.jco.org © 2009 by American Society of Clinical Oncology 4705 Information downloaded from jco.ascopubs.org and provided by University of New South Wales on October 6, 2009 from 129.94.224.120. Copyright © 2009 by the American Society of Clinical Oncology. All rights reserved. Millar et al

A Ipsilateral breast tumor recurrence B Locoregional recurrence

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

Cumulative Survival 0.2 Cumulative Survival 0.2 P = .346 P = .012

0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 Time (months) Time (months)

C Distant metastases D Breast cancer–specific death

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

Cumulative Survival 0.2 Cumulative Survival 0.2 P = .0035 P = .0482

0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 Time (months) Time (months)

Fig 1. Kaplan-Meier estimates (log-rank test) for (A) ipsilateral breast tumor recurrence (24 events), (B) locoregional recurrence (35 events), (C) distant metastases (47 events), and (D) breast cancer–specific death (37 events), according to intrinsic molecular subtype. Luminal A (blue triangle; n ϭ 394), luminal B (yellow triangle; n ϭ 23), basal (red circle; n ϭ 52), HER2 (yellow circle; n ϭ 13), unclassified (black triangle; n ϭ 16). 68 (13.6%) of 498, but there was no association with IBTR, although it Tumor subtype identifies groups with divergent behavior as- predicted LRR, distant metastases, and death, and it out-performed sociated with differing recurrence rates, times to event, and overall the basal group. In a large Danish cohort of lymph node–positive survival. Significant differences between groups was observed in patients treated with mastectomy, TNP was significantly associated terms of median time to event for all measures of outcome with greatly with LRR whether treated with radiotherapy or not.23

Table 7. Cox Proportional Hazards Multivariate Model for DDFS Variable HR 95% CI P Table 6. Cox Proportional Hazards Multivariate Model for IBTR ERϩ 0.559 0.245 to 1.272 .165 Variable HR 95% CI P PRϩ 0.631 0.316 to 1.260 .191 ERϩ 0.831 0.260 to 2.649 .754 Grade 3 1.758 0.876 to 3.530 .112 PRϩ 1.152 0.390 to 3.404 .798 Size Ͼ 20 mm 1.331 0.709 to 2.498 .373 Grade 3 3.206 1.281 to 8.024 .013 LN ϩ 3.815 2.008 to 7.250 < .001 Size Ͼ 20 mm 1.214 0.510 to 2.888 .661 LVI 2.115 1.092 to 4.097 .026 LNϩ 1.952 0.826 to 4.614 .128 Chemotherapy 0.648 0.313 to 1.33 .241 LVI 0.804 0.262 to 2.470 .704 Resolved model Marginϩ 4.508 1.248 to 16.285 .022 PR 0.523 0.287 to 0.952 .034 Resolved model Grade 3 1.912 1.046 to 3.495 .035 Grade 3 3.372 1.488 to 7.642 .004 LNϩ 3.558 1.937 to 6.536 < .001 Marginϩ 5.838 1.690 to 20.172 .005 LVI 1.977 1.054 to 3.710 .034

Abbreviations: IBTR, ipsilateral breast tumor recurrence; HR, hazard ratio; ER, Abbreviations: DDFS, distant disease–free survival; HR, hazard ratio; ER, estrogen receptor; PR, progesterone receptor; LNϩ, lymph node positive; LVI, estrogen receptor; PR, progesterone receptor; LNϩ, lymph node positive; LVI, lymphatic vascular invasion. Bold indicates signiﬁcance. lymphatic vascular invasion. Bold indicates signiﬁcance.

4706 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY Information downloaded from jco.ascopubs.org and provided by University of New South Wales on October 6, 2009 from 129.94.224.120. Copyright © 2009 by the American Society of Clinical Oncology. All rights reserved. Predicting Failure in Early Breast Cancer

to predict risk of local recurrence.25 This assay and other signatures of Table 8. Cox Proportional Hazards Multivariate Model for Breast 26-27 Cancer–Specific Death grade that predict relapse support our finding of an association of grade with IBTR. Several other gene signatures were also recently Variable HR 95% CI P assessed, with only the Wound signature being significant.28,29 Thus, ϩ ER 0.742 0.278 to 1.979 .550 the clinical utility of this approach requires further validation in de- PRϩ 0.477 0.209 to 1.088 .786 Grade 3 1.978 0.868 to 4.505 .104 fined patient cohorts. Size Ͼ 20 mm 1.927 0.905 to 4.104 .089 In summary, this study identifies that the molecular subtype of LNϩ 3.984 1.850 to 8.581 .001 breast cancer, as approximated by the five-biomarker panel, identifies LVI 2.858 1.336 to 6.113 .007 differences in behavior for IBTR, LRR, DDFS, and death after BCT. Chemotherapy 0.624 0.230 to 1.697 .356 This additional information may assist in planning ongoing manage- IBTR 2.800 0.994 to 7.889 .051 ment and suggests that those more aggressive subtypes that have Ͻ Age 50 years 1.237 0.502 to 3.049 .643 shorter recurrence times and most events occurring within 5 years Resolved model PR 0.369 0.189 to 0.722 .004 (HER-2, basal, and unclassified) should have more frequent breast Size Ͼ 20 mm 2.178 1.050 to 4.521 .037 imaging and follow-up (eg, every 6 months for the first 2 years). In LNϩ 3.342 1.597 to 6.993 .001 addition for these high risk subtypes, it may also be prudent to con- LVI 2.773 1.330 to 5.780 .007 sider the addition of a local cavity boost, which was of benefit in three IBTR 3.608 1.341 to 9.706 .011 previous randomized studies.30-32 Although tumor subtype is of less Abbreviations: HR, hazard ratio; ER, estrogen receptor; PR, progesterone predictive value than existing histopathologic parameters, such as receptor; LN, lymph node; LVI, lymphatic vascular invasion; IBTR, ipsilateral grade and lymph node status, it does provide further information to breast tumor recurrence. Bold inidcates significance. complement these indices and may be useful in routine practice to help better inform both clinician and patient about their anticipated outcome after BCT. shortened recurrence times for the more aggressive subtypes: basal, HER-2, and unclassified. However, although they do identify behav- ioral differences in terms of biology, their role as predictive markers AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS appears to be inferior to traditional pathologic variables such as high OF INTEREST grade, tumor size, lymph node status, lymphatic invasion, and hormone receptor status. Luminal A and unclassified groups were asso- The author(s) indicated no potential conflicts of interest. ciated with improved or poor outcome for LRR, distant metastases, and breast cancer–specific death. For IBTR, only grade 3 and positive margins were of predictive value, although there was a strong trend for AUTHOR CONTRIBUTIONS reduced risk with luminal A subtype (P ϭ .051). One of the main shortcomings of this study in assessing the Conception and design: Ewan K.A. Millar, Peter H. Graham, predictive value of molecular subtype is the relatively low number of Sandra A. O’Toole, Catriona M. McNeil, Lois Browne, events. This may reflect the very low rate of margin involvement Robert L. Sutherland Provision of study materials or patients: Ewan K.A. Millar, Peter H. (3.4%), eliminating an important contributing factor for IBTR. Thus, Graham, Lois Browne, Sarah Eggleton, Julia Beretov, Constantine we would expect lower recurrence rates but increased sensitivity to Theocharous, Anne Capp, Elias Nasser, John H. Kearsley, Geoff Delaney, intrinsic biologic predictors of IBTR risk. The relative success of BCT George Papadatos, Christopher Fox, Robert L. Sutherland requires large numbers of patients to be accrued to have enough events Collection and assembly of data: Ewan K.A. Millar, Peter H. Graham, to provide sufficient statistical power. We have provided results with Sandra A. O’Toole, Lois Browne, Adrienne L. Morey, Sarah Eggleton, 5- and 10-year event rates, but in a predominantly luminal A, T1 Julia Beretov, Constantine Theocharous, Anne Capp, John H. Kearsley, cohort. Longer follow-up will be needed to assess outcomes at 10 and Robert L. Sutherland Data analysis and interpretation: Ewan K.A. Millar, Peter H. Graham, 15 years. As a result of the relatively small size of the luminal B, HER-2 Sandra A. O’Toole, Catriona M. McNeil, Lois Browne, Adrienne L. and unclassified groups (23, 13, and 16 patients, respectively), the Morey, Robert L. Sutherland confidence interval for outcome estimates for these groups widens. In Manuscript writing: Ewan K.A. Millar, Peter H. Graham, Sandra A. addition, the relative importance of the luminal B and HER-2 sub- O’Toole, Catriona M. McNeil, Lois Browne, Julia Beretov, Constantine types is lessened by the fact that they would now receive anti-HER-2 Theocharous, Anne Capp, John H. Kearsley, Geoff Delaney, therapies which would alter the outcome results. Robert L. Sutherland Final approval of manuscript: Ewan K.A. Millar, Peter H. Graham, Molecular classifications of invasive carcinoma have been as- Sandra A. O’Toole, Catriona M. McNeil, Lois Browne, Adrienne L. sessed in predicting IBTR. The Oncotype Dx (Genomic Health Inc, Morey, Sarah Eggleton, Julia Beretov, Constantine Theocharous, Anne Redwood City, CA) assay which was developed to predict distant Capp, Elias Nasser, John H. Kearsley, Geoff Delaney, George Papadatos, failure in ERϩ cancers treated with tamoxifen24 has also been shown Christopher Fox, Robert L. Sutherland

cal Adjuvant Breast and Bowel Project node-positive surgery for early breast cancer on local recurrence REFERENCES adjuvant breast cancer trials. J Clin Oncol 24:2028- and 15-year survival: An overview of the randomised 2037, 2006 trials. Lancet 366:2087-2106, 2005 1. Wapnir IL, Anderson SJ, Mamounas EP, et al: 2. Clarke M, Collins R, Darby S, et al: Early 3. Perou CM, Sorlie T, Eisen MB, et al: Molecu- Prognosis after ipsilateral breast tumor recurrence Breast Cancer Trialists’ Collaborative Group: Effects lar portraits of human breast tumors. Nature 406: and locoregional recurrences in ﬁve National Surgi- of radiotherapy and of differences in the extent of 747-752, 2000 www.jco.org © 2009 by American Society of Clinical Oncology 4707 Information downloaded from jco.ascopubs.org and provided by University of New South Wales on October 6, 2009 from 129.94.224.120. Copyright © 2009 by the American Society of Clinical Oncology. All rights reserved. Millar et al

4. Sorlie T, Perou CM, Tibshirani R, et al: Gene disease: Implications for systemic therapy at the node-negative breast cancer. N Engl J Med 351: expression patterns of breast carcinomas distin- time of local relapse. J Clin Oncol 14:52-57, 1996 2817-2826, 2004 guish tumor subclasses with clinical implications. 15. Calle R, Vilcoq JR, Zafrani B, et al: Local 25. Mamounas EP, Tang G, Bryant J, et al: Asso- Proc Natl Acad SciUSA98:10869-10874, 2001 control and survival of breast cancer treated by ciation between the 21-gene recurrence score assay 5. Carey LA, Perou CM, Livasy CA, et al: Race, limited surgery followed by irradiation. Int J Radiat (RS) and risk of locoregional failure in node-negative, breast cancer subtypes and survival in the Carolina Oncol Biol Phys 12:873-878, 1986 ER-positive breast cancer: Results from NSABP breast cancer study. JAMA 295:2492-2502, 2006 16. Arriagada R, Le MG, Contesso G, et al: Pre- B-14 and NSABP B-20. Presented at the 28th Annual 6. Neilsen TO, Hsu FD, Jensen K, et al: Immu- dictive factors for local recurrence in 2006 patients San Antonio Breast Cancer Symposium Meeting, nohistochemical and clinical characterization of the with surgically resected small breast cancer. Ann December 8-11, 2005, San Antonio, TX basal-like subtype of invasive breast cancer. Clin Oncol 13:1404-1413, 2002 26. Musgrove EA, Sergio MS, Loi S, et al: Identi- Cancer Res 10:5367-5374, 2004 17. Early Breast Cancer Trialists’ Collaborative fication of functional networks of estrogen and 7. Cheang MC, Voduc D, Bajdik C, et al: Basal- Group: Favourable and unfavourable effects on long- c-Myc responsive genes and their relationship to like breast cancer defined by five biomarkers has term survival of radiotherapy for early breast cancer: response to tamoxifen therapy in breast cancer. superior prognostic value than triple-negative phe- An overview of the randomised trials. Lancet 355: PLoS ONE 3:e2987, 2008 notype. Clin Cancer Res 14:1368-1376, 2008 1757-1770, 2000 27. Sotiriou C, Wirapati P, Loi S, et al: Gene 8. Nguyen PL, Taghian AG, Katz MS, et al: 18. Fowble BL, Schultz DJ, Overmoyer B, et al: expression profiling in breast cancer: Understanding Breast cancer subtype approximated by estrogen The influence of young age on outcome in early the molecular basis of histologic grade to improve receptor, progesterone receptor, and HER-2 is asso- stage breast cancer. Int J Radiat Oncol Biol Phys prognosis. J Natl Cancer Inst 98:262-272, 2006 ciated with local and distant recurrence after breast- 30:23-33, 1994 28. Nuyten DSA, Kreike B, Hart AAM, et al: Pre- conserving therapy. J Clin Oncol 26:2373-2378, 19. Sanghani M, Balk E, Cady B, et al: Predicting dicting local recurrence after breast-conserving ther- 2008 the risk of local recurrence in patients with breast apy by gene expression profiling. Breast Cancer Res 9. Rakha EA, Reis-Filho JS, Ellis IO: Basal-like cancer: An approach to a new computer-based 8:R62, 2006 breast cancer: A critical review. J Clin Oncol 26: predictive tool. Am J Clin Oncol 30:473-480, 2007 29. van de Vijver MJ, He YD, van’t Veer LJ, et al: 2568-2581, 2008 20. Graham P, Fourquet A: Placing the boost in A gene-expression signature as a predictor of sur- 10. Millar EKA, Anderson LR, McNeil CM, et al: breast-conservation radiotherapy: A review of the vival in breast cancer. N Engl J Med 347:1999-2009, BAG-1 predicts patient outcome and tamoxifen re- role, indications and techniques for breast boost sponsiveness in ER-positive invasive ductal carci- radiotherapy. Clin Oncol 18:210-219, 2006 2002 noma of the breast. Br J Cancer 100:123-133, 2009 21. Haffty BG, Yang Q, Reiss M, et al: Locore- 30. Romestaing H, Lehingue Y, Carrie C, et al: 11. Graham P, Browne L, Capp A, et al: The St gional relapse and distant metastasis in conserva- Role of a 10 Gy boost in the conservative treatment George, Wollongong and Liverpool Breast boost tively managed triple negative early-stage breast of early breast cancer: Results of a randomised trial: 1st planned analysis at 6 year median follow-up. cancer. J Clin Oncol 24:5652-5657, 2006 clinical trial in Lyon, France. J Clin Oncol 15:963-968, Australasian Radiology 51:A85, 2007 (suppl 3) 22. Dent R, Trudeau M, Pritchard KI, et al: Triple 1997 12. Haffty BG, Fischer D, Beinfield M, et al: negative breast cancer: Clinical features and pat- 31. Bartelink H, Horiot JC, Poortmans PM, et al: Prognosis following local recurrence in the conser- terns of recurrence. Clin Cancer Res 13:4429-4434, Impact of a higher radiation dose on local control and vatively treated breast cancer patient. Int J Radiat 2007 survival in breast-conserving therapy of early breast Oncol Biol Phys 21:293-298, 1991 23. Kyndi M, Sorensen FB, Knudsen H, et al: cancer: 10-year results of the randomized boost 13. Whelan T, Clark R, Roberts R, et al: Ipsilateral Estrogen receptor, progesterone receptor, HER-2, versus no boost EORTC 22881-10882 trial. J Clin breast tumor recurrence post lumpectomy is predic- and response to post mastectomy radiotherapy in Oncol 25:3259-3265, 2007 tive of subsequent mortality: Results from a ran- high-risk breast cancer: The Danish Breast Cancer 32. Polgar C, Fodor J, Orosz Z, et al: Electron and domized trial. Int J Radiat Oncol Biol Phys 30:11-16, Cooperative Group. J Clin Oncol 26:1419-1426, high dose rate brachytherapy boost in the conserva- 1994 2008 tive treatment of stage I-II breast cancer: First 14. Haffty BG, Reiss M, Beinfield M, et al: Ipsilat- 24. Paik S, Shak S, Tang G, et al: A multigene results of the randomised Budapest boost trial. eral breast tumor recurrence as a predictor of distant assay to predict recurrence of tamoxifen treated, Strahlenther Onkol 178:615-623, 2002

■■■

4708 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY Information downloaded from jco.ascopubs.org and provided by University of New South Wales on October 6, 2009 from 129.94.224.120. Copyright © 2009 by the American Society of Clinical Oncology. All rights reserved. British Journal of Cancer (2011), 1–9 & 2011 Cancer Research UK All rights reserved 0007 – 0920/11 www.bjcancer.com

Full Paper Prediction of outcome of early ER þ breast cancer is improved using a biomarker panel, which includes Ki-67 and p53

*,1,2,3,4 4,5 1,6 5 1,7,8 1 4,5 EKA Millar , PH Graham , CM McNeil , L Browne , SA O’Toole , A Boulghourjian , JH Kearsley , 4,9 3,4,10 4,11 11 12 1,13 G Papadatos , G Delaney , C Fox , E Nasser , A Capp and RL Sutherland 1 2 Cancer Research Program, Garvan Institute of Medical Research, 384 Victoria Street, Darlinghurst, New South Wales 2010, Australia; Department of 3 Anatomical Pathology, South Eastern Area Laboratory Service, St George Hospital Kogarah, New South Wales 2217, Australia; School of Medicine and 4 Health Sciences, University of Western Sydney, Campbelltown, New South Wales, Australia; Faculty of Medicine, University of NSW, Kensington, New 5 South Wales, Australia; Department of Radiation Oncology, Cancer Care Centre, St George Hospital Kogarah, New South Wales 2217, Australia; 6 7 Department of Medical Oncology, Royal Prince Alfred Hospital, University of Sydney, Camperdown, New South Wales, Australia; Department of Diagnostic Oncology and Tissue Pathology, Royal Prince Alfred Hospital, Camperdown, New South Wales 2010, Australia; 8University of Sydney,

Camperdown, New South Wales, Australia; 9Macarthur Cancer Therapy Centre, Campbelltown, New South Wales, Australia; 10Department of Radiation 11 Oncology, Liverpool Hospital, Liverpool, UK; Department of Radiation Oncology, Wollongong Hospital, Wollongong, New South Wales, Australia; 12 13 Department of Radiation Oncology, Mater Misericordiae Hospital, Waratah, New South Wales, Australia; St Vincent’s Clinical School, Faculty of

Medicine, University of NSW, Darlinghurst, New South Wales 2052, Australia

BACKGROUND: The aim of this study is to determine whether immunohistochemical (IHC) assessment of Ki67 and p53 improves þ prognostication of oestrogen receptor-positive (ER ) breast cancer after breast-conserving therapy (BCT). In all, 498 patients with

invasive breast cancer from a randomised trial of BCT with or without tumour bed radiation boost were assessed using IHC.

METHODS: The ER þ tumours were classified as ‘luminal A’ (LA): ER þ and/or PR þ , Ki-67 low, p53, HER2 or ‘luminal B’ (LB):

ER þ and/or PR þ and/or Ki-67 high and/or p53 þ and/or HER2 þ . Kaplan–Meier and Cox proportional hazards methodology

were used to ascertain relationships to ispilateral breast tumour recurrence (IBTR), locoregional recurrence (LRR), distant metastasis-

free survival (DMFS) and breast cancer-specific survival (BCSS).

RESULTS: In all, 73 patients previously LA were re-classified as LB: a greater than four-fold increase (4.6–19.3%) compared with ER, PR,

HER2 alone. In multivariate analysis, the LB signature independently predicted LRR (hazard ratio (HR) 3.612, 95% CI 1.555–8.340,

P ¼ 0.003), DMFS (HR 3.023, 95% CI 1.501–6.087, P ¼ 0.002) and BCSS (HR 3.617, 95% CI 1.629–8.031, P ¼ 0.002) but not IBTR.

CONCLUSION: The prognostic evaluation of ER þ breast cancer is improved using a marker panel, which includes Ki-67 and p53. This

may help better define a group of poor prognosis ER þ patients with a greater probability of failure with endocrine therapy.

British Journal of Cancer advance online publication, 28 June 2011; doi:10.1038/bjc.2011.228 www.bjcancer.com & 2011 Cancer Research UK

Keywords: breast cancer; biomarker; Ki67; p53; luminal B

Oestrogen receptor-positive (ER þ ) breast cancer comprises 2005). Therefore, predicting the likely prognosis in an individual approximately 75% of all breast cancers and treatments targeting patient before treatment would allow early selection of optimal oestrogen synthesis (aromatase inhibitors) or the ER (tamoxifen) therapies, the importance of which is highlighted in the most are the most effective adjuvant therapies. Gene expression recent St Gallen guidelines for the treatment of early breast cancer profiling (GEP) studies over the past decade have established (Goldhirsch et al, 2009). molecular subtypes of ER þ luminal disease, which are charact- The abundant data derived from GEP studies have clearly erised by differences in outcome and underlying biology, largely identified the significance of genomic grade and proliferation now referred to as luminal A (LA) or luminal B (LB), the latter signatures in prognosis and response to endocrine therapy being characterised by increased proliferation and higher grade as (reviewed in detail in Musgrove and Sutherland, 2009 and Sotiriou well as lower levels of ER related genes (Perou et al, 2000; Sørlie and Pusztai, 2009). However, given the current costs of such et al, 2001). Despite the successes of endocrine therapy in reducing molecular testing, translating these findings into an economical, annual recurrences and death by 41% and 34%, respectively, reproducible and readily applicable panel for immunohistochem- resistance occurs in about 30% of patients treated with tamoxifen istry (IHC) in a routine pathology setting is a priority. Most (Early Breast Cancer Trialists’ Collaborative Group (EBCTCG), previous IHC definitions of LA and LB tumours include ER þ and/or PR þ , with HER2 positivity defining LB, creating a population size of approximately 5–10% (Cheang et al, 2008; Nguyen et al, 2008; Millar et al, 2009b; Blows et al, 2010). However, GEP studies *Correspondence: Dr EKA Millar; E-mail: [email protected] have documented the LB population to be larger than Received 4 February 2011; revised 20 May 2011; accepted 25 May 2011 this, averaging approximately 16% (ranging from 10 to 21%, Predicting failure in luminal breast cancer EKA Millar et al 2 reviewed in detail in Sorlie et al, 2003 and Hu et al, 2006), MATERIALS AND METHODS suggesting that this poorer prognosis subtype may be under- represented using this definition. This discrepancy is most likely Study subjects explained by the fact that only approximately 30% of LB cancers Training cohort Cases were drawn from the St Vincent’s Campus are in fact HER2 positive (Carey et al, 2006). Although Outcome Cohort, which comprised 292 invasive ductal carcinomas proliferation is the key discriminator of luminal tumours, the treated between February 1992 and August 2002 at St Vincent’s optimal subclassification of luminal tumours by GEP has yet to be Hospital, Sydney, Australia. Ethics approval for use of tissue and defined (Weigelt et al, 2010b). Several studies have, however, clinicopathological data was granted by the Human Research shown that intrinsic subtype as defined by IHC ‘mirrors’ Ethics Committee of St Vincent’s Hospital, Sydney (Ref. SVH H94/ the subtypes identified by GEP and that the IHC subtypes so 080 and 00/036). A more detailed description of the clinicopatho- defined have distinct clinical outcomes (Neilsen et al, 2004; Abd logical characteristics of the cohort is published elsewhere (Millar El-Rehim et al, 2005; Cheang et al, 2008, 2009; Blows et al, 2010). et al, 2009a; Lo´pez-Knowles et al, 2010). In summary, 40% of Such IHC definitions are now in common clinical usage. tumours were 420 mm, 45% were grade 3, 43% were lymph node Some recent studies have addressed the issue of a more refined þ positive, 68% were ER positive, 57% were PR positive and 18% definition of good and poor prognosis ER cancer, and used a were HER2 fluorescent in situ hybridisation (FISH) positive modified IHC definition to include assessment of the proliferation (HER2:CEP17 ratio 42.2). Median age was 54 years, and patients marker Ki-67 (Cheang et al, 2009; Cuzick et al, 2009; Hugh et al, were treated with endocrine therapy (49%), chemotherapy (38%) 2009), which results in a larger proportion of LB tumours with or both (24%). Cases were prospectively followed up for a median independent prognostic power (Cheang et al, 2009). This latter of 64 months, and the outcome events measured were as follows: study defined a Ki67 cutpoint (14%) derived from GEP analyses. recurrence (local or distant; 25%), metastasis (23%) and breast This set of biomarkers more closely resembles the Oncotype Dx þ cancer-specific death (18%). This cohort was used to identify assay of known predictive and prognostic power in ER , lymph differences in expression of several cell cycle and apoptotic node-negative cancer, which is largely driven by proliferation, markers, including Ki67 and p53 (CM McNeil et al, manuscript in HER2- and ER-related genes (Paik et al, 2004). However, a recent preparation), between LA and B cancers using the following head to head comparison of a four IHC biomarker panel of ER, PR, definitions: LA: ER þ and/or PR þ and HER2 and LB: ER þ HER2 and Ki-67 (IHC 4) has been shown to provide prognostic and/or PR þ and HER2 þ . Using the median expression levels for information, which is at least equivalent to Oncotype Dx using Ki67 and p53 as the cutpoints (5% and 10%, respectively), we were material from the ATAC trial (Cuzick et al, 2009). This important able to demonstrate a significant difference in level of expression study identifies the robustness of prognostic data, which can be between LA and LB for these antigens (P ¼ 0.0158 and P ¼ 0.0061, provided by routine IHC. Some observers support the view that respectively). Subsequently, we modified our definition of LA and GEP currently offers no more that routine IHC when combined LB to include Ki67 and p53 status as follows: ‘LA’: ER þ and/or with important morphological features (not assessable by GEP), PR þ and HER2, Ki67 low and p53 negative and ‘LB’: ER þ and/or such as lymphatic vascular invasion and lymph node status PR þ and/or HER2 þ and/or Ki67 high and/or p53 þ . Kaplan– (Weigelt and Reis-Filho, 2010). In addition, these routine analyses Meier analysis for breast cancer specific death showed a significant can be performed at a fraction of the cost of commercially available difference in outcome between these two groups of ER þ patients GEP tests. In addition, it also supports the concept that (P ¼ 0.0002) using this updated classifier (CM McNeil et al, measurement of a few well chosen protein products can identify manuscript in preparation). clinically significant patient groups (Ring et al, 2006). Histological grade is a key component of routine pathology reporting and of prognostic importance, but may, in some circumstances, be Study validation cohort In this biomarker study, tissue was affected by subjectivity, along with problems with inadequate or available from 498 patients (from a total of 688) with invasive delayed fixation, which can result in undergrading (Rakha et al, breast cancer who were enrolled into a randomised clinical trial, 2010). Incorporation of biomarkers as surrogates for molecular which compared the benefit of the addition of a local cavity boost grade into routine reporting may help more reliably define good of radiotherapy to breast-conserving therapy (BCT; Clinical Trials and poor prognosis patients, most significantly for grade 2 Registry NCT00138814). The study was conducted at St George, invasive carcinomas, which comprise 37–49% of all breast cancers Wollongong and Liverpool Hospitals, Sydney, New South Wales, (Rakha et al, 2010). Australia between 1996 and 2003 when the trial was closed to To further validate an IHC panel of markers for routine accrual. Follow-up for this analysis continued until September application in a clinical setting, we assessed a new biomarker panel 2008. Clinicopathological details are summarised in Supplemen- to differentiate good prognosis (LA) and poor prognosis (LB) tary Table 1, and have been previously published in detail Millar tumours in a cohort of predominantly ER þ early breast cancer et al (2009b). This study was approved by the Human Research patients enrolled in a randomised clinical trial of conservative Ethics Committee of the St George Hospital, Sydney, Australia (ref. surgery, post-operative whole breast radiotherapy and then no.: 96/84). The flow of patients through the trial is summarised in randomised to an additional cavity boost or not. We previously a CONSORT flow diagram (Figure 1). Patients were randomised described the clinical usefulness of a five biomarker panel (Millar using random blocking sequences set up before commencing of et al, 2009b; ER, PR, HER2, CK 5/6 and EGFR) and have further the study. Following patient consent, a person independent of the defined luminal tumours by including Ki-67 and p53 status, the study both generated the sequence and assigned participants to latter described in higher grade tumours, overexpressed more interventions as below. This was an unblinded study. frequently within LB (Sorlie, 2004; Jacquemier et al, 2008; Hugh All patients with invasive carcinoma received local excision and et al, 2009; Carey, 2010; Weigelt et al, 2010b) and as a predictor of axillary sentinel node biopsy or axillary clearance. Adjuvant endocrine resistance in some studies (Yamashita et al, 2006). chemotherapy (AC or CMF) was given to 23.7% of patients and These markers have easily available and well-characterised 44.9% received adjuvant endocrine therapy with tamoxifen. No antibodies in current use, which can be immediately applied to patients received adjuvant trastuzumab. For patients subsequently clinical practise. classified as modified ‘LA’, 49.5% received endocrine therapy and This study aimed to define the predictive value of a more refined 13.4% received chemotherapy, and those classified as modified luminal IHC biomarker signature in those patients who were ‘LB’ 55.7% received endocrine therapy and 25% received ER þ , with disease relapse and death from breast cancer as chemotherapy. Patients were randomised to whole breast radio- end-points. therapy of 50 Gy in 25 fractions or whole breast radiotherapy of

British Journal of Cancer (2011), 1 – 9 & 2011 Cancer Research UK Predicting failure in luminal breast cancer EKA Millar et al 3 CONSORT 2010 flow diagram

Enrollment *Assessed for eligibility (n =NK)

*Excluded (n =NK)

Randomized (n = 688)

Allocation Allocated to boost (n = 346) Allocated to no boost (n = 342) • Received boost (n = 338) ♦ Received no boost (n = 336) • Received no boost (n = 4): • Received boost (n = 6): error 1, patient choice 3 error 1, patient choice 4, medical 1 • Received reduced or no radiotherapy for • Received no radiotherapy (n = 2): medical reasons (n = 4) patient choice 1, medical 1

Follow-Up Lost to follow-up (n = 3) at 1, 8 and 9 years free Lost to follow-up (n = 1) because of patient of disease: withdrawal at 8 years after local recurrence. unable to contact 2 Included in analysis until loss to follow-up patient withdrawal 1 Included in analysis until loss to follow-up Tissue microarray not available for analysis Tissue microarray not available for analysis (n = 98) (n = 92)

Analysis

Tissue available for microarray analysis Tissue available for microarray analysis (n = 247) (n = 251)

Figure 1 *The trial recruited from three main centres (St George, Wollongong and Liverpool Hospitals). Although the total number of patients assessed for eligibility and excluded for all centres is not known, this data are available for the main recruiting centres at St Geroge Hospital, which contributed the majority of patients in the trial, n ¼ 546 (number assessed, n ¼ 2046; excluded, n ¼ 1500: not meeting criteria, n ¼ 943; declined to partcipate, n ¼ 235; other reasons, n ¼ 322; patients randomised in the trials, n ¼ 536).

45 Gy in 25 fractions plus a tumour bed boost of 16 Gy in eight Tissue microarray (TMA) construction, IHC and FISH fractions. Supraclavicular fields were not added unless there were four or more nodes positive. In all, 17 patients had positive TMAs were constructed from formalin-fixed paraffin-embedded margins, 65 had clearance of o1 mm and a further 86 had o2mm tissue blocks, which were available from 498 invasive carcinomas, clearance, the remainder being well clear. HER2 status was using 1 mm diameter punches with up to three cores sampled from unknown at the time of treatment. each tumour. Antibodies used in IHC were Ki-67 (1 : 100, SP6 neomarkers), p53 (1:50, DO-7; Dako, Carpentaria, CA, USA), ER (1:100, 6F11; Dako), PR (1:200, PgR 636; Dako), CK 5/6 (1:80, Study definitions MAB1602; Chemicon International, Temecula, USA), EGFR (1:100, H11; Dako). Patients were assessed at 6 weeks after radiation therapy, All staining was performed using a Dako autostainer following 6 monthly for 2 years, then annually thereafter with annual breast antigen retrieval for all antibodies except for Ki-67, which was imaging. Follow-up time for this biomarker cohort was calculated performed on a Leica (Wetzlar, Germany)/Bond Max system using from the date of the first surgical procedure to the date of the first ER2 (high pH antigen retrieval). All staining was centrally assessed event, as outlined below, or to the last known confirmed date of by one breast Pathologist (EKAM). ER and PR were assessed as breast cancer disease-free status. Median follow-up time was 84 positive if a modified ‘H score’ (i.e., percentage intensity) was months (range 1–134 months). The primary end point was time to 410. CK5/6 and EGFR were considered positive if staining of any ipsilateral breast tumour recurrence (IBTR). This included any intensity was present (i.e., 40). Tumours were considered HER2 ipsilateral in-breast recurrence (invasive or non-invasive). The positive only if they were HER2 amplified on FISH using a HER2: secondary end points were locoregional recurrence (LRR: IBTR, chromosome 17 ratio 42.2 as positive. p53 and Ki-67 were axilla, chest wall, internal mammary or supraclavicular fossa considered positive if there was 410% positive average nuclear lymph nodes) and time to distant metastases and death. staining of any intensity.

& 2011 Cancer Research UK British Journal of Cancer (2011), 1 – 9 Predicting failure in luminal breast cancer EKA Millar et al 4 Classification of intrinsic molecular phenotype prognosis modified ‘LA’ as ER þ and/or PR þ and HER2, Ki67 low and p53; and poor prognosis modified ‘LB’ as ER þ and/or Patients were initially subtyped based on the status of their PR þ and/or HER2 þ and/or Ki67 high and/or p53 þ . primary tumour as follows: ‘LA’: ER þ and/or PR þ and HER2, and ‘LB’: ER þ and/or PR þ and HER2 þ ; HER2 enriched: ER and PR and HER2 þ , and basal: ER,PR, HER2, CK 5/6 þ Five-year survival rates, univariate analysis of LA and B and/or EGFR þ , unclassified (negative for all five markers). tumours for IBTR, LRR, distant metastasis-free survival Subsequently they were re-classified as modified ‘LA’: ER þ and/or (DMFS) and breast cancer-specific survival (BCSS) PR þ and Ki-67 low, p53, HER2; modified ‘LB’: ER þ and/or PR þ and/or Ki-67 high and/or p53 þ and/or HER2 þ ; HER2 Using these updated definitions, 321 tumours (64.5%) were enriched: ER and PR and HER2 þ , and basal: ER,PR, classified as LA and 96 as LB (19.3%). Thus, 73 previously LA HER2, CK 5/6 þ and/or EGFR þ , unclassified (negative for all tumours were re-classified as LB (previously only 23 tumours were LB), five markers). that is, 4.2-fold increase (4.6–19.3%) with LB now comprising 23% of all ER þ tumours. We then examined the relative contribution of p53 and Ki67 to the updated classification of the 96 LB tumours: Statistical analyses 57 of 96 (59.4%) were p53/Ki67 þ , 19 (19.7%) were p53 þ /Ki67, 12 (12.5%) were p53 þ /Ki67 þ and 8 were Kaplan–Meier analyses for IBTR, LRR, distant disease-free þ survival and breast cancer-specific death were estimated for each p53 /Ki67 (HER2 ). subtype and compared using the log-rank test. We used Cox As previously described, no benefit of a tumour bed boost was proportional hazards univariate analysis to analyse the association observed in this group of patients (Millar et al, 2009b). At a between prognostic variables and molecular subtype with IBTR, median follow-up period of 84 months, the 5-year survival rates for LRR, metastases and breast cancer-specific death. Multivariate modified LA and modified LB, respectively, using the updated analysis (MVA) was used to construct models identifying those classifier were IBTR 99.3, 96.6%; LRR 99.7, 93.4%; DMFS 97, 87%; variables which were independently prognostic. Subsequently, and BCSS 99.7, 92.5%. Comparative analyses of the clinicopatho- step-wise removal of variables was used until resolution. Analyses logical features, crude event rates and univariate analyses of LA were performed using Statview 5.0 (Abacus systems, Berkeley, CA, and LB groups between the differing definitions are presented USA) and STATA 10.0 (StataCorp LP, College Station, TX, USA). in Tables 1 and 2. Univariate Cox proportional hazards were The ANOVA was used to assess differences in expression of target calculated for each measure of outcome for Ki67 and p53 and the antigens as continuous variables between intrinsic subtypes. modified LA and LB subtypes, which are presented with crude event rates in Table 3. Further crude event rates for modified LA and LB for lymph node negative, lymph node positive and lymphatic vascular invasion are presented in Supplementary Table 2. RESULTS As expected, the updated classification resulted in increased numbers of events for all outcomes for LB and a reduction for LA. Assessment of Ki67 and p53 expression between LA and B þ tumours This is mirrored in LB by increases in LVI and LN status, with recurrence rates and death rates two to three times that of LA. Having identified differences in Ki67 and p53 in ER þ tumours in Univariate analyses showed that modified LA is a significant our training cohort, we then assessed the difference between LA predictor for all measures of outcome including IBTR (hazard ratio and B tumours in expression level of these two antigens in our (HR) 0.314, 95% CI 0.136–0.726, P ¼ 0.007) where it previously validation cohort (n ¼ 498). Within LB tumours in this cohort, we was close to but not statistically significant (P ¼ 0.051). Modified observed significantly higher levels of Ki-67 and p53 expression LB predicted DMFS and BCSS (P ¼ 0.005 and 0.003, respectively) (P ¼ 0.0008 and 0.0048, respectively). The median average value and approached significance for IBTR and LRR (P ¼ 0.07 and for both Ki67 and p53 within the validation cohort was 10%. 0.052, respectively) where previously it was not significant for Subsequently, we modified our working definition further for good any outcome measure. Ki67 predicted outcome for all measures

Table 1 Patient tumour characteristics and event rates classified by luminal phenotype

Whole cohort, Luminal A, Luminal B, Modified luminal A, Modified luminal B, n ¼ 498 (%) n ¼ 394 (79.1%) n ¼ 23 (4.6%) n ¼ 321(64.5%) n ¼ 96 (19.3%)

Patient tumour characteristics Size o20 mm 357 (70.3) 289 (73.4) 17 (73.9) 242 (75.4) 64 (66.7) LVI+ 79 (15.8) 62 (15.7) 4 (17.4) 43 (13.4) 23 (23.9) LN+ 146 (29.0) 117 (29.6) 5 (21.7) 86 (26.7) 36 (37.5) Grade 3 145 (29.1) 65 (16.5) 16 (69.5) 26 (8.1) 55 (57.3) EIC+ 45 (9.0) 29 (7.4) 5 (21.7) 23 (7.2) 11 (11.5) Median Age 61 62 57 62 61

Events Median follow-up 84 83.5 71 84 78 IBTR 24 (4.8) 15 (3.8) 2 (8.7) 9 (2.8) 8 (8.3) LRR 35 (7.0) 20 (5.1) 2 (8.7) 11 (3.4) 11 (11.5) DMFS 47 (9.4) 30 (7.6) 2 (8.7) 16 (4.9) 16 (16.7) BCSS 37 (7.4) 23 (5.8) 2 (8.7) 11 (3.4) 14 (14.6)

Abbreviations: BCSS ¼ breast cancer-specific survival; DMFS ¼ distant metastasis-free survival; EIC ¼ extensive intraduct carcinoma; ER+ ¼ oestrogen receptor positive; IBTR ¼ ipsilateral breast tumour recurrence; LN ¼ lymph node; LRR ¼ locoregional recurrence; LVI ¼ lymphatic/vascular invasion; PR+ ¼ progesterone receptor positive. Luminal A: ER+ and/or PR+, HER2; Luminal B: ER+ and/or PR+, HER2+; modified luminal A: ER+ and/or PR+, Ki67 low and p53 and HER2; modified luminal B: ER+ and/ or PR+ and/or Ki67 high and/or p53+ and/or HER2+.

British Journal of Cancer (2011), 1 – 9 & 2011 Cancer Research UK Predicting failure in luminal breast cancer EKA Millar et al 5 Table 2 Comparative 5 and 10 year event rates for luminal A and B

IBTR LRR DM BCSD

5 Year (%) 10 Year (%) 5 Year (%) 10 Year (%) 5 Year (%) 10 Year (%) 5 Year (%) 10 year (%)

Whole cohort (n ¼ 498) 12/498 (2.4) 23/498 (4.6) 21/498 (4.2) 35/498 (6.8) 34/498 (6.8) 47/498 (9.4) 18/498 (3.6) 37/498 (7.4) 12/24 (50) 23/24 (95.8) 21/35 (60) 35/35 (100) 34/47 (72.3) 47/47 (100) 18/37 (48.6) 37/37 (100) Luminal A (n ¼ 394) 4/394 (1) 14/394 (3.6) 8/394 (2) 19/394 (4.8) 19/394 (4.8) 30/394 (7.6) 7/394 (1.8) 23/394 (5.8) 4/15 (26.6) 14/15 (93.3) 8/20 (40) 19/20 (95) 19/30 (63.3) 30/30 (100) 7/23 (30.4) 23/23 (100) Luminal B (n ¼ 23) 1/23 (4.3) 2/23 (8.7) 1/23 (4.3) 2/23 (8.6) 2/23 (8.6) 2/23 (8.6) 1/23 (4.3) 2/23 (8.6) 1/2 (50) 2/2 (100) 1/2 (50) 2/2 (100) 2/2 (100) 2/2 (100) 1/2 (50) 2/2 (100) Modified luminal A (n ¼ 321) 2/321 (0.6) 9/321 (2.8) 3/321 (0.9) 11/321 (3.4) 9/321 (2.8) 16/321 (4.9) 1/321 (0.3) 11/321 (3.4) 2/9 (22.2) 9/9 (100) 3/11 (27) 11/11 (100) 9/16 (56.3) 16/16 (100) 1/11 (9.1) 11/11 (100) Modified luminal B (n ¼ 96) 3/96 (3.1) 7/96 (7.3) 6/96 (6.3) 10/91 (10.9) 12/96 (12.5) 16/96 (16.7) 7/96 (7.3) 14/96 (14.6) 3/8 (37.5) 7/8 (87.5) 6/11 (54.5) 10/11 (90.1) 12/16 (75) 16/16 (100) 7/14 (50) 14/14 (100)

Abbreviations: BCSD ¼ breast cancer-specific death; DM ¼ distant metastasis; ER+ ¼ oestrogen receptor positive; IBTR ¼ ipsilateral breast tumour recurrence; LRR ¼ locoregional recurrence; PR+ ¼ progesterone receptor positive. Modified luminal A: ER+ and/or PR+, Ki-67 low, p53, HER2; modified luminal B: ER+ and/or PR+ and/or Ki-67 high and/or p53+ and/or HER2+. In the top row of each box, the denominator is the total number of patients within that patient group or subtype; in the bottom row of each box, the denominator is the total number of events for each group or subtype.

Table 3 Univariate crude rates and hazard ratio (Cox) for biomarkers and luminal phenotype

IBTR (n ¼ 24) LRR (n ¼ 35) DDFS (n ¼ 47) BCSS (n ¼ 37)

CR HR (95% CI) P CR HR (95% CI) P CR HR (95% CI) P CR HR (95% CI) P

Ki67 high 12/129 3.126 (1.390 – 7.029) 0.0008 19/129 3.759 (1.923 – 7.340) 0.0001 24/129 3.436 (1.926 – 6.130) o0.0001 22/129 4.948 (2.530 – 9.674) o0.0001 p53+ 3/57 1.067 (0.315 – 3.629) 0.916 5/57 1.290 (0.497 – 3.350) 0.601 11/57 2.566 (1.303 – 5.056) 0.006 11/57 3.523 (1.731 – 7.168) 0.0005 LA 15/394 0.433 (0.186 – 1.005) 0.051 20/394 0.333 (0.169 – 0.655) 0.002 30/394 0.446 (0.246 – 0.810) 0.008 23/394 0.414 (0.213 – 0.816) 0.009 LB 2/23 2.132 (0.500 – 9.098) 0.307 2/23 1.365 (0.327 – 5.697) 0.669 2/23 0.963 (0.233 – 3.971) 0.958 2/23 1.258 (0.302 – 5.234) 0.753 Modified LA 9/321 0.314 (0.136 – 0.726) 0.007 11/321 0.233 (0.113 – 0.478) o0.0001 16/321 0.263 (0.144 – 0.481) 0.0001 11/321 0.218 (0.108 – 0.441) o0.0001 Modified LB 8/96 2.217 (0.0.945 – 5.200) 0.07 11/96 2.036 (0.995 – 4.167) 0.052 16/96 2.351 (1.285 – 4.300) 0.005 14/96 2.733 (1.406 – 5.314) 0.003

Abbreviations: CI ¼ confidence interval; CR ¼ crude rate; DMFS ¼ distant metastasis-free survival; ER+ ¼ oestrogen receptor positive; HR ¼ hazard ratio; IBTR ¼ ipsilateral breast tumour recurrence; LA ¼ luminal A; LB ¼ luminal B; LRR ¼ locoregional recurrence; PR+ ¼ progesterone receptor positive. LA: ER+ and/or PR+ and HER2; LB: B ER+ and/or PR+ and HER2+; modified LA: ER+ and/or PR+, Ki-67 low, p53, HER2; modified LB: ER+ and/or PR+ and/or Ki-67 high and/or p53+ and/or HER2+. Bold typescript indicates statistical significance.

(BCSS: HR 4.98, 95% CI 2.530–9.694, Po0.0001). p53 þ predicted predicted recurrence with no other prognostic variable or intrinsic DMFS and BCSS (HR 3.523, 95% CI 1.731–7.168, P ¼ 0.0005) but subtype reaching statistical significance. not IBTR or LRR. Locoregional recurrence Kaplan–Meier analysis of intrinsic subtype Luminal B (HR 3.612, 95% CI 1.555–8.340, P ¼ 0.003), basal, Kaplan–Meier analysis (log-rank test) comparing modified LA and unclassified and extensive intraduct carcinoma positive were inde- LB alone was significant for all measures of outcome IBTR pendent predictors of outcome in the final resolved model (Table 4). P ¼ 0.02, LRR P ¼ 0.002, DMFS and BCSS both Po0.0001 (Figure 2 inserts). This classifier also showed improvement in the degree of DMFS and BCSS statistical significance between all molecular subtypes compared with the previously reported five biomarker panel, which was Luminal B was an independent predictor of adverse outcome for observed for LRR P ¼ 0.0004 (previously 0.012), DMFS Po0.0001 both metastases and breast cancer-specific death in the final (previously 0.0035) and BCSS P ¼ 0.0001 (previously 0.048) but not resolved models (LB DMFS: HR 3.023, 95% CI 1.501–6.089, for IBTR (P ¼ 0.074, previously 0.346, Figure 2). Although LA had P ¼ 0.002; BCSS: HR 3.617, 95% CI 1.629–8.031, P ¼ 0.002), along an excellent prognosis, LB had adverse survival, similar to basal, with LVI, LN positivity, basal and unclassified (Table 4). HER2-enriched and unclassified subtypes. DISCUSSION MVA for IBTR, LRR, DMFS and BCSS Oestrogen receptor-positive early breast cancer is the commonest We then constructed multivariable models of clinicopathological form of the disease and tailoring treatment to individual patients is features and intrinsic subtype to assess predictive value and a priority. It is important to identify ER þ patients with a good compare HRs between intrinsic subtypes, using modified LA as a prognosis who will receive most benefit from endocrine therapy reference group. and receive little or no benefit from chemotherapy, and, therefore, avoid any toxicity. In addition, it is also beneficial to identify Ispilateral breast tumour recurrence patients who will have little or no benefit from endocrine therapy. GEP studies have consistently identified at least two groups of Only margin status (HR 3.158, 95% CI 1.067–9.348, P ¼ 0.378) and ER þ tumours; the less favourable LB group being characterised grade (HR 3.13, 95% CI 1.4–7.012, P ¼ 0.0055) independently by higher histological grade and higher expression of proliferation

& 2011 Cancer Research UK British Journal of Cancer (2011), 1 – 9 Predicting failure in luminal breast cancer EKA Millar et al 6 Ipsilateral breast tumour recurrence Locoregional recurrence 1 1

0.8 0.8

1 0.6 0.6 1 0.8 0.8 0.6 0.6 0.4 0.4 0.4 p =0.02 0.4 0.2 p =0.002 p = 0.074 p = 0.0004 0.2

Cumulative survival 0.2 0 0.2 0 0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 0 0 0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140

Distant metastasis-free survival Breast cancer-specific survival 1 1

0.8 0.8

0.6 1 0.6 1 0.8 0.8 0.4 0.6 0.4 0.6 0.4 p <0.0001 0.4 p <0.0001 p <0.0001 0.2 p <0.0001 0.2 Cumulative survival 0.2 0.2 0 0 0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 0 0 0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 Time in months Time in months Figure 2 Kaplan–Meier estimates for ipsilateral breast tumour recurrence, locoregional recurrence, distant metastasis-free survival and breast cancer- specific survival for all intrinsic subtypes and for luminal A vs luminal B (inserts). Luminal A n ¼ 321, luminal B n ¼ 96, basal n ¼ 52, HER2 enriched n ¼ 13, unclassified n ¼ 16.

and HER2-related genes, such as MKI67, MYBBL2, CCNB1, HER2 predictor of poor prognosis in MVA for LRR, DMFS and BCSS and GRB7, and lower levels of ER-related genes. Although but not for IBTR for the whole cohort. As well as demonstrating its there is some consistency in the recognition of these differing superior predictive power over the most frequently used classifier subgroups between GEP studies, there is some doubt as to the or ER, PR, HER2 alone, we also performed additional analyses to stability of the classifiers used by different single sample predictors make a comparison with ER þ breast cancer classified by hormone (Weigelt et al, 2010b) and most assays are not yet ready for routine receptor (HR) status alone (data not shown). Some studies have clinical use (De Ronde et al, 2010). As a result, a simple and shown a significant difference in outcome between double-positive relatively cheap test using IHC surrogates would be easier to (i.e., ER þ PR þ ) and single-receptor positive HR status (i.e., transfer into clinical practise. Various combinations of markers ER þ PR or ER PR þ , Rakha et al, 2007). This latter group may have been assessed to develop a robust IHC panel for routine correspond to the LB subtype (Rakha et al, 2009). Our further pathology reporting, most recently adding Ki67 to ER, PR, HER2 to analyses of these subgroups demonstrated that HR status alone better assess proliferative luminal tumours (Cheang et al, 2009; was inferior to our updated five biomarker classifier: in univariate Hugh et al, 2009). Assessing ER þ tumours with surrogates for analysis good prognosis double-positive status (ER þ PR þ ) was molecular grade may strengthen patient selection as histological only statistically predictive for distant metastases and death (not grade can be compromised in some specimens because of sub- IBTR or LRR) and single-positive status (i.e., poor prognosis ‘LB’) optimal fixation. was not predictive for any measure of outcome in univariate Using an independent discovery cohort of 292 patients, we analysis. identified a significant difference in expression in Ki-67 and p53 Our updated classification of ER þ disease also improves the within ER þ cancers, which was associated with differences statistical significance in survival between all intrinsic subtypes, in clinical outcomes (breast-cancer specific death; CM McNeil where the adverse survival and HR of our poor prognosis ‘LB’ et al, manuscript in preparation). These findings were subse- group is three times that of ‘LA’ and closer to that of HER2- quently validated in a detailed analysis of 498 early breast cancer enriched and basal subtypes. One limitation of this study is that patients, in which we compared good and poor prognosis ‘LA’ and recurrence rates may be over estimated for LB, as the prognosis of ‘LB’ IHC signatures, which included Ki-67 and p53 in addition to HER2-positive LB tumours (24% of all LB tumours) would ER, PR and HER2. This updated definition provided superior currently be modified by the benefits of Herceptin treatment predictive power and better discrimination between the two (which was not used in this study) and an underestimate for LA, as groups of luminal tumours for all measures of outcome. In all, only 44.9% of patients received adjuvant tamoxifen. An additional 73 previously LA tumours were reclassified as LB, increasing the limitation of this study is the difference in cut points used for Ki67 size of the ‘LB’ group by 4four-fold from 4.6 to 19.7% of the positivity where the training cohort median was 5% and the cohort, better reflecting GEP estimates of the size of the LB validation cohort median was 10%. Although we have identified population. Using this definition, ‘LB’ was an independent good and poor prognostic groups with our signature, the relatively

British Journal of Cancer (2011), 1 – 9 & 2011 Cancer Research UK Predicting failure in luminal breast cancer EKA Millar et al 7 Table 4 Cox proportional hazards multivariate models number of events may provide narrower confidence intervals, which along with assessment of the hazard ratio will determine the Variable HR 95% CI P likely clinical significance derived from this panel of markers. These findings suggest a potential role for this biomarker panel Locoregional recurrence in better defining groups of ER þ cancer of low and high Grade 3 1.938 0.823 –4.568 0.130 Size420 mm 0.861 0.408 –1.817 0.694 molecular grade, allowing better selection of patients for endocrine LN+ 2.188 1.054 –4.542 0.036 therapy alone or with AC. Although Ki67 alone identifies LVI 1.286 0.546 –3.026 0.564 approximately 60% of LB tumours, p53 adds a further 20% of EIC+ 3.136 1.328 –7.405 0.009 cases, 12% are positive for both markers, 8% are negative for both but HER2 positive. This study builds upon previous work (Cheang Subtype et al, 2009) using a cut point for optimal determination of ‘high’ Modified LA (reference) 1.0 Ki-67 proliferation rate at 14% through correlation with the Modified LB 2.483 0.982 –6.281 0.055 PAM50 classifier using RT–PCR. They identified a LB population, Basal 3.939 1.281 –12.114 0.017 which was 42% of the cohort (includes their LB and luminal HER2 HER2 1.931 0.382 –9.754 0.426 Unclassified 4.471 0.926 –21.59 0.062 cases). Although the cut point of 14% correlates with GEP estimates it may, in practical terms, be difficult to discern by Resolved model IHC. Ki67 has long been analysed in breast cancer cohorts with EIC+ 2.476 1.070 –5.730 0.034 varied results in terms of its predictive value. A recent review has Modified LB 3.612 1.555 –8.340 0.003 recommended its inclusion as a routine biomarker in breast cancer Basal 5.541 2.279 –13.47 o0.001 (Yerushalmi et al, 2010), but its application as a stand alone HER2 3.549 0.764 –16.51 0.106 biomarker has been debated (Stuart-Harris et al, 2008). Therefore, Unclassified 4.913 1.077 –22.42 0.040 its inclusion in a panel to help define molecular grade and better subtype ‘LA’ and ‘LB’ cancers is independently prognostic and Distant metastasis free survival Grade 3 1.100 0.529 –2.287 0.879 valuable. However, its role as a predictive marker appears less Size420 mm 1.372 0.742 –2.540 0.313 certain. A pre- and post-biopsy analysis of endocrine treated LN+ 3.822 2.036 –7.175 o0.001 breast cancer has demonstrated that only the post-treatment LVI 1.832 0.960 –3.499 0.067 tumour Ki67 (at 2 weeks) was predictive of response to endocrine therapy, whereas baseline Ki67 was not (Dowsett et al, 2007). High Subtype Ki67 status in BIG 1–98 suggested a potential benefit in selecting Modified LA (reference) 1.0 letrozole over tamoxifen in post-menopausal patients (Viale et al, Modified LB 2.872 1.326 –6.222 0.007 2008). Most recently a significant study identified that the Basal 3.273 1.139 –9.396 0.028 prognostic information provided by ‘IHC4’ (ER, PR, HER2 and HER2 1.825 0.386 –8.639 0.448 Unclassified 9.902 3.269 –29.99 o0.001 Ki-67) was at least equivalent to Oncotype Dx (Cuzick et al, 2009) and highlights the relevance of these readily available routine Resolved model pathology markers in the clinical management of breast cancer. LN+ 4.013 2.154 –7.477 o0.001 p53 overexpression in breast cancer assessed by IHC is, rather LVI 2.011 1.075 –3.764 0.029 over simplistically, assumed to act as a surrogate for TP53 mutations Modified LB 3.023 1.501 –6.089 0.002 and is associated with higher tumour grade and responsiveness to Basal 3.902 1.657 –9.191 0.002 radiotherapy, chemotherapy and endocrine therapy (Thompson and HER2 2.064 0.472 –9.026 0.336 o Lane, 2010). Although the p53 pathway is undoubtedly highly Unclassified 10.87 3.882 –30.461 0.001 complex, its assessment by IHC does appear to provide meaningful Breast cancer specific death information. p53 mutations are more frequent in the LB group Grade 3 1.307 0.570 –2.997 0.527 compared with LA (Weigelt et al, 2010a), being described in 71% of Size420 mm 1.879 0.927 –3.807 0.080 LB tumours but only 16% of LA (Sorlie, 2004). p53 currently LN+ 4.535 2.153 –9.553 o0.001 features as one of five antibodies in the Mammostrat (Clarient, Inc., LVI 2.085 1.030 –4.223 0.041 Aliso Viejo, CA, USA) IHC test shown to be of predictive value in ER þ , tamoxifen-treated early breast cancer (Ring et al, 2006; Subtype Bartlett et al, 2010). Mammostrat uses a five IHC panel (p53, Modified LA (ref) 1.0 HTF9C, CEACAM5, NDRG1, SLC7A5) with an algorithm that is Modified LB 3.084 1.280 –7.431 0.012 independent of ER and PR status to identify low-, medium- and Basal 3.780 1.155 –12.37 0.028 HER2 2.095 0.412 –10.65 0.373 high-risk groups. The initial published study (Ring et al,2006) Unclassified 8.167 1.997 –33.40 0.003 demonstrated HRs of 1.8 and 2.3 (training and validation cohorts, respectively) for high risk compared with the low and medium risks Resolved model for disease recurrence. Elevated expression of p53 was observed by LN+ 4.906 2.353 –10.22 o0.001 IHC in our cohorts and appeared to be a useful classifier and was LVI 2.518 1.267 –5.004 0.008 included in the updated definition of poor prognosis ‘LB’ cancer. Modified LB 3.617 1.629 –8.031 0.002 Although the number of events was small, additional explora- o Basal 5.715 2.173 –15.03 0.001 tory multivariate analyses for patients treated with tamoxifen alone HER2 2.907 0.641 –13.17 0.166 (n ¼ 169, 10 events) showed that the poor prognosis ‘LB’ definition Unclassified 10.37 2.801 –38.42 o0.001 retained independent prognostic significance in the final resolved Abbreviations: CI ¼ confidence interval; EIC+ ¼ extensive intraduct component of model for breast cancer specific death (HR 5.361, 95% CI 1.418– DCIS ¼ ductal carcinoma in situ;HR¼ hazard ratio; LA ¼ luminal A; LB ¼ luminal B; 20.25, P ¼ 0.013). This finding suggests that ‘LB’ has five times the LN ¼ lymph node; LVI ¼ lymphatic vascular invasion. Bold typescript indicates risk of death compared with ‘LA’ in patients treated with endocrine statistical significance. therapy. The predictive value of this classification would however require further testing within the setting of a randomised trial of wide confidence intervals, which reflect the small numbers of endocrine therapy. events, strongly suggests the importance of further independent Our updated definition of ER þ cancer translates into an IBTR- validation. Further analyses in a larger data set with a greater free survival at 5 years of 99.3% for LA and 96.6% LB, LRR-free

& 2011 Cancer Research UK British Journal of Cancer (2011), 1 – 9 Predicting failure in luminal breast cancer EKA Millar et al 8 survival 99.7 and 93.4%. A similar recent study using ER, PR and those at risk of early relapse who may benefit from more frequent Ki67 in the definition for LA and LB found local recurrence-free follow-up and early intervention with alternative therapies and/or rates at 10 years of 92% for LA and 90% for LB (Voduc et al, 2010). chemotherapy. Further, larger studies in randomised clinical trials Importantly, our findings further support the observations of this of endocrine therapy are required to assess the clinical utility of group, who found that LB was associated with increased risk of this classification and its value as a predictor of therapeutic LRR. These results highlight the role of proliferation and grade, responsiveness. mirrored by the Oncotype Dx assay (Mamounas et al, 2005), as a predictor of locoregional recurrence, and may help further refine patient selection regarding therapy for optimal locoregional ACKNOWLEDGEMENTS control. A subsequent study analysed patterns of metastases and found both LA and LB had a predilection for bone as a metastatic We thank the National Health and Medical Research Council of site and found that LB had a distant relapse rate similar to basal Australia (Program Grant 535903, Project Grant 535947, the tumours at 15 years (Kennecke et al, 2010). In summary, this study Fellowship 427601 RLS), the Cancer Institute New South Wales suggests that good and poor prognosis ER þ breast cancers can be (Translational Program Grant 10/TPG/1-04), the Cancer Australia reliably and easily discriminated using Ki67 and p53 in addition (Project Grant 626201), the Petre Foundation and the RT Hall to ER, PR and HER2 in routine pathology IHC. This definition Trust. greatly enhances the detection of poor prognosis ER þ ‘LB’ breast cancers, with an outcome closer to that of basal and HER2- enriched tumours. This approach may help more reliably define Supplementary Information accompanies the paper on British groups of ER þ patients with an excellent prognosis and identify Journal of Cancer website (http://www.nature.com/bjc)

REFERENCES

Abd El-Rehim DM, Ball G, Pinder SE, Rakha E, Paish C, Robertson JF, Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) (2005) Effects Macmillan D, Blamey RW, Ellis IO (2005) High throughput protein of chemotherapy and hormonal therapy for early breast cancer on expression analysis using tissue microarray technology of a large recurrence and 15-year survival: an overview of the randomised trials. well-characterised series identifies biologically distinct classes of breast Lancet 365: 1687–1717 cancer confirming recent cDNA analyses. Int J Cancer 116: 340–350 Goldhirsch A, Ingle JN, Gelber RD, Coates AS, Thu¨rlimann B, Senn HJ Bartlett JMS, Thomas J, Ross DT, Seitz RS, Ring BZ, Beck RA, Pedersen HC, (2009) Thresholds for therapy: highlights of the international expert Munro A, Kunkler IH, Campbell FM, Jack W, Kerr GR, Johnstone L, consensus on the primary therapy of early breast cancer. Ann Oncol 20: Cameron DA, Chetty UL (2010) Mammostrat as a tool to stratify breast 1319– 1329 cancer patients at risk of recurrence during endocrine therapy. Breast Hu Z, Fan C, Oh DS, Marron JS, He X, Qaqish BF, Livasy C, Carey LA, Cancer Res 12: R47 Reynolds E, Dressler L, Nobel A, Parker J, Ewend MG, Sawyer LR, Wu J, Blows FM, Driver KE, Schmidt MK, Broeks A, van Leeuwen FE, Wesseling J, Liu Y, Nanda R, Tretiakova M, Ruiz Orrico A, Dreher D, Palazzo JP, Cheang MC, Gelmon K, Nielsen TO, Blomqvist C, Heikkila¨ P, Heikkinen Perreard L, Nelson E, Mone M, Hansen H, Mullins M, Quackenbush JF, T, Nevanlinna H, Akslen LA, Be´gin LR, Foulkes WD, Couch FJ, Wang X, Ellis MJ, Olopade OI, Bernard PS, Perou CM (2006) The molecular Cafourek V, Olson JE, Baglietto L, Giles GG, Severi G, McLean CA, portraits of breast tumors are conserved across microarray platforms. Southey MC, Rakha E, Green AR, Ellis IO, Sherman ME, Lissowska J, BMC Genomics 7: 96 Anderson WF, Cox A, Cross SS, Reed MW, Provenzano E, Dawson SJ, Hugh J, Hanson J, Cheang MC, Nielsen TO, Perou CM, Dumontet C, Reed J, Dunning AM, Humphreys M, Easton DF, Garcı´a-Closas M, Caldas C, Krajewska M, Treilleux I, Rupin M, Magherini E, Mackey J, Martin M, Pharoah PD, Huntsman DL (2010) Subtyping of breast cancer by Vogel C (2009) Breast cancer subtypes and response to docetaxel in immunohistochemistry to investigate a relationship between subtype and node-positive breast cancer: use of an immunohistochemical definition short and long term survival: a collaborative analysis of data for 10,159 in the BCIRG 001 trial. J Clin Oncol 27: 1168–1176 cases from 12 studies. PLoS Med 7: e1000279 Jacquemier J, Charafe-Jauffret E, Monville F, Esterni B, Extra JM, Carey LA (2010) Through a glass darkly: advances in understanding breast Houvenaeghel G, Xerri L, Bertucci F, Birnbaum D (2008) Association cancer biology, 2000–2010. Clin Breast Cancer 10: 188–195 of GATA3, p53, Ki67 status and vascular peritumoral invasion are Carey LA, Perou CM, Livasy CA, Dressler LG, Cowan D, Conway K, strongly prognostic in luminal breast cancer. Breast Cancer Res 11: R23 Karaca G, Troester MA, Tse CK, Edmiston S, Deming SL, Geradts J, Kennecke H, Yerushalmi R, Woods R, Cheang MC, Voduc D, Speers CH, Cheang MC, Nielsen TO, Moorman PG, Earp HS, Millikan RC (2006) Nielsen TO, Gelmon K (2010) Metastatic behaviour of breast cancer Race, breast cancer subtypes and survival in the Carolina breast cancer subtypes. J Clin Oncol 28: 3271– 3277 study. JAMA 295: 2492–2502 Lo´pez-Knowles E, O’Toole SA, McNeil CM, Millar EKA, Qiu MR, Crea P, Cheang MC, Chia SK, Voduc D, Gao D, Leung S, Snider J, Watson M, Musgrove EA, Sutherland RL (2010) PI3K pathway activation in breast Davies S, Bernard PS, Parker JS, Perou CM, Ellis MJ, Nielsen TO (2009) cancer is associated with the basal-like phenotype and cancer-specific Ki67 index, HER2 status, and prognosis of patients with luminal B breast mortality. Int J Cancer 126(5): 1121–1131 cancer. J Natl Cancer Inst 101: 736–750 Mamounas E, Tang G, Bryant J (2005) Association between the 21-gene Cheang MC, Voduc D, Bajdik C, Leung S, McKinney S, Chia SK, Perou CM, recurrence score assay (RS) and risk of locoregional failure in node- Nielsen TO (2008) Basal like breast cancer defined by five biomarkers negative, ER-positive breast cancer: results from NSABP B-14 and has superior prognostic value than triple-negative phenotype. Clin NSABP B-20. 28th Annual San Antonio Breast Cancer Symposium Cancer Res 14: 1368–1376 December 8–11 2005; San Antonio, TX (abstract 29) Cuzick J, Dowsett M, Wale C, Salter J, Quinn E, Zabaglo L, Howell A, Millar EK, Anderson LR, McNeil CM, O’Toole SA, Pinese M, Crea P, Buzdar A, Forbes JF (2009) Prognostic value of a combined ER, PgR, Morey AL, Biankin AV, Henshall SM, Musgrove EA, Sutherland RL, Butt Ki67, HER2 immunohistochemical (IHC4) score and comparison with AJ (2009a) BAG-1 predicts patient outcome and tamoxifen responsive- the GHI recurrence score – results from TransATAC. Cancer Res 69: 503S ness in ER-positive invasive ductal carcinoma of the breast. Br J Cancer (abstract) 100: 123–133 De Ronde J, Wessels L, Wesseling J (2010) Molecular subtyping ready to Millar EK, Graham PH, O’Toole SA, McNeil CM, Browne L, Morey AL, use? Lancet Oncol 11: 306– 307 Eggleton S, Beretov J, Theocharous C, Capp A, Nasser E, Kearsley JH, Dowsett M, Smith IE, Ebbs SR, Dixon JM, Skene A, A’Hern R, Salter J, Detre Delaney G, Papadatos G, Fox C, Sutherland RL (2009b) Prediction of S, Hills M, Walsh G, IMPACT Trialists Group (2007) Prognostic value local recurrence, distant metastases, and death after breast-conserving of Ki67 expression after short term presurgical endocrine therapy for therapy in early-stage invasive breast cancer using a five-biomarker primary breast cancer. J Natl Cancer Inst 99: 167–170 panel. J Clin Oncol 27: 4701–4708

British Journal of Cancer (2011), 1 – 9 & 2011 Cancer Research UK Predicting failure in luminal breast cancer EKA Millar et al 9 Musgrove EA, Sutherland RL (2009) Biological determinants of endocrine expression patterns of breast carcinomas distinguish tumor subclasses resistance in breast cancer. Nat Rev Cancer 9: 631–643 with clinical implications. Proc Natl Acad Sci USA 98: 10869–10874 Neilsen TO, Hsu FD, Jensen K, Cheang M, Karaca G, Hu Z, Hernandez- Sorlie T, Tibshirani R, Parker J, Hastie T, Marron JS, Nobel A, Deng S, Boussard T, Livasy C, Cowan D, Dressler L, Akslen LA, Ragaz J, Johnsen H, Pesich R, Geisler S, Demeter J, Perou CM, Lønning PE, Brown Gown AM, Gilks CB, van de Rijn M, Perou CM (2004) Immunohisto- PO, Børresen-Dale AL, Botstein D (2003) Repeated observation of breast chemical and clinical characterization of the basal-like subtype of tumor subtypes in independent gene expression data sets. Proc Natl Acad invasive breast cancer. Clin Cancer Res 10: 5367–5374 Sci USA 100: 8418–8423 Nguyen PL, Taghian AG, Katz MS, Niemierko A, Abi Raad RF, Boon WL, Sotiriou C, Pusztai L (2009) Gene-expression signatures in breast cancer. Bellon JR, Wong JS, Smith BL, Harris JR (2008) Breast cancer subtype N Engl J Med 360: 790–800 approximated by estrogen receptor, progesterone receptor, and HER-2 Stuart-Harris R, Caldas C, Pinder SE, Pharoah P (2008) Proliferation is associated with local and distant recurrence after breast conserving markers and survival in early breast cancer: a systematic review and therapy. J Clin Oncol 26: 2373–2378 meta-analysis of 85 studies in 32 825 patients. Breast J 17: 323–334 Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, Baehner FL, Walker MG, Thompson AM, Lane DP (2010) p53 transcriptional pathways in breast Watson D, Park T, Hiller W, Fisher ER, Wickerham DL, Bryant J, cancer: the good, the bad and the complex. J Pathol 220: 401–403 Wolmark N (2004) A multigene assay to predict recurrence of Viale G, Giobbie-Hurder A, Regan MM, Coates AS, Mastropasqua MG, tamoxifen-treated, node-negative breast cancer. N Engl J Med 351: Dell’Orto P, Maiorano E, MacGrogan G, Braye SG, Ohlschlegel C, Neven 2817– 2826 P, Orosz Z, Olszewski WP, Knox F, Thu¨rlimann B, Price KN, Castiglione- Perou CM, Sørlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Gertsch M, Gelber RD, Gusterson BA, Goldhirsch A, Breast International Pollack JR, Ross DT, Johnsen H, Akslen LA, Fluge O, Pergamenschikov Group Trial 1–98 (2008) Prognostic and predictive value of centrally A, Williams C, Zhu SX, Lønning PE, Børresen-Dale AL, Brown PO, reviewed Ki-67 labelling index in postmenopausal women with endo- Botstein D (2000) Molecular portraits of human breast tumours. Nature crine-responsive breast cancer: results from Breast International Group 406: 747–752 trial 1– 98 comparing adjuvant tamoxifen with letrozole. J Clin Oncol 26: Rakha EA, El Sayed ME, Green AR, Paish EC, Powe DG, Gee J, 5569– 5575 Nicholson RI, Lee AHS, Robertson JFR, Ellis IO (2007) Biologic and Voduc KD, Cheang MCU, Tyldesley S, Gelmon K, Nielsen TO, Kennecke H clinical characteristics of breast cancer with single hormone receptor- (2010) Breast cancer subtypes and the risk of local and regional relapse. positive phenotype. J Clin Oncol 25: 4772–4778 J Clin Oncol 28: 1684– 1691 Rakha EA, Reis-Filho JS, Baehner F, Dabbs DJ, Decker T, Eusebi V, Fox SB, Weigelt B, Baehner FL, Reis-Filho JS (2010a) The contribution of gene Ichihara S, Jacquemier J, Lakhani SR, Palacios J, Richardson AL, expression profiling to breast cancer classification, prognostication and Schnitt SJ, Schmitt FC, Tan PH, Tse GM, Badve S, Ellis IO (2010) Breast prediction: a retrospective of the last decade. J Pathol 220: 263–280 cancer prognostic classification in the molecular era: the role of WeigeltB,MackayA,A’hernR,NatrajanR,TanDS,DowsettM,AshworthA, histological grade. Breast Cancer Res 12: 207 Reis-Filho JS (2010b) Breast cancer molecular profiling with single Rakha EA, Reis-Filho JS, Ellis IO (2009) Combinatorial biomarker sample predictors: a retrospective analysis. Lancet Oncol 11: 339–349 expression in breast cancer. Breast Can Res Treat 120: 293–308 Weigelt B, Reis-Filho JS (2010) Molecular profiling currently offers no more Ring BZ, Seitz RS, Beck R, Shasteen WJ, Tarr SM, Cheang MC, Yoder BJ, than tumour morphology and basic immunohistochemistry. Breast Budd GT, Nielsen TO, Hicks DG, Estopinal NC, Ross DT (2006) Cancer Res 12(Suppl 4): 55 Novel prognostic immunohistochemical biomarker panel for estrogen Yamashita H, Toyama T, Nishio M, Ando Y, Hamaguchi M, Zhang Z, receptor-positive breast cancer. J Clin Oncol 24: 3039–3047 Kobayashi S, Fujii Y, Iwase H (2006) p53 protein accumulation predicts Sorlie T (2004) Molecular portraits of breast cancer: tumour subtypes as resistance to endocrine therapy and decreased post-relapse survival in distinct disease entities. Eur J Cancer 40: 2667– 2675 metastatic breast cancer. Breast Cancer Res 8: R48 Sørlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie T, Yerushalmi R, Woods R, Ravdin PM, Hayes MM, Gelmon KA (2010) Ki67 Eisen MB, van de Rijn M, Jeffrey SS, Thorsen T, Quist H, Matese JC, in breast cancer: prognostic and predictive potential. Lancet Oncol 11: Brown PO, Botstein D, Eystein Lønning P, Børresen-Dale AL (2001) Gene 174–183

& 2011 Cancer Research UK British Journal of Cancer (2011), 1 – 9

Supplementary Table 1. Patient baseline characteristics, treatments and outcomes. Characteristic No of patients (%) Median Range Length of follow-up (months) 498 84 1-134 Age (years) 61 24-84 Tumor size (mm) 16 1-60 T1a (1- 5) 4 (0.8) T1b (6 -10) 77(16.3) T1c (11-20) 270(54.2) T2 (21-50) 136(27.3) T3 (>50) 1(0.2) Tumor Grade 1 167 (33.5) 2 185 (37.1) 3 145 (29.1) Lymph node metastases 146(29.3) N0 339(69.9) N1 (1-3) 128 (25.7) N2 (4-10) 17(3.5) N3 (>10) 2 (0.4) LN unsampled 12 (2.4) ER+ 393 (78.9) PR+ 334( 68.3) HER-2 amplified (FISH) 36 (7.2) Luminal A 394 (79.1) Modified Luminal A 321 (64.5%) Luminal B 23 (4.6) Modified Luminal B 96 (19.3%) Basal-like 52 (10.4) HER-2 13 (2.6) Unclassified 16 (3.2) Triple negative 68 (13.6) Margin + 17 (3.4) Cavity boost positive 247 (49.5) Cavity boost negative 251 (50.5) Endocrine therapy 223 (44.7) Chemotherapy 117 (23.4) Endocrine & chemotherapy 48 (9.6) Patients with IBTR 24 (4.8) Patients with LRR 35 (7) Patients with distant metastases 47 (9.4) Breast cancer specific deaths 37 (7.4) 5 year IBTR free survival 97.4% 5 year LRR free survival 95.6% 5 year DDFS 92.9% 5 year breast cancer-specific survival 96.3% SupplementaryTable2.CrudeeventratesforLuminalAandLuminalBaccordingtolymphnodeandlymphaticvascularstatus.

IBTRLRR DistantMetastases BreastCancerDeath

n 5yr(%) 10yr(%) 5yr(%) 10yr(%) 5yr(%) 10yr(%) 5yr(%) 10yr(%)

LNͲ LA223 1/223(0.4) 6/223(2.7) 2/223(0.9) 8/223(3.6) 2/223(2.7) 6/223(2.7)) 1/223(0.4) 4/223(1.8)

LB60 1/60(1.7) 2/60(3.3) 2/60(3.3) 4/60(6.7) 4/60(6.7) 5/60(8.3) 2/60(3.3) 3/60(5)

LN+ LA86 1/86(1.2) 3/86(3.5) 1/86(1.2) 3/86(3.5) 7/86(8.1) 10/86(11.6) 0/86(0) 7/86(8.1)

LB36 2/36(5.5) 5/36(13.9) 4/36(11.1) 6/36(16.7) 8/36(22.2) 11/36(30.5) 6/36(16.7) 11/36(30.5)

LVI+ LA43 0/43(0) 0/43(0) 0/43(0) 0/43(0) 3/43(6.9) 4/43(9.3) 0/43(0) 3/43(6.9)

LB23 1/23(4) 3/23(13) 4/23(17.4) 5/23(21.7) 5/23(21.7) 8/23(34.7) 4/23(17.4) 8/23(34.7)

IBTR:ipsilateralbreasttumorrecurrence,LRR:locoregionalrecurrence,LNͲ:lymphnodenegative,LN+:lymphnodepositive,LVI+:lymphaticvascular invasionpositive,LA:luminalAER+and/orPR+,HER2Ͳ,KI67low,p53Ͳ;LB:luminalB:ER+and/orPR+andHER2+and/orKi67highand/orp53+.LNstatus wasnotassessedin12LApatients.

Radiation Oncology Australasian Radiology (2006) 50, 578–582

Supraclavicular radiotherapy must be limited laterally by the coracoid to avoid significant adjuvant breast nodal radiotherapy lymphoedema risk

P Graham, R Jagavkar†, L Browne and E Millar‡ Cancer Care Centre, St George Hospital, University of New South Wales, Sydney, New South Wales, Australia

SUMMARY

This cross-sectional study aimed to investigate the effect of supraclavicular fossa (SCF) radiotherapy volumes as well as patient characteristics and nodal pathology on the development of lymphoedema. Ninety-one women who had received SCF nodal radiotherapy after axillary dissection were evaluated. Lymphoedema was defined by two measurements: limb volume difference 200 mL, or circumference difference 10 cm proximal or distal to the olecranon .2 cm. On univariate analysis, the addition of axillary to SCF radiotherapy, increasing width of the SCF field, increasing age, presence of extracapsular extension of nodal involvement and use of hormone treatment was associated with lymphoedema by either one or both definitions. For both definitions of lymphoedema, on multivariate analysis, increasing nodal radiotherapy volume remained significant (P = 0.02 to 0.007), as did increased age (P = 0.05 to 0.001). We conclude that conventionally fractionated SCF radiotherapy limited laterally by the coracoid process has a lymphoedema risk similar to that expected from axillary dissection alone and a lower risk than wider SCF fields with or without an axillary boost.

Key words: breast lymphoedema, radiotherapy.

INTRODUCTION often used to avoid the potential lymphoedema risk of axillary Adjuvant postmastectomy loco-regional radiotherapy confers irradiation in addition to dissection. However, the evidence that substantial local control and also potential survival advantages. SCF irradiation does not add to the lymphoedema risk of axillary The necessity of treating all nodal volumes to achieve these dissection is limited. This study was conducted to assess how benefits is uncertain and remains the subject of ongoing stu- SCF radiotherapy volumes affect the risk of lymphoedema. dies. One of the major potential morbidities from this treatment approach is lymphoedema. Lymphoedema rates of approxi- METHODS mately 10% occur after axillary irradiation alone or after axillary This study was undertaken with ethics approval from the dissection alone; however, the frequency increases to over South Eastern Sydney Area Health Service. All participants 30% when surgery and radiotherapy are combined.1 Supra- gave written consent. One hundred and ninety-seven women clavicular fossa (SCF) irradiation without axillary irradiation is recorded as having received nodal as well as breast or chest

P Graham MB BS, FRANZCR, Cert Biothics, Grad Dip Med Stat; R Jagavkar MB BS, FRANZCR; L Browne BSc, PhD, Grad Dip Med Stat, AStat; E Millar FRCPath, FRCPA. Correspondence: Associate Professor P Graham, Cancer Care Centre, Short Street, St George Hospital, Kogarah, NSW 2217, Australia. Email: [email protected] Present address: †R Jagavkar, Department of Radiotherapy, St Vincents Hospital, Sydney, New South Wales, Australia. ‡E Millar, Department of Anatomical Pathology, South Eastern Laboratory Service, St George Hospital, Sydney, New South Wales, Australia. Submitted 16 January 2006; accepted 10 April 2006. doi: 10.1111/j.1440-1673.2006.01658.x

ª 2006 The Authors Journal compilationª 2006 Royal Australian and New Zealand College of Radiologists SUPRACLAVICULAR RADIOTHERAPY LYMPHOEDEMA 579 wall radiotherapy between 1993 and 2000 on the department field limited by the coracoid process. The remaining 78 patients database were invited to attend for formal measurement of arm were considered to have had at least partial axillary irradiation if circumferences. Circumference measurements were taken the SCF field extended lateral to the coracoid process. An AB every 10 cm from the extended fingertips. From these, volumes posterior field was used in 56 patients with a dose of 50 Gy were calculated according to the disc volume method.2 Lym- prescribed to the midplane in 24 fractions. The humeral head phoedema was defined as .200 mL difference in calculated was shielded whenever fields extended over the humerus. volume (VOLD).3 A secondary measure defined lymphoedema as 2 cm difference in circumferences 10 cm above or below Statistical methods the olecranon (DPD).4 As patients were not randomized into treatment groups, com- Selection of patients for nodal radiotherapy and extent was parison of patient baseline characteristics were by one-way by treating clinician discretion; however, cases were routinely ANOVA or by x2 test. The primary a priori hypothesis was that presented for discussion at a multidisciplinary team meeting. SCF-only irradiation resulted in less lymphoedema compared Department guidelines included a threshold of .3 lymph nodes with SCF plus axillary irradiation. Volume difference and DPD positive or a combination of other factors such as lympho- were analysed using logistic regression. Univariate correlations vascular invasion with medial tumours when fewer than four are reported. Multivariate analysis initially included all variables nodes were involved. Criteria for adding a posterior axillary boost with P , 0.25 and in a backwards stepwise process non- (AB) included no axillary surgery or a small sample, .70% of significant variables were excluded. lymph nodes positive, .10 lymph nodes involved or extracapsular extension. After 1999, extracapsular extension was disre- RESULTS garded as an indication at the multidisciplinary team meetings. Of 197 women recorded to have received nodal radiotherapy, Most postmastectomy radiotherapy patients underwent nodal 106 were still alive and also agreed to attend for measurements. radiotherapy but the extent was variable. This was particularly Of these, six had not received SCF nodal radiotherapy and the case for those with T stage .2 but node negative. were excluded. A further nine were excluded, as the indication Photons (6 MeV) were used for all radiotherapy fields. for nodal radiotherapy was management of the axilla without Supraclavicular fossa fields were matched to tangential chest surgery. None of the assessed women had evidence of loco- wall fields using a half beam block on the SCF and couch rota- regional failure. Volume difference measurements were obtained tion on the tangential fields. Supraclavicular fossa fields for 85 women and DPD measurements for 89. Baseline patient received 50 Gy in 25 fractions typically prescribed to a depth and treatment characteristics are summarized in Table 1. of 2 cm. The medial border extended to the lateral border of the Overall, the rate of VOLD-defined lymphoedema was 36/85 pedicles. Thirteen patients had the lateral border of the SCF (42%) and DPD-defined lymphoedema was 40/89 (45%).

Table 1. Baseline patient demographics, pathology and treatment

Group SCF Wide SCF Axillary boost All P-value n 13 22 56 91 (if less than 0.25) Mean Range Mean Range Mean Range Mean Range

Age (years) 43 33–81 57 25–83 58 34–90 56 25–90 0.003 Body mass index 26 18–33 27 19–43 27 16–40 27 16–43 — Nodes sampled 17 12–27 20 10–37 16 1–33 17 1–37 — Nodes positive 4 0–9 3 0–11 9 0–30 7 0–30 0.004 Width of field (cm) 9.5 8–11 15.5 10–18 16.5 11–20 15.3 8–20 ,0.001 Time since XRT (years) 3.7 1.5–6.9 3.4 1.4–7.8 4.6 0.8–7.6 4.2 0.8–7.8 0.008

n % n % n % n %

Dominant hand (Y : N)† 9:4 69:31 11:10 52:48 29:27 52:48 49:41 54:46 — Tumour stage (3,4:1,2) 4:9 31:69 5:17 23:77 13:43 23:77 22:69 24:76 — Nodal status (positive : negative) 10:3 77:23 16:6 73:27 53:3 95:5 79:12 87:13 0.02 Number of positive nodes 0:8:5 0:62:38 1:9:12 5:41:54 19:26:11 34:46:20 20:43:28 22:47:31 0.002 (.10:4–10:0–3) Extracapsular extension (Y : N) 5:7 42:58 8:14 36:64 44:9 83:17 57:30 66:34 ,0.001 Chemotherapy (Y : N) 11:2 85:15 13:9 59:41 34:22 61:39 58:33 64:36 0.24 Hormone therapy (Y : N) 7:6 54:46 18:4 82:18 43:13 77:23 68:23 75:25 0.16

†Dominant hand indicates if surgery was on the side of the dominant hand. Number of cases with missing values: body mass index 4, dominant hand 1, extracapsular extension 4. SCF, supraclavicular fossa; XRT, radiotherapy; Y : N, yes : no; —, not applicable.

ª 2006 The Authors Journal compilationª 2006 Royal Australian and New Zealand College of Radiologists 580 P GRAHAM ET AL.

Table 2. Presence of lymphoedema scored by VOLD and 10 cm DPD

SCF Wide SCF Axillary boost P-value n % (95%CI†) n % (95%CI) n % (95%CI)

VOLD 1/13 8 (0–36) 7/21 33 (11–55) 28/51 55 (41–69) 0.003 DPD 2/13 15 (2–45) 6/22 27 (7–47) 32/54 59 (46–73) 0.002

†Binomial exact. CI, confidence interval; DPD, distal/proximal circumference difference; SCF, supraclavicular fossa; VOLD, volume difference.

The use of axillary irradiation increased lymphoedema rates section. Consequently, they may elect to omit the properly significantly whether defined by the use of a posterior AB field dissected axilla from irradiation volumes even when regional (P = 0.004) or subdivided by any irradiation lateral to the cora- radiotherapy is used, on the assumption that this avoids the coid (P = 0.002; Table 2). The severity of lymphoedema was increased lymphoedema risk. Although this appears to be a greater in wide-field SCF or SCF 1 AB (Table 3). The results of reasonable assumption, there is very little evidence in the pub- univariate analysis of other potential predictive factors are in lished reports which specifically supports this assumption. Tables 4 and 5. Other factors associated with lymphoedema Our study has limited numbers, but is unique in that it spe- by both definitions were increased age, extracapsular extension cifically addresses the technical or geometric issues of SCF and field width. Hormone therapy was also associated with radiotherapy and also has used full arm volume measurements DPD. Multivariate analysis resulted in axillary irradiation and to evaluate the end-point of lymphoedema. The result does increasing age remaining the only significant factors for lympho- support the hypothesis that SCF radiotherapy volumes, which edema measured by VOLD or DPD (Table 6). fully exclude axillary irradiation, will lower the lymphoedema risk substantially. Our study suggests that SCF fields which stray DISCUSSION more laterally than the coracoid will increase the risk of lympho- In women who have not had any nodal radiotherapy, expected edema above that of fields restricted medial to a boundary rates of lymphoedema are low. In our own prospective random- which we defined radiologically by the coracoid process. Com- ized breast boost trial, lymphoedema has been monitored using pared with our breast boost series of patients treated without the DPD definition.5 Of 217 women available for 4-year assess- SCF radiotherapy, SCF radiotherapy without axillary radio- ment (approximating the median follow up in this lymphoedema therapy does not appear to give a substantially higher risk of study) who had undergone axillary dissection but no nodal lymphoedema than axillary dissection without SCF radiotherapy. radiotherapy, 21 (10%) had lymphoedema. The push for senti- One study which compared lymphoedema rates for SCF nal node surgery in early stage breast cancer relates to the fields versus SCF fields plus a posterior AB reported no signif- substantial morbidity of axillary dissection, including lymphoe- icant difference with rates of lymphoedema 6 of 15 (40%) and dema. For more advanced breast cancer, more extensive radio- 42 of 136 (31%), respectively.1 However, the SCF fields in that therapy is commonly used. In addition to the well established study were not restricted to volumes medial to the coracoid and local control advantages of loco-regional radiotherapy, there routine shielding of the humeral head was required. That study are several, large randomized trials which show survival advant- did not use limb volumes for comparison, but did subdivide ages using full nodal radiotherapy (internal mammary chain, lymphoedema rates into minimal, moderate and severe axilla and SCF) as well as chest wall irradiation and systemic (2 cm, 2–4 cm and .4 cm difference in arm circumferences therapy.6 These reported relatively low morbidity rates attached 10 cm above or below the olecranon). Five of six (83%) versus to the radiotherapy, possibly because of less extensive nodal 14 of 42 (33%) (P = 0.03; Fisher’s exact test) had only mild surgery reflected in lower numbers of lymph nodes in the dis- lymphoedema for SCF versus SCF 1 AB, respectively. This sections. Many radiation oncologists and patients remained indirectly supports our observation that a SCF field, which concerned by other evidence of substantial increases in lympho- includes tissue lateral to the coracoid, has lymphoedema con- edema when axillary irradiation is added to axillary dis- sequences intermediate between SCF only versus SCF 1 AB.

Table 3. Severity of lymphoedema

SCF Wide SCF Axillary boost P-value

Volume difference (cc) n Mean n Mean n Mean 13 35 21 235 51 314 0.03 Distal or proximal difference .4cm:4cm n % n % n % 0:13 0:100 2:20 9:91 12:42 22:78 0.09

SCF, supraclavicular fossa.

ª 2006 The Authors Journal compilation ª 2006 Royal Australian and New Zealand College of Radiologists SUPRACLAVICULAR RADIOTHERAPY LYMPHOEDEMA 581

Table 4. Univariate analysis of lymphoedema scored by volume dif- Table 6. Multivariate models ference n Odds 95% P-value Global n Odds 95% P- Global ratio confidence P-value ratio confidence value P-value interval interval Lymphoedema scored by volume difference Treatment field 85 — — — 0.003* Treatment field 85 — — — — Wide SCF versus AB 21:51 0.41 0.14–1.18 0.10 — Wide SCF 21:51 0.39 0.13–1.16 0.09 — SCF versus AB 13:51 0.07 0.01–0.57 0.01 — versus AB Age (years) 85 1.05 1.01–1.08 0.008* — SCF versus AB 13:51 0.10 0.01–0.88 0.04 — Body mass index 82 1.03 0.95–1.11 0.50 — Age (years) 85 1.04 1.00–1.07 0.05 0.02 Number of nodes sampled 85 0.93 0.87–1.00 0.06** — Lymphoedema scored by proximal or distal measures Width of field (cm) 85 1.19 1.01–1.41 0.04* — Treatment field 89 — — — — Time since XRT (years) 85 1.11 0.86–1.44 0.42 — Wide SCF 22:54 0.19 0.05–0.64 0.008 — Dominant hand versus 45:39 1.28 0.53–3.06 0.58 — versus AB non-dominant hand SCF versus AB 13:54 0.24 0.04–1.43 0.12 — T 3,4 versus 1,2 20:65 1.15 0.42–3.16 0.78 — Age (years) 89 1.07 1.03–1.12 0.001 0.007 Number of nodes positive 85 — — — 0.73 AB, axillary boost; SCF, supraclavicular fossa; —, not applicable. 4–10 versus 0–3 38:27 0.95 0.35–2.59 0.92 — .10 versus 0–3 20:27 1.45 0.45–4.66 0.53 — ECE versus no ECE 52:29 2.43 0.91–6.47 0.075** — Another study reported arm volume differences of .300 mL in Chemotherapy versus 54:31 0.83 0.34–2.03 0.69 — 3 of 30 (10%) women treated with SCF and parasternal radio- none therapy compared with 21 of 57 (37%) who also had axillary Hormone therapy versus 63:22 2.42 0.84–7.00 0.102 — radiotherapy.7 The technique was substantially different and none unconventional in fractionation to either our own or the study , , *P 0.05, **P 0.10. AB, axillary boost; ECE, extracapsular nodal of Chua et al., as it involved 6 fractions to a posterior axilla field extension; SCF, supraclavicular fossa; XRT, radiotherapy; —, not and 9 fractions to the anterior SCF field to a total dose of 45 Gy applicable. in 15 fractions.1 Hinrichs et al. reported a non-significant increase in lymphoedema rates with SCF irradiation 29 versus 16% without SCF irradiation; however, the paper provides no information on the radiotherapy technique or field definition, Table 5. Univariate analysis of lymphoedema scored by proximal or particularly the lateral extent of the field.8 The addition of a pos- distal measure terior AB was significant with rates of lymphoedema 47% with n Odds 95% P-value Global AB versus 23% without AB. Another recent publication that ratio confidence P-value reported increased lymphoedema with SCF nodal irradiation interval did not distinguish this from SCF plus AB in the summary Treatment field 89 — — — 0.002* because the SCF-only patients represented a small subgroup Wide SCF versus AB 22:54 0.26 0.09–0.76 0.014 — of patients treated with nodal radiotherapy.9 More importantly, SCF versus AB 13:54 0.13 0.03–0.62 0.011 — however, the SCF technique included coverage of the axilla, as Age (years) 89 1.07 1.03–1.11 ,0.001* — Body mass index 85 1.05 0.97–1.13 0.28 — the field extended over the humeral head, irrespective of whether Number of nodes sampled 89 0.97 0.91–1.03 0.32 — an AB was used. None the less, this series reported no lympho- Width of field (cm) 89 1.27 1.06–1.51 0.008* — edema in the (wide) SCF group without an AB. Zero lympho- Time since XRT (years) 89 1.27 0.98–1.63 0.07** — edema seems a little improbable as the risk with axillary surgery Dominant hand versus 47:41 0.94 0.40–2.17 0.88 — alone was approximately 2% and may reflect underscoring as non-dominant hand T stage 3,4 versus 1,2 21:68 1.15 0.43–3.07 0.78 — a result of the retrospective nature of the report. More com- Number of nodes positive 89 — — — 0.29 monly, studies report higher rates of lymphoedema after axillary 4–10 versus 0–3 42:27 1.82 0.67–4.96 0.24 — dissection, generally closer to 5–10% or even higher.4,9–14 .10 versus 0–3 20:27 2.44 0.74–8.04 0.14 — Although this study supports the hypothesis that SCF ir- ECE versus no ECE 56:29 3.63 1.33–9.85 0.012* — radiation only avoids the morbidity of axillary irradiation, it does Chemotherapy versus 58:31 0.54 0.22–1.31 0.17 — not answer the question of whether this approach also reduces none Hormone therapy versus 66:23 3.01 1.06–8.59 0.039* — the potential survival or quality-of-life benefits that accrue from none regional adjuvant radiotherapy. Following modified radical

*P , 0.05, **P , 0.10. AB, axillary boost; ECE, extracapsular nodal mastectomy without radiotherapy, axillary failure following ade- extension; SCF, supraclavicular fossa; XRT, radiotherapy; —, not quate axillary dissection is much less common than SCF failure, applicable. which ranks second after chest wall failure that dominates

ª 2006 The Authors Journal compilationª 2006 Royal Australian and New Zealand College of Radiologists 582 P GRAHAM ET AL. loco-regional failure patterns (13%, 35%, 46%, respectively with axillary dissection and radiation. Breast Cancer Res Treat with 6% internal mammary chain).15,16 In addition, the majority 1992; 21: 139–45. 5. Graham P, Capp A, Fox C et al. Why a breast boost should remain of isolated axillary failures (73–100%) are salvaged by further a controversial aspect of routine breast conservation management surgery, whereas isolated SCF failures achieve durable sal- in Australia and New Zealand in 2002. Australas Radiol 2003; 47: 1,16–19 vage in only a small minority of patients. Finally, SCF fail- 44–9. ure has a high likelihood of severe uncontrolled symptoms such 6. Van de Steene J, Soete G, Storme G. Adjuvant radiotherapy for as brachial plexopathy and severe lymphoedema. If the rate of breast cancer significantly improves overall survival: the missing SCF failure is 2–9% depending on factors such as degree of link. Radiother Oncol 2000; 55: 263–72. 7. Segerstrom K, Bjerle P, Graffman S, Nystrom A. Factors that influ- nodal involvement and if 60% fail to be salvaged, then uncon- ence the incidence of brachial oedema after treatment of breast trolled symptoms as a result of SCF failure can be expected to cancer. Scand J Plast Reconstr Surg Hand Surg 1992; 26: 223–7. occur in 1–6% of patients with node-positive disease.1,7 Provided 8. Hinrichs CS, Watroba NL, Rezaishiraz H et al. Lymphedema sec- that SCF-only irradiation does not add to the lymphoedema risk ondary to postmastectomy radiation: incidence and risk factors. of axillary dissection, the risk of morbid SCF nodal failure clearly Ann Surg Oncol 2004; 11: 573–80. 9. Powell SN, Taghian AG, Kachnic L, Coen JJ, Assaad SI. Risk of balances the morbidity risks of SCF irradiation that may occur if lymphedema after regional nodal irradiation with breast conserva- added to breast or chest wall irradiation and axillary surgery. The tion therapy. Int J Radiat Oncol Biol Phys 2003; 55: 1209–15. most serious risk of SCF radiotherapy is brachial plexopathy, 10. Ozaslan C, Kuru B. Lymphedema after treatment of breast cancer. which has an incidence in the order of 1% or less.1 Am J Surg 2004; 187: 69–72. Of all the other putative risk factors for lymphoedema 11. Kissin MW, Querci-della-Rovere G, Easton D, Westbury G. Risk of analysed in this study, only age remained significant on multivar- lymphoedema following the treatment of breast cancer. Br J Surg 1986; 73: 580–84. iate analysis. In this respect, our findings are very similar to those 12. Lilegren G, Holmberg L. Arm morbidity after sector resection and 9 of Powell et al. The similarity of findings for both definitions of axillary dissection with or without post-operative radiotherapy in lymphoedema is reassuring, both for investigators who rely on breast cancer stage I. Results from a randomized trial. Uppsala- the simpler DPD measurement as VOLD is impractical in larger Orebro Breast Cancer Study Group. Eur J Cancer 1997; 33: studies, and for confidence in the general conclusion of this study. 193–9. 13. Hoe AL, Iven D, Royle GT, Taylor I. Incidence of arm swelling We conclude that limiting the lateral border of SCF radiother- following axillary clearance for breast cancer. Br J Surg 1992; apy by the coracoid process will probably reduce the risk of post- 79: 261–2. mastectomy nodal irradiation lymphoedema to a level, which is 14. Siegal BM, Mayzel KA, Love SM. Level I and II axillary dissection similar to that of axillary dissection without radiotherapy. Our in the treatment of early-stage breast cancer. An analysis of 259 subgroup of restricted SCF patients is small and larger confirma- consecutive patients. Arch Surg 1990; 125: 1144–7. tory investigations of this aspect of radiotherapy are merited. 15. Rangan AM, Ahern V, Yip D, Boyages J. Local recurrence after mastectomy and adjuvant CMF: implications for adjuvant radiation therapy. Aust N Z J Surg 2000; 70: 649–55. 16. Fisher B, Redmond C, Fisher ER et al. Ten-year results of a randomized clinical trial comparing radical mastectomy and total REFERENCES mastectomy with or without radiation. N Engl J Med 1985; 312: 1. Chua B, Ung O, Boyages J. Competing considerations in regional 674–81. nodal treatment for early breast cancer. Breast J 2002; 8: 15–22. 17. Fowble B, Solin LJ, Schultz DJ, Goodman RL. Frequency, sites of 2. Latchford S, Casley-Smith JR. Estimating limb volumes and relapse, and outcome of regional node failures following conser- changes in peripheral edema from circumferences measured at vative surgery and radiation for early breast cancer. Int J Radiat different intervals. Lymphology 1997; 30: 161–4. Oncol Biol Phys 1989; 17: 703–10. 3. Carati CJ, Anderson SN, Gannon BJ, Piller NJ. Treatment of post- 18. Jackson SM. Carcinoma of the breast: the significance of supra- mastectomy lymphedema with low-level laser therapy. A double clavicular lymph node metastases. Clin Radiol 1966; 17: 107–14. blind, placebo-controlled trial. Cancer 2003; 98: 1114–22. 19. McKinna F, Gothard L, Ashley S, Ebbs SR, Yarnold JR. Lymphatic 4. Gerber L, Lampert M, Wood C et al. Comparison of pain, motion, relapse in women with early breast cancer: a difficult management and edema after modified radical mastectomy vs local excision problem. Eur J Cancer 1999; 35: 1065–9.

ª 2006 The Authors Journal compilation ª 2006 Royal Australian and New Zealand College of Radiologists British Journal of Cancer (2009) 100, 123 – 133 & 2009 Cancer Research UK All rights reserved 0007 – 0920/09 $32.00 www.bjcancer.com

BAG-1 predicts patient outcome and tamoxifen responsiveness in ER-positive invasive ductal carcinoma of the breast

,1,2 1 1,3 1,4 1 5 4 1,6,7 EKA Millar* , LR Anderson , CM McNeil , SA O’Toole , M Pinese , P Crea , AL Morey , AV Biankin , 1,6 1,6 1,6 1,6 SM Henshall , EA Musgrove , RL Sutherland and AJ Butt

1Cancer Research Program, Garvan Institute of Medical Research, Darlinghurst, Sydney, New South Wales 2010, Australia; 2Department of Anatomical

Pathology, South Eastern Area Laboratory Service, St George Hospital, Kogarah, New South Wales 2217, Australia; 3Department of Medical Oncology, 4 Westmead Hospital, University of Sydney, Westmead, New South Wales 2145, Australia; Department of Pathology (SydPath), St Vincent’s Hospital, 5 Darlinghurst, Sydney, New South Wales 2010, Australia; Department of Surgical Oncology, St Vincent’s Hospital, Darlinghurst, Sydney, New South Wales 6 7 2010, Australia; Faculty of Medicine, St Vincent’s Clinical School, University of NSW, New South Wales 2052, Australia; Division of Surgery, Bankstown

Hospital, Bankstown, New South Wales 2200, Australia

BAG-1 (bcl-2-associated athanogene) enhances oestrogen receptor (ER) function and may influence outcome and response to

endocrine therapy in breast cancer. We determined relationships between BAG-1 expression, molecular phenotype, response to

tamoxifen therapy and outcome in a cohort of breast cancer patients and its influence on tamoxifen sensitivity in MCF-7 breast

cancer cells in vitro. Publically available gene expression data sets were analysed to identify relationships between BAG-1 mRNA

expression and patient outcome. BAG-1 protein expression was assessed using immunohistochemistry in 292 patients with invasive

ductal carcinoma and correlated with clinicopathological variables, therapeutic response and disease outcome. BAG-1-overexpressing

MCF-7 cells were treated with antioestrogens to assess its effects on cell proliferation. Gene expression data demonstrated a

consistent association between high BAG-1 mRNA and improved survival. In ER þ cancer (n ¼ 189), a high nuclear BAG-1

expression independently predicted improved outcome for local recurrence (P ¼ 0.0464), distant metastases (P ¼ 0.0435), death

from breast cancer (P ¼ 0.009, hazards ratio 0.29, 95% CI: 0.114–0.735) and improved outcome in tamoxifen-treated patients

(n ¼ 107; P ¼ 0.0191). BAG-1 overexpression in MCF-7 cells augmented antioestrogen-induced growth arrest. A high BAG-1

expression predicts improved patient outcome in ER þ breast carcinoma. This may reflect both a better definition of the hormone-

responsive phenotype and a concurrent increased sensitivity to tamoxifen.

British Journal of Cancer (2009) 100, 123–133. doi:10.1038/sj.bjc.6604809 www.bjcancer.com

Published online 9 December 2008 & 2009 Cancer Research UK

Keywords: breast cancer; prognosis; response marker; BAG-1; tamoxifen sensitivity

Breast cancer is a heterogeneous disease with considerable therapy at an earlier stage in the disease process with potential variability in clinical outcome, the prognosis and management of survival benefits. In addition, they may also identify key Molecular Diagnostics which is largely based on histopathological features accompanied mechanisms involved in antioestrogen resistance/sensitivity. by established markers of hormone receptor status, oestrogen and Gene expression profiling has identified intrinsic molecular progesterone receptors (oestrogen receptor (ER), progesterone phenotypes of breast cancer that subclassify ER þ tumours into receptor (PR)), and HER-2 amplification (Sorlie et al, 2001; two main subtypes that predict outcome: luminal A and B, which Goldhirsch et al, 2007). Oestrogen receptor-positive disease can be distinguished by the presence of increased proliferation, comprises approximately 70% of cases and therapies targeting HER-2 amplification and a less favourable prognosis in the latter oestrogen synthesis or the ER are the most effective treatments, group (Sorlie et al, 2001). Signatures that predict outcome in ER þ with adjuvant tamoxifen reducing the annual risk of recurrence disease treated with tamoxifen (Ma et al, 2004; Jansen et al, 2005; and death by up to 47 and 26% respectively (Early Breast Cancer Loi et al, 2008) have been useful in identifying potential new Trialists’ Collaborative Group, 2005) and reducing the risk of predictive biomarkers. However, such molecular testing is contralateral disease by 50% (Fisher et al, 1998). However, the expensive and there is often little overlap between signatures benefits of treatment are limited by intrinsic or acquired from different studies. Furthermore, translating these findings into resistance, which occurs in approximately 40% of ER þ breast clinically useful biomarkers suitable for routine pathology practice cancers (Howell et al, 2005). New predictive biomarkers of is a priority. Ideally, this would be performed using immuno- hormone responsiveness and disease outcome are needed to histochemistry, which is more cost-effective and more easily improve selection of patients for optimal ‘targeted’ endocrine introduced within the existing infrastructure. However, this approach is often limited by the lack of commercially available antibodies for many of these genes. *Correspondence: Dr EKA Millar; E-mail: [email protected] BAG-1 (bcl-2-associated athanogene) is a pro-survival protein Received 16 September 2008; revised 12 November 2008; accepted 12 that can influence diverse biological processes including nuclear November 2008; published online 9 December 2008 hormone receptor function, apoptosis, signal transduction and BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 124 protein turnover (reviewed in Cutress et al (2002)). BAG-1 exists as transferred to the final output file without further processing. One three protein isoforms. The specific ability of the long isoform, BAG-1 probe set was available from each cohort and expression BAG-1L (p50), which possesses a nuclear localisation sequence not data were analysed for frequency distribution of mRNA and its present in the other isoforms, to upregulate the transcriptional association with patient outcome. activity of both ERa and ERb up to five-fold in MCF-7 breast cancer cells (Cutress et al, 2003), is of potential functional and Patient characteristics prognostic significance. BAG-1 is expressed in most normal human tissues (Takayama et al, 1998), and its overexpression BAG-1 protein expression was assessed by immunohistochemistry has been described not only in breast cancer, but also in other in tumours from a cohort of 292 patients diagnosed with invasive human malignancies including squamous cell carcinoma of the ductal breast carcinoma and treated by a single surgeon (PC) head and neck (Shindoh et al, 2000), chronic lymphocytic between February 1992 and August 2002. Formalin-fixed, paraffin- leukaemia (Kitada et al, 1998) and prostate cancer (Maki et al, embedded tissue was retrieved from St Vincent’s Public Hospital 2007), in which it is associated with a poor prognosis. However, its (Sydpath) and St Vincent’s Private Hospital (Douglas Hanly Moir role as a predictive marker in breast cancer has not been Pathology), Sydney, Australia. All tumours were classified as established. Several studies have attempted to relate BAG-1 protein invasive ductal carcinoma of no special type and graded using expression to disease outcome with inconsistent results, which standardised histological criteria (Elston and Ellis, 1991). Lymph may have been the result of low patient numbers, low rates of node status was assessed by axillary sampling and histological ER þ tumours (ER þ rates of 35–52% rather than a currently examination. Follow-up intervals were calculated from the date of expected rate of B70%) and incomplete pathological, clinical and definitive procedure (biopsy/lumpectomy/mastectomy) to the date treatment information. However, the improved prognosis asso- of last-recorded follow-up (median 64 months, range 0–152 ciated with a high BAG-1 expression has earlier been demonstrated months). Patients less than 50 years of age with node-positive, in three studies although with differences in subcellular localisa- ER tumours or tumours larger than 3 cm received adjuvant tion of BAG-1 expression that is, cytoplasmic (Turner et al, 2001), chemotherapy (cyclophosphamide, methotrexate and 5-fluorour- nuclear (Cutress et al, 2003), and cytoplasmic or nuclear (Nadler acil or adriamycin and cyclophosphamide (AC)). Patients with et al, 2008). More recently, BAG-1 featured as one of the 16 cancer- ER þ tumours who were more than 50 years of age received 5 specific genes included in the Oncotype Dx assay (Paik et al, 2004), years of tamoxifen therapy. Breast cancer-specific survival was which predicts distant failure in ER þ , lymph node-negative defined as date of definitive procedure to date of death due to patients treated with tamoxifen using PCR of formalin-fixed breast cancer. Patients who died of causes unrelated to breast paraffin-embedded (FFPE) material. In addition, this assay has cancer were considered as censored at the time of death. Deaths also been used to predict the potential benefit of chemotherapy from unknown causes were excluded from analysis of disease- (Paik et al, 2006) in this group of patients. specific survival. Recurrences were confirmed by imaging and/or As ER-mediated regulation of cell growth, proliferation and histology. Locoregional recurrences were defined as of the survival are key components of breast cancer development, the ipsilateral breast, chest wall, axilla or supraclavicular fossa. Distant role of BAG-1 as a predictive and prognostic marker in breast relapses and metastases were defined as disease in the lungs, liver, oeua Diagnostics Molecular cancer requires further investigation. Consequently, we aimed to brain or distant lymph nodes. These data were obtained from define the relationship of BAG-1 expression with outcome and annual review of patient files or cancer registry data. Tissue response to therapy in a large cohort of early breast cancer patients microarrays (TMAs) of FFPE tumour tissue blocks were of uniform histological type with well-documented treatment and constructed with approximately 80 1 mm cores per slide. Each follow-up data. patient was represented by two to six 1 mm cores. Prior approval for this study was obtained from the Human Research Ethics Committee of St Vincent’s Hospital, Sydney (HREC SVH H94/080, MATERIALS AND METHODS HREC 06336 SVH H00 036). BAG-1 mRNA expression and outcome Immunohistochemistry Publically available gene expression data from two published studies (van de Vijver et al, 2002; Naderi et al, 2007) of breast Four-micron sections were cut from each TMA, mounted on s cancer outcome were analysed to initially identify a potential SuperFrost Plus glass slides and baked for 2 h at 791C, then relationship between BAG-1 mRNA levels and prognosis. The dewaxed by passage through xylene (two 5 min washes), cleared cohorts chosen for these analyses were of similar clinicopatholo- and rehydrated in graded alcohol (100, 95 and 70%) ending in a gical composition to our clinical cohort. The study by Naderi et al distilled water wash. Antigen retrieval was performed using DAKO (2007) comprised 135 patients, 70% of whom were ER þ with a solution (pH 6.0) (s1699; DAKO, Carpentaria, CA, USA) in a median follow-up of 132 months (range 16–160 months). Data pressure cooker (DAKO Pascal Decloaker) for 60 s, followed by were generated using Agilent Human 1A arrays, which were cooling gently for 15 min in a running water bath. Following a available as raw scanner data files and sourced from Array Express thorough wash in distilled water, endogenous peroxidase activity (http://www.ebi.ac.uk/) accession E-UCON-1. Using the limma R was eliminated with 3% hydrogen peroxide for 5 min. Slides were package (R Development Core Team, 2007), background-sub- incubated with BAG-1 mouse monoclonal antibody raised against tracted data were normalised by the global LOESS technique full-length human BAG-1 protein (clone 3.10G3E2; Santa Cruz applied to non-control spots only. To combine information from Biotechnology Inc., Santa Cruz, CA, USA) at a dilution of 1 : 50 for duplicate dye-swap arrays, a linear model was fitted to the 45 min at room temperature. Following buffer wash, detection normalised data using limma (Smyth, 2005). Model fit coefficients employed DAKO Envision þ mouse secondary reagent (DAKO) for each sample were then used as final expression estimates, for 30 min at room temperature, followed by DAKO DAB þ expressed relative to a pooled reference RNA. The second data set, chromagen (DAKO) for 10 min. Slides were then rinsed in water sourced from van de Vijver et al (2002), comprised 295 patients, and counterstained with haematoxylin, dehydrated through graded 76% of which were ER þ , with a median follow-up of 93.6 months ethanol, cleared in xylene and mounted. Normal colon was (range 0.6–220 months). Data were generated using Rosetta NKI- employed as a control tissue that showed positive staining in spotted oligonucleotide arrays and were downloaded from http:// basal crypt cell nuclei and negative staining in the muscularis microarray-pubs.stanford.edu/wound_NKI/ explore.html as log 2- mucosae. A further negative control substituted isotype-matched transformed values in a text table format. Raw data were directly mouse IgG1 at 1 : 100 in place of the BAG-1 monoclonal antibody.

British Journal of Cancer (2009) 100(1), 123 – 133 & 2009 Cancer Research UK BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 125 Oestrogen receptor, PR, cytokeratin 5/6 and EGFR were also (3.10G3E2; Clone Chemicon International Inc., Billerica, MA, USA) stained using the following antibodies: ER, 1 : 100 (clone 6F11; confirmed BAG-1 expression. b-Actin (Sigma, St Louis, MO, USA) DAKO); PR, 1 : 200 (clone PgR 636; DAKO); CK5/6, 1 : 80 (clone was used as a loading control. MAB1602; Chemicon International, Temecula, CA, USA); and EGFR, 1 : 100 (clone H11; DAKO). HER-2 FISH was assessed in the S-phase analysis Australian National Reference Laboratory (Department of Patho- logy, St Vincent’s Hospital, Sydney) using the Vysis PathVysion Exponentially growing MCF-7 cells expressing BAG-1 or vector- HER-2 DNA dual-colour probe kit. An HER2 : chromosome 17 alone control were treated with 1 mmol l 1 4-hydroxytamoxifen ratio42.2 was classified as HER2 amplification. (Sigma) or 10 nmol l 1 ICI 182780 (7a-[9-(4,4,5,5,5-pentafluoro- All assessments of immunohistochemical staining were pentylsulphinyl) nonyl] estra-1,3,5,(10)-triene-3,17b-diol), which performed by observers blinded to the clinical and molecular data was a kind gift from Dr Alan Wakeling (Astra-Zeneca Pharma- and patient outcome. Nuclear and cytoplasmic staining for BAG-1 ceuticals, Alderly Park, Cheshire, UK), or vehicle (ethanol) for was assessed by an experienced breast pathologist (EKAM) and 24 h. Cells were harvested and S phase was analysed by propidium described in terms of the intensity (0: negative, 1 þ : weak, 2 þ : iodide staining and flow cytometry. moderate and 3 þ : strong) and percentage of cells staining positive. From these indices, a simplified ‘H score’ (i.e., intensity percentage of positive nuclei) was calculated for each Statistical analyses core and a mean and median score for each parameter calculated Statistical analyses were performed using Statview 5.0. Software for each tumour (range of two to six cores per patient). Oestrogen (Abacus Systems, Berkeley, CA, USA). A P-value o0.05 was receptor and PR (both double scored) were assessed as positive if accepted as statistically significant. BAG-1 mRNA and protein they had a simplified H score of 410. CK5/6 and EGFR (both expression and its association with clinicopathological variables double scored) were assessed as positive if there was any positive and intrinsic molecular phenotype of breast cancer were tested by cytoplasmic or membranous staining present at any intensity. applying the w2-test of association in contingency tables. Kaplan– Meier and Cox proportional hazards model were used for Definition of intrinsic molecular phenotype of breast univariate analysis and the latter for multivariate analyses. Those cancer factors that were prognostic in univariate analysis were then assessed in a multivariable model to identify factors that were This was assessed immunohistochemically using criteria similar to independently prognostic and those that were the result of those recently described by Cheang et al (2008) but using FISH to confounding variables. determine HER-2 status as follows: luminal A ¼ ER þ and/or PR þ , HER-2; luminal B ¼ ER þ and/or PR þ , HER-2 þ ; HER-2 ¼ ER and PR, HER-2 þ ; basal-like ¼ ER,PR, RESULTS HER-2, CK5/6 þ and/or EGFR þ ; unclassified ¼ negative for all five markers. BAG-1 mRNA expression and outcome To identify an association between BAG-1 gene expression levels Cell culture studies and patient outcome, we examined two published breast cancer gene expression data sets. A frequency distribution of BAG-1 The human ER þ breast cancer cell line, MCF-7, was routinely mRNA expression was used to apply serial cut points using maintained in RPMI-1640 medium supplemented with 5% foetal sequential Kaplan–Meier analysis (log-rank test) to minimise the calf serum, 10 mgml 1 insulin and 2.92 mg ml 1 glutamine under P-value and maximise the difference in survival between the two standard conditions. A cDNA insert encoding human BAG-1 (cat groups of high and low expressions. Using this approach, no. SC107955; OriGene Technologies Inc., Rockville, MD, USA) statistical significance was assessed for death using univariate was cloned into the retroviral vector pMSCV-IRES-GFP (Caldon Kaplan–Meier and Cox proportional hazards analysis (Table 1). et al, 2008). MCF-7 cells transiently expressing the murine The Wound/NKI data set of 295 patients contained a high BAG-1 ecotropic receptor were infected with BAG-1 retrovirus as expression group of 234 patients (79.3%), which was associated described earlier (Debnath et al, 2003). Green fluorescent with improved prognosis in Cox and Kaplan–Meier univariate protein-positive cells were sorted to homogeneity by flow analysis (P ¼ 0.0005, Table 1A and Figure 1A). High BAG-1 Molecular Diagnostics cytometry. Cell lysates were collected as described earlier (Prall expression was not significant in a multivariate model that et al, 1997). Subsequent western blotting using a BAG-1 antibody incorporated standard clinicopathological variables (Table 1B).

Table 1 Association between BAG-1 mRNA expression and breast cancer outcome

High BAG-1 expression n (%) HR 95% CI P-value

(A) Cox univariate analysis for high BAG-1 expression from publicly available gene expression data sets Wound/NKI (van de Vijver et al, 2002) 234/295 (79.3%) 0.439 0.277– 0.697 0.0005 Naderi (Naderi et al, 2007) 108/135 (80%) 0.412 0.212– 0.843 0.0151

(B) Cox multivariate analysis for the Wound cohort (n ¼ 295) Grade42 2.266 1.361– 3.774 0.0017 Size420 mm 1.678 1.039– 2.710 0.0343 ER positive 0.549 0.323– 0.933 0.0267 HER-2 positive 2.319 1.267– 4.244 0.0064 BAG-1 high 0.911 0.530– 1.567 0.7363

CI ¼ confidence interval; ER ¼ oestrogen receptor; HR ¼ hazards ratio.

& 2009 Cancer Research UK British Journal of Cancer (2009) 100(1), 123 – 133 BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 126

A 1 B 1

0.8 n=234 0.8 n=108

0.6 0.6 n=27 0.4 0.4 n=61 0.2 Cumulative survival Cumulative 0.2 survival Cumulative P=0.0003 P=0.0120 0 0 0 20406080100120 140 160 180 0 20 40 60 80 100 120 140 160 Months Months Figure 1 Relationship between BAG-1 mRNA expression and patient outcome. Kaplan–Meier analysis (log-rank test) for breast cancer-specific death in the Wound/NKI (A) and Naderi (B) cohorts. High BAG-1 (K); low BAG-1 (J).

The Naderi cohort of 135 patients contained a high expression 276) of the cohort as ‘high’ BAG-1 expressers and 22% (62 out of group of 108 patients (80%), again associated with a favourable 276) as ‘low’ BAG-1 expressers. This cut point appeared to outcome in Kaplan–Meier (P ¼ 0.0120) and Cox univariate represent a real split in the protein expression data, which matched analyses (P ¼ 0.0151, Table 1A and Figure 1B) but not in that observed from our analysis of the mRNA expression levels. multivariate analysis (data not shown). Using these cut points This distribution was not apparent in the frequency distribution to define high and low expression, further analyses were conducted of nuclear ‘H’ scores. Consequently, we adopted the percentage to determine association between high BAG-1 expression and of positively staining nuclei as the index for further analysis of clinicopathological parameters. In the Wound cohort, high BAG-1 association with outcomes. was associated with ER þ ,PRþ , low histological grade and HER-2 negativity (Po0.0001) but there was no association with tumour size (P ¼ 0.0862) or lymph node status (P40.999). Similarly, the Correlation of BAG-1 expression with clinicopathological Naderi cohort also showed positive associations between high features and intrinsic molecular subtype BAG-1 expression and ER þ (P ¼ 0.0014), HER-2 negativity The relationship between nuclear and cytoplasmic expression of (P ¼ 0.0044) and low histological grade (P ¼ 0.0061) but not with BAG-1 and standard clinicopathological features of the disease are tumour size or lymph node status (P ¼ 0.081 and P ¼ 0.106, oeua Diagnostics Molecular summarised in Table 2A. High expression of BAG-1 showed a respectively). These findings support an association of high BAG-1 significant positive correlation with low histological grade, ER and expression with a luminal A phenotype and improved survival. PR positivity (Po0.0001) and was correlated negatively with HER-2 amplification status (P ¼ 0.001) and the triple-negative phenotype Immunohistochemical analysis of BAG-1 protein (Po0.0001). These findings were apparent for both nuclear and expression in normal breast tissue and invasive ductal cytoplasmic staining at a cut point of 40% positivity of any intensity, carcinoma but with a higher degree of statistical significance for nuclear staining. High nuclear BAG-1 expression was also strongly correlated Representative immunohistochemistry staining patterns and with a luminal A intrinsic phenotype: 73% (154 out of 211) of BAG-1 intensities of BAG-1 are illustrated in Figure 2A–I. Similar ‘high’ were luminal A (Po0.0001, w2-test), but there was no patterns of staining were observed in normal terminal duct lobular correlation with luminal B (P ¼ 0.956). A strong negative correlation units adjacent to cancer (n ¼ 24, 20 patients) and in reduction wasobservedwiththeHER-2(5%,11outof213BAG-1highareof mammoplasty specimens (n ¼ 20, 14 patients). Nuclear staining HER-2 phenotype, Po0.0001) and the basal-like phenotype (7%, 14 was observed in all cases, with a mean of 54% of epithelial cells out of 213 BAG-1 high are basal-like, Po0.0001). (range 10–90%) showing weak-to-strong (1–3 þ ) intensity. Cytoplasmic staining was also present in 63% of cases with a þ range of 0–100% of cells showing 1 or 2 staining. BAG-1 expression and outcome In our cohort of 292 patients, 276 invasive ductal carcinomas were available for BAG-1 analysis due to loss of some tissue cores In the whole cohort (n ¼ 276), high nuclear BAG-1 expression was during processing of the TMAs. Staining was of variable intensity, associated with a favourable prognosis for all measures of outcome which ranged from negative to strong (0–3 þ ) and demonstrated in univariate analysis: local recurrence (P ¼ 0.002), distant both cytoplasmic and nuclear staining in keeping with the known metastases (Po0.0001) and breast cancer-specific death subcellular localisation of the various BAG-1 isoforms (Cutress (Po0.0001, Table 3A and Figure 3). Furthermore, high nuclear et al, 2002). There was, however, no direct correlation between BAG-1 expression was also an independent predictor of outcome nuclear and cytoplasmic expression when modelled as continuous in multivariate analysis for distant metastases (P ¼ 0.0455, variables (R ¼ 0.476). Sixteen out of 276 cases (5.7%) showed no Table 3B) but not for local recurrence or death. To assess whether nuclear staining, whereas 26 cases (9.4%) showed no cytoplasmic BAG-1 was an independent prognostic variable and not the result staining. When assessed for the percentage of positively staining of confounding by other variables, Cox proportional hazards nuclei, there appeared to be two distinct sub-populations, which models were constructed with step-wise removal of redundant could be dichotomised at a cutoff value of 40% positively staining variables until resolution. The resolved multivariate model is nuclei at any intensity (Figure 2J). Cytoplasmic staining displayed presented in Table 3C. High cytoplasmic expression of BAG-1 was a similar pattern (Figure 2K), with most tumours showing at least also associated with improved outcome for local recurrence weak positivity but again with two distinct populations that could (P ¼ 0.0092), distant metastases (P ¼ 0.0013) and death be identified using a 40% cutoff value. By applying the selected cut (P ¼ 0.0046) on Kaplan–Meier univariate analysis, but was not point of 440% mean nuclear staining, we defined 78% (214 out of significant in multivariate analysis.

British Journal of Cancer (2009) 100(1), 123 – 133 & 2009 Cancer Research UK BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 127 ABC

DEF

GHI

J 90 K 90 80 80 70 70 60 60 50 50 40 40 30 30 20 20 Number of patients 10 10 0 0 02040 60 80 100 0 20 40 60 80 100

Percentage of cells with positively Percentage of cells with positively Molecular Diagnostics staining nuclei staining cytoplasm Figure 2 Representative images of BAG-1 immunohistochemistry. (A) Negative staining in high-grade invasive ductal carcinoma (IDC), 400. (B) Weak (1 þ ) nuclear staining in low-grade IDC, 400. (C) Moderate (2 þ ) nuclear and weak (1 þ ) cytoplasmic staining in low-grade IDC, with strong nuclear staining in an adjacent normal duct (arrow). (D) Moderate (2 þ ) cytoplasmic and negative nuclear staining in high-grade IDC. (E) Strong (3 þ ) nuclear and moderate (2 þ ) cytoplasmic staining in high-grade IDC. (F) Strong (3 þ ) nuclear and weak (1 þ ) cytoplasmic staining, weak nuclear staining in normal duct (arrow), 400. (G) Strong 3 þ nuclear staining. (H) Moderate nuclear staining in normal acini. (I) Normal colon, control tissue, which shows moderate positive nuclear staining in basal crypt cell nuclei and negative staining in mucularis mucosae. Frequency distribution of BAG-1 nuclear (J) and cytoplasmic (K) staining using immunohistochemistry in 276 invasive ductal carcinomas. There are two distinct populations that can be dichotomised using a cut point of 40% (arrow), which segregates the cohort into high- and low-expressing subgroups.

Given the relationship between BAG-1 expression and ER status, ER-negative tumours (n ¼ 85), nuclear staining of BAG-1 was not we next assessed the association with outcome in the ER þ and associated with any index of outcome in univariate analysis. As our ER subgroups. Within ER-positive tumours (n ¼ 189), high data demonstrated a strong relationship between high BAG-1 nuclear BAG-1 expression was an independent predictor of expression, ER positivity and the luminal A phenotype, we outcome in both univariate and multivariate analyses (Figure 3 assessed whether BAG-1 expression was associated with a and Table 3D). In the multivariate model employed, which differential response to adjuvant tamoxifen therapy. The data incorporated standard pathological indicators of outcome: tumour reported in Figure 3 demonstrate that patients treated with size, grade, nodal status, PR and HER-2, the resolved model, which tamoxifen (n ¼ 107), whose tumours had a high nuclear BAG-1 eliminates redundant variables, retained HER-2, PR and BAG-1 expression, showed an improved outcome in univariate Kaplan– (Table 3E). Cytoplasmic staining was not significant in univariate Meier analysis for local recurrence (P ¼ 0.032), distant metastases analysis in this group of patients. In the smaller subgroup of (P ¼ 0.019) and breast cancer-specific death (P ¼ 0.038).

& 2009 Cancer Research UK British Journal of Cancer (2009) 100(1), 123 – 133 BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 128 Table 2 Clinicopathological features of the breast cancer cohort and association with BAG-1 expression

Nuclear BAG-1 Cytoplasmic BAG-1

Positive Negative P-value Positive Negative P-value

(A) Clinicopathological characteristics and associations with BAG-1 expression Age 450 135 39 0.979 135 39 0.252 o50 79 23 85 17

Grade 1 and 2 133 17 o0.0001 129 21 0.005 381459135

Size 420 mm 81 31 0.086 82 30 0.027 o20 mm 133 31 138 26

Nodal status Positive 90 30 0.424 98 22 0.429 Negative 121 32 119 34

HER-2 Positive 31 20 0.001 32 19 0.001 Negative 180 40 184 36

ER Positive 169 20 o0.0001 166 23 o0.0001 Negative 44 41 52 33

PR Positive 146 13 o0.0001 141 18 o0.0001 Negative 67 49 78 38

CK5/6 Positive 19 14 0.003 23 10 0.127 Negative 195 48 197 46 oeua Diagnostics Molecular

Triple negative Positive 25 23 o0.0001 31 17 0.0039 Negative 187 37 186 38

(B) Treatment and survival data n (%) Endocrine therapy 144/292 (49.3) Chemotherapy 111/292 (38.0) Endocrine and chemotherapy 71/292 (24.3) Recurrences 75/292 (25.7) Distant metastases 68/292 (23.3) Deaths 67/292 (22.9) Breast cancer-specific deaths 52/292 (17.8) 5-year disease-free survival 74.0% 5-year metastasis-free survival 76.8% 5-year breast cancer-specific survival 86.0%

BAG-1 overexpression and antioestrogen sensitivity MCF-7 BAG-1 cells were treated with 1 mmol l 1 4-hydroxyta- in vitro moxifen, 10 nmol l 1 ICI 182780 or vehicle for 24 h and the percentage of S-phase cells determined by flow cytometry. To provide some potential mechanistic insights into the relation- Antioestrogen-induced cell cycle arrest was enhanced in MCF-7 ship between high BAG-1 expression and improved outcome in BAG-1 cells compared with vector control (Figure 4B). Treatment ER þ tamoxifen-treated patients, we assessed the effect of BAG-1 of MCF-7 BAG-1 cells with the pure oestrogen antagonist ICI overexpression on oestrogen/antioestrogen sensitivity in ER þ 182780 (Po0.005) or the active metabolite of tamoxifen, MCF-7 breast cancer cells. A pool of high BAG-1-expressing MCF- 4-hydroxytamoxifen (Po0.05), in replicate experiments demon- 7 cells was isolated and overexpression of the three major protein strated a significantly enhanced cell cycle arrest as measured by a isoforms (BAG-1L, BAG-1M and BAG-1S) was confirmed by decrease in S phase compared with control cells (Figure 4C). western blotting (Figure 4A). BAG-1 was also overexpressed in a panel of ER þ breast cancer cell lines compared with normal and immortalised breast epithelial cells (Figure 4A), and thus this high- DISCUSSION expressing pool of MCF-7 cells represented an appropriate model to study the biological consequences of BAG-1 overexpression and The recent characterisation of molecular phenotypes of breast was used for all further analyses. cancer defines biological subgroups, independent of histological

British Journal of Cancer (2009) 100(1), 123 – 133 & 2009 Cancer Research UK BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 129 Table 3 Cox univariate and multivariate analyses

n (%) HR 95% CI P-value

(A) Whole-cohort clinicopathological variables (n ¼ 292), univariate analysis of breast cancer-specific death Age450 184/292 (63) 1.427 0.799– 2.551 0.229 Grade42 132/291 (45) 3.100 1.865– 5.163 o0.0001 Size420 mm 117/291 (40) 2.730 1.678– 4.443 o0.0001 LN positive 125/289 (43) 3.968 2.346– 6.774 o0.0001 HER-2 positive 51/273 (18) 2.459 1.463– 4.134 0.0007 ER positive 192/280 (68) 0.395 0.243– 0.642 0.0002 PR positive 161/282 (57) 0.238 0.140– 0.406 o0.0001 BAG-1 high 214/276 (78) 0.364 0.222– 0.598 o0.0001

(B) Whole-cohort (n ¼ 276) distant metastases, Cox multivariate analysis Grade42 1.398 0.751– 2.567 0.2948 Size420 mm 1.564 0.937– 2.610 0.0873 LN status 3.372 1.934– 5.880 o0.0001 HER-2 1.853 1.066– 3.220 0.0287 ER 0.990 0.525– 1.868 0.9745 PR 0.405 0.212– 0.776 0.0064 BAG-1 high 0.559 0.317– 0.989 0.0455

(C) Whole-cohort distant metastases, Cox multivariate analysis, resolved model LN status 3.597 2.097– 6.168 o0.0001 HER-2 1.973 1.158– 3.361 0.0125 PR 0.329 0.186– 0.584 0.0001 BAG-1 high 0.586 0.344– 0.998 0.0493

(D) ER+ (n ¼ 189) breast cancer-specific death, Cox multivariate analysis Grade42 1.529 0.600– 3.896 0.3730 Size420 mm 1.053 0.428– 2.591 0.9100 LN status 1.471 0.571– 3.795 0.4250 HER-2 5.578 2.036– 15.286 0.0008 PR 0.293 0.119– 0.721 0.0076 BAG-1 high 0.290 0.114– 0.735 0.0090

(E) ER+ breast cancer-specific death, Cox multivariate analysis, resolved model HER-2 6.725 2.7 – 16.644 o0.0001 PR 0.239 0.104– 0.547 0.0007 BAG-1 high 0.302 0.122– 0.744 0.0093

CI ¼ confidence interval; ER ¼ oestrogen receptor; HR ¼ hazards ratio. type, which provides a further insight into the disease at a prognosis for local recurrence, distant metastases and death. functional level. Luminal A cancers defined by the presence of ER Furthermore, for breast cancer-specific death, BAG-1 expression and/or PR positivity and HER-2 negativity form a favourable was of superior predictive power to tumour grade, tumour size and prognostic group. However, further defining this group may lymph node status. This group of patients has a strong positive provide new insights into the underlying biology of oestrogen correlation with a luminal A phenotype and low histological grade, sensitivity/resistance and provide clinically useful markers for which suggests that BAG-1 may be a useful surrogate marker of Molecular Diagnostics routine clinical practice. This study demonstrates that a high intact ER signalling and identifies those tumours maintaining a BAG-1 expression identifies a good prognosis group of cancers luminal A-differentiated phenotype. Therefore, BAG-1 is a marker with a luminal A phenotype, which may have enhanced therapeutic with potentially useful prognostic applications in ER þ disease. sensitivity to tamoxifen. Outcome studies, published earlier, of BAG-1 expression using We first addressed the question of identifying an association immunohistochemistry have shown inconsistent results but with a between BAG-1 mRNA expression levels and patient outcome in trend towards improved prognosis with high expression levels. two independent cohorts, which are broadly equivalent to our However, its role as a predictive biomarker has not yet been fully validation cohort in terms of clinicopathological characteristics. defined or adequately validated. The first published study (Tang Using serially determined cut points, we identified two populations et al, 1999) of 140 patients included both early and metastatic of patients with high and low BAG-1 expressions, which correlated disease, ER status was unknown in 38% of patients and only 35% with patient outcome. Thus, the high BAG-1 mRNA expression, of patients were ER þ . Consequently, ER and PR were excluded found within the top 80% of patients, is associated with a from multivariate analysis, which showed that an elevated nuclear favourable outcome. Correspondingly, the frequency distribution BAG-1 expression was associated with shorter disease-free and of immunohistochemically detected BAG-1 protein expression in overall survival, although BAG-1 was not significant in univariate our clinical cohort identifies two distinct subgroups of patients of analysis. These findings were not replicated in a subsequent study similar proportions to those identified in the gene expression by the same group of investigators (Tang et al, 2004). The second profiling analyses. High BAG-1 protein expression, defined as study of 122 patients (Turner et al, 2001) consisted predominantly greater than 40% positive nuclear staining of any intensity, of pre-menopausal patients (mean age, 54 years), only 41% of identified 78% of patients with a good prognosis. The predictive whom had ER þ cancers, and lymph node status was unknown in value of high BAG-1 expression was greatest in ER þ cancer in 48% of cases. In addition, well-documented prognostic indicators, which a high nuclear expression was an independent predictor of such as tumour size, grade, ER, PR and HER-2, were not significant

& 2009 Cancer Research UK British Journal of Cancer (2009) 100(1), 123 – 133 BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 130 Whole cohort ER+ ER+ treated with tamoxifen A 1 n=214 1 n=169 1 n=98 0.8 0.8 0.8 0.6 0.6 0.6 0.4 n=62 0.4 n=20 0.4 n=9 P=0.002 P=0.006 P=0.032 0.2 0.2 0.2 BAG-1 high BAG-1 high BAG-1 high Cumulative survival Cumulative 0 BAG-1 low survival Cumulative 0 BAG-1 low survival Cumulative 0 BAG-1 low 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100

B 1 1 1 n=214 n=169 n=98 0.8 0.8 0.8

0.6 0.6 0.6

0.4 n=62 0.4 n=20 0.4 n=9 P<0.0001 P=0.007 P=0.019 0.2 0.2 0.2 BAG-1 high BAG-1 high BAG-1 high

Cumulative survival Cumulative survival Cumulative 0 survival Cumulative 0 0 BAG-1 low BAG-1 low BAG-1 low 0 20 40 60 80 100 0 20406080100 0 20406080100

C 1 n=214 1 n=169 1 n=98 0.8 0.8 0.8 0.6 0.6 0.6 n=9 0.4 n=62 0.4 0.4 P<0.0001 P=0.0005 n=20 P=0.038 0.2 0.2 0.2

Cumulative survival Cumulative 0 BAG-1 high survival Cumulative BAG-1 high survival Cumulative BAG-1 high BAG-1 low 0 BAG-1 low 0 BAG-1 low oeua Diagnostics Molecular 0 604020 80 100 120 140 04020 8060100 120 140 0 4020 8060100 120 140 Time in months Time in months Time in months Figure 3 Relationship between nuclear BAG-1 protein expression by immunohistochemistry and patient outcome. Kaplan–Meier analyses (log-rank test) for (A) local recurrence, (B) distant metastases and (C) breast cancer-specific death in the whole cohort, ER þ subgroup and ER þ patients treated with tamoxifen stratified by high (K) and low (J) BAG-1 expression.

in univariate analysis. In this study, elevated cytoplasmic BAG-1 expression. This study, therefore, confirms the findings described expression was associated with improved prognosis in a multi- earlier of improved prognosis with high nuclear BAG-1 expression variate model that included ER, BCL-2 and stage. In a more described by Cutress et al (2003) and represents a detailed analysis homogeneous and representative cohort of early breast cancer, of BAG-1 expression and its potential relationship with therapeutic Cutress et al (2003) described improved prognosis in univariate responsiveness in a large cohort of uniform histological type with analysis with high nuclear, but not high cytoplasmic, BAG-1 well-documented clinical outcome. expression in a cohort of 138 patients, 60% of whom were ER þ . The finding of improved responsiveness to tamoxifen and better All patients were treated with surgery and endocrine therapy patient outcome associated with a high expression of BAG-1, a without chemotherapy. The largest and the most recent study of pro-survival antiapoptotic protein, is somewhat counter-intuitive 517 patients (Nadler et al, 2008) used image analysis-based but is mirrored by several studies identifying the overexpression of assessment of immunofluorescent staining and found that both BCL-2, a major target of BAG-1, also being consistently associated high nuclear and cytoplasmic expression of BAG-1 was associated with improved prognosis in low-grade ER þ tumours (Callagy with improved prognosis in the whole cohort and in lymph node- et al, 2006) and also in patients treated with tamoxifen (Linke et al, positive patients only in univariate analysis, with a strong 2006). Furthermore, the strong relationship between high nuclear correlation with ER, PR and Bcl-2. However, again in this study, BAG-1 expression and improved patient outcome reported by only 52% of patients were ER þ , with a predominance of large Cutress et al (2003) emanated from a cohort of tamoxifen-treated tumours (59%42 cm), and details on histological grade and patients. BAG-1, BCL-2 and ER feature among the 16 cancer- treatment were not available. Two other studies were unable to related genes of the Oncotype Dx assay (Paik et al, 2004), which identify any association with outcome in patients treated with predicts distant failure in ER þ lymph node-negative patients hormonal therapy (Townsend et al, 2002) or in a cohort with treated with tamoxifen. The derived recurrence-score algorithm, advanced breast cancer treated with chemotherapy (Sjostrom et al, which is largely weighted towards proliferation-related genes, 2002). There are many possible explanations for these discordant assigns a negative value to the BAG-1 mRNA expression level, in findings: differences in the composition of the clinical cohorts, turn supporting our observation of improved prognosis with high incomplete clinical information, different antigen retrieval expression level. Several other gene expression profiling studies methods, differing monoclonal antibodies used in the detection have identified signatures predictive of outcome in ER þ disease of BAG-1 and divergent cut points used to determine a high or low treated with tamoxifen (Ma et al, 2004; Jansen et al, 2005; Loi et al,

British Journal of Cancer (2009) 100(1), 123 – 133 & 2009 Cancer Research UK BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 131 A 2008). BAG-1 did not feature among genes within these signatures, although there is often a limited overlap in signatures between studies (Miller, 2007). Interestingly, an expression profiling study (Cleator et al, 2006) of pre-treatment biopsies from 40 patients

Normal Immortalised ER+ treated with AC chemotherapy identified BAG-1, BCL-2 and ER among a diverse group of 178 genes overexpressed in sensitive tumours. Thus, the potential role of BAG-1 as a predictive marker of therapeutic responsiveness to both endocrine and chemother-

MCF-7 vector MCF-7 BAG-1 184 184 A1 T-47D ZR-75 MB-134 MB-361 MB-468 MCF-7 apy requires further investigation. This is best performed within BAG-1L the context of randomised clinical trials and these studies are BAG-1 BAG-1M ongoing. BAG-1S The BAG-1 gene is located on chromosome 9p12 and is expressed as three protein isoforms generated through alternative -Actin initiation sites from a single mRNA (Packham et al, 1997). The overexpression of BAG-1 has been described in several breast Vector BAG-1 cancer cell lines (Takayama et al, 1998; Brimmell et al, 1999), with the three isoforms demonstrating differing intracellular localisa- B 400 + Vehicle 600 + Vehicle tions: BAG-1L is predominantly nuclear, BAG-1S is predominantly %G1=56.4 %G1=54.8 cytoplasmic and BAG-1M is present in both the cellular compart- 300 ments (Brimmell et al, 1999). This differential subcellular %S=32.8 400 %S=32.5 %G =8.8 %G =10.8 localisation of BAG-1 isoforms is altered under different experi- 200 2 2 mental conditions, possibly representing a regulatory mechanism 200 for protein activity: BAG-1M relocalises from the cytoplasm to the 100 nucleus following heat shock (Zeiner et al, 1999) and when bound 0 0 to the glucocorticoid receptor, possibly downregulating the 0 200 400 600 8001000 0 200 400 600 8001000 receptor (Schneikert et al, 1999). Nuclear-to-cytoplasmic reloca- lisation of BAG-1M has also been observed during epidermal and neuronal differentiation and during breast epithelial involution,

Cell number 2000 + ICI 182780 1200 + ICI 182780 both in vitro and in vivo (Takayama et al, 1998; Schorr et al, 1999; %G =88.6 %G1=76.3 1 Kermer et al, 2002). BAG-1 possesses a range of pro-survival 900 1500 %S=8.9 %S=14.6 properties through its ability to interact with diverse downstream %G =6.5 %G =5.8 600 2 1000 2 target molecules, originally described by its ability to bind to and enhance the activity of the antiapoptotic protein BCL-2 mediated 300 500 by its binding to the heat-shock proteins HSP70 and HSC70 (Takayama et al, 1995). Differential staining and subcellular 0 0 localisation of the isoforms have not been investigated in the 0 200 400 600 8001000 0 200 400 600 8001000 normal breast or breast cancer and are impaired by the absence of DNA content isoform-specific antibodies, all current studies in breast detecting 1.2 ‘total’ BAG-1 expression. The mechanism of overexpression of C BAG-1 in breast cancer is also not known although in prostate 1.0 cancer it is amplified in 7.4% of hormone-refractory cancers (Maki 0.8 ** et al, 2007). Our data with normal and malignant breast epithelial Vector cell lines confirm overexpression in ER þ carcinoma cell lines 0.6 ** BAG1 compared with normal epithelial cells. The BAG-1S isoform 0.4 appeared to be preferentially overexpressed, but there was 0.2 evidence for elevated expression of all three isoforms in a cell Molecular Diagnostics S-phase fold change S-phase fold line-specific manner. The relative contribution of the respective 0 isoforms to the relationships reported here must await further ICI 182780 4-OHT EtOH studies with isoform-specific antibodies. (10 nM) (1 μM) A key target of BAG-1 is ERa, which when bound by the BAG-1L Antioestrogen treatment isoform increases its transcriptional activity by up to five-fold in Figure 4 Effects of modulating BAG-1 expression on antioestrogen MCF-7 cells (Cutress et al, 2003). The ability of high nuclear BAG-1 sensitivity in MCF-7 breast cancer cells. (A) Immunoblot analysis of cell expression to predict improved outcome in ER þ cancer and also lysates from MCF-7 retrovirally infected pools stably overexpressing BAG-1 in those treated with tamoxifen is of potential mechanistic wild type, two normal breast epithelial and six ER þ breast cancer cell lines. importance as it suggests that it may have a role in responsiveness b -Actin was used as a loading control. Each of the three major BAG-1 to adjuvant endocrine therapy. In our subgroup of ER þ patients protein isoforms are indicated: BAG-1L, BAG-1M and BAG-1S. (B) treated with tamoxifen (n ¼ 107), high BAG-1 expression predicted Representative DNA histograms of MCF-7 cells stably overexpressing BAG-1 compared with control cells after treatment with 10 nmol l 1 ICI improved prognosis, which may indicate sensitivity to therapy or 182780 or vehicle for 24 h. Differences in scale are due to slight differences possibly a better definition of a good prognostic luminal A group in the number of events recorded. (C) Proliferating MCF-7 cells stably of patients. To address the question of whether BAG-1 expression overexpressing BAG-1 were treated with 1 mmol l 1 4-hydroxytamoxifen, could confer enhanced sensitivity to antioestrogen treatment in 10 nmol l 1 ICI 182780 or vehicle for 24 h. Cells were harvested and vitro, we analysed cell cycle arrest in BAG-1-overexpressing MCF-7 S phase was analysed by propidium iodide staining and flow cytometry. The cells. The data presented here demonstrate a significant increase in decrease in S phase was graphed as fold change relative to vehicle-treated sensitivity to the induction of cell cycle arrest by both tamoxifen vector control cells. The bar histograms represent the mean±s.e.m. for o and the pure steroidal antioestrogen ICI 182780. replicate samples from five independent experiments. *P 0.05; As antioestrogen therapy, targeted at oestrogen synthesis **Po0.005. (aromatase inhibitors), or the ER (tamoxifen), is the single most- effective treatment for women with hormone receptor-positive

& 2009 Cancer Research UK British Journal of Cancer (2009) 100(1), 123 – 133 BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 132 disease, the ability to predict likely success or failure of these role in a routine pathology setting as a marker for better defining therapies would enable potential alternative therapeutic strategies luminal A breast cancers and as a therapeutic response marker for to be targeted to a group of patients most likely to fail on ER-targeted therapy with tamoxifen or aromatase inhibitors. tamoxifen or aromatase therapy up-front or at an earlier stage in treatment, which may result in improved outcome. The ability of BAG-1 to predict responsiveness to antioestrogen therapy now merits further investigation by examining the relationship between ACKNOWLEDGEMENTS expression and response in large randomised clinical trials of endocrine therapy in ER þ patients. This study was supported by grants from the Cancer Institute In summary, we have demonstrated that the high BAG-1 NSW, the National Health and Medical Research Council of expression is associated with the luminal A phenotype, is an Australia (NHMRC), the Petre Foundation, the RT Hall Trust. We independent predictor of outcome and may indicate enhanced thank Ms Joanne Scorer for assistance in preparing the paper and responsiveness to tamoxifen in ER þ invasive ductal carcinoma. Drs Davendra Segara, Dianne Adams, Andrew Field and Ms Alice These effects may be related to the ability of BAG-1 overexpression Boulghourjian for assistance with the collection and processing of to confer increased sensitivity to antioestrogens in vitro. These the tissue, the establishment of the breast cancer database and data findings suggest that BAG-1 immunohistochemistry may have a management.

REFERENCES

Brimmell M, Burns JS, Munson P, McDonald L, O’Hare MJ, Lakhani SR, LC, Klijn JG, Berns EM (2005) Molecular classification of tamoxifen- Packham G (1999) High level expression of differentially localized BAG-1 resistant breast carcinomas by gene expression profiling. J Clin Oncol 23: isoforms in some oestrogen receptor-positive human breast cancers. 732–740 Br J Cancer 81: 1042–1051 Kermer P, Krajewska M, Zapata JM, Takayama S, Mai J, Krajewski S, Reed Caldon CE, Swarbrick A, Lee CS, Sutherland RL, Musgrove EA (2008) The JC (2002) Bag1 is a regulator and marker of neuronal differentiation. Cell helix-loop-helix protein Id1 requires cyclin D1 to promote the Death Differ 9: 405–413 proliferation of mammary epithelial cell acini. Cancer Res 68: 3026– 3036 Kitada S, Andersen J, Akar S, Zapata JM, Takayama S, Krajewski S, Wang Callagy GM, Pharoah PD, Pinder SE, Hsu FD, Nielsen TO, Ragaz J, Ellis IO, HG, Zhang X, Bullrich F, Croce CM, Rai K, Hines J, Reed JC (1998) Huntsman D, Caldas C (2006) Bcl-2 is a prognostic marker in breast Expression of apoptosis-regulating proteins in chronic lymphocytic cancer independently of the Nottingham Prognostic Index. Clin Cancer leukemia: correlations with in vitro and in vivo chemoresponses. Blood Res 12: 2468–2475 91: 3379–3389 Cheang MC, Voduc D, Bajdik C, Leung S, McKinney S, Chia SK, Perou CM, Linke SP, Bremer TM, Herold CD, Sauter G, Diamond C (2006) Nielsen TO (2008) Basal-like breast cancer defined by five biomarkers A multimarker model to predict outcome in tamoxifen-treated breast has superior prognostic value than triple-negative phenotype. cancer patients. Clin Cancer Res 12: 1175–1183

oeua Diagnostics Molecular Clin Cancer Res 14: 1368–1376 Loi S, Haibe-Kains B, Desmedt C, Wirapati P, Lallemand F, Tutt AM, Cleator S, Tsimelzon A, Ashworth A, Dowsett M, Dexter T, Powles T, Gillet C, Ellis P, Ryder K, Reid JF, Daidone MG, Pierotti MA, Berns EM, Hilsenbeck S, Wong H, Osborne CK, O’Connell P, Chang JC (2006) Gene Jansen MP, Foekens JA, Delorenzi M, Bontempi G, Piccart MJ, Sotiriou C expression patterns for doxorubicin (Adriamycin) and cyclophospha- (2008) Predicting prognosis using molecular profiling in estrogen mide (cytoxan) (AC) response and resistance. Breast Cancer Res Treat receptor-positive breast cancer treated with tamoxifen. BMC Genomics 95: 229–233 9: 239 Cutress RI, Townsend PA, Brimmell M, Bateman AC, Hague A, Packham G Ma XJ, Wang Z, Ryan PD, Isakoff SJ, Barmettler A, Fuller A, Muir B, (2002) BAG-1 expression and function in human cancer. Br J Cancer 87: Mohapatra G, Salunga R, Tuggle JT, Tran Y, Tran D, Tassin A, Amon P, 834–839 Wang W, Wang W, Enright E, Stecker K, Estepa-Sabal E, Smith B, Cutress RI, Townsend PA, Sharp A, Maison A, Wood L, Lee R, Brimmell M, Younger J, Balis U, Michaelson J, Bhan A, Habin K, Baer TM, Brugge J, Mullee MA, Johnson PW, Royle GT, Bateman AC, Packham G (2003) The Haber DA, Erlander MG, Sgroi DC (2004) A two-gene expression ratio nuclear BAG-1 isoform, BAG-1L, enhances oestrogen-dependent tran- predicts clinical outcome in breast cancer patients treated with scription. Oncogene 22: 4973–4982 tamoxifen. Cancer Cell 5: 607–616 Debnath J, Muthuswamy SK, Brugge JS (2003) Morphogenesis and Maki HE, Saramaki OR, Shatkina L, Martikainen PM, Tammela TL, van oncogenesis of MCF-10A mammary epithelial acini grown in three- Weerden WM, Vessella RL, Cato AC, Visakorpi T (2007) Overexpression dimensional basement membrane cultures. Methods 30: 256–268 and gene amplification of BAG-1L in hormone-refractory prostate Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) (2005) Effects cancer. J Pathol 212: 395–401 of chemotherapy and hormonal therapy for early breast cancer on Miller WR (2007) Do classical oestrogen markers predict for clinical recurrence and 15-year survival: an overview of the randomised trials. response to endocrine therapy? Breast Cancer Res 9: S19 Lancet 365: 1687–1717 Naderi A, Teschendorff AE, Barbosa-Morais NL, Pinder SE, Green AR, Elston CW, Ellis IO (1991) Pathological prognostic factors in breast cancer. Powe DG, Robertson JF, Aparicio S, Ellis IO, Brenton JD, Caldas C (2007) I. The value of histological grade in breast cancer: experience from a A gene-expression signature to predict survival in breast cancer across large study with long-term follow-up. Histopathology 19: 403–410 independent data sets. Oncogene 26: 1507–1516 Fisher B, Costantino JP, Wickerham DL, Redmond CK, Kavanah M, Cronin Nadler Y, Camp RL, Giltnane JM, Moeder C, Rimm DL, Kluger HM, Kluger WM, Vogel V, Robidoux A, Dimitrov N, Atkins J, Daly M, Wieand S, Y (2008) Expression patterns and prognostic value of Bag-1 and Bcl-2 in Tan-Chiu E, Ford L, Wolmark N (1998) Tamoxifen for prevention of breast cancer. Breast Cancer Res 10: R35 breast cancer: report of the National Surgical Adjuvant Breast and Bowel Packham G, Brimmell M, Cleveland JL (1997) Mammalian cells express two Project P-1 Study. J Natl Cancer Inst 90: 1371–1388 differently localized Bag-1 isoforms generated by alternative translation Goldhirsch A, Wood WC, Gelber RD, Coates AS, Thurlimann B, Senn HJ initiation. Biochem J 328(Part 3): 807–813 (2007) Progress and promise: highlights of the international expert Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, Baehner FL, Walker MG, consensus on the primary therapy of early breast cancer 2007. Ann Oncol Watson D, Park T, Hiller W, Fisher ER, Wickerham DL, Bryant J, 18: 1133–1144 Wolmark N (2004) A multigene assay to predict recurrence of tamoxifen- Howell A, Cuzick J, Baum M, Buzdar A, Dowsett M, Forbes JF, Hoctin-Boes G, treated, node-negative breast cancer. N Engl J Med 351: 2817–2826 Houghton J, Locker GY, Tobias JS (2005) Results of the ATAC (Arimidex, Paik S, Tang G, Shak S, Kim C, Baker J, Kim W, Cronin M, Baehner FL, Tamoxifen, Alone or in Combination) trial after completion of 5 years’ Watson D, Bryant J, Costantino JP, Geyer Jr CE, Wickerham DL, adjuvant treatment for breast cancer. Lancet 365: 60–62 Wolmark N (2006) Gene expression and benefit of chemotherapy in Jansen MP, Foekens JA, van Staveren IL, Dirkzwager-Kiel MM, Ritstier K, women with node-negative, estrogen receptor-positive breast cancer. Look MP, Meijer-van Gelder ME, Sieuwerts AM, Portengen H, Dorssers J Clin Oncol 24: 3726– 3734

British Journal of Cancer (2009) 100(1), 123 – 133 & 2009 Cancer Research UK BAG-1 predicts outcome in ER þ breast cancer EKA Millar et al 133 Prall OW, Sarcevic B, Musgrove EA, Watts CK, Sutherland RL (1997) Takayama S, Krajewski S, Krajewska M, Kitada S, Zapata JM, Kochel K, Estrogen-induced activation of Cdk4 and Cdk2 during G1-S phase Knee D, Scudiero D, Tudor G, Miller GJ, Miyashita T, Yamada M, Reed progression is accompanied by increased cyclin D1 expression and JC (1998) Expression and location of Hsp70/Hsc-binding anti-apoptotic decreased cyclin-dependent kinase inhibitor association with cyclin protein BAG-1 and its variants in normal tissues and tumor cell lines. E-Cdk2. J Biol Chem 272: 10882–10894 Cancer Res 58: 3116–3131 R Development Core Team (2007) R: A Langauge and Environment for Takayama S, Sato T, Krajewski S, Kochel K, Irie S, Millan JA, Reed JC Statistical Computing. In R Foundation for Statistical Computing: (1995) Cloning and functional analysis of BAG-1: a novel Bcl-2-binding Vienna, Austria protein with anti-cell death activity. Cell 80: 279–284 Schneikert J, Hubner S, Martin E, Cato AC (1999) A nuclear action of the Tang SC, Beck J, Murphy S, Chernenko G, Robb D, Watson P, Khalifa M eukaryotic cochaperone RAP46 in downregulation of glucocorticoid (2004) BAG-1 expression correlates with Bcl-2, p53, differentiation, receptor activity. J Cell Biol 146: 929–940 estrogen and progesterone receptors in invasive breast carcinoma. Breast Schorr K, Li M, Krajewski S, Reed JC, Furth PA (1999) Bcl-2 gene family Cancer Res Treat 84: 203–213 and related proteins in mammary gland involution and breast cancer. Tang SC, Shehata N, Chernenko G, Khalifa M, Wang X (1999) Expression of J Mammary Gland Biol Neoplasia 4: 153–164 BAG-1 in invasive breast carcinomas. J Clin Oncol 17: 1710–1719 Shindoh M, Adachi M, Higashino F, Yasuda M, Hida K, Nishioka T, Ono M, Townsend PA, Dublin E, Hart IR, Kao RH, Hanby AM, Cutress RI, Takayama S, Reed JC, Imai K, Totsuka Y, Kohgo T (2000) BAG-1 Poulsom R, Ryder K, Barnes DM, Packham G (2002) BAG-i expression expression correlates highly with the malignant potential in early lesions in human breast cancer: interrelationship between BAG-1 RNA, (T1 and T2) of oral squamous cell carcinoma. Oral Oncol 36: 444– 449 protein, HSC70 expression and clinico-pathological data. J Pathol 197: Sjostrom J, Blomqvist C, von Boguslawski K, Bengtsson NO, Mjaaland I, 51–59 Malmstrom P, Ostenstadt B, Wist E, Valvere V, Takayama S, Reed JC, Turner BC, Krajewski S, Krajewska M, Takayama S, Gumbs AA, Carter D, Saksela E (2002) The predictive value of bcl-2, bax, bcl-xL, bag-1, fas, and Rebbeck TR, Haffty BG, Reed JC (2001) BAG-1: a novel biomarker fasL for chemotherapy response in advanced breast cancer. Clin Cancer predicting long-term survival in early-stage breast cancer. J Clin Oncol Res 8: 811– 816 19: 992–1000 Smyth GK (2005) Limma: linear models for microarray data. In van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Voskuil DW, Bioinformatics and Computational biology solutions using R and Schreiber GJ, Peterse JL, Roberts C, Marton MJ, Parrish M, Atsma D, Bioconductor, Gentlemen R, Carey V, Dudoit S, Irizarry R, Huber W Witteveen A, Glas A, Delahaye L, van der Velde T, Bartelink H, (eds), pp 397– 420. Springer: New York Rodenhuis S, Rutgers ET, Friend SH, Bernards R (2002) A gene- Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie T, expression signature as a predictor of survival in breast cancer. N Engl J Eisen MB, van de Rijn M, Jeffrey SS, Thorsen T, Quist H, Matese JC, Med 347: 1999–2009 Brown PO, Botstein D, Eystein Lonning P, Borresen-Dale AL (2001) Gene Zeiner M, Niyaz Y, Gehring U (1999) The hsp70-associating protein Hap46 expression patterns of breast carcinomas distinguish tumor subclasses binds to DNA and stimulates transcription. Proc Natl Acad Sci USA 96: with clinical implications. Proc Natl Acad Sci USA 98: 10869–10874 10194–10199 Molecular Diagnostics

& 2009 Cancer Research UK British Journal of Cancer (2009) 100(1), 123 – 133 Oncogene (2011) 1–12 & 2011 Macmillan Publishers Limited All rights reserved 0950-9232/11 www.nature.com/onc ORIGINAL ARTICLE Identiﬁcation of PUMA as an estrogen target gene that mediates the apoptotic response to tamoxifen in human breast cancer cells and predicts patient outcome and tamoxifen responsiveness in breast cancer

CG Roberts1,9, EKA Millar1,2,3,9, SA O’Toole1,4,5,6, CM McNeil1,7,8, GM Lehrbach1, M Pinese1, P Tobelmann1, RA McCloy1, EA Musgrove1,3, RL Sutherland1,3 and AJ Butt1,3

1Cancer Research Program, Garvan Institute of Medical Research, Darlinghurst, New South Wales, Australia; 2Department of Anatomical Pathology, South Eastern Area Laboratory Service, St George Hospital, Kogarah, New South Wales, Australia; 3School of Medicine, University of Western Sydney, Campbelltown, New South Wales, Australia; 4St Vincent’s Clinical School, Faculty of Medicine, University of New South Wales, Sydney, New South Wales, Australia; 5Department of Tissue Pathology and Diagnostic Oncology, Royal Prince Alfred Hospital, Camperdown, New South Wales, Australia; 6Central Clinical School, Sydney Medical School, University of Sydney, Sydney, New South Wales, Australia; 7Department of Medical Oncology, Westmead Hospital, Westmead, New South Wales, Australia and 8Western Clinical School, University of Sydney, Sydney, New South Wales, Australia

Recognition of the pivotal role of estrogen in the aetiology Oncogene advance online publication, 7 March 2011; of breast cancer has led to the development of antiestro- doi:10.1038/onc.2011.36 gens (AE), such as tamoxifen (TAM) as effective therapies for the treatment and prevention of this disease. Keywords: PUMA; breast cancer; tamoxifen However, despite their widespread clinical efficacy, response to AEs is often short-lived, and acquired or innate therapeutic resistance remains a major obstacle in the successful treatment of breast cancer. Thus, delineat- Introduction ing the intracellular pathways that mediate the cellular response to estrogen could potentially lead to new, more Breast cancer remains the most common cancer of effective approaches to the treatment of breast cancer, women and the second most frequent cause of cancer particularly endocrine-resistant disease. Here, we have deaths, despite a major decline in breast cancer identified the BCL-2 homology 3 (BH3)-only, pro- mortality in past decades. Aberrations in molecular apoptotic regulator, PUMA (p53 upregulated modulator pathways regulating estrogen synthesis and action are of apoptosis) as an estrogen target gene that is acutely primary etiological factors in the pathogenesis of breast downregulated in response to estrogen in breast cancer cancer (Jordan, 2009). This dependence on the estrogen cell lines, independently of their p53 status. PUMA is drive to cell proliferation and increased cell survival, and transcriptionally upregulated following treatment with an understanding of the underlying molecular mechan- TAM, and knock down of PUMA expression in these cells isms, has led to the routine use of the estrogen receptor attenuates the apoptotic response to TAM. Furthermore, (ER) as a biomarker of hormone responsiveness, and low PUMA expression in breast carcinomas is signifi- the development of antiestrogens (AEs), for example, cantly associated with breast cancer-specific death tamoxifen (TAM), and aromatase inhibitors (AIs) as (P ¼ 0.0014 and P ¼ 0.0115, for mRNA and protein, effective therapies for the treatment and prevention of respectively), and worse outcome in TAM-treated patients breast cancer (Swaby et al., 2007). Indeed, the routine (mRNA, P ¼ 1.49e-05). These findings suggest that the use of adjuvant endocrine therapy is one of the major dysregulation of apoptotic signaling pathways such as contributors to the recent decline in breast cancer those executed through PUMA, can significantly impact mortality that has occurred preferentially in ER-positive on both the progression and therapeutic responsiveness of disease (Hurvitz and Pietras, 2008). However, despite its breast cancer. Moreover, they provide a convincing widespread clinical efficacy, response is often short- rationale for exploring new therapeutic approaches lived, and resistance to endocrine therapy remains a involving endocrine and non-endocrine therapies that major obstacle to the successful treatment of this disease target apoptotic pathways as an effective strategy for (Musgrove and Sutherland, 2009). A major challenge in tackling endocrine refractory disease. understanding the molecular basis of endocrine resistance is that although estrogen has been implicated as a major factor in the development and progression of Correspondence: Dr AJ Butt, Apoptosis Research Group, Cancer breast cancer, the downstream effectors of its mitogenic Research Program, Garvan Institute of Medical Research, 384 and pro-survival actions remain to be fully character- Victoria Street, Darlinghurst, New South Wales 2010, Australia. E-mail: [email protected] ized. Thus, the identification and elucidation of intra- 9These authors contributed equally to this work. cellular pathways that mediate the cellular response to Received 30 June 2010; revised and accepted 19 January 2011 estrogen could potentially lead to new, more effective PUMA expression and breast cancer CG Roberts et al 2 approaches to the treatment of breast cancer, particu- referred to hereafter as PUMA), whose role as an larly endocrine-resistant disease. estrogen target gene has not previously been explored. In addition to its well-characterized mitogenic ac- PUMA is a proapoptotic, BH3-only, BCL-2 family tions, estrogen has direct, inhibitory effects on the member and an essential mediator of p53-dependent and apoptotic machinery in breast cancer cells, including -independent apoptosis, induced by a diverse range of transcriptional repression of the proapoptotic molecules signals, including genotoxic stress, deregulated onco- BAK, BIK and caspase-9 (Frasor et al., 2003), and gene expression and growth factor withdrawal (Han increased expression of anti-apoptotic BCL-2 through et al., 2001; Nakano and Vousden, 2001; Yu et al., estrogen-response elements (EREs) in the BCL-2 coding 2001). Once activated, typically by transcriptional sequence (Perillo et al., 2000). There is also evidence of upregulation, PUMA acts at the mitochondria orches- ligand-independent cross-talk between estrogen and trating an intrinsic apoptotic response through the growth factor intracellular signaling cascades, particu- activation of BAX and/or BAK. Current evidence larly those mediated by the erbB family of receptor suggests that PUMA initiates apoptosis by sequestering tyrosine kinases, leading to the activation of survival anti-apoptotic binding partners, such as BCL-2 and pathways through activation of phosphatidylinositol 3- BCL-XL away from BAX and BAK, although some kinase (Butt et al., 2005). Although the involvement of suggestion of direct interaction has also been proposed aberrant apoptotic signaling in the endocrine response is (reviewed in Yu and Zhang, 2009). Once BAX and/or relatively understudied (Butt et al., 2007), such a BAK are activated, PUMA-induced apoptosis follows concept is strongly supported by several clinical studies the classical mitochondrial pathway of membrane that have revealed associations between expression of depolarisation, release of apoptogenic molecules such the death receptor, Fas and Fas ligand (Reimer et al., as cytochrome c and SMAC, and caspase activation. 2002), and expression of the proapoptotic protein, BAD Here we report that PUMA is transcriptionally (Cannings et al., 2007) and resistance to TAM downregulated in response to estradiol in human breast (Musgrove and Sutherland, 2009), and others that have cancer cells. In vitro analyses have demonstrated that reported a decrease in expression of survival genes in PUMA is robustly upregulated following treatment with breast cancers following neoadjuvant treatment with TAM, with evidence that it may influence the apoptotic AIs (Ellis et al., 2003). More recently, the intracellular response to TAM. Furthermore, low PUMA mRNA domain (4ICD) of the receptor tyrosine kinase, HER4/ and protein expression is associated with poor patient ErbB-4 was identified as an effector of TAM-induced outcome in cohorts of breast cancer patients. Thus in apoptosis through its activity as a BCL-2 homology 3 summary, this study has identified PUMA as a (BH3)-only, proapoptotic protein (Naresh et al., 2008). potentially important molecule in breast cancer progres- TAM treatment disrupts the interaction between ERa sion and therapeutic response. and 4ICD, leading to an accumulation of the latter at the mitochondria where it blocks BCL-2-mediated survival signals. The clinical relevance of these observa- Results tions was emphasized by demonstrating a significant association between suppressed HER4 expression and PUMA mRNA and protein expression in human breast TAM-resistance in breast cancer patients (Naresh et al., cancer cell lines 2008). PUMA mRNA and protein expressions were examined In a recent study aimed at an unbiased identification in a panel of cancer (both ER-positive and ER- of functional classes of estrogen-responsive genes that negative), normal and immortalized breast cancer cell might be implicated in AE resistance, we defined an lines by quantitative real-time PCR and immunoblot, apoptotic/cell survival gene signature distinct from three respectively. Figure 1 demonstrates that PUMA tran- other functional signatures representing cell prolifera- scripts (Figure 1a) and protein (Figure 1b) were detected tion, cell growth and transcriptional regulation (Mus- in both normal and cancer cells to varying degrees. grove et al., 2008). This apoptotic signature not only However, although there was a trend towards increased predicted response to TAM in breast cancer patients, PUMA expression in cancer cell lines compared with but was predictive independent of the proliferative normal and immortalized breast epithelial cell lines, this signature, providing strong evidence that aberrations was not significant at either the mRNA or the protein in pathways driving apoptosis/survival define a distinct, level. In addition, no significant difference was observed clinically-relevant mechanism of the endocrine respon- in PUMA mRNA expression between cell lines expres- siveness in a subset of patients (Musgrove et al., 2008). sing mutant or wild-type p53 protein, but there were Indeed, our more recent studies have demonstrated that significantly higher levels of PUMA protein in wild type individual genes mediating cellular survival, such as versus mutant p53-expressing cells (Figure 1b). BAG-1 can predict outcome and TAM responsiveness in breast cancer patients (Millar et al., 2008), with in vitro Downregulation of PUMA expression by estradiol in evidence that they may also have a role early on in human breast cancer cell lines breast cancer development (Anderson et al., 2010). Recently, we utilized an in vitro model in the steroid Herein, we have focused on a candidate in the hormone-responsive human breast cancer cell line apoptosis signature, PUMA/BBC3 (p53-upregulated MCF-7, to identify novel, estrogen-regulated targets in modulator of apoptosis/Bcl-2 binding component-3; breast cancer cells (Musgrove et al., 2008). Briefly,

Oncogene PUMA expression and breast cancer CG Roberts et al 3 45 NI ER + ER - 40 35 30 30 25 20 20 15 10 relative expression 10 0 5 relative expression p53 p53 0 WT mutant BT-20 T47-D SKBr3 MCF-7 BT-549 BT-483 BT-474 ZR75-1 Hs578T MCF-10A HMEC 184 HMEC 219-4 MDA-MB-468 MDA-MB-361 MDA-MB-134 MDA-MB-231 MDA-MB-436 MDA-MB-453 MDA-MB-330 MDA-MB-157 HMEC 184-A1 HMEC 184-B5

2.5 NI ER + ER -

2.0

1.5 2.5 ** 2.0 1.0 * 1.5 *

relative expression * * * 0.5 0.5 * * * * * 0 relative expression p53 p53 0 WT mutant BT-20 T47-D SKBr3 MCF-7 BT-483 BT-474 BT-549 ZR75-1 Hs578T MCF-10A HMEC 184 HMEC 219-4 MDA-MB-468 MDA-MB-361 MDA-MB-134 MDA-MB-231 MDA-MB-436 MDA-MB-453 MDA-MB-330 MDA-MB-157 HMEC 184-A1 HMEC 184-B5 Figure 1 PUMA mRNA and protein expression levels in human breast cancer cell lines compared with normal and immortalized breast cell lines. (a) PUMA mRNA expression in a panel of ER-positive and ER-negative breast cancer cell lines (black bars) and normal and immortalized breast epithelial cells (white bars), normalized to RPLPO loading control. Bars are means of triplicate samples. Representative data from two independent experiments are shown. Line indicates the mean PUMA expression level in normal and immortalized cells. (b) Densitometric analysis of PUMA protein expression in breast cancer cell lines (black bars) and normal and immortalized breast epithelial cells (white bars), normalized to b-actin loading control. Cell lines with known mutant p53 status are indicated (*). Representative data from two independent experiments are shown. Graphs represent levels of PUMA mRNA or protein in cells expressing wild-type p53 or mutant p53 protein. Error bars indicate 95% conﬁdence intervals, **Po0.01 for wild-type p53 versus mutant p53 protein.

MCF-7 cells were arrested in G1 phase by treatment also resulted in a downregulation of PUMA protein in with the AE ICI 182780, then cell cycle progression was both cell lines (Figure 2c). Densitometric analyses of reinitiated by treatment with 17b-estradiol. RNA was immunoblots from three independent experiments collected 6 h after estrogen treatment, and differential demonstrated that this effect was significant following gene expression was determined using Affymetrix 9 h of treatment in MCF-7 cells and after 12 h in T-47D GeneChip Arrays (HG-U133 Plus V2.0; Millennium (Figure 2d). This observed downregulation of PUMA Science, Box Hill, VIC, Australia) and Bayesian linear by estrogen is independent of functional p53, as MCF-7 modeling methods in the limma package. Figure 2a and T-47D cells express wild-type and mutant p53, shows the intensity of a representative probe set for respectively (Runnebaum et al., 1991). Interestingly, PUMA, which was significantly downregulated follow- estradiol-mediated PUMA downregulation also fol- ing estrogen treatment. lowed a similar time course to the downregulation of We confirmed the downregulation of PUMA expres- another BH3-only protein, BIK (Figure 2e; Hur et al., sion by estrogen over a time course of treatment in 2004), suggesting it may form part of a coordinated, estrogen-responsive MCF-7 and T-47D cells. ICI- anti-apoptotic signaling cascade acutely initiated in arrested cells were treated with 17b-estradiol (10 nM) response to estrogen. or vehicle, and PUMA mRNA and protein expression was determined by quantitative real-time PCR or PUMA is a direct transcriptional target of estrogen immunoblot, respectively. Figure 2b shows an acute The rapid effects of estrogen on PUMA mRNA led us to and sustained downregulation of PUMA mRNA within investigate if PUMA is a direct transcriptional target of 3–6 h of estradiol treatment compared with controls estrogen. Treatment with cycloheximide did not affect in both MCF-7 and T-47D cells. Estradiol treatment the estradiol-mediated downregulation of PUMA

Oncogene PUMA expression and breast cancer CG Roberts et al 4 * 1.2 120 * 1.0 80 0.8 40 0.6 Fold intensity 0 0.4 vehicle E 2 0.2 Rel. PUMA expression 1 MCF-7 T-47D 0 -+ -+E2 0.8 * * - - + + CHX ** ** * 0.6 ** ** ** MCF-7 T-47D 0.4 0.25 1.2 ** * ** ** * * Fold change ** ** 8.0 0.2 ** 0.20 1.0 0 6.0 0.8 24201612840 24201612840 0.15 time post E treatment (h) 0.6 2 4.0 0.10 0.4 0h 1h 3h 6h 9h 12h 16h 20h 24h 2.0 0.05 0.2 E2 -+ -+ -+ - +-+-+-+-+ Rel. PUMA expression PUMA Rel. luciferase activity 0 0 0 cont E2 cont E2 NT SAFB β -actin MCF-7 - + - + E2 0h 1h 3h 6h 9h 12h 20h 24h SAFB E2 -+ -+ -+ - + -+-+-+ PUMA β-actin

β-actin T-47D Figure 3 PUMA is a transcriptional target of estrogen. (a) MCF-7 cells were arrested with ICI 182780 for 48 h, then treated with 1.2 MCF-7 T-47D cycloheximide (black bars) or control (white bars) before addition 1 * of estradiol or vehicle as indicated for 8 h; levels of PUMA mRNA * ** * ** ** were determined by quantitative real-time PCR. Data shown is 0.8 * ** * the mean of triplicate samples from three independent experi- 0.6 * ments±s.e. *Po0.001 for estradiol treatment versus vehicle 0.4

Fold change treatment. (b) MCF-7 and T-47D cells were transfected with a 0.2 PUMA-luciferase reporter construct, pretreated with 10 nM ICI 0 182780 for 24 h and then stimulated with estradiol (100 nM)or 24201612840 24201612840 vehicle for 24 h. Luciferase activity was normalized to the activity time post E treatment (h) 2 of the pGL3-Basic reporter. Data shown is the mean of triplicate samples from three independent experiments±s.e. *Po0.001 for 0h 1h 3h 6h 9h 12h 16h 20h 24h estradiol treatment versus vehicle treatment. (c) MCF-7 cells were E -+-+-+- + -+-+-+ -+ 2 transfected with scaffold attachment factor (SAFB)-speciﬁc siRNA BIK or non-targeting control (NT) overnight then arrested with ICI β -actin MCF-7 182780 for 48 h before addition of estradiol or vehicle for a further 8 h; levels of PUMA mRNA were determined by quantitative real- time PCR. Data shown is the mean of triplicate samples from two 0h 1h 3h 6h 9h 12h 16h 20h 24h ± o o E2 -+-+-+- + -+-+-+ -+ independent experiments s.e. *P 0.05; **P 0.01 for estradiol BIK treatment versus vehicle treatment.

β-actin T-47D

Figure 2 Transcriptional downregulation of PUMA by estrogen mRNA expression (Figure 3a), suggesting that it is in human breast cancer cell lines. (a) PUMA mRNA expression in independent of ongoing protein synthesis. Analysis of a representative probe set from HG-U133 Plus V2.0 microarrays, the human PUMA genomic sequence 1.1 kb upstream of following 48 h pretreatment with ICI 182780, then 6 h stimulation the transcriptional start site revealed two sequences with estradiol (E2) or vehicle. (b) Cells were pretreated with ICI 0 182780 for 48 h, then stimulated with estradiol and RNA isolated similar to the ERE consensus (5 -GGTCAnnnTGACC- 0 at various time points as indicated. Samples were analyzed in 3 ). We further investigated the activation of PUMA by triplicate by reverse transcription PCR with PUMA-specific estrogen, using a reporter plasmid containing the primers and expression of PUMA is presented normalized to PUMA-promoter upstream of a luciferase cDNA in RPLPO and relative to vehicle controls at 0 h. Data shown is the mean of triplicate samples from four independent experi- the promoter-less luciferase reporter vector, pGL3 (Han ments±s.e. *Po0.01, **Po0.001 for estradiol treatment versus et al., 2001). Figure 3b demonstrates a significant vehicle treatment. (c) Representative immunoblot analysis of decrease in luciferase activity, following treatment with endogenous PUMA expression in whole cell lysates at time points estradiol compared with control cells in both MCF-7 post-estradiol treatment up to 24 h. b-actin was used as a loading control. (d) Densitometric analyses showing PUMA expression and T-47D, suggesting that PUMA is a direct transcrip- normalized to b-actin and relative to vehicle controls at 0 h, from tional target of estrogen. However, using chromatin three independent experiments±s.e. *Po0.01, **Po0.001 for immunoprecipitation (ChIP) assays to determine if estradiol treatment versus vehicle treatment. (e) Repression of estradiol binds directly to the endogenous PUMA BIK protein by estrogen in human breast cancer cells. Representa- promoter, we were unable to detect any significant tive immunoblots of endogenous BIK expression in whole cell lysates at time points post-estradiol treatment up to 24 h. b-actin binding compared with positive control ERE sequences was used as a loading control. (data not shown).

Oncogene PUMA expression and breast cancer CG Roberts et al 5 As previous studies have reported that estrogen- of PUMA knockdown on 4-OHT-induced apoptosis. mediated regulation of PUMA requires the co-repress- MCF-7 cells were treated with either non-targeting (NT) sor SAFB1 (Hammerich-Hille et al., 2010), we investi- siRNA or PUMA-specific siRNA, in the presence or gated the effects of SAFB1 suppression on PUMA absence of 4-OHT, and analyzed for PUMA expression expression following estradiol treatment. However, by immunoblot. Figure 5a demonstrates that treatment Figure 3c demonstrates that knockdown of SAFB with PUMA-specific siRNA significantly reduced ex- expression using specific siRNA did not significantly pression of PUMA compared with the NT and mock- affect PUMA repression in the presence of estradiol, transfected controls. This downregulation was main- suggesting additional, as yet undefined mechanisms may tained in the presence of 4-OHT. PUMA knockdown regulate PUMA expression in this experimental model. significantly attenuated the apoptotic response to 4- OHT compared with cells transfected with NT control TAM-induced apoptosis is associated with an (Figure 5b), comparable to levels observed in cells upregulation of PUMA overexpressing antiapoptotic BCL-2 (Figure 5c). Levels To further explore the functional consequences of of apoptosis following mock or siRNA transfection estrogen-regulated PUMA expression, we determined were typically slightly higher than the B3–4% usually the effects of the AE, TAM on PUMA expression, observed basally, and increased to approximately 30% using an established model of TAM-induced apoptosis following treatment with 4-OHT in mock and NT in breast cancer cells. Cells were treated with a controls and B15–18% in the presence of PUMA proapoptotic concentration of an active metabolite of siRNA (Supplementary Figure 1). TAM, 4-hydroxytamoxifen (4-OHT; 7.5 mM), resulting in a significant induction of apoptosis after 24 h PUMA mRNA expression and patient outcome (Figure 4a). The proapoptotic effects of 4-OHT in this The relationship between PUMA mRNA expression model are associated with a downregulation of ERa and breast cancer outcome was explored using pub- expression (Figure 4a), suggesting that they are pre- lically available gene expression data from van de Vijver dominantly mediated through ER-dependent signaling et al. (2002), chosen because of its similar clinicopatho- pathways. Similar results were obtained in T-47D cells logical composition to our clinical cohort (Millar et al., (data not shown). Using this model, we examined the 2008). Examination of the distribution frequency of expression of PUMA mRNA and protein (Figures 4b PUMA mRNA showed a normal distribution that was and c, respectively) over a time course of 4-OHT dichotomized using the median value (0), into high or treatment. Figures 4b and d demonstrate that a low expressing groups. Using this cut point, 147 out of significant induction of PUMA mRNA and protein 295 (49.8%) patients were classified as PUMA high. preceded the induction of apoptosis in 4-OHT-treated High PUMA expression correlated with ER þ , low cells. Furthermore, this induction of PUMA appears to tumor grade (1 and 2, all Po0.0001) and small tumor be specifically associated with the apoptotic response size o20 mm (P ¼ 0.023). Kaplan–Meier analysis to 4-OHT, as titration experiments in MCF-7 cells for breast cancer-specific death showed that high demonstrated that only at pro-apoptotic concentrations expression was associated with a favorable prognosis of 4-OHT was an induction of PUMA observed and low expression with a poor prognosis (P ¼ 0.0014; (Figure 4e), with similar results obtained in T-47D cells Figure 6a). Cox proportional hazards models were (data not shown). A comparison of the effects of TAM subsequently constructed to include traditional clinico- on other members of the BCL-2 family, demonstrates pathological variables, with step-wise removal of re- that although BIM is also upregulated, other members dundant variables until resolution. Using this approach, of the family are not (Figure 4f), suggesting that there is high PUMA expression remained an independent some specificity in the pro-apoptotic, AE response in predictor of outcome for breast cancer-specific death these cells. (HR 0.534, 95% CI 0.331–0.861, P ¼ 0.01; Table 1) in a Further investigations have suggested that this AE- resolved model that also included grade 3 and HER2 mediated stimulation of PUMA is reversible; when the amplification. AE, ICI 182780 is removed after 48 h treatment, PUMA We further explored the prognostic significance of mRNA levels significantly decrease (Figure 4g). How- PUMA mRNA expression in ER þ , endocrine-treated ever, interestingly, the addition of estradiol following patients using KM plotter, an online tool that incorpo- AE removal results in a decrease in PUMA expression rates publicly-available microarray data from 1809 to significantly lower levels than merely removing clinically annotated breast cancers (Gyo¨rffy et al., ICI alone. Furthermore, the effects of estradiol in this 2009). Figure 6b shows that high PUMA expression context are irreversible (Figure 4g). These data suggest remained a highly significant indicator of favorable that estrogen is directly repressing PUMA expression, prognosis in ER þ , endocrine-treated patients rather than merely counteracting the stimulatory effects (P ¼ 1.49 10 5). of AEs.

PUMA modulates the apoptotic response to 4-OHT in PUMA expression by immunohistochemistry and patient breast cancer cells outcome To further delineate a mechanistic role for TAM- PUMA protein expression was assessed by immuno- induced PUMA expression, we determined the effects histochemistry using tissue microarrays (TMAs) con-

Oncogene PUMA expression and breast cancer CG Roberts et al 6 60 MCF-7 * T-47D ** 50 ** * 12 40 * ** 30 8 * 20 ** % apoptosis 10 4 fold change fold 0 24 h 48 h 0 ERα 24201612840 24201612840 time post 4-OHT treatment (h) β-actin

MCF-7 T-47D 4-OHT 01234681624 h 0 1 3 6 9 12 16 20 24 h PUMA

β-actin

30 5 MCF-7 T-47D ** 25 4 ** ** * 20 3 ** ** 15 ** ** 10

2 % apoptosis * 5 fold change fold 1 0 2.57.5 1050 0 4-OHT conc. (μM) 24201612840 24201612840 PUMA time post 4-OHT treatment (h) β-actin

MCF-7 T-47D 4-OHT 03681624 h 0 3 6 8 16 24 h

BIMEL

BIML BIMS BIK

BAD

BID

BCL-2 N/A

β-actin

5 ** **

4 *** 3

relative expression relative 1

0 ICI EtOh 48h 48h ICI 48h/ ICI 48h/ ICI 48h/ E2 8h E2 8h/ EtOH 8h EtOH 20h

Oncogene PUMA expression and breast cancer CG Roberts et al 7 structed from tumors from a cohort of 292 patients Figures 7a–d. Cytoplasmic staining was present in 237/ diagnosed with invasive ductal breast carcinoma (Millar 268 (88.4%) of cancers and was of variable intensity, et al., 2008). From our original cohort of patients, 268 which ranged from negative to strong (0–3 þ ), with a invasive ductal carcinomas were available for analysis because of loss of some tissue cores during processing of 1 the TMAs. Representative immunohistochemistry stain- PUMA high n=148, 26 events ing patterns and intensities of PUMA are illustrated in 0.8

EtOH 4-OHT 0.6

0.4 PUMA low Mock NT PUMA Mock NT PUMA n=147, 53 events PUMA p=0.0014 cumulative survival cumulative 0.2 β-actin

0 * 5 5 5 1251007550250 200175150 4 4 * 4 Time in months 3 3 3 2 2 2 1 PUMA high 1 1 1 n=201 0.8 apoptosis fold change apoptosis fold change 0 apoptosis fold change 0 0 NT PUMA VEC BCL-2 NT BIK BIK 0.6 β-actin PUMA low Figure 5 PUMA modulates the apoptotic response to TAM in 0.4 n=198 breast cancer cells. (a) MCF-7 cells were transfected with PUMA- specific siRNA, NT control or mock transfected overnight, then p=1.49e-05 relapse free survival treated with 4-OHT or EtOH control for a further 24 h. Levels of 0.2 PUMA protein expression were determined by immunoblot. Expression of b-actin was used as a loading control. (b) Cells were treated with siRNAs as described above in the presence or absence 0 of 4-OHT for 24 h, then induction of apoptosis was determined by 0 5 10 15 M30-FITC labelling and analyzed by flow cytometry. (c) MCF-7 Time in years cells stably overexpressing BCL-2 cDNA or vector controls were treated with 4-OHT for 24 h, then apoptosis determined. (d) MCF- Figure 6 Breast cancer-specific death as a function of PUMA 7 cells treated with BIK-specific siRNA or NT, then treated with 4- mRNA expression. Kaplan–Meier analyses (log rank test) are OHT for a further 24 h, as described above. The level of apoptosis shown for (a) all patients in the NKI cohort (van de Vijver et al., induced by 4-OHT is shown as fold change relative to vehicle 2002) and (b)ERþ , endocrine-treated patients using KM plotter controls, from at least two independent experiments±s.e. *Po0.05 (http:www.kmplot.com/breast/), from publicly-available microar- for VEC or NT siRNA versus BCL-2 expressing or PUMA siRNA, ray data (Gyo¨rffy et al., 2009). The data were dichotomized at the respectively. median value into high and low expressing groups.

Figure 4 Tamoxifen-induced apoptosis is associated with an upregulation of PUMA. (a) Induction of apoptosis by 4-OHT. MCF-7 cells were incubated in the presence (black bars) or absence (white bars) of 4-OHT (7.5 mM) for 24 h, then attached and floating populations were analyzed for M30-FITC positivity by flow cytometry. Values shown are means of triplicate wells from three independent experiments±s.e. *Po0.001, **Po0.0001 for 4-OHT-treated cells versus untreated controls. Samples were also analyzed for ERa expression by immunoblotting and a representative blot is shown. b-actin was used as a loading control. (b) Cells were treated with 4-OHT and RNA isolated at various time points as indicated. Samples were analyzed in triplicate by reverse transcription PCR with PUMA-specific primers and expression of PUMA is presented normalized to RPLPO and relative to vehicle controls at 0 h. Data shown is the mean of triplicate samples from four independent experiments±s.e. *Po0.01, **Po0.001 for 4-OHT treatment versus vehicle treatment. (c) Representative immunoblot analysis of endogenous PUMA expression in whole cell lysates at time points post-4- OHT treatment up to 24 h. b-actin was used as a loading control. (d) Densitometric analyses showing PUMA expression normalized to b-actin and relative to vehicle controls at 0 h, from three independent experiments±s.e. *Po0.01, **Po0.001 for 4-OHT treatment versus vehicle treatment. (e) MCF-7 cells were incubated with various concentrations of 4-OHT as indicated for 24 h, then attached and floating populations were analyzed for M30-FITC positivity by flow cytometry. Values shown are means of triplicate wells from two independent experiments±s.e. Samples were also analyzed for PUMA expression by immunoblotting and a representative blot is shown. b-actin was used as a loading control. (f) Response of BCL-2 family members to 4-OHT treatment. Representative immunoblots of endogenous BCL-2 family members in whole cell lysates at time points post-4-OHT treatment (7.5 mM, MCF-7; 10 mM, T-47D) up to 24 h. b-actin was used as a loading control. T-47D cells do not express detectable levels of BCL-2 protein or the BIML and BIMS isoforms. (g) Effects of ICI/E2 removal on PUMA mRNA expression. Proliferating MCF-7 cells were arrested with the AE, ICI 182780 (10 nM) or EtOH for 48 h, then stimulated with estradiol (E2; 100 nM) or EtOH for 8 h and PUMA mRNA expression determined by RT-PCR, or E2 removed for a further 20 h before analysis. Data shown is the mean of triplicate samples±s.e. **Po0.01, ***Po0.001.

Oncogene PUMA expression and breast cancer CG Roberts et al 8 Table 1 Cox multivariate analysis for breast cancer speciﬁc death (NKI cohort) Variable HR 95% CI P-value

Size 420 mm 1.722 1.061–2.795 0.0278 Grade 3 2.219 1.338–3.680 0.002 HER2 ampliﬁed 2.782 1.529–5.061 0.0008 ER þ 0.647 0.381–1.099 0.107 Lymph node þ 0.887 0.561–1.402 0.607 PUMA high 0.613 0.369–1.021 0.107

Resolved model Grade 3 2.869 1.796–4.582 o0.0001 HER2 ampliﬁed 2.942 1.687–5.131 0.0001 PUMA high 0.534 0.331–0.861 0.01

Abbreviations: ER, estrogen receptor; CI, conﬁdence interval; HER, human epidermal growth factor; HR, hazards ratio; NKI, Nederlands Kanker Instituut cohort; PR, progesterone receptor; PUMA, p53- upregulated modulator of apoptosis. Bold indicates signiﬁcant P-value.

range of ‘H score’ from 0–270 (Figure 7e). We used the 100 median H score (50) as the cut point to dichotomise the data into high- and low-expressing groups. Using this 80 cut point, 132 of 268 patients (49.3%) were PUMA high. 60 In keeping with the mRNA data, high PUMA expression was positively correlated with ER þ , progesterone 40 receptor (PR) þ (both o0.0001), low tumor grade (P ¼ 0.0004), small tumor size o20 mm (P ¼ 0.0001), number of cases 20 luminal A phenotype (Po0.0001), and negatively with ¼ 0 p53 status (P 0.0152) and the basal subtype 0 15010050 200 300250 o (P 0.0001). There was no correlation with lymph node PUMA H score status, luminal B or HER2 intrinsic subtypes of breast cancer. Kaplan–Meier analysis showed high PUMA 1 expression to be associated with a good prognosis for all PUMA high recurrences (P ¼ 0.0126), distant metastases (Po0.0001) n=132, 17 events and breast cancer-speciﬁc death (Figure 7f; P ¼ 0.0115). 0.8 However, high PUMA expression was not signiﬁcant in multivariate analysis for any outcome measure (Table 2). 0.6 PUMA low n=136, 32 events 0.4

Discussion cumulative survival 0.2 p=0.0115 Estrogen is a major aetiological factor in the development and progression of breast cancer, and its mitogenic 0 effects on breast cancer cells have been well character- 160140120100806040200 180 ized (Butt et al., 2008). However, although the Time in months dysregulation of apoptotic/survival pathways is a hallmark of breast cancer like other malignancies (Hanahan Figure 7 (a–d) Representative images of PUMA immunohistochemistry in 268 invasive ductal carcincomas. Cytoplasmic staining and Weinberg, 2000), the influence of disrupted apop- was present at variable intensity that ranged from 0 (a), 1 þ (b), totic signaling on breast tumor growth and response to 2 þ (c), 3 þ (d). The intensity of staining and percentage of positive endocrine therapy is less well understood (Butt et al., cells was multiplied to form a modified ‘H’ (histo) score, the 2007). Here, we have examined the role of an estrogen distribution of which is presented in (e). (f) Kaplan–Meier analysis (log rank test) of breast cancer-specific death in 268 invasive ductal target—the BH3-only, pro-apoptotic gene, PUMA in carcinomas. The cohort was dichotomized at the median H score this context. value (50) into high and low expressing groups. Despite its established role as a p53-upregulated gene and essential mediator of p53-dependent apoptosis (Vousden, 2005), we have demonstrated that PUMA with wild-type p53 compared to those with mutant p53. mRNA is expressed across a range of breast cancer cell Interestingly, the acute downregulation of PUMA by lines, independent of their p53 status, however PUMA estradiol was observed in p53 wild-type (MCF-7) and protein expression was significantly increased in cells mutant (T-47D) cell lines, suggesting that PUMA’s

Oncogene PUMA expression and breast cancer CG Roberts et al 9 Table 2 Cox multivariate analysis for breast cancer-specific death PUMA’s pivotal role in the induction of stress- induced apoptosis and as a common target of anti- Variable HR 95% CI P-value apoptotic growth factors implicates its dysregulation as Size 420 mm 1.078 0.592–1.964 0.805 an important step in the tumorigenic process. Given that Grade 3 1.597 0.767–3.325 0.2113 functional p53 is lost in the majority of human cancers, HER2 amp. 2.709 1.461–5.025 0.0016 PUMA function is indeed compromised indirectly in ER þ 0.712 0.352–1.442 0.3457 PR þ 0.320 0.144–0.710 0.005 these malignancies, with evidence that this impinges on Lymph node þ 3.203 1.703–6.025 0.0003 the apoptotic response to irradiation and chemother- PUMA high 0.719 0.385–1.340 0.2985 apeutic drugs (Yu and Zhang, 2005). More directly, loss of PUMA expression has been reported in melanomas Abbreviations: ER, estrogen receptor; CI, confidence interval; HER, (Karst et al., 2005), a proportion of Burkitt’s lympho- human epidermal growth factor; HR, hazards ratio; PR, progesterone receptor; PUMA, p53-upregulated modulator of apoptosis. mas (Garrison et al., 2008), and interestingly, in the Bold indicates significant P-value. tumor stroma associated with breast carcinoma (Finak et al., 2008). However, the lack of spontaneous tumor formation in puma -knockout mice (Jeffers et al., 2003), regulation and hence, function is independent of p53 in and the paucity of studies showing reduced PUMA breast cancer. This appears consistent with previous expression in cancer, suggests that it is not commonly reports that have described p53-independent induction directly inactivated during the tumorigenic process. of PUMA mRNA by the glucocorticoid, dexamethasone Indeed, our studies reported herein showed no signifi- and serum withdrawal (Han et al., 2001), and an almost cant decrease in PUMA expression in breast cancer cell total attenuation of p53-independent, cytokine depriva- lines compared with normal and immortalized breast tion-induced apoptosis in PUMA-null primary myeloid epithelial cells, and a normal distribution of PUMA cells (Jeffers et al., 2003). mRNA and protein in breast cancer tissue. The rapid transcriptional downregulation of PUMA Despite this lack of evidence for a direct repression of in response to estrogen further delineates the mechan- PUMA in cancers, we did observe highly significant isms by which this anti-apoptotic growth factor correlations between reduced PUMA expression at both mediates its survival effects. Estrogen’s transcriptional the mRNA and protein level in primary breast modulation of other apoptotic mediators such as BAK, carcinomas, and breast cancer-specific death—suggest- BIK, caspase-9 (Frasor et al., 2003; Hur et al., 2004) and ing that when PUMA repression does occur, it can BCL-2 (Perillo et al., 2000) have been reported, and our impact on disease outcome. Furthermore, following the current data suggest that PUMA and other BH3-only Cox multivariate regression analysis, high PUMA proteins, such as BIK may form part of a broad, anti- mRNA remained an independent predictor of outcome apoptotic signaling cascade acutely and irreversibly for disease-specific death in a resolved model including initiated in response to estrogen in breast cancer cells. grade 3 and HER2 amplification. Thus, low PUMA Indeed, other potent survival factors such as IGF-I and expression may serve as a molecular marker of poor EGF also repress PUMA mRNA and protein expres- prognosis in breast cancer, lending further support to sion, with evidence that this is mediated through the concept that dysregulation of apoptotic pathways phosphatidylinositol-3-kinase signaling (Han et al., can significantly influence the progression of this disease 2001). We were unable to demonstrate direct binding (Butt et al., 2008). of ERa to putative EREs in the PUMA proximal Reduced PUMA expression may also have important promoter, using ChIP analysis. However, estrogen- implications for therapeutic response in breast cancers. mediated repression does not commonly occur through PUMA is rapidly induced by chemotherapeutic agents direct ERE binding, but rather indirectly through the that elicit p53-dependent, DNA damage (Han et al., sequestration of shared-transcriptional corepressors 2001), with evidence that it is necessary to educe an necessary for maintaining basal gene expression (Carroll apoptotic response to these drugs (Yu et al., 2003). et al., 2006). Indeed, the maintenance of the transcrip- However, to our knowledge, this is the first report of tional effect in the presence of the protein synthesis the p53-independent, transcriptional upregulation of inhibitor, cycloheximide would support the concept that PUMA by the AE, TAM. PUMA induction occurred PUMA is a primary, yet indirect, target of ERa action before the detectable onset of apoptosis in this cell in breast cancer cells. Interestingly, recent work by system and was specifically associated with the cyto- Hammerich-Hille et al. not only demonstrated ERa toxic, as opposed to the cytostatic, response to TAM binding to the proximal PUMA promoter at a site (the latter being observed at lower concentrations). distinct from the putative EREs, but also showed the This, together with evidence from PUMA-specific involvement of the scaffold attachment factor, SAFB1 siRNA studies, suggests that PUMA expression may, as an obligate, ERa corepressor of PUMA in response at least in part, mediate the apoptotic response to 4- to estrogen in breast cancer cells. However, in our OHT in breast cancer cells. Importantly, the prognostic experimental model system, knockdown of SAFB by significance of these in vitro observations is strongly specific siRNA did not significantly abrogate estradiol- supported by our clinical data showing a highly mediated PUMA repression, suggesting that additional, significant association between low PUMA mRNA as yet undefined mechanisms may regulate PUMA expression and worse outcome in ER þ , endocrine- expression under these conditions. treated patients.

Oncogene PUMA expression and breast cancer CG Roberts et al 10 Interestingly, manipulation of PUMA levels alone did checked for equal loading by reprobing with anti-actin not completely ameliorate TAM-induced apoptosis antibody (Sigma-Aldrich). in vitro, possibly reflecting the complex interplay between numerous apoptotic regulators in coordinating Measurement of apoptosis by flow cytometry the cytotoxic, endocrine response. Indeed, analysis of For M30 analysis, floating and attached cell populations were additional BCL-2 family proteins following TAM combined, fixed and permeabilized in ice cold 70% EtOH, then treatment demonstrated that although BIM is also resuspended in phosphate-buffered saline (PBS/0). 5% bovine upregulated, other members of the family are not. Of serum albumin (BSA) with fluorescein isothiocyanate-conju- interest is the lack of response of BIK to TAM gated M30 CytoDEATH monoclonal antibody (1:100; Alexis treatment, which differs from its induction following Biochemicals, Lausen, Switzerland) before the M30-positive (apoptotic) population was determined by flow cytometry. treatment with the pure AE, fulvestrant (Hur et al., 2004). These data suggest that there may be a differential apoptotic response pattern to AEs—a theory Real-time quantitative PCR supported by the observation that PUMA levels remain Total RNA was isolated using the RNAeasy kit (Qiagen Victoria, Australia) from cells pretreated with 10 nM ICI unchanged following fulvestrant treatment (Hur et al., 182780 (7a-[9-(4,4,5,5,5-pentafluoropentylsulfinyl) nonyl] estra- 2004). 1,3,5,(10)-triene-3,17b-diol, a kind gift of Dr Alan Wakeling, In conclusion, we have identified the pro-apoptotic Astra-Zeneca Pharmaceuticals, Alderly Park, Cheshire, UK) for regulator, PUMA as an important target of estrogen in 48 h and then stimulated with 17b-estradiol (100 nM), and was breast cancer cells, and a predictor of outcome and reverse-transcribed using the Reverse Transcription System TAM responsiveness in breast cancer patients. In (Promega, Sydney, Australia), according to the manufacturer’s addition, we have demonstrated PUMA’s upregulation instructions. Real-time PCR was performed with an ABI Prism by the AE, TAM in breast cancer cells and provided 7900HT Sequence Detection System (Applied Biosystems, evidence that it can mediate TAM’s pro-apoptotic Foster City, CA, USA) using inventoried (pre-made) Taq-Man effects in this cell system. Given PUMA’s broad probes for PUMA (Applied Biosytems). Data analyses were performed using the DCt method with RPLPO (Applied promiscuity in binding to all anti-apoptotic members Biosystems) as an internal loading control. Fold changes in of the Bcl-2 family (Chen et al., 2005) and its potent gene expression were calculated relative to untreated controls. efficacy as an apoptotic mediator, its transcriptional regulation by estrogen and AEs is likely to be of Luciferase reporter assays considerable relevance in both breast cancer progression MCF-7 and T-47D cells were transfected using Lipofectamine and therapeutic response. Thus, our data support the 2000 (Invitrogen Life Technology, San Diego, CA, USA) with significant potential of identifying approaches to en- a luciferase reporter construct containing either the PUMA hance PUMA’s activity as a therapeutic strategy to promoter (a kind gift from Thomas Chittenden, Immunogen, target, in particular, hormone refractory disease. Cambridge, MA, USA), or an ERE sequence as a positive control, in the promoter-less pGL3-Basic vector. 24 h posttransfection, cells were pretreated with 10 nM ICI 182790 for 24 h and then stimulated with 17b-estradiol (100 nM) for 24 h. Materials and methods Luciferase activity was assayed using the Dual-Luciferase Reporter Assay System (Promega) and normalized to the Cell lines and reagents activity of the pGL3-Basic reporter. The human breast cancer cell lines MCF-7 and T-47D were obtained from ATCC (American Type Culture Collection, RNA interference Manassas, VA, USA) and routinely maintained in RPMI-1640 Small interfering RNAs (siRNA) specific for PUMA, SAFB or m supplemented with 5% fetal calf serum (FCS), 10 g/ml insulin BIK (ON-TARGETplus SMARTpool, human BBC3, SAFB and 2.92 mg/ml glutamine under standard conditions. MCF-7 and BIK, respectively) and non-targeting controls (ON-TAR- cells stably overexpressing human BCL-2 have been previously GETplus siCONTROL) were purchased from Dharmacon. Cells described (Butt et al., 2006). 4-OHT was purchased from were transfected with LipofectAMINE 2000 in the presence of Sigma-Aldrich (St Louis, MO, USA). the siRNAs according to the manufacturer’s protocol.

Immunoblot analysis PUMA gene expression profiling Proteins from whole cell lysates were resolved under reducing Publically available gene expression data sourced from a conditions on 12% SDS–polyacrylamide gels using standard published study by van de Vijver et al. (2002) of breast methods. Resolved proteins were transferred to polyvinylidene cancer outcome were analysed to determine a potential fluoride (PVDF) membranes and probed with antibodies relationship between PUMA mRNA levels and prognosis. against ERa (Lab Vision, Fremont, CA, USA), PUMA (Cell The data set from the Nederlands Kanker Instituut and Signaling Technology, Danvers, MA, USA), SAFB (Sigma- designated the NKI cohort, comprised 295 patients, 76% of Aldrich), BIK (Santa Cruz Biotechnology, Santa Cruz, CA, which were ER þ , with a median follow-up of 93.6 months USA), BCL-2 (DakoCytomation, Glostrup, Denmark), BIM (range 0.6–220 months) and was of similar clinicopathological (Calbiochem, Darmstadt, Germany), BAD and BID (both BD composition to our clinical cohort. Data were generated using Biosciences Pharmingen, San Diego, CA, USA) overnight at Rosetta NKI-spotted oligonucleotide arrays and were down- 4 1C or for 2 h at room temperature. Immunoreactive protein loaded from http://microarray-pubs.stanford.edu/wound_ bands were detected by the relevant anti-IgG antibodies NKI/explore.html as log 2-transformed values in a text table conjugated with horseradish peroxidase, followed by enhanced format, as previously described (Millar et al., 2008). Raw data chemiluminescence (Pierce, Rockford, IL, USA). Blots were were directly transferred to the final output file without further

Oncogene PUMA expression and breast cancer CG Roberts et al 11 processing. Data from one probeset were available and following antibodies: ER (1:100; clone 6F11; Dako), PR (1:200; expression data were analysed for frequency distribution of clone PgR 636; Dako), CK5/6 (1:80; clone MAB1602; Chemicon mRNA and its association with patient outcome. International, Temecula, CA, USA), EGFR (1:100; clone H11; Further analyses in ER þ , endocrine-treated patients were Dako) and p53 (1:400; clone DO-7, Novocastra, UK). HER-2 carried out using KM plotter (http://www.kmplot.com/breast/), FISH was assessed in the Australian National Reference an on line tool that incorporates public microarray data from Laboratory (Department of Pathology, St Vincent’s Hospital, 1809 breast cancer patients and enables filtering based on Sydney, Australia) using the Vysis PathVysion HER-2 DNA hormone receptor status and adjuvant endocrine treatment dual-colour probe kit (Abbott Laboratories, Abbott Park, IL, (Gyo¨rffy et al., 2009). USA). A HER2:chromosome 17 ratio 42.2 was classified as HER2 amplification. ER and PR were assessed as positive if 4 Patient characteristics and clinical cancer cohort they had an H score of 10. CK5/6, and EGFR were assessed PUMA protein expression was assessed by immunohistochem- as positive if there was any positive cytoplasmic or membranous istry using TMAs constructed from tumors from a cohort of staining present at any intensity. P53 was considered positive if 4 292 patients diagnosed with invasive ductal breast carcinoma. 10% staining was present at any intensity. This cohort has been previously described in more depth elsewhere (Millar et al., 2008). Briefly, the cohort consists of Statistical analyses cases of invasive ductal carcinoma of no special type, median Statistical analyses were performed using Statview 5.0 Software age 54 (range 24–87) with a median follow-up of 64 months (Abacus Systems, Berkeley, CA, USA). For in vitro studies, (range 0–152.1). Of these, 68.6% were ER þ , 57.1% PR þ , differences between groups were evaluated by Fisher’s protected 18.7% HER-2 amplified (by FISH), 43.3% lymph node least significant difference test after analysis of variance positive. Endocrine therapy (TAM) was given to 49.3% of (ANOVA) or factorial analysis where appropriate. Po0.05 patients and chemotherapy (AC or CMF) to 38%. Before the was accepted as statistically significant. PUMA mRNA and approval for this study was obtained from the Human PUMA protein expression, and its association with clinico- Research Ethics Committee of St Vincent’s Hospital, Sydney pathological variables and intrinsic molecular phenotype of (HREC SVH H94/080, SVH H00/36). breast cancer were tested by applying the w2-test of association in contingency tables. Kaplan–Meier and Cox proportional Immunohistochemistry hazards model were used for univariate analysis and the latter 4-mm sections were cut from each TMA, mounted on for multivariate analyses. Those factors that were prognostic in SuperFrost Plus glass slides and baked for 2 h at 70 1C, then univariate analysis were then assessed in a multivariable model dewaxed by passage through xylene (two 5 min washes), to identify factors that were independently prognostic and those cleared and rehydrated in graded alcohol (100, 95 and 70%) that were the result of confounding variables. ending in a distilled water wash. Antigen retrieval was performed using Dako solution (pH 9.0, s2367, Dako, Carpentaria, CA, USA) in a pressure cooker (Dako Pascal Abbreviations Decloaker) for 30 s, followed by cooling gently for 15 min in a running water bath. Following a thorough wash in distilled AE, anti-estrogen; TAM, tamoxifen; FCS, fetal calf serum; water, endogenous peroxidase activity was eliminated with 3% 4-OHT, 4-hydroxytamoxifen; ER, estrogen receptor; ChIP, hydrogen peroxide for 5 min. Slides were incubated with chromatin immunoprecipitation; TMA, tissue microarray; PR, PUMA rabbit polyclonal antibody (#4976, Cell Signaling progesterone receptor; EGFR, epidermal growth factor Technology), 1:300 dilution for 60 min at room temperature receptor; ERE, estrogen response element; HER2, human (RT), then staining was completed in a Dako autostainer. epidermal growth factor receptor 2. Slides were then rinsed in water and counterstained with haematoxylin, dehydrated through graded ethanol, cleared in xylene and mounted. Negative tissue controls included kidney Conflict of interest and prostate, as well as isotype-matched non-specific immu- noglobulin substituted for the primary antibody. The authors declare no conflict of interest. All assessments of PUMA immunohistochemical staining were performed by an experienced breast pathologist (EKAM) blinded to the clinical and molecular data and patient outcome. Acknowledgements Cytoplasmic staining for PUMA was described in terms of the intensity (0: negative, 1 þ :weak,2þ : moderate and 3 þ : This research was supported by a National Health and strong) and percentage of cells staining positive. From these Medical Research Council (NHMRC) of Australia Program indices, a simplified ‘H score’ (that is, intensity percentage of Grant 535903, the Australian Cancer Research Fund, the RT positive staining) was calculated for each core and a mean and Hall Trust, the Petre Foundation, a Cancer Institute NSW median score for each parameter calculated for each tumor Career Development and Support Fellowship (AJB), Cancer (range of two to six cores per patient). Institute NSW Clinical Research Fellowships (EKAM and ER, progesterone receptor (PR), cytokeratin 5/6 and epider- SAO’T) and an NHMRC health professional training fellow- mal growth factor receptor (EGFR) were also stained using the ship (SAO’T).

References

Anderson LR, Sutherland RL, Butt AJ. (2010). BAG-1 overexpression Butt AJ, Caldon CE, McNeil CM, Swarbrick A, Musgrove EA, attenuates luminal apoptosis in MCF-10A mammary epithelial cells Sutherland RL. (2008). Cell cycle machinery: links with genesis and through enhanced RAF-1 activation. Oncogene 29: 527–538. treatment of breast cancer. Adv Exp Med Biol 630: 189–205.

Oncogene PUMA expression and breast cancer CG Roberts et al 12 Butt AJ, McNeil CM, Musgrove EA, Sutherland RL. (2005). receptor biology, treatment options, and future directions. Cancer Downstream targets of growth factor and oestrogen signalling and 113: 2385–2397. endocrine resistance: the potential roles of c-Myc, cyclin D1 and Jeffers JR, Parganas E, Lee Y, Yang C, Wang J, Brennan J et al. cyclin E. Endocr Rel Cancer 12: S47–S59. (2003). Puma is an essential mediator of p53-dependent and - Butt AJ, Musgrove EA, Sutherland RL. (2007). Live or let die: independent apoptotic pathways. Cancer Res 52: 5204–5207. oestrogen regulation of survival signalling in endocrine response. Jordan VC. (2009). A century of deciphering the control mechanisms Breast Cancer Res 9: 306. of sex steroid action in breast and prostate cancer: the origins of Butt AJ, Roberts CG, Seawright AA, Oelrichs PB, MacLeod JK, Liaw targeted therapy and chemoprevention. Cancer Res 69: 1243–1254. TYE et al. (2006). A novel plant toxin, persin, with in vivo activity in Karst AM, Dai DL, Martinka M, Li G. (2005). PUMA expression is the mammary gland, induces Bim-dependent apoptosis in human significantly reduced in human cutaneous melanomas. Oncogene 24: breast cancer cells. Mol Cancer Ther 5: 2300–2309. 1111–1116. Cannings E, Kirkegaard T, Tovey SM, Dunne B, Cooke TG, Bartlett Millar EKA, Anderson LR, McNeil CM, O’Toole SA, Pinese M, Crea JMS. (2007). Bad expression predicts outcome in patients treated P et al. (2008). BAG-1 predicts patient outcome and tamoxifen with tamoxifen. Breast Cancer Res Treat 102: 173–179. responsiveness in ER-positive invasive ductal carcinoma of the Carroll JS, Meyer CA, Song J, Li W, Geistlinger TR, Eeckhoute J breast. Br J Cancer 100: 123–133. et al. (2006). Genome-wide analysis of estrogen receptor binding Musgrove EA, Sergio CM, Loi S, Inman CK, Anderson LR, Alles MC sites. Nat Genet 38: 1289–1297. et al. (2008). Identification of functional networks of estrogen- and Chen L, Willis SN, Wei A, Smith BJ, Fletcher JI, Hinds MG et al. c-Myc-responsive genes and their relationship to response to (2005). Differential targeting of prosurvival Bcl-2 proteins by their tamoxifen therapy in breast cancer. PLoS One 3: e2987. BH3-only ligands allows complementary apoptotic function. Mol Musgrove EA, Sutherland RL. (2009). Biological determinants of Cell 17: 393–403. endocrine resistance in breast cancer. Nat Rev Cancer 9: 631–643. Ellis MJ, Rosen E, Dressman H, Marks J. (2003). Neoadjuvant Nakano K, Vousden KH. (2001). PUMA, a novel proapoptotic gene, comparisons of aromatase inhibitors and tamoxifen: pretreatment is induced by p53. Mol Cell 7: 683–694. determinants of response and on-treatment effect. J Steroid Biochem Naresh A, Thor AD, Edgerton SM, Torkko KC, Kumar R, Jones FE. Mol Biol 86: 301–307. (2008). The HER4/4ICD estrogen receptor coactivator and BH3- Finak G, Bertos N, Pepin F, Sadekova S, Souleimanova M, Zhao H only protein is an effector of tamoxifen-induced apoptosis. Cancer et al. (2008). Stromal gene expression predicts clinical outcome in Res 68: 6387–6395. breast cancer. Nat Med 14: 18–27. Perillo B, Sasso A, Abbondanza C, Palumbo G. (2000). 17b-estradiol Frasor J, Danes JM, Komm B, Chang KCN, Lyttle CR, Katzenel- inhibits apoptosis in MCF-7 cells, inducing bcl-2 expression via two lenbogen BS. (2003). Profiling of estrogen up- and down-regulated estrogen-responsive elements present in the coding sequence. Mol gene expression in human breast cancer cells: insights into gene Cell Biol 20: 2890–2901. networks and pathways underlying estrogenic control of prolifera- Reimer T, Koczan D, Muller H, Friese K, Thiesen H-J, Gerber B. tion and cell phenotype. Endocrinology 144: 4562–4574. (2002). Tumour Fas ligand:Fas ratio greater than 1 is an Garrison SP, Jeffers JR, Yang C, Nilsson JA, Hall MA, Rehg JE et al. independent marker of relative resistance to tamoxifen therapy in (2008). Selection against PUMA gene expression in Myc-driven B- hormone receptor positive breast cancer. Breast Cancer Res 4: R9. cell lymphomagenesis. Mol Cell Biol 28: 5391–5402. Runnebaum IB, Nagarajan M, Bowman M, Soto D, Sukumar S. Gyo¨rffy B, Lanczky A, Eklund AC, Denkert C, Budczies J, Li Q et al. (1991). Mutations in p53 as potential molecular markers for human (2009). An online survival analysis tool to rapidly assess the effect of breast cancer. Proc Natl Acad Sci USA 88: 10657–10661. 22 277 genes on breast cancer prognosis using microarray data of Swaby RF, Sharma CG, Jordan VC. (2007). SERMs for the 1809 patients. Breast Cancer Res Treat 123: 725–731. treatment and prevention of breast cancer. Rev Endocr Metab Hammerich-Hille S, Kaipparettu BA, Tsimelzon A, Creighton CJ, Disord 8: 229–239. Jiang S, Polo JM et al. (2010). SAFB1 mediates repression of van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Voskuil immune regulators and apoptotic genes in breast cancer cells. J Biol DW et al. (2002). A gene-expression signature as a predictor of Chem 285: 3608–3616. survival in breast cancer. N Engl J Med 347: 1999–2009. Han J, Flemington C, Houghton AB, Gu Z, Zambettidagger GP, Lutz Vousden KH. (2005). p53 and PUMA: a deadly duo. Science 309: RJ et al. (2001). Expression of bbc3, a pro-apoptotic BH3-only 1685–1686. gene, is regulated by diverse cell death and survival signals. Proc Yu J, Wang Z, Kinzler KW, Vogelstein B, Zhang L. (2003). PUMA Natl Acad Sci USA 98: 11318–11323. mediates the apoptotic response to p53 in colorectal cancer cells. Hanahan D, Weinberg RA. (2000). The hallmarks of cancer. Cell 200: Proc Natl Acad Sci USA 100: 1931–1936. 57–70. Yu J, Zhang L, Hwang PM, Kinzler KW, Vogelstein B. (2001). PUMA Hur J, Chesnes J, Coser KR, Lee RS, Geck P, Isselbacher KJ et al. induces the rapid apoptosis of colorectal cancer cells. Mol Cell 7: (2004). The Bik BH3-only protein is induced in estrogen-starved and 673–682. antiestrogen-exposed breast cancer cells and provokes apoptosis. Yu J, Zhang L. (2005). The transcriptional targets of p53 in apoptosis Proc Natl Acad Sci USA 101: 2351–2356. control. Biochem Biophys Res Commun 331: 851–858. Hurvitz SA, Pietras RJ. (2008). Rational management of endocrine Yu J, Zhang L. (2009). PUMA, a potent killer with or without p53. resistance in breast cancer: a comprehensive review of estrogen Oncogene 27: S71–S83.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc)

Oncogene Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155

c-Myc overexpression and endocrine resistance in breast cancerଝ Catriona M. McNeil a, C. Marcelo Sergio a, Luke R. Anderson a, Claire K. Inman a, Sarah A. Eggleton a, Niamh C. Murphy a,b, Ewan K.A. Millar a,c, Paul Crea a,d, James G. Kench a,e,f, M. Chehani Alles a, Margaret Gardiner-Garden a, Christopher J. Ormandy a,f, Alison J. Butt a,f, Susan M. Henshall a,f, Elizabeth A. Musgrove a,f, Robert L. Sutherland a,f,∗

a Cancer Research Program, Garvan Institute of Medical Research, 384 Victoria Street, Darlinghurst, NSW 2010, Australia b Sydney Cancer Centre, Royal Prince Alfred Hospital, Missenden Road, Camperdown 2006, Australia c Department of Anatomical Pathology, South East Area Laboratory Services, St. George Hospital, Gray Street, Kogarah 2216, Australia d Department of Surgery, St. Vincent’s Hospital, Victoria Street, Darlinghurst, NSW 2010, Australia e Institute of Clinical Pathology and Medical Research, Westmead Hospital, Westmead, NSW 2145, Australia f St. Vincent’s Hospital Clinical School, Faculty of Medicine, University of NSW, St. Vincent’s Hospital, Victoria Street, Darlinghurst, NSW 2010, Australia

Abstract

The oncoprotein c-Myc is frequently overexpressed in breast cancer and ectopic expression in breast cancer cell lines attenuates responses to antiestrogen treatment. Here, we review preliminary data aimed at further elucidating a potential role for c-Myc in clinical endocrine resistance in breast cancer. Immunohistochemical and semi-quantitative PCR revealed that c-Myc protein and c-myc mRNA were frequently overexpressed in both ER-positive and ER-negative breast carcinoma. Furthermore, both constitutive and inducible c-Myc overexpression in MCF-7 breast cancer cell lines markedly reduced their sensitivity to the growth inhibitory effects of the pure antiestrogen ICI 182,780. In order to identify potential downstream targets of c-Myc that mediate this effect, Affymetrix microarrays were employed to examine the patterns of gene expression shared by MCF-7 cells stimulated by estrogen, or by induction of c-Myc. Approximately 50% of estrogen target genes identiﬁed 6 h after treatment were also regulated by c-Myc. One novel target, EMU4, was transcriptionally regulated by c-Myc. In addition, there was a strong correlation between c-myc and EMU4 mRNA expression in a battery of breast cancer cell lines. These data conﬁrm that c-Myc overexpression is a common event in breast cancer, and that this is associated with resistance to antiestrogens in vitro. Furthermore, the development of an experimental paradigm for the discovery of c-Myc and estrogen target genes associated with endocrine resistance provides a framework for the discovery and validation of genes involved in estrogen signalling, and c-Myc-mediated-antiestrogen resistance. © 2006 Elsevier Ltd. All rights reserved.

Keywords: Breast cancer; c-Myc; Endocrine resistance

1. Introduction related survival has improved by ∼25% over the last 10–20 years as a result of improvements in screening and in local Breast cancer represents a major public health problem in and systemic treatment, it remains a major cause of mortal- industrialised and developing countries. While breast cancer- ity among women, accounting for 15–20% of cancer related deaths [1]. ଝ Lecture presentation at the 17th International Symposium of the Journal The female sex-steroid estrogen plays a key role in the of Steroid Biochemistry & Molecular Biology, ‘Recent Advances in Steroid control of cellular proliferation in breast cancer [2]. For the Biochemistry and Molecular Biology’ (Seefeld, Tyrol, Austria, 31 May–3 ∼70% of women whose breast cancers express estrogen June 2006). ∗ receptors (ER), modulation of estrogen synthesis (aromatase Corresponding author at: Cancer Research Program, Garvan Institute of inhibitors) and ER signalling by blockade of ER (antie- Medical Research, 384 Victoria Street, Darlinghurst, NSW 2010, Australia. Tel.: +61 2 9295 8322; fax: +61 2 9295 8321. strogens) remain the cornerstone of the systemic treatment E-mail address: [email protected] (R.L. Sutherland). of their disease. Treatments, such as tamoxifen reduce the

0960-0760/$ – see front matter © 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.jsbmb.2006.09.028 148 C.M. McNeil et al. / Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155 relative risk of disease relapse and breast cancer-related an independent predictor of survival in patients treated with death by 50% and 30%, respectively [3]. Nonetheless, many tamoxifen [20]. MYC gene amplification occurs in approxi- women treated with adjuvant endocrine therapy eventually mately 15–20% of patients with breast cancer [14]. However, relapse and die as a result of their breast cancer (in the order overexpression of c-myc mRNA and c-Myc protein occurs of 10–15% in the ATAC trial which enrolled a cohort of more frequently, generally 30–50% [19,21,22], particularly relatively good prognosis patients) [4]. in high-grade tumours [23]. While some studies link c-myc The biological basis of de novo or acquired antiestro- mRNA overexpression with adverse outcome [19], others gen resistance as it is manifest in the clinic remains to be studies do not [24]. Immunohistochemical studies have gen- fully defined, although there is evidence to support a causal erally failed to demonstrate an association between c-Myc relationship with upregulation of growth-factor signalling protein expression and outcome [25,26]. While some studies pathways that cross-talk with ER signalling [5,6], adap- show an association between c-Myc overexpression and neg- tive hypersensitivity to estrogen deprivation, changes in the ative prognostic factors such as poor differentiation and high co-expression of ER␤, and loss or mutation of ER␣ itself proliferation index [21], and others document co-operation (reviewed by Ring and Dowsett [7]). An area of research with Bcl-2 overexpression in promoting lymph node metas- that has received somewhat less attention is the role played tasis [27], at present it is difficult to draw definite conclusions by aberrant expression and activity of cell cycle regulatory regarding the prognostic significance of c-Myc protein over- proteins in the development of antiestrogen resistance. expression in breast cancer. Estrogens and antiestrogens regulate a number of genes On the basis of these in vitro and clinical studies, we pro- critical to cell cycle entry and progression at the G1/S inter- pose that altered expression of c-Myc and key downstream face [8]. Of these, one of the key effectors of estrogen sig- target genes may be involved in the development of antiestro- nalling is the transcription factor c-Myc. The expression of gen resistance. Indeed, there is now an accumulating body c-Myc is rapidly induced by estrogen treatment, and con- of evidence from our laboratory and others in support of versely, downregulated by antiestrogen treatment [9,10]. Fur- this concept, whereby inducible expression of c-Myc attenu- thermore, anti-sense oligonucleotides directed against c-Myc ates the anti-proliferative effects of antiestrogen treatment in inhibit estrogen-induced cell proliferation in a manner similar MCF-7 cells [12,28–30]. Furthermore, sustained expression to antiestrogens [10,11], while c-Myc overexpression reca- of c-Myc has been observed in the development of tamoxifen- pitulates the effects of estrogen on cell cycle progression in resistant cell lines, suggesting that these molecular changes cells arrested in G0/G1 by pre-treatment with either SERMS may contribute to clinical antiestrogen resistance [31]. or pure antiestrogens [12]. In this brief review, we summarise the experimental In clinical cohorts, MYC gene amplification has been approach, and present some preliminary data aimed at associated with the transition from in situ to invasive car- more clearly defining the role of c-Myc and selected target cinoma [13], markers of aggressive phenotype [14,15] and genes in the acquisition of antiestrogen resistance in breast poor prognosis in general [14,16–19]. It has been reported as cancer.

Fig. 1. c-Myc mRNA expression in clinical breast cancer. Expression of c-myc and the housekeeping gene 36B4 were analysed by RT-PCR in 141 primary breast tumour RNA samples using the Promega Reverse Transcription System (Promega, USA) and the Roche Lightcycler Fast Start DNA Master Sybr Green 1 kit (Roche, Germany), and a Rotor-gene RG-3000 (Corbett Research, Australia). A standard curve was generated for both c-myc and 36B4 from PCR product, and the relative expression calculated for each sample, normalised to 36B4 expression. The PCR reactions were performed in triplicate, from two independent reverse transcription reactions. C.M. McNeil et al. / Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155 149

Fig. 2. Immunohistochemistry of c-Myc expression in breast cancer. c-Myc protein expression was measured using immunohistochemistry on tissue microarrays constructed from parafﬁn-embedded blocks of inﬁltrating ductal carcinomas from 289 patients [32] treated by a single surgeon (PC). Of these patients, cores from 217 were analysable for both c-Myc and ER. c-Myc staining was performed using mouse monoclonal anti-human c-Myc, 9E10 (Dako, USA), and ER staining was performed using mouse monoclonal anti-human ER, ID5 (Dako, USA). The cohort was scored for each antibody by two independent observers (CMMcN and JGK) one of whom was a pathologist, and both of whom were blinded to patient identity and outcome. For c-Myc, cores were assessed for both nuclear and cytoplasmic staining by percentage of cells staining positive, and the intensity of staining in the range 0–3. Examples of the staining patterns are illustrated: (A) negative, (B) nuclear, and (C) cytoplasmic. Data for the whole cohort are presented in Table 1.

2. Materials and methods we employed quantitative RT-PCR and immunohistochemistry on two independent cohorts of breast cancer patients The human breast cancer tissue samples and in vitro where either RNA or tissue microarrays were available. cell culture models employed in this study have been Fig. 1 presents the mRNA expression data, relative to the described in detail in previous publications from this labo- housekeeping gene 36B4 in a cohort of 141 breast carci- ratory [12,32,33]. Speciﬁc details of individual experiments nomas. A wide range of relative c-myc mRNA expression are presented in the legends to Figs. 1–6. was observed in both ER+ and ER− breast cancers with the relative range and frequency of expression similar between the two groups. Likewise, when protein expression was eval- 3. Results and discussion uated using immunohistochemistry in a separate cohort of 217 primary breast cancer samples, c-Myc overexpression 3.1. c-myc mRNA and protein expression in clinical was observed in both ER+ and ER− cancers. As detailed breast cancer in Table 1, the prevalence of overexpression was slightly higher in the ER+ tumours when nuclear or cytoplasmic stain- In order to determine the range of c-myc mRNA and c- ing were evaluated individually or in combination, although Myc protein expression in samples of clinical breast cancer, the difference between the ER+ and ER− groups was most 150 C.M. McNeil et al. / Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155

Fig. 4. Constitutive and inducible expression of c-Myc attenuates the growth inhibitory response to antiestrogens in vitro. (A) Two independent clones of Fig. 3. c-Myc effects on cell cycle progression in antiestrogen-arrested cells. MCF-7 cells constitutively overexpressing c-Myc (∼2-fold compared with (A) MCF-7 cells inducibly expressing c-Myc under the influence of the zinc- wild type MCF-7 cells) were treated with ICI 182,780 at concentrations of inducible metallothionein promoter were arrested with 10 nM of the pure 10 nM or 100 nM, or ethanol vehicle. After 48 h, changes in S phase were antiestrogen, ICI 182,780, for 48 h and compared to similarly-treated con- measured by DNA flow cytometry and expressed as a percentage of the S trol cells bearing the empty vector only, and parental MCF-7 cells. Estrogen phase in control cells. (B) MCF-7 cells inducibly expressing c-Myc under the ␤ ␮ (100 nM 17 -estradiol) or zinc (65 M) were added to arrested cells, and influence of a zinc-inducible promoter were treated with increasing concen- 48 h later changes in cell cycle phase distribution were measured by flow trations of zinc while in exponential growth phase, and 3 h later were treated cytometry as previously described [12]. Nocodazole at a final concentration with 10 nM of the pure antiestrogen, ICI 182,780. After 24 h, changes in S of 50 ng/mL was added to inhibit cytokinesis and allow all cells that pro- phase were measured by DNA flow cytometry and expressed as a percent- gressed to accumulate in G2/M. (B) After pre-treatment for 24 h with c-Myc age of the S phase in vehicle (ethanol)-treated exponentially growing cells. siRNA or RISC-free control siRNA (Dharmacon, USA), MCF-7 cells were In parallel, protein lysates were prepared from zinc-treated cells at 3 h and treated with 10 nM ICI 182,780 for 48 h to induce cell cycle arrest. Cells were immunoblotted for c-Myc (as described previously [12]). then stimulated with estrogen (100 nM 17␤-estradiol) for 40 h and changes in cell cycle phase distribution were measured by DNA flow cytometry as described above. disease [14] and raise the question of the functional consequences of elevated c-Myc expression in ER+ tumours, striking when only nuclear expression was evaluated (77% particularly as it relates to response to endocrine therapies. versus 47%). Fig. 2 depicts the predominant staining patterns observed in the breast cancer tissue microarray cores. 3.2. c-Myc is required for estrogen-induced cell cycle Together these data demonstrate that c-myc mRNA and progression following antiestrogen arrest c-Myc protein expression are detectable in the majority of breast cancers with elevated expression observed in both ER+ To determine the relationship between c-Myc- and and ER− phenotypes. Such data challenge the view that estrogen-induced cell cycle progression, clonal cell lines c-Myc overexpression is predominantly a feature of ER− expressing c-Myc under the influence of the zinc-inducible, C.M. McNeil et al. / Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155 151

nocodazole that inhibited further cell cycle progression and re-entry into G1 phase. This effect was analogous to that seen in empty vector or parental MCF-7 cells upon the addition of estrogen (Fig. 3A). Since, these data conﬁrmed earlier experiments demonstrating that c-Myc induction was sufﬁcient for cell cycle progression in antiestrogen-arrested cells we then determined if estrogen induction of c-Myc was necessary for estrogen- induced cell cycle progression. MCF-7 cells treated with both antiestrogen and siRNA directed against c-myc remained growth-arrested upon stimulation with estrogen (Fig. 3B). These data suggest not only that c-Myc induction is capable of replicating the proliferative effects of estrogen, but also that c-Myc expression is necessary for estrogen-induced cell cycle progression.

3.3. Constitutive or inducible expression of c-Myc attenuates the growth-inhibitory response to antiestrogens in vitro

The ability of the pure antiestrogen ICI 182,780 to decrease proliferation of MCF-7 cells expressing different levels of c-Myc was assessed in two experimental systems, Fig. 5. An experimental paradigm for identifying estrogen-regulated Myc using as an endpoint the percentage of the cell population in S target genes in breast cancer cells. MCF-7 cells inducibly overexpressing phase 24–48 h after antiestrogen treatment. In the first study, c-Myc (wild type, and c-Zip, a deletion mutant of c-Myc in which the N- MCF-7 cell lines that constitutively overexpressed c-Myc at terminal transcriptional activation domain is absent), parental cells, and cells alevel∼2-fold higher than that seen in empty vector control bearing the empty vector only were growth-arrested with 10 nM ICI 182,780 cells were employed, while in the second paradigm antiestro- for 48 h. Parental cells were treated with estrogen (100 nM 17␤-estradiol) or vehicle (ethanol), and cells transfected with c-Myc (wild type or c-Zip) or gen sensitivity was tested in cell lines inducibly expressing the empty vector were treated with zinc (65 ␮M). RNA was prepared from c-Myc. Fig. 4A demonstrates the attenuation of the anti- cells harvested at 6 h post-treatment, and analysed using the Affymetrix gene proliferative effect of ICI 182,780 in two clones of MCF-7 chip platform and Bayesian linear modelling methods in the limma package cells in which c-Myc was constitutively overexpressed. These [36]. From the list of probes that were significantly regulated by estrogen data demonstrate that an ∼2-fold increase in c-Myc results (relative to vehicle-treated cells, adjusted p < 0.01), those also significantly ∼ regulated by c-Myc and not by c-Zip (relative to zinc-treated empty vector in an 50% decrease in sensitivity to the anti-proliferative cells) were further selected to yield a number of downstream candidate genes effect of ICI 182,780 in MCF-7 cells. In order to establish if for further in vitro validation studies. higher levels of c-Myc overexpression can further attenuate the anti-proliferative effects of antiestrogens on exponentially metallothionein promoter were employed as described pre- growing cells in vitro, c-Myc overexpression was induced in viously [10]. As depicted in Fig. 3A, induction of c-Myc 2 clonal cell lines expressing c-Myc under the influence of a expression with zinc was able to mimic the effects of estrogen zinc-inducible promoter as described above. c-Myc expres- action in initiating cell cycle progression in antiestrogen- sion was rapidly induced following zinc treatment and after arrested cells, such that the majority of the cell population 3 h, the cells were treated with 10 nM ICI 182,780. After 24 h exited from G0/G1 following treatment with estradiol or zinc. treatment with antiestrogen, cells with no c-Myc induction Cells were arrested in G2/M due to concurrent treatment with (i.e. 0 ␮M zinc added) demonstrated the expected decline in

Table 1 Immunohistochemical assessment of c-Myc expression in a cohort of 217 breast cancers Positive staining ≥25% at ≥2 + intensity ≥50% at ≥2 + intensity ≥75% at ≥2 + intensity

ER +ve ER −ve ER +ve ER −ve ER +ve ER −ve (n = 157 (72%)) (n = 60 (28%)) (n = 157 (72%)) (n = 60 (28%)) (n = 157 (72%)) (n = 60 (28%)) Nuclear positive 121 (77%) 28 (47%) 97 (62%) 22 (37%) 45 (29%) 8 (13%) Cytoplasmic positive 123 (78%) 41 (68%) 107 (68%) 22 (37%) 82 (52%) 33 (55%) Nuclear and cytoplasmic positive 100 (46%) 23 (38%) 78 (47%) 18 (30%) 36 (23%) 8 (13%) Positive staining was deﬁned as ≥2 + intensity. The data were analysed at cut-offs of 25%, 50% and 75% of cancer cells positive. In our cohort, approximately 30% normal breast tissue cells stained for c-Myc with 2 + intensity or greater. The threshold for ER positivity was 10% for any degree of nuclear staining. Consistent with the representative nature of the cohort, 72% of the cancers were ER positive. 152 C.M. McNeil et al. / Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155

S phase to about 30% of that observed in untreated exponentially growing cells (Fig. 4B). However, as the concentration of zinc was incrementally increased, resulting in a concurrent increase in c-Myc expression (Fig. 4B, lower panel), the S phase following antiestrogen treatment was also incrementally increased. At the highest concentration of zinc (60 ␮M), the antiestrogen-induced decline in S phase was signiﬁcantly attenuated to ∼80% of that seen in untreated cells (Fig. 4B, upper panel). Together these data clearly demonstrate that c-Myc overexpression can attenuate the growth inhibitory response to antiestrogen in a concentration-dependent manner and potentially, at even higher levels of expression, render cells completely insensitive to antiestrogens. These and other published data [28,29] further support a potential role for c-Myc overexpression in the development of endocrine resistance in breast cancer.

3.4. An experimental paradigm for identifying estrogen-regulated target genes in breast cancer

We next adapted the inducible c-Myc model to facilitate the identification of estrogen target genes that were regulated secondarily to estrogen induction of c-Myc expression. The downstream targets of both c-Myc and estrogen are numerous, reflecting their pleiomorphic cellular effects. Therefore, in order to identify genes that are downstream of both estrogen and c-Myc, gene expression profiling was undertaken following the scheme outlined in Fig. 5. MCF-7 cells expressing wild type c-Myc under the control of the metallothionein promoter were stimulated with zinc, while parental cells were treated with either estrogen or vehicle (ethanol). In addition, MCF-7 cells containing vector only, or a deletion mutant of c-Myc (c-Zip) in which the two N-terminal transcriptional activation domains, Myc box I and Myc box II, are Fig. 6. EMU4 is an estrogen and c-Myc target gene. (A) Parental MCF-7 cells were antiestrogen-pretreated (ICI 182,780, 10 nM, 48 h) then treated absent [34], were stimulated with zinc. The gene expression with estrogen (100 nM 17␤-estradiol) or vehicle (ethanol). In parallel, empty profiles following 6 h of these treatments were then deter- vector cells or cells expressing c-Myc under the control of the metalloth- mined using Affymetrix GeneChips (U133 Plus V2.0), and ionein promoter were pretreated with antiestrogen and then treated with zinc probe sets significantly regulated by estrogen (relative to ␮ (65 M). RNA was prepared, and real-time PCR was performed in triplicate vehicle-treated control cells) were identified as those with using a RotorGene RG-3000 (Corbett Research, Australia). Data analyses p < 0.01 (adjusted for multiple hypothesis testing). Probe sets were performed using the Ct method relative to a 0 h control sample. The change in the expression of EMU4 was measured at 6 and 12 h and is pre- that were also significantly regulated by c-Myc (relative to sented normalised to GAPDH level. (B) Schematic diagram illustrating the zinc-treated empty vector cells), but not by c-Zip were also EMU4 proximal promoter reporter construct. The sequence between −799 identified and compared with the estrogen-regulated probes. and +43 bp of the EMU4 gene was ligated upstream of the luciferase cDNA This analysis revealed that approximately two-thirds of the in the pGL3-basic reporter construct. The locations of the potential c-Myc binding sites (E-boxes) are indicated. MCF-7 cells were transfected with the estrogen-regulated probes were upregulated, i.e. 635 of 968 EMU4 luciferase reporter construct and increasing amounts of the c-Myc significantly regulated probe sets, and of these approximately expression construct, pCDNA3.1-cMyc. All samples were cotransfected half, i.e. 322 of 635, were also regulated by c-Myc. A signifi- with the Renilla luciferase reporter construct, pRLSV40; which was used for cant proportion of these genes had been previously identified normalisation of transfection efficiency. Cells were transfected using Lipo- as either estrogen- or c-Myc-regulated, indicating that the fectAMINE2000 (Invitrogen, Australia) and luciferase was assayed using the dual luciferase reporter assay system (Promega, USA) in accordance model can robustly identify genes of likely relevance to estro- with manufacturer’s recommendations. All transfections were performed in gen and c-Myc action. triplicate and are presented as mean ± S.D. All values are relative to the A number of candidate genes were selected for further activity of the pGL3-basic reporter. evaluation as downstream targets of estrogen and c-Myc action, on the basis of degree of upregulation, potential c- Myc binding sites in the promoter region, or other motifs suggestive of a functional role in cell proliferation. One such C.M. McNeil et al. / Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155 153

Fig. 7. EMU4 expression correlates with Myc expression in breast cancer cell lines. (A) Semi-quantitative RT-PCR was performed to assess the expression of c-myc and EMU4 mRNA in total RNA isolated from a battery of breast cancer cell lines. PCR products were generated in 20–25 cycles, run on 1% agarose gels and stained with ethidium bromide, before UV visualisation and photography. (B) Total RNA from a larger series of breast cancer cell lines was used for real-time qRT-PCR in triplicate using an ABI Prism 7900HT Sequence Detection System (Applied Biosystems, USA) with Taq-Man probes for EMU4 and c-Myc, normalised using RPLP0. Data analyses were performed using the Ct method, relative to expression in the 184 normal breast epithelial cell strain. candidate gene was a novel gene that was upregulated by both ing amounts of a c-Myc expression vector were cotransfected estrogen and c-Myc, which we have named EMU4 (Estrogen- with the EMU4 promoter luciferase reporter, an increase in and Myc-upregulated) and is discussed further below. luciferase expression was clearly apparent (Fig. 6B). These data demonstrate that c-Myc can act as a transcriptional acti- 3.5. EMU4 is a novel estrogen and c-Myc target gene vator of EMU4 gene expression, identifying EMU4 as a novel c-Myc target gene. In order to confirm regulation of EMU4 by estrogen and c-Myc, parental MCF-7 cells and clonal cell lines expressing 3.6. EMU4 and c-Myc mRNA expression are correlated inducible c-Myc (as above), were pretreated for 48 h with in breast cancer cell lines ICI 182,780 then treated with ethanol, estrogen or zinc as appropriate, and RNA was then extracted for PCR analysis. To further investigate EMU4 as a potential c-Myc target Empty vector cells were also included as controls. Quanti- gene in breast cancer, both EMU4 and c-myc mRNA expres- tative RT-PCR was performed for EMU4, 6 and 12 h after sion were analysed in a battery of human breast cancer cell treatment. At 6 h, the EMU4 mRNA level in the inducible lines using semiquantitative RT-PCR (Fig. 7A). These data c-Myc cell line was increased relative to empty vector con- demonstrated that c-Myc expression and EMU4 expression trol cells by 2.5–3-fold. This induction continued to at least are closely related. In a separate experiment using quanti- 12 h post-treatment, with a level of induction of 2–2.5-fold tative RT-PCR, a strong positive correlation between c-myc relative to control (Fig. 6A). A similar degree of induction of and EMU4 mRNA expression in this series of 16 breast car- EMU4 mRNA was observed following 6 and 12 h treatment cinoma cell lines was observed (Fig. 7B), suggesting that with estrogen validating the Affymetrix GeneChip data and EMU4 expression may be a good surrogate marker of c-Myc confirming that EMU4 is indeed both an estrogen and c-Myc activity in breast cancer. target gene in breast cancer cells. To confirm a potential role for c-Myc in the transcriptional regulation of EMU4, a luciferase reporter construct contain- 4. Conclusions ing ∼800 bp of the EMU4 proximal promoter was transfected into parental MCF-7 cells. This sequence included three puta- Our in vitro studies demonstrate that both constitutive and tive c-Myc-responsive E-box motifs (Fig. 6B). When increas- inducible overexpression of c-Myc in the estrogen-responsive 154 C.M. McNeil et al. / Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155 breast cancer cell line MCF-7 confer partial resistance to References the anti-proliferative effects of the antiestrogen ICI 182,780. Furthermore, this effect is dependent on the degree of c- [1] R. Peto, J. Boreham, M. Clarke, C. Davies, V.Beral, UK and USA breast Myc overexpression. These data are consistent with previous cancer deaths down 25% in year 2000 at ages 20–69 years, Lancet 355 (9217) (2000) 1822. studies, in which inducible expression of c-Myc re-initiated [2] G.A. Colditz, Relationship between estrogen levels, use of hormone cell cycle progression in antiestrogen-arrested cells [12,29]. replacement therapy, and breast cancer, J. Natl. Cancer Inst. 90 (11) Published data indicate that c-Myc likely exerts its prolifer- (1998) 814–823. ative effects, at least in part, through repression of cyclin- [3] EBCTCG, Early Breast Cancer Trialists’ Collaborative Group, Tamox- dependent kinase inhibitor p21WAF1/CIP1 expression [29].It ifen for early breast cancer: an overview of the randomised trials, Lancet 351 (9114) (1998) 1451–1467. remains unclear whether this occurs through direct modula- [4] A. Howell, J. Cuzick, M. Baum, A. Buzdar, M. Dowsett, J.F. Forbes, WAF1/CIP1 tion of the p21 promoter by the binding of c-Myc to G. Hoctin-Boes, J. Houghton, G.Y. Locker, J.S. Tobias, Results of the an E-box element [29], or indirectly via c-Myc titrating Sp1 ATAC (Arimidex, Tamoxifen, Alone or in Combination) trial after com- away from the Spl binding sites in the p21WAF1/CIP1 promoter pletion of 5 years’ adjuvant treatment for breast cancer, Lancet 365 [35]. However, the large number of c-Myc-regulated genes (9453) (2005) 60–62. [5] R.I. Nicholson, C. Staka, F. Boyns, I.R. Hutcheson, J.M. Gee, Growth identified here in association with estrogen-induced mitoge- factor-driven mechanisms associated with resistance to estrogen depri- nesis raises the possibility that other downstream targets of vation in breast cancer: new opportunities for therapy, Endocr. Relat. c-Myc may contribute to the proliferative effect of estrogen. Cancer 11 (4) (2004) 623–641. To this end, we have used gene expression profiling to [6] R.I. Nicholson, S.R. Johnston, Endocrine therapy—current benefits and identify genes that are coordinately regulated with c-myc limitations, Breast Cancer Res. Treat. 93 (Suppl. 1) (2005) S3–S10. [7] A. Ring, M. Dowsett, Mechanisms of tamoxifen resistance, Endocr. in association with the proliferative response to estrogen in Relat. Cancer 11 (4) (2004) 643–658. breast cancer cells. One such target is the novel gene, EMU4. [8] S.F. Doisneau-Sixou, C.M. Sergio, J.S. Carroll, R. Hui, E.A. Musgrove, The level of EMU4 mRNA was correlated with the level of R.L. Sutherland, Estrogen and antiestrogen regulation of cell cycle pro- c-myc expression following estrogen treatment in a battery gression in breast cancer cells, Endocr. Relat. Cancer 10 (2) (2003) of human breast cancer cell lines. Further work is required to 179–186. [9] D. Dubik, T.C. Dembinski, R.P. Shiu, Stimulation of c-myc onco- evaluate the biological role of this novel c-Myc target and its gene expression associated with estrogen-induced proliferation of relevance, if any, to c-Myc action in breast cancer. human breast cancer cells, Cancer Res. 47 (24 part 1) (1987) 6517– Our data also confirm a broad range of c-Myc expression 6521. in human breast cancer at both the RNA and protein level. [10] J.S. Carroll, A. Swarbrick, E.A. Musgrove, R.L. Sutherland, Mecha- Furthermore, this spectrum of expression is displayed across nisms of growth arrest by c-myc antisense oligonucleotides in MCF-7 − breast cancer cells: implications for the antiproliferative effects of antie- both ER+ and ER tumours. Thus, in the light of our in vitro strogens, Cancer Res. 62 (11) (2002) 3126–3131. data, aberrant expression of c-Myc in ER+ disease may be [11] P.H. Watson, R.T. Pon, R.P. Shiu, Inhibition of c-myc expression by of clinical relevance in predicting sensitivity to antiestrogen phosphorothioate antisense oligonucleotide identifies a critical role for therapy. The in vitro studies described here have led to the c-myc in the growth of human breast cancer, Cancer Res. 51 (15) (1991) identification of a number of novel estrogen and c-Myc-target 3996–4000. [12] O.W. Prall, E.M. Rogan, E.A. Musgrove, C.K. Watts, R.L. Sutherland, genes associated with the proliferative response to estrogen c-Myc or cyclin D1 mimics estrogen effects on cyclin E-Cdk2 activation in breast cancer cells and are likely to be of relevance in and cell cycle reentry, Mol. Cell. Biol. 18 (8) (1998) 4499–4508. addressing the important problem of antiestrogen resistance, [13] E.C. Robanus-Maandag, C.A. Bosch, P.M. Kristel, A.A. Hart, I.F. and in contributing further to our broader understanding of Faneyte, P.M. Nederlof, J.L. Peterse, M.J. van de Vijver, Association the biology of the estrogen signalling pathways. of c-Myc amplification with progression from the in situ to the invasive stage in c-Myc-amplified breast carcinomas, J. Pathol. 201 (1) (2003) 75–82. [14] S.L. Deming, S.J. Nass, R.B. Dickson, B.J. Trock, c-Myc amplifica- Acknowledgements tion in breast cancer: a meta-analysis of its occurrence and prognostic relevance, Br. J. Cancer 83 (12) (2000) 1688–1695. Work in this laboratory is supported by grants from the fol- [15] C. Ruiz, S. Seibt, K. Al Kuraya, A.K. Siraj, M. Mirlacher, P. Schraml, R. Maurer, H. Spichtin, J. Torhorst, S. Popovska, R. Simon, G. Sauter, lowing organisations: Cancer Institute NSW, R.T. Hall Trust, Tissue microarrays for comparing molecular features with proliferation Australian Cancer Research Foundation (ACRF), Associ- activity in breast cancer, Int. J. Cancer 118 (9) (2006) 2190–2194. ation for International Cancer Research (AICR) and the [16] E.M. Berns, J.G. Klijn, M. Smid, I.L. van Staveren, M.P.Look, W.L.van National Health and Medical Research Council of Australia Putten, J.A. Foekens, TP53 and MYC gene alterations independently (NH&MRC). Catriona McNeil is a Cancer Institute NSW predict poor prognosis in breast cancer patients, Genes Chromosomes Cancer 16 (3) (1996) 170–179. Scholar and the recipient of an NH&MRC Postgraduate [17] C.M. Schlotter, U. Vogt, U. Bosse, B. Mersch, K. Wassmann, c-Myc, Scholarship. Niamh Murphy is the recipient of a Sydney not HER-2/neu, can predict recurrence and mortality of patients with Breast Cancer Institute Fellowship. Elizabeth Musgrove, Ali- node-negative breast cancer, Breast Cancer Res. 5 (2) (2003) R30–R36. son Butt and Susan Henshall are Cancer Institute NSW [18] K. Al-Kuraya, P. Schraml, J. Torhorst, C. Tapia, B. Zaharieva, H. Research Fellows. Robert Sutherland is a Senior Principal Novotny, H. Spichtin, R. Maurer, M. Mirlacher, O. Kochli, M. Zuber, H. Dieterich, F. Mross, K. Wilber, R. Simon, G. Sauter, Prognostic rel- Research Fellow of the NH&MRC. We thank Joanne Scorer evance of gene amplifications and coamplifications in breast cancer, for assistance in the preparation of the figures. Cancer Res. 64 (23) (2004) 8534–8540. C.M. McNeil et al. / Journal of Steroid Biochemistry & Molecular Biology 102 (2006) 147–155 155

[19] A. Scorilas, T. Trangas, J. Yotis, C. Pateras, M. Talieri, Determination [28] M. Venditti, B. Iwasiow, F.W. Orr, R.P. Shiu, c-Myc gene expression of c-myc amplification and overexpression in breast cancer patients: alone is sufficient to confer resistance to antiestrogen in human breast evaluation of its prognostic value against c-erbB-2, cathepsin-D and cancer cells, Int. J. Cancer 99 (1) (2002) 35–42. clinicopathological characteristics using univariate and multivariate [29] S. Mukherjee, S.E. Conrad, c-Myc suppresses p21WAF1/CIP1 expression analysis, Br. J. Cancer 81 (8) (1999) 1385–1391. during estrogen signaling and antiestrogen resistance in human breast [20] S.P. Linke, T.M. Bremer, C.D. Herold, G. Sauter, C. Diamond, A mul- cancer cells, J. Biol. Chem. 280 (18) (2005) 17617–17625. timarker model to predict outcome in tamoxifen-treated breast cancer [30] C.M. McNeil, C.M. Sergio, A.J. Butt, E.A. Musgrove, R.L. Suther- patients, Clin. Cancer Res. 12 (4) (2006) 1175–1183. land, Overexpression of c-Myc confers anti-estrogen resistance in [21] R. Naidu, N.A. Wahab, M. Yadav, M.K. Kutty, Protein expression and breast cancer cells, Breast Cancer Res. Treat. 94 (Suppl. 1) (2005) molecular analysis of c-myc gene in primary breast carcinomas using S162. immunohistochemistry and differential polymerase chain reaction, Int. [31] S. Hiscox, W.G. Jiang, K. Obermeier, K. Taylor, L. Morgan, R. Burmi, J. Mol. Med. 9 (2) (2002) 189–196. D. Barrow, R.I. Nicholson, Tamoxifen resistance in MCF7 cells pro- [22] P. Chrzan, J. Skokowski, A. Karmolinski, T. Pawelczyk, Amplification motes EMT-like behaviour and involves modulation of beta-catenin of c-myc gene and overexpression of c-Myc protein in breast cancer and phosphorylation, Int. J. Cancer 118 (2) (2006) 290–301. adjacent non-neoplastic tissue, Clin. Biochem. 34 (7) (2001) 557–562. [32] E.Y. Sum, D. Segara, B. Duscio, M.L. Bath, A.S. Field, R.L. Suther- [23] J. Blancato, B. Singh, A. Liu, D.J. Liao, R.B. Dickson, A. Thiantanawat, land, G.J. Lindeman, J.E. Visvader, Overexpression of LMO4 induces B.J. Long, A.M. Brodie, Correlation of amplification and overexpres- mammary hyperplasia, promotes cell invasion, and is a predictor of sion of the c-myc oncogene in high-grade breast cancer: FISH, in situ poor outcome in breast cancer, Proc. Natl. Acad. Sci. 102 (21) (2005) hybridisation and immunohistochemical analyses, Br. J. Cancer 90 (8) 7659–7664. (2004) 1612–1619. [33] R. Hui, J.R. Ball, R.D. Macmillan, F.S. Kenny, O.W. Prall, D.H. Camp- [24] M.G. Le, M.C. Mathieu, S. Douc-Rasy, M.L. Le Bihan, H. Adb El bell, A.L. Cornish, R.A. McClelland, R.J. Daly, J.F. Forbes, R.W. All, M. Spielmann, G. Riou, c-Myc, p53 and bcl-2, apoptosis-related Blamey, E.A. Musgrove, J.F. Robertson, R.I. Nicholson, R.L. Suther- genes in infiltrating breast carcinomas: evidence of a link between bcl- land, EMS1 gene expression in primary breast cancer: relationship to 2 protein over-expression and a lower risk of metastasis and death in cyclin D1 and oestrogen receptor expression and patient survival, Onco- operable patients, Int. J. Cancer 84 (6) (1999) 562–567. gene 17 (8) (1998) 1053–1059. [25] K.I. Bland, M.M. Konstadoulakis, M.P. Vezeridis, H.J. Wanebo, Onco- [34] B. Mukherjee, S.D. Morgenbesser, R.A. DePinho, Myc family oncopro- gene protein co-expression. Value of Ha-ras, c-myc, c-fos, and p53 teins function through a common pathway to transform normal cells in as prognostic discriminants for breast carcinoma, Ann. Surg. 221 (6) culture: cross-interference by Max and trans-acting dominant mutants, (1995) 706–718. Genes Dev. 6 (8) (1992) 1480–1492. [26] G.M. Callagy, P.D. Pharoah, S.E. Pinder, F.D. Hsu, T.O. Nielsen, J. [35] A.L. Gartel, X. Ye, E. Goufman, P. Shianov, N. Hay, F. Najmabadi, Ragaz, I.O. Ellis, D. Huntsman, C. Caldas, Bcl-2 is a prognostic marker A.L. Tyner, Myc represses the p21(WAF1/CIP1) promoter and inter- in breast cancer independently of the Nottingham Prognostic Index, acts with Sp1/Sp3, Proc. Natl. Acad. Sci. 98 (8) (2001) 4510– Clin. Cancer Res. 12 (8) (2006) 2468–2475. 4515. [27] A. Sierra, X. Castellsague, A. Escobedo, A. Moreno, T. Drudis, A. [36] G.K. Smyth, Limma: linear models for microarray data, in: R. Gentle- Fabra, Synergistic cooperation between c-Myc and Bcl-2 in lymph node man, V. Carey, S. Dudoit, R. Irizarry, W. Huber (Eds.), Bioinformat- progression of T1 human breast carcinomas, Breast Cancer Res. Treat. ics and Computational Biology Solutions Using R and Bioconductor, 54 (1) (1999) 39–45. Springer, New York, 2005, pp. 397–420. Research Article

Cyclin D1b Is Aberrantly Regulated in Response to Therapeutic Challenge and Promotes Resistance to Estrogen Antagonists

Ying Wang,7 Jeffry L. Dean,1,2,7 Ewan K.A. Millar,4,5 Thai Hong Tran,1,2 Catriona M. McNeil,4 Craig J. Burd,6 Susan M. Henshall,4 Fransiscus E. Utama,1,2 Agnieszka Witkiewicz,1,3 Hallgeir Rui,1,2 Robert L. Sutherland,4 Karen E. Knudsen,1,2,7 and Erik S. Knudsen1,2,7

1Kimmel Cancer Center; 2Department of Cancer Biology and 3Department of Pathology, Thomas Jefferson University, Philadelphia Pennsylvania; 4Cancer Research Program, Garvan Institute of Medical Research, St. Vincent’s Hospital, Sydney, NSW; 5Department of Anatomical Pathology, South Eastern Area Laboratory Service, St. George Hospital, Sydney, NSW; 6National Institutes of Environmental Health Science, Research Triangle Park, North Carolina; and 7Department of Cell and Cancer Biology, University of Cincinnati, Cincinnati, Ohio

Abstract signals to promote entry into the cell cycle (7, 8). Given this central Cyclin D1 is a key mediator of cell cycle progression that is role in modulating cellular proliferation, it is not surprising that aberrantly regulated in multiple cancers, especially in breast deregulation of cyclin D1 is a common facet of human cancer. cancers. A number of studies have indicated that a polymor- Cyclin D1 was found to be the site of Prad1 rearrangement in phism in a splice donor site in the cyclin D1 gene is associated parathyroid cancer and is involved in a large fraction of with alternative splicing and the production of the alternative translocations occurring in mantle cell lymphoma (1, 9, 10). cyclin D1b transcript. Furthermore, this polymorphism is Furthermore, amplification of the cyclin D1 gene is found in breast selectively associated with disease outcomes. However, rela- cancer at relatively high frequency and many primary breast tively little is known regarding the protein product of the tumors overexpress cyclin D1 protein in the absence of a clear alternatively spliced message, cyclin D1b. Using antibodies underlying genetic mechanism (2, 4, 11). These studies indicate specific for cyclin D1b, it was found that this protein is readily that cyclin D1 function may be particularly relevant in breast detectable in a number of cancer cell lines and primary breast tumorigenesis. Consistent with this hypothesis, ectopic expression cancers. Whereas cyclin D1b interacts with cyclin-dependent of cyclin D1 in mouse mammary models can drive tumor kinase 4 (CDK4), it is relatively inefficient at mediating RB development, whereas deletion of the cyclin D1 gene potently phosphorylation and cell cycle progression in model systems protects against mammary tumorigenesis (12, 13). due to the lack of exon 5 of cyclin D1–encoded sequences. Cyclin D1 is a critical regulator of cyclin-dependent kinase 4 However, cyclin D1b protein levels are not significantly (CDK4) function that contributes to cell cycle progression. The attenuated by DNA damage or antiestrogen treatment, cyclin D1 protein directly associates with CDK4 and CDK6 and indicating that the protein may have significant effect on the stimulates catalytic activity of the complex (14, 15). Interestingly, response to such therapeutic modalities. Whereas enforced whereas most CDK/cyclin complexes phosphorylate multiple expression of cyclin D1b was not sufficient to abrogate DNA substrates harboring S-P or T-P motifs, cyclin D1–associated damage checkpoint responses, it did efficiently overcome cell kinase activity is relatively specific toward the retinoblastoma cycle arrest mediated by antiestrogen therapeutics. This tumor suppressor protein (RB), and the related proteins p107 and action of cyclin D1b was not associated with effects on p130 (14, 16, 17). Cyclin D1-catalyzed phosphorylation leads to the estrogen receptor activity, but was rather dependent on disruption of RB activity, thus promoting cell cycle progression functional association with CDK4. Combined, these studies (16). RB represents a seminal target for cyclin D1 function, as indicate that the cyclin D1b protein is aberrantly regulated RB-deficient cells efficiently bypass the requirement for D-type and could contribute to therapeutic failure in the context of cyclins and CDK4 activity (18). Thus, it is believed that a principle ER-positive breast cancer. [Cancer Res 2008;68(14):5628–38] means through which cyclin D1 acts to control cell cycle progression is via the phosphorylation of RB. However, a number Introduction of additional mechanisms of cyclin D1 function are known to contribute to cell cycle regulation and tumorigenesis. Particularly, Cyclin D1 is a key mediator of cell cycle progression that is cyclin D1 can serve as a transcriptional modulatory protein leading aberrantly regulated in multiple cancers (1–4). This protein was to the stimulation of specific transcription factors (e.g., estrogen initially identified based on complementation of yeast cyclin receptor; refs. 19, 20), whereas mediating repression of others mutants and as a delayed early gene during mitogen stimulated (e.g., androgen receptor; refs. 21, 22). Recent studies have suggested proliferation (1, 4–6). It is now clear that cyclin D1 serves to that such functions of cyclin D1 could be important in mammary integrate the activity of multiple mitogenic signaling cascades. This gland development and hormone-dependent cancers (23, 24). function of cyclin D1 is critical for appropriate cell cycle regulation Whereas the conventional cyclin D1 protein that is encoded by as ectopic expression of cyclin D1 can promote the G -S transition, 1 five exons at the cyclin D1 locus has been extensively studied, it is whereas cyclin D1 deficiency compromises the ability of mitogenic now clear that variant cyclin D1 proteins are also encoded from this locus (1, 2, 4, 25). Specifically, an alternatively spliced form of Requests for reprints: Erik S. Knudsen, 233 South 10th Street, BLSB-Room cyclin D1 is produced, which has been termed cyclin D1b (26–28). 1002, Philadelphia, PA 19107. Phone: 513-558-8885; E-mail: eknudsen@ This variant is encoded by the first four exons, but due to a lack of kimmelcancercenter.org. I2008 American Association for Cancer Research. splicing excludes exon 5 encoded sequences and includes doi:10.1158/0008-5472.CAN-07-3170 sequences derived from the fourth intron (26, 28). This splicing

Cancer Res 2008; 68: (14). July 15, 2008 5628 www.aacrjournals.org Cyclin D1b Promotes Resistance to Estrogen Antagonists event is believed to be modulated by a polymorphism at the splice Cells were fixed and BrdUrd-positive nuclei were visualized as previously donor site and both the polymorphism and cyclin D1b expression described (35). have been associated with enhanced cancer risk and poor clinical Adenoviral infections and bivariate flow cytometry and 3-(4,5- outcome (4, 25, 26, 29). dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide analyses. MCF-7 cells were infected with adenovirus encoding cyclin D1 or cyclin Due to the importance of cyclin D1 in cell cycle control, the D1b. GFP adenovirus, a kind gift from Gustavo Leone (Ohio State University), protein is subject to complex regulation of subcellular localization was used as a control. The infections were performed at a calculated and stability. Primarily, protein localization and stability are multiplicity of infection of 50 to 100 for f95% to 100% infection efficiency modulated by phosphorylation of T286 (1, 30, 31), which is present after 16 h (as determined by plaque assay in 293 cells). Cells were treated with in exon 5 encoded sequences of cyclin D1 and are thus absent in CDDP (16 and 32 Amol/L) or vehicle for an additional 16 h or cells were the cyclin D1b protein. As a result, the cyclin D1b protein remains washed with PBS extensively and switched to medium with 10% CDT serum exclusively nuclear and escapes this mode of regulation of cyclin in the presence of 1 nmol/L 4-hydroxy tamoxifen or 1 Amol/L ICI for 36 or D1 (32, 33). Prior studies had shown that mutation of T286 resulted 72 h. For flow cytometry analysis, Ad-GFP,Ad-D1, and Ad-D1b infected MCF-7 in enhanced transformation in NIH-3T3 cells (31), and correspond- cells were pulse labeled with BrdUrd for 1 h. Bivariate flow cytometry analysis ingly, cyclin D1b harbors enhanced transforming potential relative of DNA synthesis (BrdUrd incorporation) and DNA content (propidium iodide, PI) was performed. At least 20,000 gated events were collected for each to cyclin D1 (32, 33). These studies have fueled speculation that sample. The percentage of BrdUrd-positive cells was quantified. cyclin D1b expression could represent a potent event associated For measuring proliferation, 5,000 cells infected with cyclin D1 or cyclin with tumorigenesis (4, 25). D1b encoding adenoviruses were plated per well in 96-well flat-bottomed Here, we probed the regulation of endogenous cyclin D1b plates in quadruplicate. Cells were treated with tamoxifen or ICI at the time protein and assessed the functional activity of cyclin D1b in the of plating to a total volume of 90 AL. At 24, 48, 72, 96, and 120 h postplating context of therapeutic modalities used in the treatment of breast cell number/viability was determined by adding 3-(4,5-dimethylthiazol-2- cancers. These studies show that the regulation of cyclin D1b is yl)-2,5-dibenzyltetrozolium bromide (MTT) reagent (50 Agin10AL PBS) to distinct from that elicited on cyclin D1 and that cyclin D1b is a each well. Cells were incubated for 4 h, then solubilized with 110 AL of 10% critical modulator of the response to estrogen antagonists used in Triton X-100 (Amresco), 0.1 N HCl (Fisher) in anhydrous isopropanol the treatment of this disease. (Fisher). Colorimetric analysis was performed at 570 nm. Statistical analysis. All statistical analysis was performed using Prism 4.0c software (GraphPad Software, Inc.) Results were analyzed for statistical Materials and Methods significance using the Student’s unpaired t test and SE. For all analyses, Cell culture and drug treatment. MCF-7, ZR751, T47D, SW13, C33A, P < 0.05 was considered significant. SAOS-2, and U2OS cells were maintained in DMEM. BT549, A427, H596, PCR-RFLP analysis and reverse transcription–PCR. Genomic DNA H520, and H2172 cells were maintained in RPMI. HeLa cells were was extracted from cell culture lines. The isolated DNA was quantified, and maintained in IMEM at 37jCin5%CO2. Both DMEM and RPMI were 100 ng of DNA were amplified as described (36). Products were digested supplemented with 10% fetal bovine serum (FBS), IMEM with 5% FBS, with ScrFI restriction enzyme (Biolab) and resolved by agarose gel 100 unit/mL penicillin-streptomycin, and 2 mmol/L L-glutamine. For electrophoresis. Transcript specific reverse transcription–PCR (RT-PCR) culture in steroid hormone-depleted media, MCF-7 cells were maintained was as previously described (36). in phenol red–free DMEM containing 10% charcoal dextran–treated (CDT) Immunoblots and antibodies. Immunoblotting was performed by FBS with addition of 1 nmol/L 4-hydroxy tamoxifen (Sigma-Aldrich) or following standard biochemical techniques. Antibodies against the follow- 1 Amol/L ICI 182,780 (Tocris Bioscience) for indicated time courses. MCF-7 ing proteins were used: cyclin D1 (Ab-3; NeoMarkers), cyclin D1 (DSC-6; Cell cells were treated with a final concentration of 5 Amol/L MG132 (Sigma) Signaling), RB (BD Biosciences PharMingen), phosphorylated RB (Ser780; for 6 h, cisplatin (CDDP; Redford) with a final concentration of either 16 Cell Signaling), CDK4 (C-22; Santa Cruz), GFP (Santa Cruz), p21Cip1 or 32 Amol/L for 16 h, or PD33291 (Pfizer) at a final concentration of (C-19; Santa Cruz), p27Kip1 (BD Biosciences PharMingen), h-tubulin (Santa 250 nmol/L (provided by Pfizer). Cruz), Lamin B (Santa Cruz), actin (Santa Cruz), and cyclin A (Santa Cruz). Plasmids. The RB, CDK4, H2B-GFP, GFP–cyclin D1, and GFP–cyclin D1b Cyclin D1b antibody against cyclin D1b–specific peptide sequence expression plasmids have been described previously (33). GFP–cyclin D1b– YRGRHLVPRKCRGWCQGPQG was generated by Bethyl Laboratories (37). KE mutant was generated by the Quickchange Site-Directed Mutagenesis kit The immunoactivity of cyclin D1b antibody was determined by Western according to the manufacturer (Stratagene) using pEGFP–cyclin D1b as a blots using purified glutathione S-transferase (GST)–cyclin D1 and GST– template. GFP-cyclin D1-DI4 was generated by PCR amplification of cyclin cyclin D1b proteins. Quantification of resultant proteins was carried out D1 exon 1 to exon 4 sequences using pcDNA3.1-HA-cyclin D1 as template. using SDS-PAGE with bovine serum albumin (Sigma) as a standard curve. The resulting fragment was inserted into pCR2.1 as directed by the Band density was measured using SYPRO Red Protein Gel Stain kit manufacturer (Invitrogen). The fragment was subsequently removed from (Cambrex Bio Science Rockland, Inc.). Known concentrations of GST-D1 the pCR vector using BamH1/EcoR1 digestion and inserted into the BglII and GST-D1b were subjected to SDS-PAGE and transferred to an Immobilon and EcoR1 sites of the pEGFP-C1 vector (Clontech). All mutants generated/ membrane (Millipore). GST-D1 and GST-D1b proteins were detected by used were confirmed by sequencing and primer sequences are available cyclin D1 (DCS-6; Ab-3) and cyclin D1b specific antibodies. Proteins were upon request. The ERE-Luciferase, cytomegalovirus h-galactosidase visualized using horseradish peroxidase–conjugated secondary antibody reporter plasmids and the human ERa plasmid have been described (Pierce) along with enhanced chemiluminescence (Perkin-Elmer Life previously (34). Sciences). The relative protein band intensities were determined using Transfection and 5-bromo-2’deoxyuridine incorporation. Plasmids Image J software (version 1.24t). were transfected using the lipid-based transfection reagent FuGENE 6 Reporter assays. MCF-7 cells were cotransfected with cytomegalovirus (Roche Molecular Biochemical) in accordance with the manufacturer’s h-galactosidase reporter plasmid, ERE-Luciferase reporter plasmid, ERa, recommended protocol. In the 5-bromo-2’deoxyuridine (BrdUrd; Roche) and cyclin D1 and cyclin D1b expression plasmids or empty vector at a incorporation experiments, 3 105 MCF-7 cells were plated in six-well 0.5:1:1:1.5 ratio. At 24 h posttransfection, cells were washed with PBS, media culture dishes containing coverslips and incubated overnight. Cotransfec- were replaced, and cells were treated with the indicated concentration of A A tions were done using 1.8 g input DNA plus 0.2 gH2B-GFP DNA. After estradiol (E2) or ethanol vehicle for 24 h. After stimulation, cells were 16 h, cells were washed with PBS and replaced in phenol red–free DMEM harvested, and luciferase activity was determined and normalized against containing 10% CDT with either 1 nmol/L 4-hydroxy tamoxifen or 1 Amol/L h-galactosidase activity for transfection efficiency. Data were collected from ICI was added. Cells were then pulse labeled with 35 Ag/mL BrdUrd for 6 h. three independent experiments. www.aacrjournals.org 5629 Cancer Res 2008; 68: (14). July 15, 2008 Cancer Research

Immunohistochemistry and automated quantitative analyses. Im- under these conditions. In contrast, the YW2 antibody specifically munohistochemistry analyses were performed at two different array recognized cyclin D1b (Fig. 1B). Having confirmed the specificity of platforms with slightly different staining and quantitation protocols:. the antibodies, the relative sensitivity of Ab-3 and YW2 antibodies First, sections of a tissue array generated by cutting edge matrix assembly was determined by blotting the same membrane with both (38) that contained 80 invasive breast carcinoma tissues and 20 normal antibodies simultaneously (Fig. 1B, bottom). These analyses breast tissue were used. Briefly, after deparaffinization and rehydration of array sections, antigen retrieval was performed by microwave treatment in indicated that the cyclin D1b–specific antibody had heightened citrate buffer (pH 6; DAKO). Sections were blocked with 10% goat serum reactivity for cyclin D1b, as compared against the AB3 antibody and followed by incubation of primary cyclin D1b at a dilution of 1:50 for 1 that recognizes cyclin D1. h. Sections were then washed thrice with TBS and subsequently incubated Using the YW2 and Ab3 antibody, the levels of cyclin D1b and D1 with a mouse cytokeratin antibody (DAKO) for 1 h. The cyclin D1b proteins were evaluated in multiple cell lines (Fig. 1C, top). There antibodies were detected using an antirabbit horseradish peroxidase– was clear heterogeneity in the levels of cyclin D1 and cyclin D1b conjugated secondary antibody (DAKO EnVision-Plus), followed by between cell lines. Interestingly, relative levels of expression were incubation with Tyramide-Cy5 (Perkin-Elmer). Cytokeratin was visualized not a direct reflection of the A-870 polymorphism that is believed by further incubating the sections with a mouse secondary antibody to influence the production of cyclin D1b (Fig. 1C, bottom). conjugated to Alexa 488 (Molecular Probes). Finally, all sections were Strikingly, the breast cancer lines MCF-7, BT549, and Zr-75-1 stained with 4¶,6-diamidino-2-phenylindole (DAPI; Vector) for nuclear visualization. Automated quantitative analysis was performed using the expressed detectable levels of cyclin D1b, suggesting that cyclin AQUA/PM2000 Imaging Platform (HistoRx) as described (39). Tissue array D1b could be particularly relevant in reference to breast cancer slides were scanned and images of each breast cancer tissue were captured biology. To compare the relative amount of cyclin D1 and cyclin at different channels detecting FITC/Alexa 488, Cy5, or DAPI. AQUA D1b in a given tumor line, the antibody preparation as used for software was then used to identify epithelial masks based on FITC-positive Fig. 1B was used. Surprisingly, although cyclin D1b is clearly cytokeratin-expressing cells. AQUA scores for cyclin D1b representing detected in MCF-7 cells, based on the higher affinity of the cyclin average signal intensities within epithelial cells. D1b antibody, it represents <10% of the cyclin D1 protein present in Second, immunohistochemistry for cyclin D1b was performed on the cell (Fig. 1D). Similarly, in Zr-75-1 cells, cyclin D1b represents a formalin-fixed paraffin-embedded tissue microarrays (TMA) containing minor portion of the total cyclin D1 protein. This finding may 1-mm diameter cores derived from 175 patients with invasive ductal explain the relative difficulty in detecting endogenous cyclin D1b carcinoma of no special type treated at St. Vincent’s Hospital. Sctions (4 Am thick) were mounted on Superfrost Plus adhesion slides (Lomb Scientific). with reagents that also detect cyclin D1 (27, 28). Interestingly, Slides were baked at 78jC before being deparaffinized in xylene and BT549 cells express undetectable levels of cyclin D1, as has been rehydrated through serial alcohol solutions (100%, 95%, and 70%). previously reported. However, expression of cyclin D1b is clearly Endogenous peroxidase activity was blocked using 3% H2O2 for 5 min, detectable. Thus, a heterogenous expression of cyclin D1b is found followed by protein block serum–free (DAKO Corporation) for 30 min. in breast cancer cell lines and can, in specific circumstances, Antigen retrieval was performed in a pressure cooker at pH 9.0 (Dako represent a significant fraction of the cyclin D1 protein present in retrieval solution S2367) for 10 s. Sections were incubated with cyclin D1b the cell. primary antibody at room temperature at a dilution of 1:150 for 30 min, To determine the expression of cyclin D1b in primary tumors, a followed by detection using Envision+ Rabbit (DAKO) for 30 min at room cohort of 19 normal biopsy specimens and 65 breast cancers temperature and visualization with 3,3¶-diaminobenzidine+ (DAKO) for 10 specimens were analyzed by immunohistochemistry quantitatively min. Sections were then counterstained with Mayer’s hematoxylin (DAKO), A dehydrated, and mounted for analysis. The negative control used using AQUA technology. As shown in Fig. 2 , relatively low levels of concentration-matched nonspecific rabbit IgG in place of the primary cyclin D1b were observed in normal mammary tissue. However, antibody. Stained slides were assessed by an experienced breast pathologist cyclin D1b was broadly expressed in breast cancer specimens and (E.K.A.M.) and scored for the percentage and intensity of cells which its expression was significantly increased in primary tumors showed nuclear staining: 0 negative, 1+ weak, 2+ moderate, 3+ strong. An (P < 0.001). Consistent results were observed in an independent average intensity score was calculated as the mean intensity from two to six cohort of 175 invasive ductal carcinomas, which exhibited cores per patient. heterogeneous levels of cyclin D1b expression (Fig. 2C). Thus, cyclin D1b expression is a feature of breast cancer and its relative level of expression could be related to pathologic features. Results The inefficient catalytic activity of cyclin D1b is due to a lack Cyclin D1b expression in cancer cells and primary breast of exon 5 encoded sequences. Because cyclin D1 proteins are cancer. The cyclin D1 and cyclin D1b proteins differ only in their believed to function in concert with CDK4, the interaction of distinct COOH terminal domains (26). In the case of cyclin D1, the endogenous cyclin D1b with CDK4 was analyzed by coimmuno- exon 5 encoded sequence is the epitope for many of the commonly precipitation (Fig. 3A). Whereas the negative control antibody used commercial antibodies. Conversely, cyclin D1b harbors a (DBF4) failed to precipitate CDK4, the cyclin D1b antibody resulted unique COOH terminal domain that is encoded by intron in significant precipitation of CDK4. Thus, these analyses show 4 containing sequences (Fig. 1A). A 20–amino acid peptide from that the endogenous cyclin D1b protein does associate with CDK4 intron 4 encoded sequences was used to immunize rabbits, and and thus could contribute to cell cycle control in tumor cells by resultant sera was affinity purified. To test the specificity and functioning through the canonical cyclin D1–RB pathway. To relative sensitivity of this reagent (YW2 antibody), GST–cyclin D1 probe the function of cyclin D1b in catalyzing RB phosphorylation and GST–cyclin D1b proteins were purified from bacteria to serve and cell cycle progression with CDK4, the SAOS-2 cell culture as a source of known antigen. The DCS-6 antibody recognizes an model was used. SAOS-2 cells are a well established model for epitope common to cyclin D1 and cyclin D1b and recognized both analyzing cyclin function (16, 33, 40) and lack the expression of RB proteins with equivalent sensitivity (Fig. 1B, top). The Ab-3 and correspondingly express very low/undetectable levels of cyclin antibody recognizes an epitope in the COOH terminus of cyclin D1 and cyclin D1b (Fig. 1C). As such, it is an ideal model for D1 and correspondingly only recognized the GST–cyclin D1 protein evaluating the functional relationships regarding the control of RB

Cancer Res 2008; 68: (14). July 15, 2008 5630 www.aacrjournals.org Cyclin D1b Promotes Resistance to Estrogen Antagonists

Figure 1. Cyclin D1b expression in different cancer cell lines and association with CDK4 in MCF-7 cells. A, diagram depicting genomic structures and splicing of cyclin D1 gene; exons indicated by boxes and introns by lines. The sequence VSEGDVPGSLAGAYRGRHLVPRKCRGWCQGPQG was derived from intron 4. Cyclin D1b antibody was derived from the sequence indicated in the underlined portion. B, YW2-D1b specific antibody efficiently recognizes cyclin D1b protein. Different amounts of GST-D1 and GST-D1b were resolved by SDS-PAGE. GST-D1 and GST-D1b were detected by immunoblotting with cyclin D1 (DCS-6) antibody (top) and YW2-D1b–specific antibody (middle) alone. YW2-D1b and D1 (Ab3) antibodies were used to blot the membranes separately. The separated membranes were exposed together (bottom). C, total protein isolated from the indicated cancer cell lines was resolved by SDS-PAGE. Cyclin D1b (first panel) or cyclin D1 (second panel) expression was detected by immunoblotting with cyclin D1b or D1 (Ab-3) antibody, respectively. Equal loading was confirmed by h-tubulin immunoblot (third panel). Genotype of exon 4–intron 4 boundary polymorphism (bottom). DNA was isolated from indicated cell lines and analyzed as described in Materials and Methods. D, endogenous cyclin D1b and D1 expression in MCF-7, BT549, and Zr-75-1 cells was determined by immunoblot. SAOS-2 cells were used as negative control (top). Equal loading was confirmed by lamin B immunoblot (bottom). phosphorylation by exogenous cyclin D1 proteins. SAOS-2 cells through which they can mediate phosphorylation. To determine were cotransfected with RB, CDK4, and the indicated GFP–cyclin if the effects observed with cyclin D1b were occurring through D1 expression plasmid (Fig. 3C). The use of GFP fusions allowed us CDK4 interactions, cyclin D1b was engineered to contain the to clearly determine that the level of expression in all transfected K112E (KE) mutation (Fig. 3B). This mutation limits the ability of populations was both of equal level (Fig. 3C) and confirm that cyclin D1 to interact with CDK4 (41, 42). The cyclin D1b KE mutant percentage of transfected cells was equivalent (not shown). The was severely compromised for the phosphorylation of RB (Fig. 3C, ectopic expression of CDK4 and cyclin D1 induced phosphorylation lane 3). The ability to phosphorylate RB directly correlated with of RB with high efficiency, such that virtually all of the RB protein cell cycle progression, as measured by BrdUrd incorporation, in was hyperphosphorylated (Fig. 3C, lane 1), and cell cycle this model (Fig. 3D). Whereas both CDK4/cyclin D1b promoted progression was stimulated (Fig. 3D). Consistent with previously BrdUrd incorporation, this activity is compromised relative to published work (33), cyclin D1b proved to be a poor mediator of RB CDK4/cyclin D1 (P < 0.05). In contrast, cyclin D1b KE was phosphorylation (Fig. 3C, lane 2). The differing activity between incapable of promoting cell cycle progression (Fig. 3D). Thus, the cyclin D1 and cyclin D1b could be due to several possible activity of cyclin D1b in this model is apparently dependent on mechanisms. A stimulatory motif in the exon 5 encoded sequences CDK4 activity. To confirm that CDK4 activity is critical for the could be responsible for the enhanced activity of cyclin D1. effects of cyclin D1b in this model, PD-332991 (a highly specific Conversely, an inhibitory motif encoded in intron 4 could be inhibitor of CDK4) was used. As shown in Fig. 3E, PD-332991 mediating the reduced activity of cyclin D1b. To differentiate strongly inhibited CDK4/cyclin D1b induced RB phosphorylation between these possibilities, we developed the cyclin D1-DI4 allele, and cell cycle progression. Together, these data indicate that cyclin which encodes only exons 1 to 4 (Fig. 3B). This mutant behaved in D1b acts through CDK4 to modulate cell cycle progression. a manner virtually identical to cyclin D1b (Fig. 3C, lane 4). Thus, Cyclin D1b is subjected to aberrant regulation. The afore- the exon 5 encoded sequences are important for stimulating cyclin mentioned studies indicate that the exon 5 encoded sequences of D1 activity against RB. Cyclins harbor multiple mechanisms cyclin D1 are critical for catalytic activity. However, these same www.aacrjournals.org 5631 Cancer Res 2008; 68: (14). July 15, 2008 Cancer Research sequences are also important in the regulation of cyclin D1 protein damage (top). These results were apparent with both endogenous levels and localization (1, 31). Therefore, modulation of endogenous and exogenously expressed GFP–cyclin D1 protein. Surprisingly, cyclin D1b protein levels was analyzed in response to stresses that neither endogenous cyclin D1b nor GFP–cyclin D1b levels were influence the stability and regulation of cyclin D1 protein. The altered after cisplatin exposure (Fig. 4B, bottom). The influence of cyclin D1 protein has a relatively short half-life and its abundance CDDP on protein levels was not associated with alteration in RNA can be dramatically enhanced via the inhibition of proteasomal levels, as there was no effect of CDDP on the RNA levels of either degradation (43, 44). Consistent with these published studies, cyclin D1 transcript. Thus, cyclin D1b evades turnover downstream treatment of MCF-7 cells with the proteasome inhibitor MG132 led from DNA damage signaling pathways. to accumulation of both endogenous cyclin D1 and a GFP–cyclin Lastly, we determined the influence of hormonal regulation on D1 introduced into the cells via transfection (Fig. 4A, top). In cyclin D1 variant expression. MCF-7 cells are an ER-positive breast contrast, MG132 had little effect on cyclin D1b abundance, either cancer cell line and dependent on estrogen and ER activity for with the endogenous protein or ectopically expressed GFP–cyclin proliferation (45, 46). Estrogen withdrawal and estrogen antago- D1b (Fig. 4A, bottom). nists,, such as tamoxifen and ICI, result in the attenuation of cyclin In contrast with proteasomal inhibitors, DNA damage has been D1 expression (2, 45–47), which is believed to be important for the shown to mediate cyclin D1 degradation via NH2 terminal cell cycle inhibition achieved by such therapeutic modalities. regulatory motifs, which are present in both cyclin D1 and cyclin Correspondingly, levels of cyclin D1 were diminished with both D1b (43). As shown in Fig. 4B, cyclin D1 levels were efficiently tamoxifen and ICI treatment (Fig. 4C). Surprisingly, during this down-regulated as a consequence of cisplatin-induced DNA same time course of estrogen antagonism, the levels of cyclin D1b

Figure 2. Cyclin D1b is expressed in primary breast cancer. A, a TMA section consisting of normal tissue and breast cancer tissue was costained for pan-cytokeratins (green) and cyclin D1b (red) expression and visualized by immunofluorescence. Representative images of normal mammary tissue and two tumor sections with high or low D1b expression are shown. Summary of automated quantitative imaging of cyclin D1b levels within cytokeratin-positive benign or malignant epithelial cells. B, an independent tissue array containing two to six cores of invasive ductal carcinoma from 175 patients was stained for cyclin D1b expression by immunohistochemistry and scored by a pathologist. Representative images depicting the absence of staining (a) and intensity scores of 1+ (b), 2+ (c), and 3+ (d) are shown. The percentage of breast cancers exhibiting a given intensity score was quantified.

Cancer Res 2008; 68: (14). July 15, 2008 5632 www.aacrjournals.org Cyclin D1b Promotes Resistance to Estrogen Antagonists

were not significantly reduced and in fact became elevated. Consistent with published studies, cyclin D1 RNA levels were diminished in response to estrogen antagonists (47–51). Similarly, cyclin D1b RNA levels were attenuated with estrogen antagonists.

Figure 3. The functions of cyclin D1b strongly associate with CDK4 and its intron 4 regions. A, MCF-7 cells were harvested, and cell lysate was prepared and subjected to immunoprecipitation (IP) with specific cyclin D1b, CDK4, and Dbf4 antibodies. Input lysate and the resultant precipitated proteins were resolved by SDS-PAGE. CDK4 proteins were detected by immunoblotting. B, diagram depicting the mutant structures of cyclin D1b gene. The cyclin Figure 4. Distinct behaviors of cylin D1b protein compared with cyclin D1. D1b–KE is a mutation of K112E. The mutant D1b-DI4 is encoded by the first four A, MCF-7 cells were transfected with GFP–cyclin D1b or D1 expression exons of cyclin D1 gene and the COOH terminal intron 4 has been deleted. plasmids. Cells were treated with 5 Am MG-132 for 6 h. Total cell lysate was Both D1b mutants are NH2 terminal GFP tagged proteins. C, SAOS-2 cells were resolved by SDS-PAGE. Endogenous D1 (left, top) and D1b (left, low) proteins cotransfected with the indicated expression plasmids. Total cellular protein were detected by immunoblotting. Exogenous GFP–cyclin D1 (right, top) and was resolved by SDS-PAGE. RB, hyperphosphorylated RB, GFP-D1, D1b, GFP–cyclin D1b (right, low) were blotted with GFP antibody. Lamin B was and D1b mutant proteins were detected by immunoblotting. D, SAOS-2 cells used as equal loading control. B, total protein or RNA was isolated from MCF-7 were cotransfected with RB, CDK4, and the GFP-D1b, GFP-D1, and GFP-D1b cells treated with CDDP for 16 h and resolved by SDS-PAGE. Western blot mutants for 40 h and labeled with BrdUrd for 6 h. Cells were fixed and analysis for endogenous D1 (left, top), D1b (left, bottom), and exogenous immunostained for relative BrdUrd incorporation to monitor S phase progression. GFP-D1 (right, top), GFP-D1b (right, bottom) was performed. Lamin B was used Data shown are from two independent experiments with at least 200 cells as a control for equal loading. RNA levels were evaluated by transcript counted for each experiment. *, P < 0.01 compared with RB + Vector; specific RT-PCR, glyceraldehyde-3-phosphate dehydrogenase (GAPDH) is a **, P < 0.05 compared with RB + D1 + CDK4. E, SAOS-2 cells were control for equal loading (bottom). C, MCF-7 cells were treated with either cotransfected and treated with vehicle (DMSO) or PD-0332991 as indicated. 10 9 mol/L Tam or 10 6 mol/L ICI for up to 72 h. Total cell lysates were At 40 h posttransfections, cells were labeled with BrdUrd for 6 h. Cells were harvested at indicated intervals. The effects of tamoxifen (left) and ICI (right) fixed and relative BrdUrd incorporation was determined from two independent upon expression of cyclin D1 and cyclin D1b protein were detected by Western experiments; *, P < 0.05 compared with RB + CDK4 (left). Lysates were blot. Lamin B was used as equal loading control. RNA levels were evaluated by prepared and RB detected by immunoblotting (right). transcript specific RT-PCR, GAPDH is a control for equal loading (bottom). www.aacrjournals.org 5633 Cancer Res 2008; 68: (14). July 15, 2008 Cancer Research

Figure 5. Cyclin D1b overexpression in MCF-7 cells does not affect the DNA damage checkpoint. A, MCF-7 cells were infected with adenovirus encoding cyclin D1b, cyclin D1, and GFP (control) for 24 h. Cells were then treated with CDDP for 16 h. Total cell lysates were harvested and analyzed by immunoblotting with antibodies to RB, cyclin A, cyclin D1, cyclin D1b, and p21Cip1. Lamin B was used as a loading control. B, MCF-7 cells were treated as described in A of same figure. After treatment with CDDP, cells were pulse labeled with BrdUrd for 1 h, followed by BrdUrd staining and propidium iodide staining. Bivariate flow cytometry analysis of DNA synthesis (BrdUrd incorporation) and DNA content (PI) was performed. Shown are representative histograms of 20,000 gated events from two independent experiments. Quantitation of the percentage of BrdUrd-positive cells are from two independent bivariate flow cytometry experiments.

Thus, the levels of cyclin D1b protein are not directly associated infected with GFP, cisplatin treatment lead to the specific loss of with its RNA levels and indicate that regulation of translation/ cyclin D1 protein while retaining cyclin D1b (Fig. 5A). Under this protein stability are critical modulators of cyclin D1b protein levels. condition, RB is dephosphorylated, suggesting that the relatively Combined, these studies indicate that the regulation of cyclin D1b low levels of cyclin D1b present in such cells are not sufficient to protein levels occurs in a manner largely distinct from cyclin D1. promote cell cycle progression. Ectopic expression of cyclin D1 was Cyclin D1b does not influence cisplatin DNA damage used to mimic the overproduction of cyclin D1 proteins observed response but does specifically deregulate response to ER- in tumors. As expected, the overall levels of cyclin D1 protein antagonists. Because cyclin D1b is refractory to degradation increased, but the accumulated protein was efficiently degraded, elicited by cisplatin damage, the ability of cyclin D1b to override consistent with our findings using GFP–cyclin D1. In the case of the cisplatin induced checkpoints in MCF-7 cells was evaluated. cyclin D1b, the exogenously expressed cyclin D1b was refractory MCF-7 cells were infected with adenoviruses that express GFP to the effects of cisplatin. However, the protein was incapable of (control), cyclin D1, or cyclin D1b proteins. At 24 hours, maintaining RB phosphorylation. Furthermore, neither cyclin D1 postinfection cells were exposed to cisplatin at the indicated doses nor cyclin D1b was sufficient to overcome the CDDP-mediated and harvested 16 hours posttreatment. As expected in the cells checkpoint (Fig. 5B and C). Analyses of p21Cip1 levels revealed a

Cancer Res 2008; 68: (14). July 15, 2008 5634 www.aacrjournals.org Cyclin D1b Promotes Resistance to Estrogen Antagonists significant up-regulation of p21Cip1 (Fig. 5A), suggesting that the progression as monitored via bivariate flow cytometry (Fig. 6B action of this CDK inhibitor may mediate the cessation of cell cycle and C). The ectopic expression of cyclin D1 was maintained in the progression even in the presence of significant cyclin D1b protein. presence of tamoxifen and ICI (Fig. 6A). However, this event was To further define a role for cyclin D1b in response to not sufficient to maintain RB phosphorylation, and correspond- antimitogenic therapeutics, the response to estrogen antagonists ingly, downstream target gene expression was repressed (e.g., cyclin was determined. For these analyses, MCF-7 cells were infected with A). Consistent with these outcomes, cell cycle inhibition was adenoviruses encoding GFP, cyclin D1, and cyclin D1b and elicited by estrogen antagonists in the presence of cyclin D1 subjected to estrogen deprivation in combination with tamoxifen expression (Fig. 6B and C). In contrast, exogenous expression of or ICI exposure. Consistent with the literature and previous cyclin D1b maintained RB phosphorylation in the presence of both findings (45–51), both estrogen antagonists led to down-regulation tamoxifen and ICI. Consistent with these findings, cyclin D1b of cyclin D1 levels concomitant with RB dephosphorylation in expression maintained RB target gene expression (e.g., cyclin A control (GFP-infected) cells (Fig. 6A). This effect of estrogen levels) in the presence of tamoxifen and ICI. These molecular antagonists was similarly associated with the cessation of cell cycle findings were further corroborated as cyclin D1b expressing cells

Figure 6. Cyclin D1b overexpression in MCF-7 cells imparts resistance to antiestrogen therapy. A, MCF-7 cells were infected with adenovirus encoding cyclin D1b and cyclin D1, GFP (control) for 24 h. Next, cells were cultured in complete medium, CDT/Tam (left), or CDT/ICI (right) for another 36 h. Cells were harvested and analyzed by immunoblotting with antibodies to RB, cyclin A, cyclin D1, cyclin D1b, and p27Kip1. Lamin B was used as a loading control. B, MCF-7 cells were treated as described in A of the same figure. Bivariate flow cytometry was performed. Representative histograms of 20,000 gated events from two independent experiments. Quantitation of the percentage of BrdUrd-positive cells are from two independent bivariate flow cytometry experiments; *, P < 0.05; ***, P < 0.001 compared with GFP. C, MCF-7 cells infected with adenoviruses encoding either cyclin D1 or cyclin D1b were plated and cultured in FBS, CDT/ Tam, or CDT/ICI for the indicated time and then processed for MTT analyses. Four replicates were performed, and the SD in the data are shown.

www.aacrjournals.org 5635 Cancer Res 2008; 68: (14). July 15, 2008 Cancer Research efficiently bypassed the cell cycle inhibition mediated by tamoxifen or ICI (Fig. 6B and C). Whereas p21Cip1 is implicated in mediating DNA damage response pathways, p27Kip1 is crucial for response to estrogen antagonists (52). Analyses of p27Kip1 expression showed significant up-regulation in both GFP and cyclin D1–transduced cells with tamoxifen or ICI exposure (Fig. 6A). In contrast, cyclin D1b ectopic expression reduced the levels of p27Kip1 in MCF-7 cells and blunted the up-regulation observed upon treatment of estrogen antagonists (Fig. 6A). To determine whether the cyclin D1b transduced cells retained the capacity to proliferate in the presence of estrogen antagonists, MTT analysis was performed. MCF-7 cells were infected with cyclin D1 and cyclin D1b encoding adenoviruses. Cells were then cultured either in the presence of full steroid (FBS) or in the absence of steroids and supplemented with tamoxifen or ICI. As shown in Fig. 6D, infected cells expressing cyclin D1 or cyclin D1b proliferated with similar kinetics in the presence of estrogen. However, in the presence of estrogen antagonists cyclin D1b provided a significant proliferative advantage over cyclin D1 (Fig. 6D). Thus, exogenous expression of cyclin D1b is capable of potently promoting both cell cycle progression and proliferation in the presence of estrogen antagonism. Bypass of tamoxifen response is dependent on CDK4 interaction. The cyclin D1 protein has been postulated to contribute to deregulation of estrogen dependence via two mechanisms. First, cyclin D1 can function as an estrogen receptor coactivator, and it has been suspected that this function could contribute to tamoxifen resistance (19, 20, 53). Therefore, the influence of cyclin D1 and cyclin D1b on ER activity was evaluated. Using reporter assays, it was confirmed that cyclin D1 cooperates to Figure 7. Effects of cyclin D1b on the transcriptional activity of ER and stimulate the transcriptional activity of the ER (19). However, cyclin requirement for CDK4 activity in response to antiestrogen therapies. A, MCF-7 cells were cotransfected and treated as described in Materials and Methods D1b does not share this activity and is thus not competent in ER with cytomegalovirus h-galactosidase reporter and ERE-Luciferase reporter coactivation (Fig. 7A). This result is not surprising, as cyclin D1b plasmids. Cells were harvested and assayed for luciferase activity, which lacks the LXXL motif that has been shown to be critical for cyclin D1 was normalized to h-galactosidase activity. Data are from three independent experiments. B, MCF-7 cells were cotransfected with H2B-GFP and indicated function in ER regulation. Thus, these findings suggest that cyclin plasmids for 16 h. Then, cells were washed with PBS and replaced in phenol D1 function in this system could involve CDK4 activity (3, 54). red–free DMEM containing 10% CDT with either 109 mol/L 4-hydroxy tamoxifen or 106 mol/L ICI for 36 h. Cells were pulsed with BrdUrd for 6 h, and BrdUrd Therefore, cyclin D1b mutants were used to assess the effect of incorporation was monitored via indirect immunofluoresence. Two independent CDK4 interaction on overcoming estrogen antagonists. MCF-7 cells experiments with at least 200 cells counted per experiment. Error bars, SD. were transfected with expression plasmids encoding the indicated *, P < 0.05 compared with vector. proteins and subsequently challenged by culture in tamoxifen or ICI containing media. Under this condition, vector transfected cells that cyclin D1b transcript levels are quite high in tumors, incurred an inhibition of BrdUrd incorporation. Consistent with the supporting the speculation that cyclin D1b may function in preceding data, the cell cycle inhibitory effects of tamoxifen or ICI tumorigenesis (29, 55). Here, we quantitatively interrogated the could be largely reversed via the expression of cyclin D1b. The cyclin affinity and specificity of an antibody specific to cyclin D1b. D1DI4 protein behaved in a manner consistent with cyclin D1b. The Utilizing this reagent, it is apparent that cyclin D1b protein is KE mutant, which fails to stimulate CDK4 activity, was deficient in expressed in a significant fraction of tumor cell lines. Furthermore, bypassing tamoxifen-mediated or ICI-mediated cell cycle inhibition cyclin D1b protein levels are clearly elevated in primary breast (Fig. 7B). Therefore, these data suggest that cyclin D1b uses CDK- cancers. Interestingly, in a number of asynchronously proliferating dependent mechanisms to stimulate cell cycle progression in the cell lines, the levels of cyclin D1b represent <10% of the total cyclin presence of estrogen antagonists. D1 protein in the cell line. This result is surprising given the relatively similar levels of cyclin D1 and cyclin D1b transcripts were detected in cell lines and tumor specimens (26, 56). Thus, it would Discussion seem that the alternatively spliced cyclin D1b transcript, while The cyclin D1 locus has been critically implicated in tumori- produced, is not efficiently translated. This finding also explains genesis (1, 2, 4, 25). Whereas the activity of the canonical encoded why the detection of cyclin D1b protein is difficult in the absence of cyclin D1 protein is well established, surprisingly little is known specialized reagents for detection (27, 32). However, there are clear regarding the cyclin D1b protein, which is produced as a result of exceptions, wherein cyclin D1b seems to be the predominant cyclin alternative splicing. The polymorphism that is believed to D1 species in specific cell lines (e.g., BT549). Strikingly, the levels of contribute to the expression of cyclin D1b transcript has an allele cyclin D1b could play significant roles in tumor behavior under frequency of 40%, and thus could represent a significant participant specific conditions, wherein translation of the cyclin D1b transcript in cancers driven by cyclin D1 gene amplification or translocations could become more favorable or the levels of the protein become (4, 25). Interestingly, using RT-PCR approaches, it has become clear predominant.

Cancer Res 2008; 68: (14). July 15, 2008 5636 www.aacrjournals.org Cyclin D1b Promotes Resistance to Estrogen Antagonists

The regulation of the cyclin D1 protein levels is subject to cell cycle arrest in the presence of cisplatin damage (60). Strikingly, complex transcriptional and posttranscriptional mechanisms. overexpression of cyclin D1b is not sufficient to overcome the Mitogenic signaling pathways are believed to effect cyclin D1 effects of cisplatin at the molecular level or bypass the DNA protein levels primarily via the regulation of cyclin D1 RNA levels damage checkpoint. As such, the alterations in cyclin D1b stability and modulation of protein stability via the T286 phosphorylation are not apparently influencing the response to this form of site (1, 4, 25, 30, 57). This regulatory motif is compromised in challenge. However, the overexpression of cyclin D1b did efficiently tumors harboring specific mutations and in synthetic alleles which overcome the effects of estrogen antagonists. This finding suggests surgically mutate the T286 site (31, 58, 59). Therefore, it is perhaps that high levels of cyclin D1b expression could have a deleterious not surprising that regulation of cyclin D1b protein stability is effect on response to hormone therapies. Additional studies will be distinct from that observed with cyclin D1. However, it has been required to define this action of cyclin D1b in primary tumors. reported that sites in the NH2 terminus of cyclin D1 modulate The mechanism through which cyclin D1b exerts its influence in protein stability after DNA damage (43). As such, it was surprising cancer has remained largely obscure. Prior studies indicate that that cyclin D1b escaped degradation mediated by cisplatin damage, cyclin D1b is a more potent oncogene in classic fibroblast whereas cyclin D1 was readily degraded under identical conditions. transformation models and has a propensity to impart anchor- Presumably, this enhanced stability is not a reflection of differences age-independent growth to cells (32, 33). This response is in nucleo/cytoplasmic shuttling as the T286A allele of cyclin D1, associated with the constitutive nuclear localization of the protein. which is exclusively nuclear, is sensitive to DNA damage–mediated However, the mechanism through which this occurs is not clear, degradation (43). Thus, the mechanism through which cyclin D1b particularly because cyclin D1b has impaired catalytic activity evades such signaling is at present unknown. It is well appreciated relative to cyclin D1 (33). We show that this impairment is due to that cyclin D1 levels are regulated by multiple mechanisms via the lack of the COOH terminal region. Thus, paradoxically, the exon estrogen antagonists in ER-positive breast cancer cells (45–51). 5 encoded sequences of cyclin D1 impart enhanced catalytic Surprisingly, cyclin D1b also evades this form of regulation and in activity against critical substrates of CDK4, yet also enhance fact cyclin D1b levels are stimulated in the presence of estrogen susceptibility to degradation mediated via a variety of antimito- antagonists. As such, the ratio of cyclin D1b to cyclin D1 can be genic signaling cascades. Whereas it is appealing to speculate that dramatically altered in various conditions of challenge (4.5-fold the effect of cyclin D1b on susceptibility to estrogen antagonists increased in DNA damage response and 11-fold increased with could be through a direct effect on ER function, we find that cyclin estrogen antagonist). However, the level of endogenous cyclin D1b D1b is not an effector of ER, and conversely the interaction of is apparently not sufficient to preclude cell cycle arrest mediated by cyclin D1b with CDK4 is crucial for this activity. These results are estrogen antagonists or DNA damaging agents in MCF-7 cells as in agreement with structure function analyses carried out in T47D these cells arrest although the expression of cyclin D1b is retained. cells (61) and suggest that aberrant modulation of the CDK4 The cyclin D1 locus is subject to amplification in primary breast assembled complexes underlies the action of cyclin D1b proteins cancer and overexpression of cyclin D1 through additional on therapeutic response. Combined, these studies underscore the mechanisms is observed in >50% of breast cancer (2, 49). Based aberrant regulation and function of cyclin D1b and indicate that on our data in primary tumors, overexpression of cyclin D1b could this protein could be particularly important in the context of contribute to the pathology of breast cancer and may be hormonal therapy for ER-positive breast cancer. particularly relevant under those conditions wherein D1b would be the predominant cyclin D1 moiety. Similar results were previously observed with primary prostate cancer specimens, Acknowledgments wherein the cyclin D1b transcript was specifically induced in tumor Received 8/16/2007; revised 1/15/2008; accepted 3/10/2008. tissue versus matched nonneoplastic controls. Under conditions of Grant support: Supported by NIH grants R01-CA104213 (to E.S. Knudsen), R01- CA101841 (to H. Rui), and CA099996 (to K.E. Knudsen); National Health and Medical cisplatin treatment, MCF-7 cells incur a strong antiproliferative Research Council of Australia, The Cancer Institute NSW, Australian Cancer Research arrest accompanied by the degradation of cyclin D1 and Foundation, RT Hall Trust, and The Petre Foundation (to R.L. Sutherland). The costs of publication of this article were defrayed in part by the payment of page dephosphorylation of RB. The status of RB is critically important charges. This article must therefore be hereby marked advertisement in accordance for this response as MCF-7 cells rendered RB-deficient fail to elicit with 18 U.S.C. Section 1734 solely to indicate this fact.

References 8. Lukas J, Bartkova J, Bartek J. Convergence of 13. Wang TC, Cardiff RD, Zukerberg L, Lees E, Arnold A, mitogenic signalling cascades from diverse classes Schmidt EV. Mammary hyperplasia and carcinoma in 1. Diehl JA. Cycling to cancer with cyclin D1. Cancer Biol of receptors at the cyclin D-cyclin-dependent kinase- MMTV-cyclin D1 transgenic mice. Nature 1994;369:669–71. Ther 2002;1:226–31. pRb-controlled G1 checkpoint. Mol Cell Biol 1996;16: 14. Matsushime H, Ewen ME, Strom DK, et al. Identifi- 2. Arnold A, Papanikolaou A. Cyclin D1 in breast cancer 6917–25. cation and properties of an atypical catalytic subunit pathogenesis. J Clin Oncol 2005;23:4215–24. 9. Arnold A, Motokura T, Bloom T, et al. PRAD1 (cyclin (p34PSK-J3/cdk4) for mammalian D type G1 cyclins. Cell 3. Sherr CJ. D-type cyclins. Trends Biochem Sci 1995;20:187–90. D1): a parathyroid neoplasia gene on 11q13. Henry Ford 1992;71:323–34. 4. Knudsen KE, Diehl JA, Haiman CA, Knudsen ES. Hosp Med J 1992;40:177–80. 15. Matsushime H, Quelle DE, Shurtleff SA, Shibuya M, Cyclin D1: polymorphism, aberrant splicing and cancer 10. Motokura T, Bloom T, Kim HG, et al. A novel cyclin Sherr CJ, Kato JY. D-type cyclin-dependent kinase risk. Oncogene 2006;25:1620–8. encoded by a bcl1-linked candidate oncogene. Nature activity in mammalian cells. Mol Cell Biol 1994;14: 5. Sherr CJ. Mammalian G1 cyclins and cell cycle 1991;350:512–5. 2066–76. progression. Proc Assoc Am Physicians 1995;107:181–6. 11. Bartkova J, Lukas J, Muller H, Lutzhoft D, Strauss M, 16. Ewen ME, Sluss HK, Sherr CJ, Matsushime H, Kato J, 6. Xiong Y, Connolly T, Futcher B, Beach D. Human D- Bartek J. Cyclin D1 protein expression and func- Livingston DM. Functional interactions of the retino- type cyclin. Cell 1991;65:691–9. tion in human breast cancer. Int J Cancer 1994;57: blastoma protein with mammalian D-type cyclins. Cell 7. Lukas J, Pagano M, Staskova Z, Draetta G, Bartek J. Cyclin 353–61. 1993;73:487–97. D1 protein oscillates and is essential for cell cycle 12. Yu Q, Geng Y, Sicinski P. Specific protection against 17. Kato J, Matsushime H, Hiebert SW, Ewen ME, progression in human tumour cell lines. Oncogene 1994;9: breast cancers by cyclin D1 ablation. Nature 2001;411: Sherr CJ. Direct binding of cyclin D to the 707–18. 1017–21. retinoblastoma gene product (pRb) and pRb www.aacrjournals.org 5637 Cancer Res 2008; 68: (14). July 15, 2008 Cancer Research

phosphorylation by the cyclin D-dependent kinase phorylation, and cellular transformation. J Biol Chem decreased retinoblastoma protein phosphorylation. CDK4. Genes Dev 1993;7:331–42. 2003;278:30339–47. Mol Endocrinol 1995;9:1804–13. 18. Lukas J, Bartkova J, Rohde M, Strauss M, Bartek J. 34. Hess-Wilson JK, Boldison J, Weaver KE, Knudsen KE. 48. Musgrove EA, Hamilton JA, Lee CS, Sweeney KJ, Cyclin D1 is dispensable for G1 control in retinoblas- Xenoestrogen action in breast cancer: impact on ER- Watts CK, Sutherland RL. Growth factor, steroid, and toma gene-deficient cells independently of cdk4 activity. dependent transcription and mitogenesis. Breast Cancer steroid antagonist regulation of cyclin gene expression Mol Cell Biol 1995;15:2600–11. Res Treat 2006;96:279–92. associated with changes in T-47D human breast 19. Neuman E, Ladha MH, Lin N, et al. Cyclin D1 35. Strobeck MW, Fribourg AF, Puga A, Knudsen ES. cancer cell cycle progression. Mol Cell Biol 1993;13: stimulation of estrogen receptor transcriptional activity Restoration of retinoblastoma mediated signaling to 3577–87. independent of cdk4. Mol Cell Biol 1997;17:5338–47. Cdk2 results in cell cycle arrest. Oncogene 2000;19: 49. Buckley MF, Sweeney KJ, Hamilton JA, et al. 20. Zwijsen RM, Wientjens E, Klompmaker R, van der 1857–67. Expression and amplification of cyclin genes in human Sman J, Bernards R, Michalides RJ. CDK-independent 36. Sawa H, Ohshima TA, Ukita H, et al. Alternatively breast cancer. Oncogene 1993;8:2127–33. activation of estrogen receptor by cyclin D1. Cell 1997; spliced forms of cyclin D1 modulate entry into the cell 50. Carroll JS, Prall OW, Musgrove EA, Sutherland RL. A 88:405–15. cycle in an inverse manner. Oncogene 1998;16:1701–12. pure estrogen antagonist inhibits cyclin E-Cdk2 activity 21. Knudsen KE, Cavenee WK, Arden KC. D-type cyclins 37. Burd CJ, Petre CE, Morey LM, et al. Cyclin D1b in MCF-7 breast cancer cells and induces accumulation complex with the androgen receptor and inhibit its variant influences prostate cancer growth through of p130-2F4 complexes characteristic of quiescence. transcriptional transactivation ability. Cancer Res 1999; aberrant androgen receptor regulation. Proc Natl Acad J Biol Chem 2000;275:38221–9. 59:2297–301. Sci U S A 2006;103:2190–5. 51. Hui R, Finney GL, Carroll JS, Lee CS, Musgrove EA, 22. Petre CE, Wetherill YB, Danielsen M, Knudsen KE. 38. LeBaron MJ, Crismon HR, Utama FE, et al. Ultrahigh Sutherland RL. Constitutive overexpression of cyclin D1 Cyclin D1: mechanism and consequence of androgen density microarrays of solid samples. Nat Methods 2005; but not cyclin E confers acute resistance to antiestro- receptor co-repressor activity. J Biol Chem 2002;277: 2:511–3. gens in T-47D breast cancer cells. Cancer Res 2002;62: 2207–15. 39. Dolled-Filhart M, Ryden L, Cregger M, et al. 6916–23. 23. Lamb J, Ramaswamy S, Ford HL, et al. A mechanism Classification of breast cancer using genetic algo- 52. Cariou S, Donovan JC, Flanagan WM, Milic A, of cyclin D1 action encoded in the patterns of gene rithms and tissue microarrays. Clin Cancer Res 2006; Bhattacharya N, Slingerland JM. Down-regulation of expression in human cancer. Cell 2003;114:323–34. 12:6459–68. p21WAF1/CIP1 or p27Kip1 abrogates antiestrogen- 24. Landis MW, Pawlyk BS, Li T, Sicinski P, Hinds PW. 40. Hinds PW, Mittnacht S, Dulic V, Arnold A, Reed SI, mediated cell cycle arrest in human breast cancer cells. Cyclin D1-dependent kinase activity in murine develop- Weinberg RA. Regulation of retinoblastoma protein Proc Natl Acad Sci U S A 2000;97:9042–6. ment and mammary tumorigenesis. Cancer Cell 2006;9: functions by ectopic expression of human cyclins. Cell 53. Zwijsen RM, Buckle RS, Hijmans EM, Loomans CJ, 13–22. 1992;70:993–1006. Bernards R. Ligand-independent recruitment of steroid 25. Knudsen KE. The cyclin D1b splice variant: an old 41. Hinds PW,Dowdy SF, Eaton EN, Arnold A, Weinberg RA. receptor coactivators to estrogen receptor by cyclin D1. oncogene learns new tricks. Cell Div 2006;1:15. Function of a human cyclin gene as an oncogene. Proc Natl Genes Dev 1998;12:3488–98. 26. Betticher DC, Thatcher N, Altermatt HJ, Hoban P, Acad Sci U S A 1994;91:709–13. 54. Sherr CJ. Cell cycle control and cancer. Harvey Lect Ryder WD, Heighway J. Alternate splicing produces a 42. Dowdy SF, Hinds PW, Louie K, Reed SI, Arnold A, 2000;96:73–92. novel cyclin D1 transcript. Oncogene 1995;11:1005–11. Weinberg RA. Physical interaction of the retinoblastoma 55. Howe D, Lynas C. The cyclin D1 alternative tran- 27. Hosokawa Y, Joh T, Maeda Y, Arnold A, Seto M. protein with human D cyclins. Cell 1993;73:499–511. scripts [a] and [b] are expressed in normal and Cyclin D1/PRAD1/BCL-1 alternative transcript [B] 43. Agami R, Bernards R. Distinct initiation and malignant lymphocytes and their relative levels are protein product in B-lymphoid malignancies with maintenance mechanisms cooperate to induce G1 cell influenced by the polymorphism at codon 241. Haema- t(11;14)(q13;q32) translocation. Int J Cancer 1999;81: cycle arrest in response to DNA damage. Cell 2000;102: tologica 2001;86:563–9. 616–9. 55–66. 56. Hosokawa Y, Arnold A. Mechanism of cyclin D1 28. Hosokawa Y, Gadd M, Smith AP, Koerner FC, 44. Bagui TK, Mohapatra S, Haura E, Pledger WJ. (CCND1, PRAD1) overexpression in human cancer cells: Schmidt EV, Arnold A. Cyclin D1 (PRAD1) alternative P27Kip1 and p21Cip1 are not required for the formation analysis of allele-specific expression. Genes Chromo- transcript b: full-length cDNA cloning and expression of active D cyclin-cdk4 complexes. Mol Cell Biol 2003;23: somes Cancer 1998;22:66–71. in breast cancers. Cancer Lett 1997;113:123–30. 7285–90. 57. Alao JP. The regulation of cyclin D1 degradation: 29. Bala S, Peltomaki P. CYCLIN D1 as a genetic modifier 45. Foster JS, Henley DC, Bukovsky A, Seth P, roles in cancer development and the potential for in hereditary nonpolyposis colorectal cancer. Cancer Wimalasena J. Multifaceted regulation of cell cycle therapeutic invention. Mol Cancer 2007;6:24. Res 2001;61:6042–5. progression by estrogen: regulation of Cdk inhibitors 58. Benzeno S, Lu F, Guo M, et al. Identification of 30. Gladden AB, Diehl JA. Location, location, location: and Cdc25A independent of cyclin D1-4 function. Mol mutations that disrupt phosphorylation-dependent the role of cyclin D1 nuclear localization in cancer. J Cell Cell Biol 2001;21:794–810. nuclear export of cyclin D1. Oncogene 2006;25:6291–303. Biochem 2005;96:906–13. 46. Prall OW, Sarcevic B, Musgrove EA, Watts CK, 59. Moreno-Bueno G, Rodriguez-Perales S, Sanchez-Estevez C, 31. Alt JR, Cleveland JL, Hannink M, Diehl JA. Phosphor- Sutherland RL. Estrogen-induced activation of Cdk4 et al. Cyclin D1 gene (CCND1) mutations in endometrial ylation-dependent regulation of cyclin D1 nuclear and Cdk2 during G1-S phase progression is accompa- cancer. Oncogene 2003;22:6115–8. export and cyclin D1-dependent cellular transformation. nied by increased cyclin D1 expression and decreased 60. Bosco EE, Wang Y, Xu H, et al. The retinoblastoma Genes Dev 2000;14:3102–14. cyclin-dependent kinase inhibitor association with tumor suppressor modifies the therapeutic response of 32. Lu F, Gladden AB, Diehl JA. An alternatively spliced cyclin E-Cdk2. J Biol Chem 1997;272:10882–94. breast cancer. J Clin Invest 2007;117:218–28. cyclin D1 isoform, cyclin D1b, is a nuclear oncogene. 47. Watts CK, Brady A, Sarcevic B, deFazio A, Musgrove 61. Bindels EM, Lallemand F, Balkenende A, Verwoerd D, Cancer Res 2003;63:7056–61. EA, Sutherland RL. Antiestrogen inhibition of cell cycle Michalides R. Involvement of G1-S cyclins in estrogen- 33. Solomon DA, Wang Y, Fox SR, et al. Cyclin D1 splice progression in breast cancer cells in associated with independent proliferation of estrogen receptor-positive variants. Differential effects on localization, RB phos- inhibition of cyclin-dependent kinase activity and breast cancer cells. Oncogene 2002;21:8158–65.

Cancer Res 2008; 68: (14). July 15, 2008 5638 www.aacrjournals.org Oncogene (2009) 28, 1812–1820 & 2009 Macmillan Publishers Limited All rights reserved 0950-9232/09 $32.00 www.nature.com/onc SHORT COMMUNICATION Cyclin D1b protein expression in breast cancer is independent of cyclin D1a and associated with poor disease outcome

EKA Millar1,2, JL Dean3,4, CM McNeil1, SA O’Toole1, SM Henshall1, T Tran3,4, J Lin3,4, A Quong3, CES Comstock3,4, A Witkiewicz3,4,5, EA Musgrove1, H Rui3,4, L LeMarchand6, VW Setiawan7, CA Haiman7, KE Knudsen3,4,8, RL Sutherland1 and ES Knudsen3,4

1Cancer Research Program, Garvan Institute of Medical Research, Sydney, New South Wales, Australia; 2Department of Anatomical Pathology South Eastern Area Laboratory Service, St George Hospital, Sydney, New South Wales, Australia; 3Kimmel Cancer Center, Thomas Jefferson University, Philadelphia, PA, USA; 4Department of Cancer Biology, Thomas Jefferson University, Philadelphia, PA, USA; 5Department of Pathology, Thomas Jefferson University, Philadelphia, PA, USA; 6Epidemiology Program, Cancer Research Center of Hawaii, Honolulu, HI, USA; 7Department of Preventive Medicine/Norris Comprehensive Cancer Center, Keck School of Medicine, University of Southern California, Los Angeles, CA, USA and 8Department of Urology, Thomas Jefferson University, Philadelphia, PA, USA

Aberrant expression of cyclin D1 protein is a common Aberrant cellular proliferation is an inherent component feature of breast cancer. However, the CCND1 gene of human cancer. Correspondingly, the deregulation of encodes two gene products, cyclin D1a and cyclin D1b, cell-cycle control pathways is an exceedingly common which have discrete mechanisms of regulation and impact occurrence in cancer (Malumbres and Barbacid, 2001; on cell behavior. A polymorphism at nucleotide 870 in the Diehl, 2002; Cobrinik, 2005; Knudsen, 2006). The CCND1 gene, rs603965, influences the relative production retinoblastoma tumor suppressor pathway is believed of the encoded proteins and can impart increased risk for to function to restrain inappropriate proliferation and is tumor development. Here, the impact of both the G/A870 inactivated at high-frequency in tumors (Cobrinik, 2005; polymorphism and cyclin D1b protein production on Knudsen and Knudsen, 2006). The retinoblastoma breast cancer risk, disease phenotype and patient outcome protein limits proliferation by functioning as a tran- was analysed. In a large multiethnic case–control study, scriptional co-repressor of genes that are required for the G/A870 polymorphism conferred no significant risk DNA replication and mitotic progression (Blais and for breast cancer overall or by stage or estrogen receptor Dynlacht, 2007; Iaquinta and Lees, 2007). After (ER) status. However, the cyclin D1b protein was found to mitogenic signaling, this function of retinoblastoma is be upregulated in breast cancer, independent of cyclin D1a disrupted through CDK (cyclin-dependent kinase)- levels, and exhibited heterogeneous levels in breast cancer mediated phosphorylation. Particularly, the cyclin D1 specimens. High cyclin D1a expression inversely corre- protein is rate-limiting for the initiation of retinoblas- lated with the Ki67 proliferation marker and was not toma phosphorylation and subsequent inactivation associated with clinical outcome. In contrast, elevated which facilitates cellular division (Cobrinik, 2005). cyclin D1b expression was independently associated with Correspondingly, the levels of cyclin D1 protein are adverse outcomes, including recurrence, distant metastasis directly modulated by mitogenic and antiproliferative and decreased survival. Interestingly, cyclin D1b was signaling pathways (Diehl, 2002; Arnold and Papaniko- particularly associated with poor outcome in the context laou, 2005; Knudsen et al., 2006). Thus, cyclin D1 is a of ER-negative breast cancer. Thus, specific cyclin D1 key regulator of cell-cycle progression, and functions as isoforms are associated with discrete forms of breast an oncogene in specific model systems. cancer and high cyclin D1b protein levels hold prognostic Multiple studies indicate that cyclin D1 plays an potential. important role in mammary biology and tumorigenesis. Oncogene (2009) 28, 1812–1820; doi:10.1038/onc.2009.13; Mice deficient in cyclin D1 harbor surprisingly few published online 16 March 2009 defects; however, there is a pronounced deficit in mammary gland development (Sicinski et al., 1995). Keywords: cyclin D1b; breast cancer; polymorphism; Correspondingly, cyclin D1-null mice are refractory to estrogen receptor; prognosis mammary tumor development driven by multiple oncogenes (Yu et al., 2001), whereas the transgenic expression of cyclin D1 is sufficient to induce mammary tumors in mice (Wang et al., 1994). Consistent with these functional studies, the cyclin D1 protein is Correspondence: Professor ES Knudsen, Kimmel Cancer Center, overexpressed in >50% of breast tumors (Buckley Thomas Jefferson University, BLSB Room 1002, 233 South 10th et al., 1993; Gillett et al., 1994, 1996; Diehl, 2002; Street, Philadelphia, PA 19107, USA. Sutherland and Musgrove, 2004; Arnold and Papani- E-mail: [email protected] Received 30 October 2008; revised 11 December 2008; accepted 16 kolaou, 2005). The basis for the overexpression of cyclin December 2008; published online 16 March 2009 D1 has not been fully determined, as the cyclin D1 locus Cyclin D1b expression associated with poor outcome EKA Millar et al 1813 is amplified in only 10–20% of breast cancer. Further- efficient for directing splicing (Knudsen et al., 2006). more, the prognostic value of cyclin D1 overproduction Findings from a number of laboratories have documen- in breast cancer patients remains somewhat controver- ted a role for the G/A870 polymorphism in cancer risk sial (Gillett et al., 1994, 1996; Arnold and Papanikolaou, (Betticher et al., 1995; Matthias et al., 1998; Knudsen 2005; Jirstrom et al., 2005; Roy and Thompson, 2006; et al., 2006; Pabalan et al., 2008). Furthermore, specific Rudas et al., 2008). studies have documented a trend toward allele-specific It is now apparent that cyclin D1 exists in two expression of cyclin D1b (Betticher et al., 1995; Holley isoforms: the conventional cyclin D1 (referred to as et al., 2001); however, there are clearly modifiers of the cyclin D1a) and cyclin D1b (Betticher et al., 1995; splicing, as the cyclin D1b transcript can be observed in Knudsen, 2006; Knudsen et al., 2006). The cyclin D1b individuals or tumors harboring only the G-allele (Bala variant arises as a consequence of alternative splicing of and Peltomaki, 2001; Carrere et al., 2005; Krieger et al., the CCND1 transcript. This event leads to the loss of 2006; Gupta et al., 2008). To specifically investigate the exon 5-encoded sequences and a unique C terminus that impact of the G/A870 polymorphism, a large multi- arises from translation of intron 4 (Betticher et al., ethnic, case–control study consisting of 1376 invasive 1995). Importantly, exon 5 harbors a number of breast cancer cases and 2583 controls was examined regulatory motifs that are required for appropriate (Table 1). In this population, the rs603965 polymorph- cyclin D1a regulation and protein turnover, including a ism was not associated with breast cancer risk, disease putative PEST domain and the threonine 286 phospho- stage or ER status (Table 1), and was not modified by rylation site, which is a critical effector of the subcellular the age of diagnosis, body mass index, postmenopausal localization and oncogenic potential of cyclin D1a hormone use or family history of breast cancer (P-values (Diehl, 2002; Lu et al., 2003; Solomon et al., 2003; for interaction >0.18, data not shown). These findings Knudsen et al., 2006). Consistent with these known indicate that the G/A870 polymorphism is not a strong structural alterations, cyclin D1b is constitutively predictor of breast cancer risk or associated with the nuclear and has enhanced activity for transforming specific clinicopathological parameters evaluated in this fibroblastic cells (Lu et al., 2003; Solomon et al., 2003). cohort. Furthermore, cyclin D1b plays a potent role in modifying the requirement for anchorage dependence (Holley et al., 2005), and can effectively contribute to Relationship between cyclin D1b and cyclin D1a the bypass of estrogen antagonists in cell culture models expression of estrogen receptor (ER)-positive breast cancer (Wang et al., 2008). These findings, in particular, led us to As there is an incomplete, and likely complex, link evaluate the impact of cyclin D1b in breast cancer. between the polymorphism and the production of cyclin D1b protein, direct analysis of this protein is pivotal to understanding its potential role in breast cancer. Here, G/A870 polymorphism (rs603965) and breast cancer risk we used a cyclin D1b-specific antibody, which has been utilized in a number of published studies, in which The production of cyclin D1b can be modulated by a immunoblotting and immunostaining approaches have common G/A polymorphism at nucleotide 870 shown heterogeneity in protein expression in cell lines (rs603965) of the CCND1 gene (Betticher et al., 1995; and tissue specimens (Burd et al., 2006; Marzec et al., Holley et al., 2001; Knudsen, 2006; Knudsen et al., 2006; Sanchez et al., 2008; Wang et al., 2008). However, 2006). This nucleotide is at a splice donor site, and to further confirm the specificity of the antibodies for although the G-allele represents an ideal consensus immunostaining, antibody reactivity against MCF-7 sequence for splicing, the A-allele is predicted to be less and U2OS cells was compared. Earlier studies have

Table 1 The association between rs603965 in cyclin D1 and breast cancer risk in the MEC Cyclin D1 genotype Cases Controls ORa (95% CI) (n,%) (n,%)

n ¼ 1376* n ¼ 2583* All cases Localized cases Advanced cases ER+ cases ER cases n ¼ 956 n ¼ 328 n ¼ 849 n ¼ 252

All racial/ethnic groups GG 409 (32.9) 869 (36.5) 1.00 1.00 1.00 1.00 1.00 AG 609 (49.0) 1111 (46.7) 1.08 (0.92–1.27) 1.11 (0.93–1.33) 1.05 (0.80–1.37) 1.18 (0.97–1.43) 0.96 (0.70–1.31) AA 226 (18.2) 401 (16.8) 1.07 (0.87–1.33) 1.09 (0.86–1.38) 0.98 (0.68–1.41) 1.17 (0.91–1.50) 1.16 (0.78–1.74) Per allele 1.04 (0.94–1.16) 1.05 (0.94–1.18) 1.00 (0.84–1.19) 1.09 (0.97–1.24) 1.06 (0.87–1.30) P ¼ 0.43 P ¼ 0.39 P ¼ 0.99 P ¼ 0.16 P ¼ 0.58

Abbreviations: CI, conﬁdence interval; OR, odds ratio; MEC, Multiethnic Cohort. aAdjust for race/ethnicity and age (o55, 55–64 and 65+ years). *Numbers do not add up to total because of missing genotype data. P-values for heterogeneity by race/ethnicity X0.07.

Oncogene Cyclin D1b expression associated with poor outcome EKA Millar et al 1814 GFP Immunostaining DAPI

α αD1a DAPI αD1b DAPI D1 U20S MCF7 GFP-D1a Cyclin D1a MCF7 αD1b Cyclin D1b

U20S Lamin B αD1 GFP-D1b

αD1b

CK/dapi D1b/dapi CK/D1b CK/D1b/dapi CK/dapi D1a/dapi CK/D1a CK/D1a/dapi Normal Normal Case 3 Case 1 Case 4 Case 2

50 6

40 5 4 30 3 20 2

Mean D1b intensity 10 Mean D1a intensity 1

0 0 Normal Cancer Normal Cancer Figure 1 Cyclin D1a and cyclin D1b expression is elevated in breast cancer. (a) U2OS and MCF-7 cells were stained with cyclin D1a or cyclin D1b antibodies as indicated (left panel). Representative images taken at equal exposures for cyclin D1/cyclin D1b are shown. U2OS and MCF-7 cell lysates were subjected to immunoblotting with the indicated antibodies (right panel). (b) U2OS cells were transfected with the GFP-cyclin D1a or GFP-cyclin D1b expression plasmids as indicated. These cell populations were then stained with cyclin D1a or cyclin D1b antibodies. Representative images taken at equal exposure are shown. (c) Slides were co-stained for cyclin D1b and cytokeratin (CK) to detect epithelial cells. DAPI staining was used to detect all nuclei in the section. Representative images of speciﬁc breast cancer cases and normal controls are shown (top panel). Quantitation of the D1b signal in the CK-positive compartment was carried out and data were obtained from 39 normal specimens and 150 invasive ductal carcinomas (bottom panel; error bars indicate 95% CI, Po0.001). Student’s t-test was carried out. (d) Slides were co-stained for cyclin D1a and cytokeratin (CK) to detect epithelial cells. DAPI staining was utilized to detect all nuclei in the section. Representative images of speciﬁc breast cancer cases and normal controls are shown (top panel). Quantitation of the D1a signal in the CK-positive compartment was carried out from 44 normal specimens and 148 invasive ductal carcinomas (bottom panel; error bars indicate 95% CI, Po0.001). Student’s t-test was carried out.

shown that MCF-7 cells express robust levels of both Immunostaining with the cyclin D1a and cyclin D1b cyclin D1 isoforms, whereas the levels of both proteins antibodies showed specific signals only in those cells are barely detectable in U2OS cells (Wang et al., 2008). transfected with appropriate GFP-cyclin D1a/D1b As shown in Figure 1a, both cyclin D1a and D1b expression plasmids, indicating isoform specificity of nuclear reactivities were readily apparent in MCF-7 the antibodies (Figure 1b). cells, but only minimal background staining was These antibodies were first used to evaluate a cohort observed in U2OS cells. Thus, the amount of immuno- containing 44 normal breast tissue specimens and 150 reactivity observed in cytological analyses is cancer cases, as summarized in Supplementary Table 1. consistent with the levels determined by immuno- The staining for cyclin D1b was quantified using the blotting (Figure 1a). To confirm that there was no AQUA fluorescence-based immunohistochemistry plat- cross-reactivity for the antibodies, U2OS cells were form (Dolled-Filhart et al., 2006). Cyclin D1b reactivity transfected with expression plasmids for green fluores- is located in the cytokeratin-positive carcinoma cells and cent protein (GFP)-cyclin D1a or GFP-cyclin D1b. is restricted from the stroma (Figure 1c). Moreover,

Oncogene Cyclin D1b expression associated with poor outcome EKA Millar et al 1815 cyclin D1b exhibits a nuclear staining pattern in all same tumor. Therefore, the expression of cyclin D1a positive-staining sections analysed (Figure 1c). Overall, was determined in parallel on the same quantitative the levels of cyclin D1b expression were heterogeneous platform. In this context, cyclin D1a also showed between tumors, suggesting that alterations in the levels elevated expression among breast cancer specimens of cyclin D1b expression could be associated with (Figure 1d, Po0.001), showing an overall distribution specific features of breast cancer (Figure 1c). Impor- profile similar to cyclin D1b. tantly, these analyses show that cyclin D1b protein These findings suggested that the expression of cyclin levels are upregulated in breast cancer relative to normal D1b could be simply a by-product of elevated levels of breast tissue (Po0.001) and are consistent with recent cyclin D1. Such a concern has also been noted with findings (Wang et al., 2008). As cyclin D1b is respect to the low-molecular-weight forms of cyclin E transcribed from the same gene as cyclin D1a, the observed in breast cancer specimens (Spruck et al., association with breast cancer may be merely reflective 2006). Using quantitative reverse transcription PCR of the overall overexpression of cyclin D1 in the analyses of the cyclin D1 transcripts in a limited number

35 N=19 cancer cases 250 N=143 cancer cases 30 200 25

20 150

15 100

10 D1b Intensity Relative D1b RNA 50 5 0 0 01020 30 40 0 5025 12510075 Relative D1a RNA D1a Intensity

Case 5 Case 6 cyclin D1b cyclin D1a

50 10

40 8

30 6

20 4 Mean cyclin D1a

Mean cyclin D1b 10 2

0 0 1 2 3 1 2 3 Grade Grade Figure 2 Cyclin D1b defines a population of breast cancer distinct from cyclin D1a. (a) Quantitative reverse transcription PCR was performed to determine the relative levels of cyclin D1a- and cyclin D1b-specific RNA, wherein signal from non-neoplastic was set to ‘1’ (left panel). The absolute intensity of cyclin D1b vs cyclin D1a in a total of 143 breast cancer specimens is shown (right panel). (b) Specific cases show preferentially the expressions of cyclin D1b (case 5) and cyclin D1a (case 6). (c) Cyclin D1b or cyclin D1a intensity was determined as a function of tumor grade (error bars indicate 95% CI). Cyclin D1b levels are associated with increased tumor grade (P ¼ 0.014) and cyclin D1a levels are not associated with increased tumor grade (P>0.5). ANOVA statistical test was carried out.

Oncogene Cyclin D1b expression associated with poor outcome EKA Millar et al 1816 of breast tumor specimens (n ¼ 19), we observed that cell culture models (Lu et al., 2003; Solomon et al., 2003; there was not a strong direct relationship between the Wang et al., 2008), the association of either cyclin D1 levels of cyclin D1a and cyclin D1b RNA (Figure 2a). variant was evaluated against the proliferation marker Furthermore, quantitative assessment of staining for Ki67 (Figure 3a). These analyses showed that levels of cyclin D1a vs cyclin D1b in each tumor indicated that cyclin D1a staining were inversely correlated with Ki67 protein levels are not directly correlated (Figure 2a). (assessed by the Mann–Whitney non-parametric test), Although many cases expressed both cyclin D1a and consistent with the earlier reported relationship with the cyclin D1b proteins, in specific cases, the expression of a ER-positive, low-proliferation phenotype (Barnes and single cyclin D1 species appeared to predominate, as Gillett, 1998; Roy and Thompson, 2006). In contrast, shown in representative stained chromagen sections cyclin D1b levels were not associated with Ki67 staining. (Figure 2b). Thus, although both cyclin D1a and cyclin These findings further indicated that cyclin D1a and D1b levels are elevated in breast cancer, the relative D1b may be disparately associated with breast cancer abundance of each species could reflect distinct subtypes subtypes and resultant clinical outcomes. of the disease. In univariate analysis, high cyclin D1a levels were not To initially investigate the relationships between the associated with recurrence, metastases or cancer-specific different cyclin D1 proteins and selected properties of death (P>0.05) (Figure 3b). In contrast, high levels of breast cancer, these tumors were evaluated for other cyclin D1b were associated with adverse patient out- molecular markers associated with disease. These come for recurrence (P ¼ 0.0055), distant metastases analyses revealed no relationship of cyclin D1b with (P ¼ 0.0079) and breast cancer-specific death ER status, progesterone receptor (PR) expression, or (P ¼ 0.0091) (Figure 3b). Furthermore, in univariate Her2/Neu (data not shown). However, there was a analysis incorporating traditional prognostic variables, modest quantitative relationship of increasing cyclin including tumor size, grade, lymph node status, ER and D1b levels with increased grade (r ¼ 0.258, P ¼ 0.017), PR status and HER2 amplification, high expression of and the difference in cyclin D1b levels between grade 1 cyclin D1b was a statistically significant predictor of and grade 3 invasive ductal carcinoma was statistically poor outcome from breast cancer (HR (hazard ratio): significant (P ¼ 0.014, Figure 2c). In contrast, cyclin 2.164, 95% CI (confidence interval): 1.195–3.921, D1a levels were not statistically different between breast P ¼ 0.011) (Figure 3c). To assess if cyclin D1b was an cancer cases grouped by grade (P ¼ 0.108, Figure 2c). independent prognostic factor in this cohort, and not Thus, on the basis of the differential expression patterns the result of other confounding variables, Cox propor- of cyclins D1a and D1b across cancer cases, cyclin D1b tional hazard models were constructed with stepwise may harbor prognostic power in breast cancer that is removal of redundant variables until resolution distinct from cyclin D1a. (Figure 3c). The initial model incorporating all the significant variables on univariate analysis showed that high expression of cyclin D1b remained significant on multivariate analysis (HR 2.378, 95% CI 1.252-4.515, Cyclin D1b expression and breast cancer outcome P ¼ 0.0081; Figure 3c). The resolved model showed that high levels of cyclin D1b remain associated with poor To specifically define the impact of cyclin D1b vs D1a survival together with the well-validated prognostic expression on clinical outcome, expression levels were markers, such as lymph node status, HER2 amplifica- analysed in an independent cohort of 175 primary breast tion and PR status. Thus, cyclin D1b protein levels are cancers with a median follow-up of 75 months independently associated with poor survival in breast (Supplementary Table 2). The levels of cyclin D1 protein cancer. reactivity were quantified as described in the Supple- To determine whether the effects of cyclin D1b on mentary Materials and Methods, and the slides were survival were modified by other factors associated with scored by an expert breast cancer pathologist (EKAM). breast cancer pathogenesis, specific subset analyses were On this platform, high nuclear cyclin D1b expression carried out. Initially, as cyclin D1b is related in structure (that is, average intensity >2 þ ) was not associated and function to cyclin D1a, the four subsets of relative with clinicopathological features (assessed by the levels of the isoforms were compared (Figure 3d). Mann–Whitney non-parametric test), including tumor These data showed that cyclin D1b was predictive for size (P ¼ 0.21), age (P ¼ 0.67), lymph node status disease outcomes irrespective of the relative cyclin D1a (P ¼ 0.08), ER (P ¼ 0.59), PR (P ¼ 0.57) or HER2 levels. Subsequently, as high levels of cyclin D1b protein amplification (w2-test, P ¼ 0.79). With this scoring are observed in both ER-positive and ER-negative criterion, tumors exhibiting high cyclin D1b levels were tumors, the impact of cyclin D1b on both forms of not significantly associated with grade III disease disease was analysed. As shown in Figure 3e, ER- (P ¼ 0.34). Furthermore, the expression of cyclin D1b positive tumors with high levels of cyclin D1b tended to was not associated with the level of other cell-cycle have a poor prognosis, but this result did not reach regulatory factors (for example, cyclin E and p27Kip1), statistical significance. In contrast, in ER-negative including cyclin D1a (P ¼ 0.46), in agreement with the tumors, high-level expression of cyclin D1b was data in Figure 2. associated with poor outcome for recurrence As both cyclin D1a and cyclin D1b have been (P ¼ 0.0051), distant metastases (P ¼ 0.004) and breast associated with stimulating proliferative responses in cancer death (P ¼ 0.0054) (Figure 3e). Thus, cyclin D1b

Oncogene Cyclin D1b expression associated with poor outcome EKA Millar et al 1817 is associated with poor disease outcome in breast cancer polymorphism on breast cancer risk. Owing to the size in a manner that is not modified by the relative level of and ethnic heterogeneity of our cohort, we believe that a cyclin D1a; however, the association of high cyclin D1b consensus is arising related to the modest impact of the levels with disease is most apparent in the context of G/A870 polymorphism in breast cancer, which contrasts ER-negative disease. with the stronger influence observed in other tumor types (Pabalan et al., 2008). It is now clear from a number of studies that modifiers Summary external to the polymorphism are associated with the accumulation of cyclin D1b. For example, factors The relevance of G/A870 polymorphism to breast associated with chromatin remodeling and translation cancer has been the subject of several earlier studies potently modulate the abundance of the cyclin D1b (Krippl et al., 2003; Ceschi et al., 2005; Shu et al., 2005; isoform (Batsche et al., 2006; Sanchez et al., 2008). Naidu et al., 2008; Yu et al., 2008). Although the overall Specifically, loss of the Brm chromatin remodeling conclusions of these studies have been disparate, a protein results in a substantial increase in the cyclin recent meta-analysis of published cases suggested that D1b transcript as a result of less efficient splicing. the polymorphism is weakly related to increased breast Through an independent mechanism, the Ewing’s cancer risk (Pabalan et al., 2008). Our findings are in sarcoma oncogene and the ets family transcription agreement, in revealing a negligible impact of the factor (FLI1) results in the upregulation of cyclin D1b

a 80 p=0.027 80 p=NS b 70 70 60 60 RECURRENCE RECURRENCE 50 50 1 1 40 40 0.8 0.8 30 30 0.6 20 20 0.6 expression

Average Ki67 10 10 0.4 0.4 0 0 p=0.0055 0.2 p=NS 0.2 High Cyclin D1b -10 -10 High Cyclin D1a 0 Low Cyclin D1a Low Cyclin D1b High D1a Low D1a High D1b Low D1b Cumulative Survival 0 0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160 Time (months) c Univariate and Multivariate Coxproportional hazards analysis Time (months) of cyclin D1b and clinicopatholoical variables for breast cancer-related death Univariate DISTANT METASTASIS DISTANT METASTASIS Hazard ratio 95% CI p value 1 1 IIIedarG 3.520 1.930 - 6.418 <0.0001 0.8 0.8 Size > 20 mm 2.468 1.415 - 4.304 0.0015 0.6 0.6 Lymph node positive 3.691 2.025 - 6.729 <0.0001 HER2 amplified 3.491 1.956 - 6.229 <0.0001 0.4 0.4 ER positive 0.300 0.172 - 0.524 <0.0001 0.2 p=NS 0.2 p=0.0079 PR positive 0.170 0.087 - 0.333 <0.0001 0 Cyclin D1b high 2.164 1.195 - 3.921 0.0109 Cumulative Survival 0 0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160 Multivariate Time (months) Time (months) Hazard ratio 95% CI p value IIIedarG 1.772 0.784 - 4.006 0.1690 BREAST CANCER DEATH BREAST CANCER DEATH Size > 20 mm 1.350 0.699 - 2.604 0.3713 1 1 Lymph node positive 2.725 1.375 - 5.401 0.0041 HER2 amplified 2.873 1.434 - 5.757 0.0029 0.8 0.8 ER positive 0.744 0.331 - 1.672 0.4740 0.6 0.6 PR positive 0.346 0.142 -0.840 0.0191 0.4 0.4 Cyclin D1b high 2.378 1.252 - 4.515 0.0081 0.2 p=NS 0.2 p=0.0091 Resolved Multivariate Model 0 Cumulative Survival 0 Hazard Ratio 95% CI p value 0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160 Lymph node positive 2.732 1.400 - 5.332 0.0032 Time (months) Time (months) HER2 amplified 3.248 1.662 - 6.347 0.0006 PR positive 0.247 0.119 - 0.514 0.0002 Cyclin D1b high 2.535 1.354 - 4.745 0.0036

Figure 3 Cyclin D1b is associated with poor outcome in breast cancer. (a) Relationship between cyclin D1a and cyclin D1b levels and the proliferation marker Ki67 were monitored. Box plots depict data for cyclin D1a low (n ¼ 73), cyclin D1a high (n ¼ 74), dichotomized on the median H-score, and cyclin D1b low (n ¼ 98) and cyclin D1b high (n ¼ 50) dichotomized at an intensity >2 þ , as related to the percentage of Ki67-positive cells. A signiﬁcant inverse correlation between cyclin D1a levels and Ki67 was detected (P ¼ 0.0256) using the Mann–Whitney non-parametric test. (b) The Kaplan–Meier analysis for cyclin D1a and cyclin D1b expression was carried out in the entire cohort. Graphs represent analyses carried out for recurrence, distant metastasis and breast cancer-related death. Left panel: high cyclin D1a expressors (n ¼ 82) are represented by and low cyclin D1a expressors (n ¼ 87) are represented by J. Right panel: high cyclin D1b expressors (n ¼ 52) are represented by and low cyclin D1b expressors (n ¼ 123) are represented by J. Statistical analyses were carried out using the log-rank test. (c) Summary of statistical analyses of cyclin D1b by univariate and multivariate analyses. Details of the analyses are provided in the Supplementary Materials and methods. (d) The Kaplan– Meier analysis was carried out for distinct cyclin D1a/D1b-isoform subsets in the entire cohort. High cyclin D1a/high cyclin D1b expressors (n ¼ 31) are represented by J, high cyclin D1a/low cyclin D1b expressors (n ¼ 51) are represented by þ , low cyclin D1a/high cyclin D1b expressors (n ¼ 21) by , and low cyclin D1a/low cyclin D1b expressors (n ¼ 66) are represented by &. Statistical analyses were carried out using the log-rank test. (e) Kaplan–Meier graphs for cyclin D1b expression stratiﬁed by ER status. Graphs represent analyses carried out for recurrence, distant metastasis and breast cancer-related death among ER-positive or ER-negative cases. High cyclin D1b expressors are represented by and low cyclin D1b expressors by J. Statistical analyses were carried out using the log-rank test.

Oncogene Cyclin D1b expression associated with poor outcome EKA Millar et al 1818 RECURRENCE 1 ER-positive ER-negative

0.8 RECURRENCE RECURRENCE 1 1 0.6 0.8 0.8 0.4 High D1a/High D1b (n=31) 0.6 0.6 0.2 High D1a/Low D1b (n=51) Cumulative Survival Cumulative Cumulative Survival Cumulative 0.4 Low D1a/High D1b (n=21) p=0.0051 0.4 0 Low D1a/Low D1b (n=66) 0.2 Low Cyclin D1b 0.2 Low Cyclin D1b Cumulative Survival Cumulative 0 20 40 60 80 100 120 140 160 Survival Cumulative p=NS p=0.0051 0 High Cyclin D1b 0 High Cyclin D1b Time (months) 0 20 40 60 80 100 120 140 160 0 20406080100120140160 DISTANT METASTASIS Time (months) Time (months) 1

0.8 DISTANT METASTASIS DISTANT METASTASIS 1 1 0.6 0.8 0.8 0.4 0.6 0.6

0.2 0.4 0.4 Cumulative Survival Cumulative p=0.0054 0 0.2 0.2 Cumulative Survival Cumulative p=NS Survival Cumulative p=0.0040 0 20 40 60 80 100 120 140 160 0 0 Time (months) 0 20 40 60 80 100 120 140 160 0 20406080100120140160 Time (months) Time (months) BREAST CANCER DEATH 1 BREAST CANCER DEATH BREAST CANCER DEATH 0.8 1 1

0.6 0.8 0.8 0.6 0.6 0.4 0.4 0.4 0.2 p=0.0039 Cumulative Survival Cumulative 0.2 0.2 0 Cumulative Survival Cumulative p=NS Survival Cumulative p=0.0054 0 0 0 20 40 60 80 100 120 140 160 Time (months) 0 20 40 60 80 100 120 140 160 0 20406080100120140160 Time (months) Time (months) Figure 3 Continued.

through the alteration of transcript elongation. In transcripts from primary breast cancer. These analyses addition, a dissociation between the G/A870 genotype show that the expressions of cyclin D1a and cyclin D1b and the relative expression of cyclin D1b has been are not related by means of a simple linear relationship. observed in multiple tumor types, including colorectal, Such a finding is supported by our analyses of cyclin mantle cell lymphoma and non-small-cell lung cancer D1b levels in breast cancer, and similar analyses in (Bala and Peltomaki, 2001; Carrere et al., 2005; Krieger mantle cell lymphoma (Carrere et al., 2005; Krieger et al., 2006; Gupta et al., 2008). In this context, it is et al., 2006) and prostate cancer (Comstock et al., 2009). important to appreciate that both genotypic and Consistent with these results, in cell culture models, the histochemical based analyses have merit in dissecting levels of cyclin D1 isoforms are differentially regulated an impact on cancer risk or disease outcome. However, by ER antagonists and DNA damage signals (Wang owing to the lack of direct correlation between genotype et al., 2008). Thus, the relative production of cyclin D1a and protein production, direct analyses of the cyclin and D1b may reflect differences in the tumor-specific D1b protein product is perhaps the only means to define signaling pathways or tissue context of the tumor. As the impact of the protein on disease. such, outcomes associated with cyclin D1b are not As cyclin D1a and cyclin D1b proteins are closely merely a reflection of an overabundance of cyclin D1 related (they derive from the same primary transcript), it protein. is critical to discern whether the expressions of cyclin The importance of cyclin D1 as a marker for outcome D1a and D1b are coupled or independent factors. A in breast cancer is somewhat controversial. Although clear advantage in the analyses of cyclin D1b is the >50% of breast cancers express cyclin D1 at elevated possible use of cyclin D1a- and D1b-specific antibodies. levels, a significant majority of these tumors are of the As described earlier, and further validated herein, the ER-positive luminal subtype, which is generally asso- available reagents can clearly differentiate between ciated with better prognosis (Gillett et al., 1994, 1996; cyclin D1a and D1b (Burd et al., 2006; Marzec et al., Barnes and Gillett, 1998; Roy and Thompson, 2006). 2006; Sanchez et al., 2008; Wang et al., 2008). Using From our analyses, we found that cyclin D1a levels were such reagents, we quantitatively determined the levels of elevated predominantly in ER-positive tumors, were each protein unambiguously, and these results were inversely correlated with Ki67 and had little impact on further supported by quantitative analyses on RNA disease outcome in the cohort analysed when dichot-

Oncogene Cyclin D1b expression associated with poor outcome EKA Millar et al 1819 omized about the median into high and low expressors. with ER-positive cancers, high levels of cyclin D1b were In contrast, elevated levels of cyclin D1b protein strongly associated with poor disease outcome in ER- were independent of a variety of clinicopathological negative breast cancers. As this form of the disease is features, including Ki67 and ER. However, in marked characterized by relatively low cyclin D1a expression, contrast to cyclin D1a, elevated cyclin D1b was this finding suggests that cyclin D1b may be particularly associated with poor overall survival, recurrence and relevant in the etiology, progression and the ultimate metastasis independent of other commonly analysed poor prognosis associated with ER-negative breast prognostic variables. Thus, cyclin D1b identifies a cancer. Presumably, the differing survival could reflect unique subset of tumors associated with increased an important impact of cyclin D1b on therapeutic disease progression. In keeping with cyclin D1b acting interventions involved in the treatment of ER-negative independently of cyclin D1a, the association of cyclin breast cancer. However, this possibility remains to be D1b with disease outcome was not modulated by the explored in preclinical models. status of cyclin D1a. In addition, cyclin D1b levels were Combined, these studies provide critical insight into independent of Her2. However, owing to the limited the association of the G/A870 polymorphism and cyclin number of Her2-positive tumors in our cohorts, we were D1 isoform expression with risk and progression of unable to address whether cyclin D1b status influenced breast cancer, and in so doing show that cyclin D1b disease outcome in this subset. In contrast, stratification levels are independently associated with adverse disease of patients based on ER status provides a compelling outcome. rationale for investigating cyclin D1b function in discrete subtypes of breast cancer. Among ER-positive Acknowledgements breast cancers, high levels of cyclin D1b, although trending toward poor outcome, were not significant. On The authors thank all of the laboratory and administrative the basis of the preclinical studies (Wang et al., 2008), it staff that contributed to the editing and preparation of the will be crucial to specifically interrogate the association manuscript. This work was supported by the following grants: of cyclin D1b levels to the response to endocrine the National Health and Medical Research Council of Australia (Grant no. 276408), the Cancer Institute NSW, the therapies (for example, tamoxifen) in a setting free of Petre Foundation and the RT Hall Trust, the National cytotoxic agents. Clearly, larger studies will be required Institutes of Health Grant CA104213 (ESK), the National to specifically define the influence of cyclin D1b levels in Institutes of Health Grant CA099996 (KEK). We thank Drs ER-positive breast cancer and address the critical Paul Crea and Davendra Segara of St Vincent’s Clinic for question of response to hormonal therapy. In contrast access to patient material and data.

References

Arnold A, Papanikolaou A. (2005). Cyclin D1 in breast cancer Cobrinik D. (2005). Pocket proteins and cell cycle control. Oncogene pathogenesis. J Clin Oncol 23: 4215–4224. 24: 2796–2809. Bala S, Peltomaki P. (2001). Cyclin D1 as a genetic modifier Comstock CES, Augello MA, Pe Benito R, Karch J, Tran TH, Utama in hereditary nonpolyposis colorectal cancer. Cancer Res 61: FE et al. (2009). Cyclin D1 splice variants: polymorphism, risk, and 6042–6045. isoform specific regulation in prostate cancer. Clin Cancer Res Barnes DM, Gillett CE. (1998). Cyclin D1 in breast cancer. Breast (submitted). Cancer Res Treat 52: 1–15. Diehl JA. (2002). Cycling to cancer with cyclin D1. Cancer Biol Ther 1: Batsche E, Yaniv M, Muchardt C. (2006). The human SWI/SNF 226–231. subunit Brm is a regulator of alternative splicing. Nat Struct Mol Dolled-Filhart M, McCabe A, Giltnane J, Cregger M, Camp RL, Biol 13: 22–29. Rimm DL. (2006). Quantitative in situ analysis of beta-catenin Betticher DC, Thatcher N, Altermatt HJ, Hoban P, Ryder WD, expression in breast cancer shows decreased expression is associated Heighway J. (1995). Alternate splicing produces a novel cyclin D1 with poor outcome. Cancer Res 66: 5487–5494. transcript. Oncogene 11: 1005–1011. Gillett C, Fantl V, Smith R, Fisher C, Bartek J, Dickson C et al. Blais A, Dynlacht BD. (2007). E2F-associated chromatin modifiers (1994). Amplification and overexpression of cyclin D1 in breast and cell cycle control. Curr Opin Cell Biol 19: 658–662. cancer detected by immunohistochemical staining. Cancer Res 54: Buckley MF, Sweeney KJ, Hamilton JA, Sini RL, Manning DL, 1812–1817. Nicholson RI et al (1993). Expression and amplification of cyclin Gillett C, Smith P, Gregory W, Richards M, Millis R, Peters G et al. genes in human breast cancer. Oncogene 8: 2127–2133. (1996). Cyclin D1 and prognosis in human breast cancer. Int J Burd CJ, Petre CE, Morey LM, Wang Y, Revelo MP, Haiman CA Cancer 69: 92–99. et al. (2006). Cyclin D1b variant influences prostate cancer growth Gupta VK, Feber A, Xi L, Pennathur A, Wu M, Luketich JD et al. through aberrant androgen receptor regulation. Proc Natl Acad Sci (2008). Association between CCND1 G/A870 polymorphism, allele- USA 103: 2190–2195. specific amplification, cyclin D1 expression, and survival in Carrere N, Belaud-Rotureau MA, Dubus P, Parrens M, de Mascarel esophageal and lung carcinoma. Clin Cancer Res 14: 7804–7812. A, Merlio JP. (2005). The relative levels of cyclin D1a and D1b Holley SL, Heighway J, Hoban PR. (2005). Induced expression of alternative transcripts in mantle cell lymphoma may depend more human CCND1 alternative transcripts in mouse Cyl-1 knockout on sample origin than on CCND1 polymorphism. Haematologica fibroblasts highlights functional differences. Int J Cancer 114: 90: 854–856. 364–370. Ceschi M, Sun CL, Van Den Berg D, Koh WP, Yu MC, Probst- Holley SL, Parkes G, Matthias C, Bockmuhl U, Jahnke V, Leder K Hensch N. (2005). The effect of cyclin D1 (CCND1) G870A- et al. (2001). Cyclin D1 polymorphism and expression in patients polymorphism on breast cancer risk is modified by oxidative stress with squamous cell carcinoma of the head and neck. Am J Pathol among Chinese women in Singapore. Carcinogenesis 26: 1457–1464. 159: 1917–1924.

Oncogene Cyclin D1b expression associated with poor outcome EKA Millar et al 1820 Iaquinta PJ, Lees JA. (2007). Life and death decisions by the E2F populations: a meta-analysis. Cancer Epidemiol Biomarkers Prev transcription factors. Curr Opin Cell Biol 19: 649–657. 17: 2773–2781. Jirstrom K, Stendahl M, Ryden L, Kronblad A, Bendahl PO, Stal O Roy PG, Thompson AM. (2006). Cyclin D1 and breast cancer. Breast et al. (2005). Adverse effect of adjuvant tamoxifen in premenopausal 15: 718–727. breast cancer with cyclin D1 gene ampliﬁcation. Cancer Res 65: Rudas M, Lehnert M, Huynh A, Jakesz R, Singer C, Lax S et al. 8009–8016. (2008). Cyclin D1 expression in breast cancer patients Knudsen ES, Knudsen KE. (2006). Retinoblastoma tumor suppressor: receiving adjuvant tamoxifen-based therapy. Clin Cancer Res 14: where cancer meets the cell cycle. Exp Biol Med (Maywood) 231: 1767–1774. 1271–1281. Sanchez G, Bittencourt D, Laud K, Barbier J, Delattre O, Auboeuf D Knudsen KE. (2006). The cyclin D1b splice variant: an old oncogene et al. (2008). Alteration of cyclin D1 transcript elongation learns new tricks. Cell Div 1: 15. by a mutated transcription factor up-regulates the oncogenic Knudsen KE, Diehl JA, Haiman CA, Knudsen ES. (2006). Cyclin D1: D1b splice isoform in cancer. Proc Natl Acad Sci USA 105: polymorphism, aberrant splicing and cancer risk. Oncogene 25: 6004–6009. 1620–1628. Shu XO, Moore DB, Cai Q, Cheng J, Wen W, Pierce L et al. (2005). Krieger S, Gauduchon J, Roussel M, Troussard X, Sola B. (2006). Association of cyclin D1 genotype with breast cancer risk and Relevance of cyclin D1b expression and CCND1 polymorphism in survival. Cancer Epidemiol Biomarkers Prev 14: 91–97. the pathogenesis of multiple myeloma and mantle cell lymphoma. Sicinski P, Donaher JL, Parker SB, Li T, Fazeli A, Gardner H et al. BMC Cancer 6: 238. (1995). Cyclin D1 provides a link between development and Krippl P, Langsenlehner U, Renner W, Yazdani-Biuki B, Wolf G, oncogenesis in the retina and breast. Cell 82: 621–630. Wascher TC et al. (2003). The 870G>A polymorphism of the cyclin Solomon DA, Wang Y, Fox SR, Lambeck TC, Giesting S, Lan Z et al. D1 gene is not associated with breast cancer. Breast Cancer Res (2003). Cyclin D1 splice variants. Differential effects on localization, Treat 82: 165–168. RB phosphorylation, and cellular transformation. J Biol Chem 278: Lu F, Gladden AB, Diehl JA. (2003). An alternatively spliced cyclin 30339–30347. D1 isoform, cyclin D1b, is a nuclear oncogene. Cancer Res 63: Spruck C, Sun D, Fiegl H, Marth C, Mueller-Holzner E, Goebel G 7056–7061. et al. (2006). Detection of low molecular weight derivatives of cyclin Malumbres M, Barbacid M. (2001). To cycle or not to cycle: a critical E1 is a function of cyclin E1 protein levels in breast cancer. Cancer decision in cancer. Nat Rev Cancer 1: 222–231. Res 66: 7355–7360. Marzec M, Kasprzycka M, Lai R, Gladden AB, Wlodarski P, Sutherland RL, Musgrove EA. (2004). Cyclins and breast cancer. Tomczak E et al. (2006). Mantle cell lymphoma cells express J Mammary Gland Biol Neoplasia 9: 95–104. predominantly cyclin D1a isoform and are highly sensitive to Wang TC, Cardiff RD, Zukerberg L, Lees E, Arnold A, Schmidt EV. selective inhibition of CDK4 kinase activity. Blood 108: 1744–1750. (1994). Mammary hyperplasia and carcinoma in MMTV-cyclin D1 Matthias C, Branigan K, Jahnke V, Leder K, Haas J, Heighway J et al. transgenic mice. Nature 369: 669–671. (1998). Polymorphism within the cyclin D1 gene is associated with Wang Y, Dean JL, Millar EK, Tran TH, McNeil CM, Burd CJ et al. prognosis in patients with squamous cell carcinoma of the head and (2008). Cyclin D1b is aberrantly regulated in response to therapeutic neck. Clin Cancer Res 4: 2411–2418. challenge and promotes resistance to estrogen antagonists. Cancer Naidu R, Yip CH, Taib NA. (2008). Polymorphisms of HER2 Res 68: 5628–5638. Ile655Val and cyclin D1 (CCND1) G870A are not associated with Yu CP, Yu JC, Sun CA, Tzao C, Ho JY, Yen AM. (2008). Tumor breast cancer risk but polymorphic allele of HER2 is associated with susceptibility and prognosis of breast cancer associated with the nodal metastases. Neoplasma 55: 87–95. G870A polymorphism of CCND1. Breast Cancer Res Treat 107: Pabalan N, Bapat B, Sung L, Jarjanazi H, Francisco-Pabalan O, 95–102. Ozcelik H. (2008). Cyclin D1 Pro241Pro (CCND1-G870A) poly- Yu Q, Geng Y, Sicinski P. (2001). Speciﬁc protection against breast morphism is associated with increased cancer risk in human cancers by cyclin D1 ablation. Nature 411: 1017–1021.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc)

Oncogene IJC International Journal of Cancer

PI3K pathway activation in breast cancer is associated with the basal-like phenotype and cancer-speciﬁc mortality

Elena Lo´pez-Knowles1, Sandra A. O’Toole1,2,3,4, Catriona M. McNeil1,5, Ewan K.A. Millar1,6,7,8, Min R. Qiu1,9, Paul Crea10, Roger J. Daly1,4, Elizabeth A. Musgrove1,4 and Robert L. Sutherland1,4

1 Cancer Research Program, Garvan Institute of Medical Research, Darlinghurst, Sydney, NSW 2010, Australia 2 Department of Anatomical Pathology, Royal Prince Alfred Hospital, Camperdown, NSW 2050, Australia 3 Central Clinical School, University of Sydney, Sydney, NSW 2050, Australia 4 St Vincent’s Clinical School, Faculty of Medicine, University of New South Wales, Sydney, NSW 2052, Australia 5 Department of Medical Oncology, Westmead Hospital, Sydney, NSW 2145, Australia 6 Department of Anatomical Pathology, South Eastern Area Laboratory Service, St George Hospital, Kogarah, NSW 2217, Australia 7 Department of Pathology, School of Medicine, University of Western Sydney, Sydney, Australia 8 School of Medical Sciences, Faculty of Medicine, University of New South Wales, Sydney, Australia 9 Department of Anatomical Pathology, Sydpath, St Vincent’s Hospital, Darlinghurst, NSW 2010, Australia 10 Department of Surgical Oncology, St Vincent’s Hospital, Darlinghurst, NSW 2010, Australia

Breast cancer is a common malignancy with current biological therapies tailored to steroid hormone (ER, PR) and HER2 receptor status. Understanding the biological basis of resistance to current targeted therapies and the identification of new potential therapeutic targets is an ongoing challenge. The PI3K pathway is altered in a high proportion of breast cancers and may contribute to therapeutic resistance. We undertook an integrative study of mutational, copy number and expression analyses of key regulators of the PI3K pathway in a cohort of 292 invasive breast cancer patients with known treatment outcomes. The alterations identified in this cohort included PIK3CA mutations (12/168, i.e. 7%), PIK3CA copy number gain (28/209, i.e. 14%), PTEN loss (73/258, i.e. 28%) and AKT activation (62/258, i.e. 24%). Overall at least 1 parameter was altered in 72% (139/193) of primary breast cancers. PI3K pathway activation was significantly associated with ER negative (p 5 0.0008) and PR negative (p 5 0.006) status, high tumor grade (p 5 0.032) and a ‘‘basal-like’’ phenotype (p 5 0.01),

where 92% (25/27) of tumors had an altered pathway. In univariate analysis, PI3K pathway aberrations were associated with Cancer Cell Biology death from breast cancer; however, this relationship was not maintained in multivariate analysis. No association was identified between an activated pathway and outcome in tamoxifen- or chemotherapy-treated patients. We concluded that >70% of breast cancers have an alteration in at least 1 component of the PI3K pathway and this might be exploited to therapeutic advantage especially in ‘‘basal-like’’ cancers.

The phosphatidylinositol 3 kinase (PI3K) pathway regulates receptor. Once recruited to the membrane, the p110a many cellular functions, mainly associated with cell prolifera- catalytic subunit, encoded by PIK3CA, phosphorylates phos- tion, survival and migration.1 Activation of PI3K can occur phatidylinositol biphosphate to generate phosphatidylinositol in response to a variety of extracellular signals through triphosphate (PIP3), which recruits phosphoinositide-depend- growth factor receptor- or integrin-mediated pathways. Upon ent kinase 1 (PDK1) and V-Akt murine thymoma viral onco- receptor activation, the p85 regulatory subunit of Class 1a gene homolog 1 (AKT) to the cell membrane. PDK1 and PI3K is recruited to phosphotyrosine residues within the PDK2 (likely the TORC2 complex) then phosphorylate AKT leading to the regulation of an array of downstream pathways including forkhead rhabdomyosarcoma, GSK3b and BAD Key words: breast cancer, PI3K, PTEN, prognosis, basal-like with consequent effects on cell proliferation, cell metabolism phenotype and apoptosis, respectively. PIP3 is a substrate for phospha- Grant sponsors: National Health and Medical Research Council of tase and tensin homologue deleted on chromosome 10 Australia (NHMRC), Cancer Institute NSW, Petre Foundation, RT (PTEN), a well-described tumor suppressor protein, which Hall Trust dephosphorylates PIP3, inactivating the PI3K pathway. DOI: 10.1002/ijc.24831 Alterations in the PI3K pathway have been documented History: Received 10 May 2009; Accepted 3 Aug 2009; Online 14 in many human cancers.2 In breast cancer, PIK3CA activating Aug 2009 mutations, in exons 9 and 20, have been identiﬁed in 24% Correspondence to: Robert L. Sutherland, Cancer Research (range, 12–40%) of patient specimens.3–16 These mutations in Program, Garvan Institute of Medical Research, 384 Victoria St., the p110a catalytic subunit can activate the pathway and Darlinghurst, NSW 2010, Australia, Fax: þ61-2-92958321, contribute to mammary tumor progression.15 PIK3CA ampli- E-mail: [email protected] ﬁcation has only been described in 2 breast cancer studies

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC 1122 PI3K pathway activation and breast cancer outcome

with a frequency of 1% and 9%, respectively.12,16 Molecular the breast from patients treated by a single surgeon (P.C.) alterations and loss of PTEN have also been described in between February 1992 and August 2002 at St. Vincent’s 24% (range, 4–48%) of breast cancers5,6,14,17–23 and Hospital, Sydney, Australia. Ethics approval was granted for increased activation of AKT has been identified in 41% the use of pathology specimens and cognate clinicopathologi- (range, 9–76%) of breast cancer patients.5,17,19,21,22,24–28 De- cal data (Human Research Ethics Committee of St. Vincent’s spite the number of studies exploring the PI3K pathway in Hospital, Sydney). A more detailed description of the clinico- breast cancer, the association of specific molecular alterations pathological characteristics of the cohort is published else- with clinicopathological markers and outcome is often con- where.32 Forty percent of tumors were >20 mm, 45% were tradictory and unclear with a consensus yet to be reached. Grade >2, 43% were lymph node positive, 68% were ER pos- This may be due in part to the widely different frequencies itive, 57% were PR positive and 18% were HER2 FISH posi- of specific aberrations reported in published studies. tive (>2.2 ratio of HER2:chr.17 centrosome). Median age was PIK3CA mutations have been associated with positive hor- 54 years and patients were treated with endocrine therapy mone receptor status in 2 studies,3,14 but not in others,7,11,15 (49%), chemotherapy (38%) or both (24%). Cases were pro- and with both favorable9 and poor patient outcome.10 PTEN spectively followed-up for a median of 64 months and the loss has been associated with high grade and negative hor- outcome events measured were as follows: recurrence (local mone receptor status in some studies17,20 but not in or distant) (25%), metastasis (23%) and all deaths were others,22,23 whereas AKT activation has been associated with recorded but only breast cancer-related deaths (18%) were high tumor grade,19 PR negativity and HER2 positive status28 considered for survival analyses. but other studies have identified no such associations.12,17 Furthermore, there has been no published integrative study DNA extraction determining the combined effects of aberrations in these 3 One core from each tumor block, from an area previously pivotal members of the pathway. selected by a specialist breast pathologist (S.A.O’T.) as the Breast cancer remains the most common invasive cancer most representative of the tumor, was used to extract DNA. of women in industrialized nations and is second only to The Agencourt formapure 96-well format kit (Beckman lung cancer as the leading cause of cancer-related death in Coulter, Beverly, MA), a paramagnetic bead-based technol- women. Breast cancer is heterogenous but can be classified ogy, was used to extract DNA from formalin-fixed, paraffin- into at least 4 subtypes by gene expression profiling: Luminal embedded tissue from 288 individual patients. A significant A, Luminal B, HER2þ and ‘‘basal-like’’ tumors,29,30 with number of cases (n ¼ 120 of 288, i.e. 42%) were unsuitable Cancer Cell Biology each subgroup reflecting a different biology, behavior and for the mutational analysis, and the main reasons were the outcome. A surrogate classification of these subgroups has poor quality of DNA caused by years of storage of these tis- been attempted using immunohistochemistry, whereby sues, the fixation protocol employed or the small amount of patients were subclassified according to hormone receptor tissue sometimes available for DNA extraction. (ER, PR), HER2, cytokeratin 5/6 (CK5/6) and epidermal growth factor receptor (EGFR) status.31 Current breast cancer PIK3CA mutational analysis treatment is tailored in great measure by this subtype classifi- Primers were designed to avoid amplification of a known cation. Hormone receptor positive cancers are treated pre- PIK3CA pseudogene for exons 9 and 20 of the PIK3CA gene: dominantly with endocrine therapy (ERþ/PRþ/HER2)or 9F CTGTAAATCATCTGTGAATC, 9R ATTTTAGCACTTA Trastuzumab (HER2þ), but despite the advances in treat- CCTGTGAC; 20F CAATCTTTTGATGACATTGC, 20R TG ment and outcome, understanding the development of thera- GAATCCAGAGTGAGCTTT. These 2 exons contain the 3 peutic resistance and developing strategies to combat it is a hotspots for PIK3CA mutations in cancer. Codons 542 and continuing challenge. ‘‘Basal-like’’ cancers (triple negative for 545 in exon 9 and codon 1047 in exon 20. PCR reactions ER, PR and HER2 and positive for CK5/6 or EGFR) have no were undertaken using 10–50 ng of DNA, 0.4 lmol/l of each identified target for biological therapy and have recently been primer, 200 lmol/l deoxynucleotide triphosphates, 3 mmol/l

a focus of studies searching for potential new therapeutic tar- MgCl2,1 PCR II buffer and 1.5 U of Amplitaq Gold DNA gets. Thus, the aim of our study was to analyze the overall Polymerase (Applied Biosystems, Foster City, CA). The PCR status of the PI3K pathway, by defining mutational status in conditions used were as follows: 94C (10 min) for 1 cycle, exons 9 and 20, copy number and expression of key mole- 94C (30 sec), 50C (30 sec), 72C (30 sec) for 40 cycles and cules within the pathway, in a cohort of breast cancer a final extension step of 72C (10 min). All PCR products patients of known phenotype, treatment response and out- were sequenced with the ABI 3100 Genetic analyzer (Applied come status. Biosystems, Foster City, CA). All mutations were confirmed by analyzing the product of a second, independent PCR. Material and Methods Patients and tumors PIK3CA gene copy number Cases were drawn from the St. Vincent’s Campus Outcome A subset of tumors (n ¼ 209) was analyzed for PIK3CA gene Cohort, which comprised 292 invasive ductal carcinomas of copy number by real time quantitative PCR (qPCR) on the

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC Lo´pez-Knowles et al. 1123

Roche Lightcycler 480 platform (Roche Diagnostics; F. Hoff- istry protocols, scoring and analysis for p27Kip1, p21Cip1/Waf1, mann-La Roche Ltd., Mannheim, Germany). RPLPO, located cyclin E1, cyclin D1 and Ki67 are described elsewhere37 at 12q24, was used as an endogenous control, as previously (McNeil C.M., unpublished data). described.33 RPLPO yielded similar results as the use of a gene copy number control from an unaltered region on the Immunohistochemical scoring same chromosome as PIK3CA. Primers for both genes were Scoring was completed by 4 independent investigators (E.L.- designed by the Roche primer software using the universal K., S.A.O’T., E.K.A.M., M.R.Q.) blinded to the clinical and probe library system. qPCR reactions were completed using pathological information, 3 of whom were specialist breast 50 ng of DNA, 0.2 lmol/l of each primer, 0.2 ll of probe pathologists; in cases of discrepancy a consensus was reached and 5 ll of master mix. qPCR conditions used were as fol- by conferencing. All tissue areas were examined for both in- lows: 95C (7 min) for 1 cycle, 95C (5 sec), 50–60C (0.4C tensity (1þ to 3þ) and the proportion of positive cells. A per second) (30 sec), 72C (15 sec) for 50 cycles and a final histoscore was calculated by multiplying the percentage of extension step of 40C (30 sec). The efficiency of both assays positively stained cells and each category of staining intensity was analyzed from the slope of a standard curve; RPLPO and (Fig. 1). PTEN showed cytoplasmic localization and loss PIK3CA qPCR assays had an efficiency of 95% and 97%, of expression was defined as the lack of any tumor cell stain- respectively. Fold changes were calculated with the Delta Ct ing. The data were dichotomized as present or absent stain- method. Fold changes above the mean 6 1 SD of all the ing. pAKT demonstrated predominantly cytoplasmic staining, samples analyzed, i.e. >1.381, were considered copy number although in a few cases nuclear staining was apparent. For gain. This is in agreement with a previously published copy immunostaining of pAKT the mean histoscore value was number threshold based on a pool of normal breast tis- used to determine the cutoff value of H > 22. sue.34,35 The coefficient of variation for triplicate estimates was <0.04. Statistical analyses To assess the independence of 2 categorical variables, the Tissue specimens chi-square test was applied when the variables were dichoto- A total of 18 tissue microarrays (TMA) containing 2 cores of mous, the Mann–Whitney test was applied when one variable each tumor sample were constructed from the formalin-fixed, was dichotomous and the other continuous and the Spear- paraffin-embedded tumor material from each patient in the man Rank correlation was applied when both variables were cohort as previously reported.32 TMAs were produced using continuous. Kaplan–Meier survival curves and Cox propor- Cancer Cell Biology the MTA-1 Manual Tissue Arrayer (Beecher Instruments, tional hazard ratio (HR) were estimated to obtain risks of re- Woodland, CA, USA). currence, metastasis and death, after adjusting for other confounder variables. The results were considered significant at Immunohistochemistry the two-sided p of 0.05 level. Statview version 5.0 was used Three-micrometer sections of each TMA were cut, deparaffi- for the analysis (Abacus Systems, Berkeley, CA). nized and used for immunostaining. The following antibodies were used: pAKT (Ser473) rabbit monoclonal antibody clone Results 736E11 (Biocare Medical, Concord, CA, USA) and PTEN PIK3CA genetic alterations and PTEN loss mouse monoclonal antibody clone 6H2.1 (Dako, Denmark). To assess the frequency of genetic lesions in the PIK3CA For pAKT, antigen retrieval was performed using DAKO so- gene or loss of PTEN protein, both central players in the lution pH 6.0 (s1699; DAKO, Carpentaria, CA) at 125C for PI3K pathway, we analyzed PIK3CA mutations by sequenc- 4 min in a Pascal pressure chamber (DAKO), the antibody ing, PIK3CA copy number by qPCR and PTEN expression was incubated for 90 min at a 1:50 dilution. For PTEN, anti- levels by immunohistochemistry. gen retrieval was performed using DAKO solution pH 9.0 Overall, 12 of 168 (7%) breast cancers assessed harbored (s2367; DAKO) at 125C for 30 sec in a Pascal pressure PIK3CA mutations. All mutations were at previously reported chamber, and antibody incubation was 90 min at a 1:200 hotspot codons in exons 9 and 20, i.e. E542K (n ¼ 1), E545K dilution. A DAKO autostainer was used for immunostaining (n ¼ 2), Q546R (n ¼ 1) and H1047R (n ¼ 8). (DAKO). As secondary antibody a FLEX envision system Copy number gain was identified in 28 of 209 (14%) can- (DAKO) was used to improve staining. Reactions were devel- cers analyzed. Mutations and copy number gains were almost oped using diaminobenzidine and sections were counter- exclusive events with only 1 cancer encompassing both muta- stained, dehydrated and mounted. For pAKT, sections incu- tion and copy number gain of the PIK3CA gene (p ¼ 0.748). bated with nonimmune serum (IgG) and LNCAP prostate Neither mutations nor copy number gain were associated cancer cells treated with LY29402 were used as negative con- with any clinicopathological parameter or outcome. trols36 and untreated LNCAP cells as positive controls. For PTEN, a negative regulator of the PI3K pathway, was lost PTEN, sections of human breast and prostate tumors were in 73 of 258 (28%) cancers and was associated with high tu- used as positive controls and the same sections incubated mor grade (p ¼ 0.009), tumor size >20 mm (p ¼ 0.009), ER with IgG were used as negative controls. Immunohistochem- negativity (p < 0.001) and PR negativity (p ¼ 0.0002) (Table 1).

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC 1124 PI3K pathway activation and breast cancer outcome

Figure 1. Immunohistochemical staining of parafﬁn-embedded breast cancer tissue. Photomicrographs for PTEN expression: negative (IgG) and positive controls (a) and tumor scoring ranging from negative to intensities 1þ,2þ and 3þ (c). Photomicrographs for pAKT expression: negative (IgG) and positive controls (b) and tumor scoring ranging from negative to intensities 1þ,2þ and 3þ (d).

PTEN loss was not associated with recurrence, metastasis or way. These 193 patients had similar clinicopathological char- breast cancer-speciﬁc death. acteristics as the overall 292 tumors and therefore were rep-

Cancer Cell Biology resentative of the whole cohort (data not shown). Figure 2 describes the overlap between the alterations in the 4 param- AKT activation eters analyzed. Over 75% (105/139) of cancers had only 1 As a measure of PI3K pathway activation, we also examined altered parameter, 23% (32/139) had 2 alterations and only pAKT expression levels by immunohistochemistry. pAKT 1% (2/139) had 3 altered parameters. This measure of path- was positive in 62 of 258 (24%) specimens and was positively way activation was associated with high tumor grade (p ¼ associated with high tumor grade (p ¼ 0.011), ER negativity 0.032), ER negative (p ¼ 0.0008) and PR negative (p ¼ (p ¼ 0.040), PR negativity (p ¼ 0.014) and HER2 positivity 0.006) status (Table 1). (p ¼ 0.016). pAKT was also associated with Ki67, a marker Of the 193 patients, with at least 1 altered parameter or of cell proliferation (p ¼ 0.005) (Table 1) and breast cancer- wild type for all four, 124 had data for all 4 parameters ana- specific death (p ¼ 0.03). lyzed. The clinicopathological characteristics of this subgroup of 124 patients were not statistically significantly different to PI3K pathway activation and clinicopathological the 193 group of patients (data not shown). All statistical parameters analysis were then undertaken for this group to determine To assess the overall status of the PI3K pathway, we consid- relationships with clinicopathological parameters, subtypes ered PIK3CA mutational and copy number data, PTEN and outcome. The results and conclusions remained expression and AKT activation together. PTEN loss and unchanged when the analysis was confined to this subgroup PIK3CA mutations were mutually exclusive events (p ¼ of 124 patients (data not shown). 0.0368); all PIK3CA mutations were identified in PTEN- expressing tumors. No association was identified between PTEN loss and PIK3CA amplification (p ¼ 0.982). A negative PI3K pathway activation and breast cancer subtypes correlation was identified between expression of PTEN and When cases were distributed into 5 phenotypes according to activation of AKT by the Spearman Rank correlation (p ¼ ER/PR/HER2, CK5/6 and EGFR status,31,38 i.e. Luminal A 0.0001). Low PTEN expression was associated with high (ERþ and/or PRþ, HER2), Luminal B (ERþ and/or PRþ, AKT activation. When all 4 parameters were combined HER2þ), HER2 (ER and PR, HER2þ), ‘‘basal-like’’ (PIK3CA mutations, PIK3CA copy number gain, PTEN loss (ER,PR, HER2, CK5/6þ and/or EGFRþ) and unclassi- and AKT activation) 139 of 193 (72%) breast cancer patients fied (negative for all markers), there was no significant rela- had an alteration in 1 or more components of the PI3K path- tionship with PIK3CA mutation or amplification. PTEN loss

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC n.J Cancer: J. Int. Lo ´ pez-Knowles

Table 1. Association between the PI3K pathway and clinicopathological parameters

126 PIK3CA sequence PIK3CA copy number PTEN expression AKT expression PI3K pathway tal. et

1113 (2010) 1121–1131 , WT MUT p-Value* NC GAIN p-Value* H 5 0 H > 0 p-Value* H < 22 H > 22 p-Value* Inactive Active p-Value* All cohort 156 12 181 28 73 185 196 62 54 139 Tumor size <20 mm 93 5 96 16 33 116 112 37 34 72 >20 mm 62 7 0.214 84 12 0.707 40 68 0.009 84 24 0.627 20 66 0.176

V Tumor grade C 09UICC 2009 2 89 6 90 13 29 107 112 24 33 61 >2 66 6 0.617 91 15 0.746 44 78 0.009 84 38 0.011 21 78 0.032 LN status 0 95 7 94 14 36 101 104 33 31 70 >0 58 5 0.796 84 14 0.782 37 81 0.371 89 29 0.928 20 69 0.202 Age (years) <50 60 3 75 10 27 71 76 22 24 48 >50 96 9 0.353 106 18 0.566 46 114 0.835 120 40 0.642 30 91 0.201 ER Negative 43 5 59 12 38 39 52 25 10 61 Positive 106 7 0.351 114 16 0.369 32 144 <0.0001 140 36 0.040 43 74 0.0008 PR Negative 63 4 77 14 43 64 73 34 17 74 Positive 88 8 0.570 97 14 0.571 28 119 0.0002 120 27 0.014 36 62 0.006 HER2 Negative 113 8 136 21 59 142 157 43 42 103 Positive 29 4 0.295 35 6 0.834 11 35 0.442 29 18 0.016 10 30 0.621 Ki67 Negative 71 5 84 13 30 101 109 22 30 60 Positive 64 6 0.649 84 14 0.858 40 83 0.086 84 39 0.005 24 74 0.181

WT, wild type; MUT, mutated; NC, no change; H, histoscore; LN, lymph node status. Thresholds of positivity: ER and PR, H > 10; HER2 FISH ratio, >2.2 HER2:chr.17 centromere; Ki67 staining >5%; PIK3CA copy number is considered gain 2dCt > 1.38. PI3K pathway status is deﬁned as active if tumors present an alteration in at least one of the studied parameters. Inactive tumors are wild type for the parameters analyzed. *p-Values were obtained by chi-square analysis. 1125

Cancer Cell Biology 1126 PI3K pathway activation and breast cancer outcome -Value* p -Value* Inactive Active p

Figure 2. Venn diagram representing the overlap between AKT 22

activation measured by pAKT positive staining, PTEN loss > measured by PTEN negative staining, PIK3CA mutation analyzed by H sequencing and PIK3CA copy number gain analyzed by qPCR. The

ﬁgure shows that although there is a slight overlap between the 22 <

alterations (24%), many cancers have only 1 altered parameter. H

was associated with the ‘‘basal-like’’ phenotype (p < 0.001) 0.0001 19 10 0.19 2 25 0.01 0.001 125 30 0.01 37 66 0.005 and high PTEN expression with Luminal A cancers. High -Value* < < p pAKT levels were associated with the HER2 phenotype (p ¼

0.04) and low pAKT levels with the Luminal A phenotype 0 (p ¼ 0.01). The activated pathway as measured by PIK3CA > H Cancer Cell Biology mutations, PIK3CA copy number gain, AKT activation or ¼

PTEN loss was associated with ‘‘basal-like’’ tumors (p 0

0.01) and the inactive pathway with Luminal A cancers (p ¼ 5 0.005) (Table 2). H

PI3K pathway activation and cell cycle markers

We next analyzed the relationship between components of -Value* the PI3K pathway and key cell cycle regulators that had pre- p viously been assessed in these cancers as downstream markers of pathway activation. PTEN loss was associated copy number PTEN expression AKT expression PI3K pathway with increased cyclin E1 expression (p ¼ 0.019) and low Cip1/Waf1 expression of cyclin D1 (p ¼ 0.019), p21 (p ¼ 0.019) PIK3CA and p27Kip1 (p ¼ 0.024). AKT activation was associated with increased cyclin E1 (p ¼ 0.008) and p21Cip1/Waf1 expression (p ¼ 0.0005). An activated PI3K pathway was associated with high expression of cyclin E1 (p ¼ 0.009), but not with -Value* NC GAIN expression of cyclin D1 (p ¼ 0.158) or the Cdk inhibitors p p27Kip1 (p ¼ 0.112) and p21Cip1/Waf1 (p ¼ 0.724) (Fig. 3). sequence

PI3K pathway activation, disease outcome and treatment In a univariate analysis an aberrant PI3K pathway was asso- PIK3CA

ciated with breast cancer-specific death (HR ¼ 4.17, 95% CI WT MUT ¼ 1.28–13.59, p ¼ 0.017) (Fig. 4a). However, when the cohort was stratified on ER status the relationship Association between the PI3K pathway and breast cancer subtypes approached significance for ERþ tumors (n ¼ 117) (HR ¼ 6.61, 95% CI ¼ 0.86–50.38, p ¼ 0.068) but not ER tumors ‘‘Basal like’’ 16 1 0.74 25 3 0.62 19 10 HER2Luminal A 13 89 2 7 0.41 0.71 101 16 15 4 0.70 0.38 30 6 125 15 0.95 12 9 0.04 4 15 0.49 Luminal BUnclassified 16 6 2 0 0.59 0.46 19 9 2 3 0.55 0.24 5 9 21 6 0.28 0.65 17 12 9 3 0.22 0.65 6 2 15 12 0.93 0.24 ¼ ¼ ¼ ¼ p -Values were obtained by chi-square analysis. (n 71) (HR 1.40, 95% CI 0.32–6.04, p 0.649). In a Table 2. WT, wild type; MUT,wild mutated; type NC, for no the change;* parameters H, analyzed. histoscore. PI3K pathway status is defined as active if tumors present an alteration in at least one of the studied parameters. Inactive tumors are

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC Lo´pez-Knowles et al. 1127

Figure 3. Correlation between expression of cell cycle markers cyclin D1, cyclin E1, p21Cip1/Waf1 and p27Kip1 and PI3K pathway status. Box plots from a Mann–Whitney analysis of the association between an active or inactive PI3K pathway and expression, measured by immunohistochemistry, are shown. Cancer Cell Biology

Figure 4. Kaplan–Meier curves illustrating the relationship between an activated PI3K pathway and disease progression and treatment. (a) Recurrence, metastasis and cancer-specific death stratified by an active (A: alteration of PIK3CA, PTEN or AKT) or inactive (I: wild type for parameters studied) pathway. Number of patients at risk, hazard ratios, 95% CI and p value of the association by log rank testing are shown. (b) Breast cancer-specific death in patients treated with adjuvant endocrine (tamoxifen) or chemotherapy and untreated patients, stratified by an active or inactive pathway.

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC 1128 PI3K pathway activation and breast cancer outcome

Table 3. Multivariate analysis for clinicopathological parameters, hormone receptor status and PI3K pathway

Variable Hazard ratio 95% conﬁdence interval p-Value Univariate Grade >2 3.520 1.930–6.418 <0.0001 Size >20 mm 2.468 1.415–4.304 0.0015 Lymph node >0 3.691 2.025–6.729 <0.0001 Age >50 years 1.427 0.799–2.551 0.2297 ERþ 0.300 0.172–0.524 <0.0001 PRþ 0.170 0.087–0.333 <0.0001 HER2 FISHþ 3.491 1.956–6.229 <0.0001 AKT activation 1.891 1.065–3.356 0.029 Activated pathway 4.174 1.281–13.598 0.017 Multivariate Model 1 Grade >2 2.053 0.952–4.429 0.066 Lymph node status >0 3.309 1.601–6.841 0.001 PRþ 0.273 0.119–0.628 0.002 HER2 FISHþ 2.037 1.017–4.078 0.045 Activated pathway 2.612 0.796–8.570 0.113 Activated pathway 3.738 1.092–12.793 0.035 Model 2 AKT activation 1.293 0.669–2.499 0.444

multivariate analysis incorporating all clinicopathological and Discussion molecular parameters reported in Table 3, PI3K pathway Previous articles indicate that the PI3K pathway is altered in activation was not an independent predictor of breast cancer- a high proportion of breast cancer patients.3,17,39 Here, we Cancer Cell Biology related death in a model that included lymph node status, tu- report the first combined analysis of several pivotal molecules mor grade, PR and HER2 status (HR ¼ 2.61, 95% CI ¼ within the PI3K pathway in a single cohort of breast cancer 0.80–8.57, p ¼ 0.113). However, when tested in a multivari- patients in an attempt to derive an overall measure of path- ate model against pAKT the activated pathway was a better way activation and relate this to clinicopathological features prognostic marker than pAKT alone (Table 3, model 2). and outcome of the disease. We detected PIK3CA mutations Next, patients were distributed into 3 treatment categories, or copy number gain, PTEN loss or AKT activation in >70% i.e. endocrine therapy, chemotherapy and no adjuvant treat- of breast cancer patients, an association between an active ment, and the associations between an active PI3K pathway pathway and worse outcome and a high prevalence of altera- and breast cancer-specific death were assessed. In patients tions in the ‘‘basal-like’’ phenotype, which identifies the PI3K treated with endocrine therapy (n ¼ 93), an activated path- pathway as a potential therapeutic target in this subset of way was not associated with breast cancer-specific death (HR breast cancers. ¼ 3.79, 95% CI ¼ 0.50–28.79, p¼ 0.197). A similar result We detected a relatively low frequency of PIK3CA muta- was apparent in the chemotherapy-treated patients (n ¼ 75) tions, i.e. 7% compared to earlier estimates of 12–40%.3–16 (HR ¼ 1.97, 95% CI ¼ 0.58–6.76, p ¼ 0.279). However, in The use of sequencing is the standard method to detect the group of patients that received no adjuvant treatment (n mutations and although pyrosequencing has a higher analyti- ¼ 67), an activated pathway was associated with a markedly cal sensitivity which could aid in the mutation detection in inferior outcome to patients with no pathway aberrations DNA from tumor tissue contaminated with normal cells, all where no patient relapsed. There is no p value as there are previous breast cancer studies have used sequencing as their no events in the inactive pathway group (Fig. 4b). detection method. Some additional mutations could be pres- The same association between an active PI3K pathway ent in other exons not analyzed but exons 9 and 20 contain and breast cancer-specific death was then calculated as crude the highest frequency of PIK3CA mutations in cancer.14 Each 5-year survival rate. For endocrine treated patients the crude mutation was confirmed in an independent PCR reaction. 5-year survival rate was 69% (11/16), for chemotherapy Our Australian cohort is representative of previously pub- treated patients it was 89% (16/18) and for patients with no lished population-based breast cancer cohorts, with respect to adjuvant treatment it was 54% (6/11), of all breast cancer- clinicopathological parameters such as tumor grade, tumor specific death events recorded in each therapy group of size, ER, PR and HER2 status and outcome (McNeil C.M., patients. unpublished data). Therefore, neither the methodology nor

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC Lo´pez-Knowles et al. 1129 the cohort characteristics explain this lower proportion of association was between AKT activation, PTEN loss and PIK3CA mutations. However, the frequency of copy number pathway activation with increased cyclin E expression. This gain was 14%, higher than that detected in previous studies could be a rate limiting molecule in cell cycle progression (1–9%).12,16 PIK3CA mutation and amplification were almost due in part to a decrease of GSK3b phosphorylation, which mutually exclusive events and since they represent independ- also leads to stabilized cyclin E protein levels.43 ent mechanisms for elevating PI3K kinase activity, a higher The PI3K pathway was not predictive of response to en- frequency of copy number gain could perhaps explain the docrine therapy in our cohort of breast cancer patients; how- lower prevalence of PIK3CA mutations in this cohort com- ever, an association was identified when only PIK3CA status pared to those in other reports. Also, the poor quality of the and AKT activation were combined (data not shown). This DNA derived from paraffin embedded samples and the fail- latter relationship is in agreement with previous data from ure to enrich for tumor cells before sequencing could be a in vitro studies which show that activation of PI3K/AKT reason for the low frequency of mutations reported here. We pathway in breast cancer cells activates ERa and protects identified AKT activation in 24% of patients, slightly lower breast cancer cells from tamoxifen-induced apoptosis.44 Other levels that some publications although the range of published studies have shown an association between individual mem- data is highly variable5,17,19,21,22,24–28 and a frequency of bers of the PI3K pathway and tamoxifen resistance.6,26,45 De- PTEN loss of 28%, similar to the average of previous reports spite 1 report showing PTEN loss was associated with relapse on breast cancer patients.5,6,14,17–23 in patients treated with tamoxifen,6 we found no association, In our study, PTEN loss was exclusive of PIK3CA muta- possibly because in our cohort PTEN is commonly lost in tions, as described previously,14 suggesting that these altera- hormone receptor negative tumors and was thus less com- tions represent redundant mechanisms for activating the mon in endocrine-treated patients. Analysis of pAKT and PI3K pathway. Recent data have shown that the enhanced PIK3CA may be beneficial in identifying tamoxifen-resistant proliferation and survival resulting from PTEN loss are unaf- patients and this requires further testing in the context of a fected by PIK3CA knockdown and highlight an important randomized treatment trial where treatment and outcome role for the p110b isoform.40,41 This may reflect different ba- data are more robust. However, a poor outcome was also sal activities for p110a and b. However, it should be noted observed in patients not treated with adjuvant therapy, sug- that in a recent paper, PTEN loss and PIK3CA mutation gesting that aberrations in the PI3K pathway are associated were often concordant in breast cancer patients,3 indicating with poor outcome independent of treatment, i.e. they are additional work is required to clarify the functional contribu- prognostic rather than predictive markers. Cancer Cell Biology tion of these pathway aberrations as recently highlighted in Another important discovery from the study was the asso- the review of Yuan and Cantley.42 Furthermore, these workers3 ciation between an activated PI3K pathway and the ‘‘basal- identified that PTEN loss, but not PIK3CA mutation, was sig- like’’ phenotype of breast cancers. In HER2 cancers an acti- nificantly associated with AKT activation. We report similar vated pathway described by PIK3CA mutations or PTEN loss findings, where PIK3CA mutation and/or copy number gain has been previously associated with a poor prognosis after was not associated with AKT S473 phosphorylation, in con- trastuzumab therapy,46,47 confirming the importance of the trast to PTEN loss. These findings raise the possibility that al- PI3K pathway in this subgroup of cancers and pointing to ternative PI3K effectors to AKT, such as Rac, play an impor- the need for further characterization of PI3K pathway inhibi- tant role in PIK3CA mutated cancers. tors as complementary treatment strategies for trastuzumab- To examine the overall activation status of the pathway, resistant disease. In ‘‘basal-like’’ cancers the PI3K pathway is PIK3CA mutational status, PIK3CA copy number gain, PTEN emerging as an attractive candidate for targeted therapy. Pre- loss and activation of AKT as measured by pAKT levels were vious studies have shown that PTEN inactivation leads to the combined and by this definition >70% of cancers had an acti- development of mammary tumors that display pathological vated pathway. Only 25% of cancers had overlapping altera- and molecular features associated with the ‘‘basal-like’’ sub- tions, emphasizing the need of an integrative analysis of the type of breast cancer48,49 and 2 further studies have shown PI3K pathway. This activation was associated with a worse an association of cyclin E overexpression and p27Kip1 loss patient outcome although it was not an independent predictor with a basal phenotype.50,51 Furthermore, a recent publication of relapse or death from breast cancer implying that the has reported that PTEN alterations and an active PI3K path- reported associations with tumor grade (data not shown) and way are associated with ‘‘basal-like’’ breast cancers.52 These hormone receptor status were likely confounding. Pathway acti- data, together with the results presented here, point to the vation was identified in half of the hormone receptor-positive importance of further analyzing PI3K pathway aberrations in patients and was associated with an adverse outcome, although ‘‘basal-like’’ cancers and the effect of specific PI3K inhibitors this did not reach statistical significance. In hormone receptor- on this cancer subtype. negative disease pathway activation was identified in a higher In conclusion, over 70% of breast cancers in our study percentage of cases (85%) but was not associated with outcome. had an altered PI3K pathway which was associated with Confirmation of the activated pathway was measured by breast-cancer specific death, and we identified this pathway studying selected downstream cell cycle targets. The strongest as a potential therapeutic target in ‘‘basal-like’’ breast cancers.

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC 1130 PI3K pathway activation and breast cancer outcome

Further studies of the integrated PI3K pathway need to be Acknowledgements undertaken in larger cohorts of breast cancer patients, pref- The authors thank Ms. Alice Boulghourjian for assistance with processing erably in the context of randomized treatment trials, to con- the tissue, Ms. Anne Holliday for data management and Assoc. Prof. Adri- ﬁrm these ﬁndings. enne Morey for help with the HER2 FISH analysis.

References

1. Bader AG, Kang S, Zhao L, Vogt PK. 10. Li SY, Rong M, Grieu F, Iacopetta B. negative breast cancer patients aged <55 Oncogenic PI3K deregulates transcription PIK3CA mutations in breast cancer are years. Cell Oncol 2007;29:25–35. and translation. Nat Rev Cancer 2005;5: associated with poor outcome. Breast 20. Lee JS, Kim HS, Kim YB, Lee MC, Park 921–9. Cancer Res Treat 2006;96:91–5. CS, Min KW. Reduced PTEN expression is 2. Hennessy BT, Smith DL, Ram PT, Lu Y, 11. Buttitta F, Felicioni L, Barassi F, Martella associated with poor outcome and Mills GB. Exploiting the PI3K/AKT C, Paolizzi D, Fresu G, Salvatore S, angiogenesis in invasive ductal carcinoma pathway for cancer drug discovery. Nat Cuccurullo F, Mezzetti A, Campani D, of the breast. Appl Immunohistochem Mol Rev Drug Discov 2005;4:988–1004. Marchetti A. PIK3CA mutation and Morphol 2004;12:205–10. 3. Stemke-Hale K, Gonzalez-Angulo AM, histological type in breast carcinoma: high 21. Noh WC, Kim YH, Kim MS, Koh JS, Kim Lluch A, Neve RM, Kuo WL, Davies M, frequency of mutations in lobular HA, Moon NM, Paik NS. Activation of the Carey M, Hu Z, Guan Y, Sahin A, carcinoma. J Pathol 2006;208:350–5. mTOR signaling pathway in breast cancer Symmans WF, Pusztai L, et al. An 12. Wu G, Xing M, Mambo E, Huang X, Liu J, and its correlation with the integrative genomic and proteomic analysis Guo Z, Chatterjee A, Goldenberg D, Gollin clinicopathologic variables. Breast Cancer of PIK3CA, PTEN, and AKT mutations SM, Sukumar S, Trink B, Sidransky D. Res Treat 2008;110:477–83. in breast cancer. Cancer Res 2008;68: Somatic mutation and gain of copy 22. Panigrahi AR, Pinder SE, Chan SY, Paish 6084–91. number of PIK3CA in human breast EC, Robertson JF, Ellis IO. The role of 4. Liedtke C, Cardone L, Tordai A, Yan K, cancer. Breast Cancer Res 2005;7:R609–16. PTEN and its signalling pathways, Gomez HL, Figureoa LJ, Hubbard RE, 13. Levine DA, Bogomolniy F, Yee CJ, Lash A, including AKT, in breast cancer; an Valero V, Souchon EA, Symmans WF, Barakat RR, Borgen PI, Boyd J. Frequent assessment of relationships with other Hortobagyi GN, Bardelli A, et al. PIK3CA- mutation of the PIK3CA gene in ovarian prognostic factors and with outcome. activating mutations and chemotherapy and breast cancers. Clin Cancer Res 2005; J Pathol 2004;204:93–100. sensitivity in stage II-III breast cancer. 11:2875–8. 23. Tsutsui S, Inoue H, Yasuda K, Suzuki K, Breast Cancer Res 2008;10:R27. 14. Saal LH, Holm K, Maurer M, Memeo L, Higashi H, Era S, Mori M. Reduced 5. Lerma E, Catasus L, Gallardo A, Peiro G, Su T, Wang X, Yu JS, Malmstrom PO, expression of PTEN protein and its

Cancer Cell Biology Alonso C, Aranda I, Barnadas A, Prat J. Mansukhani M, Enoksson J, Hibshoosh H, prognostic implications in invasive ductal Exon 20 PIK3CA mutations decreases Borg A, et al. PIK3CA mutations correlate carcinoma of the breast. Oncology 2005;68: survival in aggressive (HER-2 positive) with hormone receptors, node metastasis, 398–404. breast carcinomas. Virchows Arch 2008; and ERBB2, and are mutually exclusive 24. Akcakanat A, Sahin A, Shaye AN, Velasco 453:133–9. with PTEN loss in human breast MA, Meric-Bernstam F. Comparison of 6. Perez-Tenorio G, Alkhori L, Olsson B, carcinoma. Cancer Res 2005;65:2554–9. Akt/mTOR signaling in primary breast Waltersson MA, Nordenskjold B, Rutqvist 15. Bachman KE, Argani P, Samuels Y, tumors and matched distant metastases. LE, Skoog L, Stal O. PIK3CA mutations Silliman N, Ptak J, Szabo S, Konishi H, Cancer 2008;112:2352–8. and PTEN loss correlate with similar Karakas B, Blair BG, Lin C, Peters BA, 25. Andre F, Nahta R, Conforti R, Boulet T, prognostic factors and are not mutually Velculescu VE, et al. The PIK3CA gene is Aziz M, Yuan LX, Meslin F, Spielmann M, exclusive in breast cancer. Clin Cancer Res mutated with high frequency in human Tomasic G, Pusztai L, Hortobagyi GN, 2007;13:3577–84. breast cancers. Cancer Biol Ther 2004;3: Michiels S, et al. Expression patterns and 7. Benvenuti S, Sartore-Bianchi A, Di 772–5. predictive value of phosphorylated AKT in Nicolantonio F, Zanon C, Moroni M, 16. Campbell IG, Russell SE, Choong DY, early-stage breast cancer. Ann Oncol 2008; Veronese S, Siena S, Bardelli A. Oncogenic Montgomery KG, Ciavarella ML, Hooi CS, 19:315–20. activation of the RAS/RAF signaling Cristiano BE, Pearson RB, Phillips WA. 26. Kirkegaard T, Witton CJ, McGlynn LM, pathway impairs the response of metastatic Mutation of the PIK3CA gene in ovarian Tovey SM, Dunne B, Lyon A, Bartlett JM. colorectal cancers to anti-epidermal growth and breast cancer. Cancer Res 2004;64: AKT activation predicts outcome in breast factor receptor antibody therapies. Cancer 7678–81. cancer patients treated with tamoxifen. Res 2007;67:2643–8. 17. Bose S, Chandran S, Mirocha JM, Bose N. J Pathol 2005;207:139–46. 8. Barbareschi M, Buttitta F, Felicioni L, The Akt pathway in human breast cancer: 27. Rojo F, Najera L, Lirola J, Jimenez J, Cotrupi S, Barassi F, Del Grammastro M, a tissue-array-based analysis. Mod Pathol Guzman M, Sabadell MD, Baselga J, Ferro A, Dalla Palma P, Galligioni E, 2006;19:238–45. Ramon y Cajal S. 4E-binding protein 1, a Marchetti A. Different prognostic roles of 18. Depowski PL, Rosenthal SI, Ross JS. Loss cell signaling hallmark in breast cancer mutations in the helical and kinase of expression of the PTEN gene protein that correlates with pathologic grade and domains of the PIK3CA gene in breast product is associated with poor outcome in prognosis. Clin Cancer Res 2007;13:81–9. carcinomas. Clin Cancer Res 2007;13: breast cancer. Mod Pathol 2001;14:672–6. 28. Tokunaga E, Kimura Y, Oki E, Ueda N, 6064–9. 19. Janssen EA, Soiland H, Skaland I, Futatsugi M, Mashino K, Yamamoto M, 9. Maruyama N, Miyoshi Y, Taguchi T, Gudlaugson E, Kjellevold KH, Nysted A, Ikebe M, Kakeji Y, Baba H, Maehara Y. Tamaki Y, Monden M, Noguchi S. Soreide JA, Baak JP. Comparing the Akt is frequently activated in HER2/neu- Clinicopathologic analysis of breast cancers prognostic value of PTEN and Akt positive breast cancers and associated with with PIK3CA mutations in Japanese expression with the Mitotic Activity Index poor prognosis among hormone-treated women. Clin Cancer Res 2007;13:408–14. in adjuvant chemotherapy-treated node- patients. Int J Cancer 2006;118:284–9.

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC Lo´pez-Knowles et al. 1131

29. Sorlie T, Perou CM, Tibshirani R, Aas T, 37. Millar EK, Dean JL, McNeil CM, O’Toole 46. Berns K, Horlings HM, Hennessy BT, Geisler S, Johnsen H, Hastie T, Eisen MB, SA, Henshall SM, Tran T, Lin J, Quong A, Madiredjo M, Hijmans EM, Beelen K, Linn van de Rijn M, Jeffrey SS, Thorsen T, Comstock CE, Witkiewicz A, Musgrove SC, Gonzalez-Angulo AM, Stemke-Hale K, Quist H, et al. Gene expression patterns of EA, Rui H, et al. Cyclin D1b protein Hauptmann M, Beijersbergen RL, Mills breast carcinomas distinguish tumor expression in breast cancer is independent GB, et al. A functional genetic approach subclasses with clinical implications. Proc of cyclin D1a and associated with poor identifies the PI3K pathway as a major Natl Acad Sci USA 2001;98:10869–74. disease outcome. Oncogene 2009;28: determinant of trastuzumab resistance 30. Sotiriou C, Neo SY, McShane LM, Korn 1812–20. in breast cancer. Cancer Cell 2007;12: EL, Long PM, Jazaeri A, Martiat P, Fox SB, 38. Millar EK, Graham PH, O’Toole SA, 395–402. Harris AL, Liu ET. Breast cancer McNeil CM, Browne L, Morey AL, 47. Ginestier C, Adelaide J, Goncalves A, classification and prognosis based on gene Eggleton S, Beretov J, Theocharous C, Repellini L, Sircoulomb F, Letessier A, expression profiles from a population- Capp A, Nasser E, Kearsley JH, et al. Finetti P, Geneix J, Charafe-Jauffret E, based study. Proc Natl Acad Sci USA 2003; Prediction of local recurrence, distant Bertucci F, Jacquemier J, Viens P, et al. 100:10393–8. metastases and death following breast- ERBB2 phosphorylation and trastuzumab 31. Cheang MC, Voduc D, Bajdik C, Leung S, conserving therapy in early-stage invasive sensitivity of breast cancer cell lines. McKinney S, Chia SK, Perou CM, Nielsen breast cancer using a five biomarker panel. Oncogene 2007;26:7163–9. TO. Basal-like breast cancer defined by five J Clin Oncol 2009;27:(in press). 48. Dourdin N, Schade B, Lesurf R, Hallett M, biomarkers has superior prognostic value 39. Tokunaga E, Kimura Y, Mashino K, Oki E, Munn RJ, Cardiff RD, Muller WJ. than triple-negative phenotype. Clin Cancer Kataoka A, Ohno S, Morita M, Kakeji Y, Phosphatase and tensin homologue deleted Res 2008;14:1368–76. Baba H, Maehara Y. Activation of PI3K/ on chromosome 10 deficiency accelerates 32. Millar EK, Anderson LR, McNeil CM, Akt signaling and hormone resistance in tumor induction in a mouse model of O’Toole SA, Pinese M, Crea P, Morey AL, breast cancer. Breast Cancer 2006;13: ErbB-2 mammary tumorigenesis. Cancer Biankin AV, Henshall SM, Musgrove EA, 137–44. Res 2008;68:2122–31. Sutherland RL, Butt AJ. BAG-1 predicts 40. Jia S, Liu Z, Zhang S, Liu P, Zhang L, Lee 49. Saal LH, Gruvberger-Saal SK, Persson C, patient outcome and tamoxifen SH, Zhang J, Signoretti S, Loda M, Roberts Lovgren K, Jumppanen M, Staaf J, Jonsson responsiveness in ER-positive invasive TM, Zhao JJ. Essential roles of PI(3)K- G, Pires MM, Maurer M, Holm K, Koujak ductal carcinoma of the breast. Br J Cancer p110beta in cell growth, metabolism and S, Subramaniyam S, et al. Recurrent gross 2009;100:123–33. tumorigenesis. Nature 2008;454:776–9. mutations of the PTEN tumor suppressor 33. de Cremoux P, Tran-Perennou C, 41. Wee S, Wiederschain D, Maira SM, Loo A, gene in breast cancers with deficient DSB Brockdorff BL, Boudou E, Brunner N, Miller C, deBeaumont R, Stegmeier F, Yao repair. Nat Genet 2008;40:102–7. Magdelenat H, Lykkesfeldt AE. Validation YM, Lengauer C. PTEN-deficient cancers 50. Foulkes WD, Brunet JS, Stefansson IM, of real-time RT-PCR for analysis of human depend on PIK3CB. Proc Natl Acad Sci Straume O, Chappuis PO, Begin LR, breast cancer cell lines resistant or sensitive USA 2008;105:13057–62. Hamel N, Goffin JR, Wong N, Trudel M,

to treatment with antiestrogens. Endocr 42. Yuan TL, Cantley LC. PI3K pathway Kapusta L, Porter P, et al. The prognostic Cancer Cell Biology Relat Cancer 2003;10:409–18. alterations in cancer: variations on a implication of the basal-like (cyclin E high/ 34. Kita D, Yonekawa Y, Weller M, Ohgaki H. theme. Oncogene 2008;27:5497–510. p27 low/p53þ/glomeruloid-microvascular- PIK3CA alterations in primary (de novo) 43. Liang J, Slingerland JM. Multiple roles of proliferationþ) phenotype of BRCA1- and secondary glioblastomas. Acta the PI3K/PKB (Akt) pathway in cell cycle related breast cancer. Cancer Res 2004;64: Neuropathol 2007;113:295–302. progression. Cell Cycle 2003;2:339–45. 830–5. 35. Jung V, Kindich R, Kamradt J, Jung M, 44. Campbell RA, Bhat-Nakshatri P, Patel NM, 51. Voduc D, Nielsen TO, Cheang MC, Muller M, Schulz WA, Engers R, Constantinidou D, Ali S, Nakshatri H. Foulkes WD. The combination of high Unteregger G, Stockle M, Zimmermann R, Phosphatidylinositol 3-kinase/AKT- cyclin E and Skp2 expression in breast Wullich B. Genomic and expression mediated activation of estrogen receptor cancer is associated with a poor prognosis analysis of the 3q25-q26 ampliﬁcation unit alpha: a new model for anti-estrogen and the basal phenotype. Hum Pathol reveals TLOC1/SEC62 as a probable target resistance. J Biol Chem 2001;276: 2008;39:1431–7. gene in prostate cancer. Mol Cancer Res 9817–24. 52. Marty B, Maire V, Gravier E, Rigaill G, 2006;4:169–76. 45. Tokunaga E, Kataoka A, Kimura Y, Oki E, Vincent-Salomon A, Kappler M, Lebigot I, 36. Carson JP, Kulik G, Weber MJ. Mashino K, Nishida K, Koga T, Morita M, Djelti F, Tourdes A, Gestraud P, Hupe P, Antiapoptotic signaling in LNCaP prostate Kakeji Y, Baba H, Ohno S, Maehara Y. Barillot E, et al. Frequent PTEN genomic cancer cells: a survival signaling pathway The association between Akt activation and alterations and activated independent of phosphatidylinositol 30- resistance to hormone therapy in phosphatidylinositol 3-kinase pathway in kinase and Akt/protein kinase B. Cancer metastatic breast cancer. Eur J Cancer basal-like breast cancer cells. Breast. Cancer Res 1999;59:1449–53. 2006;42:629–35. Res 2008;10:R101.

Int. J. Cancer: 126, 1121–1131 (2010) VC 2009 UICC Inositol polyphosphate 4-phosphatase II regulates PI3K/Akt signaling and is lost in human basal-like breast cancers

Clare G. Fedelea, Lisa M. Oomsa, Miriel Hoa, Jessica Vieusseuxa, Sandra A. O’Tooleb,c, Ewan K. Millarb,d, Elena Lopez- Knowlesb, Absorn Sriratanaa, Rajendra Gurunga, Laura Bagliettoe, Graham G. Gilese, Charles G. Baileyf, John E. J. Raskof,g, Benjamin J. Shieldsa, John T. Pricea, Philip W. Majerush,1, Robert L. Sutherlandb, Tony Tiganisa, Catriona A. McLeani, and Christina A. Mitchella,1 aDepartment of Biochemistry and Molecular Biology, Monash University, Clayton, Victoria 3800, Australia; bCancer Research Program, Garvan Institute of Medical Research, Darlinghurst, New South Wales 2010, Australia; cDepartment of Anatomical Pathology, Royal Prince Alfred Hospital, Camperdown, New South Wales 2050, Australia; dSouth East Area Laboratory Services, St. George Hospital, Kogarah, New South Wales 2217, Australia; eCancer Epidemiology Centre, The Cancer Council Victoria, Carlton, Victoria 3053, Australia; fCentenary Institute, Camperdown, New South Wales 2050, Australia; gCell and Molecular Therapies, Royal Prince Alfred Hospital, Camperdown, New South Wales 2050, Australia; hDepartment of Hematology, Washington University, St. Louis, MO 63110; and iDepartment of Anatomical Pathology, Alfred Hospital, Prahran, Victoria 3181, Australia

Contributed by Philip W. Majerus, October 27, 2010 (sent for review September 3, 2010)

Inositol polyphosphate 4-phosphatase-II (INPP4B) is a regulator of liferation, and orthotopic tumor formation in mice (9, 10). How- the phosphoinositide 3-kinase (PI3K) signaling pathway and is ever, to date analysis of human tumors has revealed little evidence implicated as a tumor suppressor in epithelial carcinomas. INPP4B that INPP4A functions as a tumor suppressor. In contrast, the loss of heterozygosity (LOH) is detected in some human breast related INPP4B may function as a tumor suppressor in epithelial cancers; however, the expression of INPP4B protein in breast cancer carcinomas. Short hairpin RNA (shRNA)-targeting of INPP4B subtypes and the normal breast is unknown. We report here leads to cell transformation in human mammary epithelial cells that INPP4B is expressed in nonproliferative estrogen receptor (HMECs) (11) and INPP4B knockdown in HMECs promotes Akt (ER)-positive cells in the normal breast, and in ER-positive, but not activation and anchorage-independent growth (12). Deletion of INPP4B negative, breast cancer cell lines. INPP4B knockdown in ER-positive the chromosome region occurs in some primary human INPP4B breast cancer cells increased Akt activation, cell proliferation, and breast cancers (13) and loss of heterozygosity (LOH) of is xenograft tumor growth. Conversely, reconstitution of INPP4B ex- frequently observed in BRCA1 mutant and hormone receptor- pression in ER-negative, INPP4B-null human breast cancer cells re- negative breast cancers (12). Furthermore, loss of INPP4B protein duced Akt activation and anchorage-independent growth. INPP4B expression in breast and ovarian cancer is associated with de- protein expression was frequently lost in primary human breast creased patient survival (12). To date, however, the expression of INPP4B protein in normal breast and in human breast cancer carcinomas, associated with high clinical grade and tumor size subtypes, relative to clinicopathologic variables, remains to be and loss of hormone receptors and was lost most commonly in determined. aggressive basal-like breast carcinomas. INPP4B protein loss was Invasive breast carcinomas can be categorized into distinct sub- also frequently observed in phosphatase and tensin homolog types, luminal A, luminal B, HER2 positive and basal-like, on the (PTEN)-null tumors. These studies provide evidence that INPP4B basis of expression of various clinocopathologic markers (14–16). functions as a tumor suppressor by negatively regulating normal Luminal A and luminal B breast cancer subtypes express estrogen and malignant mammary epithelial cell proliferation through reg- receptor (ER) and/or progesterone receptor (PgR), whereas ulation of the PI3K/Akt signaling pathway, and that loss of INPP4B HER2-positive and basal-like subtypes are hormone receptor neg- protein is a marker of aggressive basal-like breast carcinomas. ative and are in general more aggressive and confer a worse prognosis compared with luminal-type breast carcinomas (15, 16). The phosphatidylinositol 3,4-bisphosphate cellular mechanisms underlying these observations, however, are poorly defined and current research is focused on identifying new he phosphoinositide 3-kinase (PI3K) signaling pathway pro- biomarkers to improve breast cancer classification and treatment. Tmotes cell proliferation and survival. In response to extracel- In this study, we identify INPP4B as a previously undescribed lular stimuli, activation of PI3K results in the transient production marker of hormone receptor-positive breast cancers that func- of the phosphoinositides, PtdIns(3,4,5)P3 and PtdIns(3,4)P2,at tions to control both normal breast and malignant ER-positive the plasma membrane. Both phosphoinositides can bind and ac- breast cancer cell proliferation through regulation of PI3K sig- tivate multiple downstream effectors, most importantly the pro- naling. Loss of INPP4B protein expression occurs most frequently in aggressive hormone receptor-negative basal-like breast carci- tooncogene, Akt, and both PtdIns(3,4,5)P3 and PtdIns(3,4)P2 are necessary for full Akt activation in vivo (1–3). Deregulated PI3K/ nomas, associated with high tumor grade and size, and is fre- Akt activation promotes oncogenesis and has been described in quently associated with loss of the tumor suppressor PTEN. many human cancers (4). Recently, the PI3K signaling pathway These studies identify loss of INPP4B protein as a previously has been identified as a putative therapeutic target in a range of undescribed molecular marker for basal-like breast cancers and human malignancies and there are currently a number of phase I– II clinical trials in progress investigating the efficacy of PI3K pathway inhibitors in the treatment of human cancers (4). Author contributions: C.G.F. and C. A. Mitchell designed research; C.G.F., L.M.O., M.H., PtdIns(3,4,5)P is negatively regulated via dephosphorylation by J.V., and B.J.S. performed research; A.S., R.G., L.B., G.G.G., C.G.B., J.E.J.R., P.W.M., R.L.S., 3 and C. A. McLean contributed new reagents/analytic tools; C.G.F., S.A.O., E.K.M., E.L.-K., the 3-phosphatase and tumor suppressor, phosphatase and tensin L.B., G.G.G., J.T.P., R.L.S., and C. A. McLean analyzed data; and C.G.F., P.W.M., T.T., and homolog (PTEN), to form PtdIns(4,5)P2, or by 5-phosphatases, C. A. Mitchell wrote the paper. which generate PtdIns(3,4)P2 (1). In turn, PtdIns(3,4)P2 is hy- The authors declare no conflict of interest. drolyzed by two inositol polyphosphate 4-phosphatases (type I and 1To whom correspondence may be addressed. E-mail: [email protected] or type II) (INPP4A and B, respectively) (5, 6). INPP4A expression [email protected]. protects neurons from excitotoxic cell death (7, 8) and this enzyme This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. MEDICAL SCIENCES is also a negative regulator of Akt phosphorylation, cell pro- 1073/pnas.1015245107/-/DCSupplemental. www.pnas.org/cgi/doi/10.1073/pnas.1015245107 PNAS | December 21, 2010 | vol. 107 | no. 51 | 22231–22236 provide evidence for the cooperative promotion of oncogenesis In the normal breast, ER is expressed in only 15–30% of epi- through the PI3K/Akt signaling pathway. thelial cells and inversely correlates with cell proliferation (17, 18). Costaining normal breast sections with ER and INPP4B Abs Results revealed that INPP4B was expressed in ER-positive but not ER- Expression of INPP4B Protein in Normal Human Breast. We generated negative cells (Fig. 1B and Fig. S2B). Colocalization of INPP4B monoclonal and polyclonal antibodies (Abs) to purified 6xHis- and the proliferation marker, Ki-67 in breast lobules revealed that ~ – K tagged human INPP4B, which both detected endogenous INPP4B 5 10% of lobular cells were i-67 positive and of these the ma- > C–D in ER-positive MCF-7 breast cancer cells and purified recombi- jority ( 90%) did not show INPP4B staining (Fig. 1 ). Of the INPP4B-positive cells (20% of lobular cells), the majority did not nant His-INPP4B, but not recombinant INPP4A (Fig. S1). Using K E these Abs, we assessed INPP4B protein expression in normal hu- exhibit i-67 staining (Fig. 1 ). Therefore INPP4B may suppress man mammary tissue sections (Fig. 1 and Fig. S2). In epithelial ER-positive lobular cell proliferation in the normal breast. cells comprising the mammary ducts, INPP4B expression was re- Characterization of INPP4B Protein Expression in Human Breast stricted to luminal epithelial cells and appeared absent in sur- Cancer Cell Lines. We next assessed INPP4B protein expression rounding myoepithelial cells (Fig. 1A). In control studies, no in a panel of human breast cancer cell lines. INPP4B protein was immunoreactivity was detected using mouse IgG as a negative A expressed in ER-positive (MCF-7, T47D, and BT-474) but not control (Fig. S2 ). Interestingly, in the secretory lobular units, ER-negative (MDA MB 231, Hs578T, and BT-549) cell lines INPP4B was present in only ∼20% of cells, with many cells showing A A (Fig. 2 ). Quantitative real-time reverse-transcription PCR no staining (Fig. 1 ). The heterogeneous expression pattern of (qRT-PCR) analysis confirmed INPP4B mRNA levels were fi INPP4B in these structures may indicate a speci c role for INPP4B lower in all ER-negative relative to ER-positive cells, although in mammary cell differentiation and function. a low level of INPP4B mRNA was detected in ER-negative cell lines (Fig. 2B). Furthermore, metaanalysis of human breast cancer datasets made publicly available in Oncomine (19, 20) re- A INPP4B Zoom vealed a positive association between INPP4B mRNA and the hormone receptors, ER and PgR, in human breast cancers in 11 and 3 independent studies, respectively (Fig. S3 A and B). Control studies evaluated whether the related INPP4A is also Ducts expressed in breast cancer cell lines using previously characterized INPP4A Abs (21). Endogenous INPP4A was detected in mouse brain lysates but not in breast cancer cell lines (Fig. S3 C D INPP4B Zoom and ). Collectively this data indicates INPP4B is the only PtdIns (3,4)P2 4-phosphatase expressed in breast cancer cells and suggests a correlation between INPP4B and hormone receptor status in human breast cancer.

Lobules Involvement of INPP4B in ER-Positive Breast Cancer Cell Signaling, Proliferation, and Tumor Formation. To characterize INPP4B function in human breast cancer cells, shRNA-mediated INPP4B protein knockdown was undertaken in a variant of the ER- B Zoom positive MCF-7 human breast cancer cell line, MCF-7-luc-F5, which stably expresses luciferase enzyme allowing detection of cells in vivo. MCF-7-luc-F5 cells exhibited similar INPP4B protein levels to the parental line (Fig S4A). In cell populations

Ducts INPP4B Lobules individually expressing two distinct -directed shRNAs (INPP4B KD[1] and [2]) (Fig. S4B), INPP4B protein was de- INPP4B//ER INPP4B//ER INPP4B//ER ToPro3 ToPro3 ToPro3 creased by ∼70% and 60% compared with vector controls (Fig. 3A) INPP4B INPP4B / Ki-67 and similar decreases were observed in INPP4B mRNA (Fig. S4C). C D INPP4B + E INPP4B + / Ki-67 + * 100 20 The protooncogene Akt is activated by phosphorylation on ser- * ﬁ

lls ine and threonine residues. Signi cantly, in serum-starved cells,

80 e 15

* 60 10 *** 40 ER- ER- ER-negative lobular cells 5 A ER-positive B * 20 positive negative %Ki-67-positive

INPP4B//Ki-67 *** % total lobular c 4 ToPro3 0 0 231 B 74 M 49

F-7 A mRNA 7D -4 -5 3 Fig. 1. INPP4B is expressed in nonproliferative ER-positive cells in normal 4 T MC T B MD Hs578TBT human breast. (A) Normal human breast sections were immunostained using fi μ 2 an INPP4B-speci c monoclonal antibody (3D5). (Scale bars, 50 m.) (B)Im- INPP4B INPP4B munofluorescent images of normal breast sections costained for INPP4B 1 (green), ER (red), and the nuclear marker, ToPro3 iodide (blue) (see also Fig. ER S2). Coexpression of INPP4B and ER is indicated by arrows. Arrowheads show 0 Relative 4 9 cells negative for both INPP4B and ER. (Scale bars, 25 μm, 75 μm.) (C) Immu- -7 D T 4 47 47 -5 -tubulin CF T T- 578 T fl M s no uorescent costaining of normal human breast lobules for INPP4B (green), B H B AMB231 Ki-67 (red), and ToPro3 iodide (blue) (Fig. S2). Arrows show INPP4B-positive MD cells. Outlines and asterisks mark Ki-67–positive nuclei. (Scale bar, 25 μm.) More than 100 Ki-67–positive lobular cells/section were assessed for INPP4B Fig. 2. INPP4B protein and mRNA expression in a panel of human breast expression, and the mean number of cells that were INPP4B negative (closed cancer cells lines. (A) Human breast cancer cell line lysates were immuno- bars) or INPP4B positive (open bars) ± SEM from three normal breasts is shown blotted for INPP4B, ER, and β-tubulin protein expression, revealing INPP4B in D. More than 1,000 lobular cells/section were assessed for Ki-67 and INPP4B protein is expressed in ER-positive, but not ER-negative, cell lines. (B) qRT- expression and the mean number of INPP4B-positive lobular cells that were PCR analysis of INPP4B mRNA, relative to GAPDH, in ER-positive versus ER- Ki-67 negative (closed bars) or Ki-67 positive (open bars) ± SEM from two negative human breast cancer cell lines. The graph shows the mean INPP4B normal breasts is shown in E. ***P < 0.002. mRNA expression ± SEM from two independent experiments.

22232 | www.pnas.org/cgi/doi/10.1073/pnas.1015245107 Fedele et al. 2.5 ** A 1.0 B Control INPP4B INPP4B Control INPP4B INPP4B 2.0 e

vector KD(1) KD(2) 0.8 vector KD(1) KD(2) c * 1.5

INPP4B in 0.6

INPP4B eren e -Akt/Akt f f

*** 3 0.4 1.0 -tubulin pSer473-Akt di 47 prot *** 0.2 0.5

Akt Ser Fold Relative INPP4B 0 p 0 rol ntrol nt PP4B Co INPP4BKD(1) INPP4B Co IN INPP4B vector KD(2) vector KD(1) KD(2)

2.0 ** Control vector Control vector C INPP4B KD(1) D INPP4B KD(1) Control vector INPP4B KD(1) *** 1.6 * 1.5 *** EGF (mins): 0 5 10 20 0 5 10 20

Akt 1.2 nce itive e

INPP4B 1.0 s 0.8 pSer473-Akt 0.5 rdu-po 0.4 B Fold difference

Akt Fold differ pSer473-Akt/ 0 0 24 48 0 5 10 20 Serum- EGF (mins): starved (hrs):

Control vector E F G 80 INPP4B KD(1) Control vector *** 2.0 Control INPP4B *** vector KD(1) r *** 3 60 1.5 ** umou t nd

1.0 o 40 ge c a r INPP4B KD(1) /se ns 0.5 volume (mm20 ) Ave Fold difference colony number oto

0 Ph l 0 ro nt 2.0 5.5 Co INPP4B vector KD(1) Weeks post-injection

Fig. 3. Decreased INPP4B expression enhances Akt phosphorylation and tumorigenic potential. (A) INPP4B protein knockdown in two MCF-7-luc-F5 cell populations expressing unique INPP4B-specific shRNAs (INPP4B KD[1] and [2]) was confirmed by immunoblotting. The mean INPP4B protein levels relative to vector control ± SEM from three independent experiments is shown. (B and C) Immunoblot analysis of INPP4B, pSer473-Akt, and total Akt in cells serum-starved for 24 h (B) or stimulated with EGF (C), demonstrating enhanced pSer473-Akt in INPP4B knockdown cells. The mean fold change in pSer473-Akt/total Akt relative to control ± SEM from six (B) or three (C) independent experiments is shown. (D) INPP4B knockdown and control cells were serum starved for 24 and 48 h and assessed for BrdU incorporation as a marker of proliferation. The mean fold increase in BrdU-positive cells relative to control ± SEM from three independent experiments is shown. (E) INPP4B knockdown or control cells were grown in soft agar. The number of colonies/well was determined and the mean fold increase in colony number relative to control ± SEM from two independent experiments performed in triplicate is shown. (F and G) MCF-7-luc-F5 cells expressing control vector or INPP4B-specific shRNAS were injected into the mammary fat pads of Balbc nu/nu mice in the presence of estrogen and tumor growth was analyzed by bioluminescence and caliper measurement. Bioluminescent imaging of xenograft tumors in vivo (F) and mean tumor volumes at 2 and 5.5 wk postinjection (G) indicate INPP4B knockdown cells form larger tumors in vivo. The mean tumor volumes ± SEM of five animals/group is shown. *P < 0.05, **P < 0.01, ***P < 0.005. basal phosphorylated Akt (pSer473-Akt) was increased by 1.5- to Effect of INPP4B Reconstitution on ER-Negative Breast Cancer Cell 2-fold in INPP4B knockdown cells, conditions under which Akt Growth and Akt Phosphorylation. We next evaluated the effects of activation is normally minimal (Fig. 3B). Furthermore INPP4B INPP4B protein reconstitution in INPP4B-null, ER-negative knockdown in MCF-7-luc-F5 cells resulted in increased pSer473- MDA MB 231 breast cancer cells. MDA MB 231 cells are highly Akt/Akt following 20 min EGF stimulation, relative to vector proliferative and exhibit anchorage-independent growth in soft C agar. Expression of GFP-INPP4B reduced colony formation in soft controls (Fig. 3 ). A Akt activation promotes cell proliferation. INPP4B knock- agar by 35% (Fig. 4 ) and suppressed EGF stimulated phosphorylation of Akt at Ser473 (pSer473-Akt) and Thr308 (pThr308- down cells exhibited enhanced (1.4-fold) cell proliferation in Akt) (Fig. 4 B–D), relative to GFP controls. Therefore in ER- response to serum-starvation (24–48 h), conditions under which D negative breast cancer cell lines, which have lost endogenous cells normally become quiescent (Fig. 3 ). The ability of cells to INPP4B, reconstitution of this enzyme is sufficient to reduce EGF- form colonies in soft agar and tumors in nude mice is a feature of stimulated Akt activation and anchorage-independent cell growth. transformed cells. INPP4B knockdown in MCF-7-luc-F5 cells enhanced anchorage-independent cell growth (Fig. 3E). To as- Expression of INPP4B Protein in Primary Human Breast Carcinoma sess INPP4B regulation of ER-positive breast cancer growth in Subtypes. We next assessed the expression of INPP4B protein in vivo, INPP4B knockdown or control MCF-7-luc-F5 cells were clinical human breast carcinomas. In tissue lysates from 22 pri- injected into the abdominal mammary fat pads of athymic nude mary human breast cancers, INPP4B protein was detected in all mice and palpable tumors were measured over 6 wk. Seven of of the ER-positive tumors, but few of the ER-negative tumors A the eight mice injected with INPP4B knockdown cells developed (Fig. 5 ). INPP4B protein expression was then evaluated in 107 palpable tumors that grew over the course of the experiment, primary breast carcinoma sections, collected as part of the Mel- compared with four out of eight mice injected with vector-con- bourne Collaborative Cohort Study (MCCS) (22, 23), by IHC. Notably, INPP4B protein was not detected in 22/107 (21%) of all trol cells. Furthermore, INPP4B knockdown tumors exhibited breast cancers analyzed, but adjacent normal epithelium exhibi- a 3.2-fold increase in tumor size over time, compared with a 1.9- ted INPP4B immunoreactivity in every case (Fig. 5B). In INPP4B-

F G MEDICAL SCIENCES fold increase in vector controls (Fig. 3 and ). positive tumors, protein expression was exclusively cytoplasmic

Fedele et al. PNAS | December 21, 2010 | vol. 107 | no. 51 | 22233 GFP A GFP-INPP4B A 250 * * * * * * * * GFP GFP-INPP4B 200 *** INPP4B 150 actin 100 50 ER:

Colonies/well 0 B 0 1 2 3 B GFP GFP-INPP4B EGF (min): 0 5 15 30 0 5 15 30 INPP4B pSer473-Akt

pThr308-Akt 100 C 88 MCCS dataset 80 Akt 80 SVCOC dataset

nloss(%) 60 2.0 C D 38 GFP GFP rtion tumours 40 1.5 e 1.5 GFP-INPP4B 25 22

GFP-INPP4B /Akt e t 20 Propo c *** *** * *** 11 8 10 erenc t/Akt INPP4B protei -Ak 1.0 f 1.0 k f ren 8 0 A e 0 + l -

3 sal- ld di

73 Ba HER2 Lumina Luminal 4 0.5 0.5 B A ***

r like *** Fo pThr e Fold diff pS 0 0 Fig. 5. INPP4B protein expression in primary human breast cancers. EGF (min): 0515 30 EGF (min): 0515 30 (A) Immunoblot analysis of INPP4B protein expression in a panel of ER- positive and ER-negative primary human breast cancer lysates. ER- Fig. 4. INPP4B protein reconstitution decreases Akt phosphorylation and negative tumors were predominantly INPP4B-negative (asterisks). (B) cell proliferation. (A) INPP4B-null MDA MB 231 cells expressing GFP alone or Representative IHC staining of INPP4B in primary human breast carcino- GFP-tagged INPP4B were suspended in 0.3% agar and colonies allowed to mas, scored as: 0, no expression; 1, low; 2, moderate; and 3, strong. Inset grow for 3 wk. The number of colonies/well was then determined and the image in tumor scored 0 for INPP4B shows INPP4B expression in adjacent ± mean number of colonies/well SEM from two independent experiments normal tissue. (Scale bar, 200 μm.) (C) The frequency of INPP4B protein performed in triplicate is shown. (B) Immunoblot analysis of INPP4B, loss of expression in breast cancer subtypes from two independent human – pSer473-Akt, pThr308-Akt, and total Akt in GFP or GFP-INPP4B expressing datasets [MCCS (black bars) and SVCOC (open bars)] was determined by cells in response to EGF. The mean fold difference in pSer473-Akt (C) and IHC. Loss of INPP4B was positively correlated with the basal-like subtype ± pThr308-Akt (D) normalized to total Akt relative to controls SEM from and negatively correlated with the luminal A subtype in both datasets. < three independent experiments is shown. *P = 0.05, ***P 0.005. ***P < 0.0001.

– and ranged from low, moderate, to high (staining score 1 3) (Fig. and size >20 mm (P = 0.047) and expression of proliferation 5B). Significantly, INPP4B protein deficiency correlated with loss markers, Ki-67 (P = 0.003) and cyclin E1 (P = 0.0001), and was of both ER (P < 0.0001) and PgR expression (P < 0.0001) and inversely associated with the cell cycle inhibitor, p27 (P = 0.0055) positively correlated with the basal marker, CK5/6 (P = 0.0014), fi (Table 1). Interestingly, there was no correlation between INPP4B and approached signi cance for epidermal growth factor receptor P (EGFR) expression (P = 0.06) (Table 1). Loss of INPP4B was loss and axillary lymph node involvement ( = 0.89), indicating also associated with high tumor grade (P = 0.018). There was no that INPP4B may be a regulator of tumor cell proliferation and significant correlation between HER2 amplification and INPP4B growth, but not metastasis. expression (P = 0.17). Tumors were then classified into breast − cancer subtypes as: luminal A (ER+ and/or PgR+, HER2 ), lu- Association of INPP4B Protein Loss with Altered Parameters of the − minal B (ER+ and/or PgR+, HER2+), HER2+ (ER/PgR , PI3K Pathway. The incidence of INPP4B loss of expression in − HER2+), or basal-like (ER/PgR/HER2 ,CK5/6+,and/orEGFR+), breast cancers, as reported here, is comparable to those reported as previously described (24–26). Strikingly, loss of INPP4B expres- for PIK3CA mutations (7–40%) (26, 31–36) or loss of the tumor sion was observed most frequently in the hormone receptor-nega- suppressor PTEN (13–37%) (26, 37–39). To date, most studies tive subtypes, basal-like (7/8, i.e., 88%) and HER2+ (3/8, i.e., 38%) indicate that alterations to PIK3CA and PTEN in human cancers and less frequently in the hormone receptor-positive subtypes, lu- occur in a mutually exclusive manner (39–42). We evaluated minal B (2/9, i.e., 0.22%) and luminal A (6/72, i.e., 8%) (Fig. 5C). whether loss of INPP4B was associated with alterations in other These results were independently validated by assessing components of the PI3K pathway in 267 tumors from the SVCOC INPP4B immunostaining in a more extensive cohort of 267 in- dataset that had been previously assessed for PIK3CA amplifi- vasive ductal carcinomas drawn from the St. Vincent’s Campus – cation/mutation, Akt hyperactivation, or PTEN loss (26). Nota- Outcome Cohort (SVCOC) (25 27). INPP4B protein expression bly, loss of INPP4B protein did not correlate with PIK3CA was lost in 71/267 (27%) of all tumors analyzed from this dataset, mutation/amplification (P = 0.54), indicating that these events are associated with loss of both ER and PgR expression (P < 0.0001) mutually exclusive in breast cancer (Table 1). In contrast, INPP4B and positively correlated with CK5/6 (P = 0.0003), but not HER2 amplification (P > 0.999) (Table 1). Again, INPP4B expression protein expression was lost in 49% (36/73) of PTEN-null tumors, compared with 14% (25/185) of PTEN-expressing tumors, re- was lost most frequently in basal-like carcinomas (24/30, i.e., 80%, fi P < 0.0001) and infrequently in luminal A carcinomas (16/158, i.e., vealing a signi cant positive correlation between INPP4B de- fi P < 10%, P < 0.0001) (Fig. 5C). The luminal A subtype is the least ciency and PTEN loss ( 0.0001). Whereas high pAkt alone did aggressive tumor type, associated with the best patient prognosis not significantly correlate with INPP4B loss (P = 0.493), PTEN (15, 16, 28, 29), whereas basal-like breast carcinomas are in gen- null tumors that showed high pAkt frequently exhibited INPP4B eral aggressive, exhibiting high histologic tumor grade and mitotic deficiency (P < 0.0002) (Table 1). Therefore, concomitant loss of index (30). Consistent with this phenotype, loss of INPP4B ex- both PTEN and INPP4B may contribute to Akt hyperactivation pression positively correlated with high tumor grade (P < 0.0001) and the development of aggressive forms of breast cancer.

22234 | www.pnas.org/cgi/doi/10.1073/pnas.1015245107 Fedele et al. Table 1. Association analysis of INPP4B protein loss with clinicopathologic, cell cycle, and PI3K pathway variables in human breast cancers (MCCS and SVCOC datasets) MCCS SVCOC

Total INPP4B loss Total INPP4B loss Nature of Variable (n = 107) (n = 22) P value (n = 267) (n =71) P value association

Clinical Histological grade 3 34 12 (35%) 0.018 123 48 (39%) <0.0001 Positive Size (>20mm) - - - 101 36 (33%) 0.047 Positive LN status (positive) - - - 119 31 (26%) 0.89 Nil Receptors/Markers ER status (positive) 68 5 (7%) <0.0001 178 19 (11%) <0.0001 Negative PgR status (positive) 56 3 (5%) <0.0001 150 11 (7%) <0.0001 Negative HER2 ampliﬁcation 26 8 (31%) 0.17 47 11 (23%) >0.999 Nil CK5/6 (positive) 6 5 (83%) 0.0014 62 28 (45%) 0.0003 Positive EGFR (positive) 7 3 (43%) 0.06 - - - Positive* Cell Cycle Ki-67 (high) - - - 123 40 (33%) 0.0003 Positive Cyclin E1 (high) - - - 66 29 (44%) 0.0001 Positive p27 (high) - - - 113 18 (16%) 0.0055 Negative PI3K pathway PIK3CA amp/mut - - - 26 8 (31%) 0.54 Nil PTEN loss - - - 73 36 (49%) <0.0001 Positive pAkt > mean - - - 62 17 (23%) 0.49 Nil PTEN loss/pAkt >mean - - - 17 11 (65%) 0.0002 Positive

LN, axillary lymph node involvement; ER, estrogen receptor; PgR, progesterone receptor; CK, cytokeratin; EGFR, epidermal growth factor receptor 1; amp, ampliﬁed; mut, mutated; pAkt, phospho-Akt. *Approaching signiﬁcance.

Discussion negative breast cancers (12); however, this report did not assess This study identifies INPP4B as a previously undescribed regu- expression of basal markers to discriminate between triple- lator of ER-positive normal and breast cancer cell proliferation negative and basal-like cancers or investigate INPP4B protein and tumor growth. In the normal breast INPP4B is expressed expression in breast cancer subtypes, as reported here. The predominantly in ER-positive cells that are nonproliferative. discrepancy observed between the frequency of INPP4B LOH INPP4B protein knockdown in ER-positive breast cancer cells previously reported (55%) (12) and loss of INPP4B protein ex- enhances baseline and EGF-dependent Akt phosphorylation, pression presented here (84%) in basal-like breast cancers sug- cell proliferation, anchorage independent cell growth, and xe- gests alternative mechanisms for INPP4B protein loss in addition nograft tumor formation. Conversely, in INPP4B-null, ER- to gene deletion. Interestingly, INPP4B transcription has recently negative MDA MB 231 cells, reintroduction of this PtdIns(3,4)P2 been shown to be influenced by gene hypermethylation (47). Our 4-phosphatase inhibits EGF-dependent Akt signaling and sup- studies suggest that suppression of INPP4B protein expression presses cell growth in soft agar. specifically in basal-like breast cancers may contribute to the ag- Our study is unique in investigating and reporting a correlation gressive nature of this breast cancer subtype. between INPP4B protein expression and clinicopathologic param- Like INPP4B, expression of the tumor suppressor and 3- eters in human breast cancer. Gene array studies have predicted phosphatase, PTEN, is also frequently lost in basal-like carci- a correlation between INPP4B expression and hormone receptor nomas (26, 39). Notably, we report that concomitant loss of status in human breast cancer (43, 44), and the studies presented both INPP4B and PTEN proteins is commonly observed in here support this association at the protein level. The nature of the human primary breast cancers. Concurrent PTEN and INPP4B relationship between ER and INPP4B, however, remains elusive. absence may result in the accumulation of their respective Although PI3K can negatively regulate ER protein levels by de- phosphoinositide substrates, PtdIns(3,4,5)P3 and PtdIns(3,4)P2, creasing ER protein stability (45), we noted that INPP4B protein which together promote maximal Akt activation. We report that knockdown in ER-positive MCF-7-luc-F5 cells, or reconstitution in in human breast cancers loss of INPP4B is frequently associated ER/INPP4B-negative MDA MB 231 cells, did not affect ER protein with PTEN-null tumors exhibiting high pAkt, although loss of expression (Fig. S5 A and B). Similarly we have demonstrated es- INPP4B alone did not show such an association with high trogen stimulation of MCF-7 cells, or treatment with ER antagonists does not affect the protein levels of INPP4B (Fig. S5C), despite pAkt. This is of interest as we, and others (12), have noted only INPP4B a modest increase in Akt activation following INPP4B knock- gene array studies indicating transcription may be en- – hanced by estrogen stimulation (46). Therefore the nature of the down in response to cell stimulation (1.2 1.4-fold). Therefore relationship between INPP4B and ER is likely to be complex, INPP4B may not regulate the amplitude of Akt activation fol- possibly involving signaling derived from stromal tissue (47). lowing cell stimulation but rather the duration of the signal. We report that INPP4B protein expression is lost in 84% of Consistent with this contention we noted elevated pAkt/Akt in human basal-like breast carcinomas, which are generally highly serum-starved INPP4B-depleted cells, conditions under which Akt aggressive with poor clinical outcome and are frequently asso- activation is normally minimal. Although recent studies report that ciated with breast cancer 1 (BRCA1) gene mutations (30). knockdown of INPP4B and PTEN together in immortalized nor- Basal-like carcinomas are defined as exhibiting basal cytoker- mal human mammary epithelial cell lines leads to cellular senes- atin and EGFR expression and low or absent expression of cence (12), it is likely that in human cancers loss of INPP4B and hormone receptors (30, 48). Our studies indicate that absence PTEN occurs during tumor progression in combination with ad- of INPP4B protein may be an additional marker of the basal- ditional mutations. Indeed, knockdown of the tumor suppressor like carcinoma signature profile. LOH of INPP4B has recently TP53 in INPP4B/PTEN-deficient HMECs rescues the senescence been reported to occur in 55% of human ER/PgR/HER2 triple- phenotype (12). MEDICAL SCIENCES

Fedele et al. PNAS | December 21, 2010 | vol. 107 | no. 51 | 22235 In summary, the studies reported here identify INPP4B as a pu- University School of Biomedical Sciences (SOBSB/B/2008/14) and were tative tumor suppressor that functions to regulate PI3K signaling established and analyzed as described in SI Materials and Methods. and ER-positive mammary cell proliferation. In addition, we provide evidence that loss of INPP4B is a unique marker of human Patients, Breast Carcinoma Samples, and INPP4B Expression Analysis. Ethics basal-like carcinomas. This study reports the concurrent loss of two approval was obtained from the Standing Committee on Ethics in Research phosphoinositide phosphatases in human breast cancer and pro- Involving Humans, Monash University (CF08/1142–2008000564) for all vides evidence for the cooperative promotion of oncogenesis human tissues and cognate clinicopathological data used in this study. through alterations to multiple components of the PI3K signaling Details of the human breast cancer cohorts and analysis of INPP4B are pathway. There are currently no directed therapies for the treat- described in SI Materials and Methods. ment of human basal-like cancers and tumors exhibiting loss of Further details of materials and methods can be found in SI Materials INPP4B protein with or without concurrent PTEN loss may rep- and Methods. resent ideal candidates for treatment with PI3K pathway inhibitors. ACKNOWLEDGMENTS. This work was supported by grants from the National Materials and Methods Health and Medical Research Council (491029 and 504711). We also Tumorigenicity in Mice. For xenograft tumor growth assays, all procedures acknowledge the Cancer Institute New South Wales, the R. T. Hall Trust, involving mice were approved by the animal ethics committee at the Monash and the Petre Foundation.

1. Ma K, Cheung SM, Marshall AJ, Duronio V (2008) PI(3,4,5)P3 and PI(3,4)P2 levels 24. Cheang MC, et al. (2008) Basal-like breast cancer defined by five biomarkers has superior correlate with PKB/akt phosphorylation at Thr308 and Ser473, respectively; PI(3,4)P2 prognostic value than triple-negative phenotype. Clin Cancer Res 14:1368–1376. levels determine PKB activity. Cell Signal 20:684–694. 25. Millar EK, et al. (2009) Prediction of local recurrence, distant metastases, and death 2. Franke TF, Kaplan DR, Cantley LC, Toker A (1997) Direct regulation of the Akt proto- after breast-conserving therapy in early-stage invasive breast cancer using a five- oncogene product by phosphatidylinositol-3,4-bisphosphate. Science 275:665–668. biomarker panel. J Clin Oncol 27:4701–4708. 3. Scheid MP, et al. (2002) Phosphatidylinositol (3,4,5)P3 is essential but not sufficient for 26. López-Knowles E, et al. (2010) PI3K pathway activation in breast cancer is associated protein kinase B (PKB) activation; phosphatidylinositol (3,4)P2 is required for PKB with the basal-like phenotype and cancer-specific mortality. Int J Cancer 126:1121–1131. phosphorylation at Ser-473: Studies using cells from SH2-containing inositol-5- 27. Millar EK, et al. (2009) BAG-1 predicts patient outcome and tamoxifen responsiveness – phosphatase knockout mice. J Biol Chem 277:9027 9035. in ER-positive invasive ductal carcinoma of the breast. Br J Cancer 100:123–133. 4. Liu P, Cheng H, Roberts TM, Zhao JJ (2009) Targeting the phosphoinositide 3-kinase 28. Oh DS, et al. (2006) Estrogen-regulated genes predict survival in hormone receptor- pathway in cancer. Nat Rev Drug Discov 8:627–644. positive breast cancers. J Clin Oncol 24:1656–1664. 5. Norris FA, Majerus PW (1994) Hydrolysis of phosphatidylinositol 3,4-bisphosphate by 29. Sotiriou C, et al. (2003) Breast cancer classification and prognosis based on gene expression inositol polyphosphate 4-phosphatase isolated by affinity elution chromatography. J profiles from a population-based study. Proc Natl Acad Sci USA 100:10393–10398. Biol Chem 269:8716–8720. 30. Rakha EA, Reis-Filho JS, Ellis IO (2008) Basal-like breast cancer: A critical review. J Clin 6. Norris FA, Atkins RC, Majerus PW (1997) The cDNA cloning and characterization of Oncol 26:2568–2581. inositol polyphosphate 4-phosphatase type II. Evidence for conserved alternative 31. Bachman KE, et al. (2004) The PIK3CA gene is mutated with high frequency in human splicing in the 4-phosphatase family. J Biol Chem 272:23859–23864. breast cancers. Cancer Biol Ther 3:772–775. 7. Sasaki J, et al. (2010) The PtdIns(3,4)P(2) phosphatase INPP4A is a suppressor of 32. Barbareschi M, et al. (2007) Different prognostic roles of mutations in the helical and excitotoxic neuronal death. Nature 465:497–501. – 8. Sachs AJ, David SA, Haider NB, Nystuen AM (2009) Patterned neuroprotection in the kinase domains of the PIK3CA gene in breast carcinomas. Clin Cancer Res 13:6064 6069. Inpp4a(wbl) mutant mouse cerebellum correlates with the expression of Eaat4. PLoS 33. Li SY, Rong M, Grieu F, Iacopetta B (2006) PIK3CA mutations in breast cancer are – ONE 4:e8270. associated with poor outcome. Breast Cancer Res Treat 96:91 95. 9. Vyas P, Norris FA, Joseph R, Majerus PW, Orkin SH (2000) Inositol polyphosphate 4- 34. Maruyama N, et al. (2007) Clinicopathologic analysis of breast cancers with PIK3CA – phosphatase type I regulates cell growth downstream of transcription factor GATA-1. mutations in Japanese women. Clin Cancer Res 13:408 414. Proc Natl Acad Sci USA 97:13696–13701. 35. Wu G, et al. (2005) Somatic mutation and gain of copy number of PIK3CA in human 10. Ivetac I, et al. (2009) Regulation of PI(3)K/Akt signalling and cellular transformation by breast cancer. Breast Cancer Res 7:R609–R616. inositol polyphosphate 4-phosphatase-1. EMBO Rep 10:487–493. 36. Campbell IG, et al. (2004) Mutation of the PIK3CA gene in ovarian and breast cancer. 11. Westbrook TF, et al. (2005) A genetic screen for candidate tumor suppressors Cancer Res 64:7678–7681. identifies REST. Cell 121:837–848. 37. Bose S, Chandran S, Mirocha JM, Bose N (2006) The Akt pathway in human breast 12. Gewinner C, et al. (2009) Evidence that inositol polyphosphate 4-phosphatase type II cancer: A tissue-array-based analysis. Mod Pathol 19:238–245. is a tumor suppressor that inhibits PI3K signaling. Cancer Cell 16:115–125. 38. Pérez-Tenorio G, et al. (2007) PIK3CA mutations and PTEN loss correlate with similar 13. Naylor TL, et al. (2005) High resolution genomic analysis of sporadic breast cancer using prognostic factors and are not mutually exclusive in breast cancer. Clin Cancer Res 13: array-based comparative genomic hybridization. Breast Cancer Res 7:R1186–R1198. 3577–3584. 14. Bertucci F, et al. (2005) Gene expression profiling identifies molecular subtypes of 39. Saal LH, et al. (2005) PIK3CA mutations correlate with hormone receptors, node inflammatory breast cancer. Cancer Res 65:2170–2178. metastasis, and ERBB2, and are mutually exclusive with PTEN loss in human breast 15. Sørlie T, et al. (2001) Gene expression patterns of breast carcinomas distinguish tumor carcinoma. Cancer Res 65:2554–2559. – subclasses with clinical implications. Proc Natl Acad Sci USA 98:10869 10874. 40. Stemke-Hale K, et al. (2008) An integrative genomic and proteomic analysis of 16. Sorlie T, et al. (2003) Repeated observation of breast tumor subtypes in independent PIK3CA, PTEN, and AKT mutations in breast cancer. Cancer Res 68:6084–6091. – gene expression data sets. Proc Natl Acad Sci USA 100:8418 8423. 41. Abubaker J, et al. (2007) PIK3CA mutations are mutually exclusive with PTEN loss in 17. Russo J, Ao X, Grill C, Russo IH (1999) Pattern of distribution of cells positive for diffuse large B-cell lymphoma. Leukemia 21:2368–2370. α estrogen receptor and progesterone receptor in relation to proliferating cells in the 42. Wang Y, et al. (2007) High prevalence and mutual exclusivity of genetic alterations in mammary gland. Breast Cancer Res Treat 53:217–227. the phosphatidylinositol-3-kinase/akt pathway in thyroid tumors. J Clin Endocrinol 18. Capuco AV, Ellis S, Wood DL, Akers RM, Garrett W (2002) Postnatal mammary ductal Metab 92:2387–2390. growth: three-dimensional imaging of cell proliferation, effects of estrogen 43. West M, et al. (2001) Predicting the clinical status of human breast cancer by using treatment, and expression of steroid receptors in prepubertal calves. Tissue Cell 34: gene expression profiles. Proc Natl Acad Sci USA 98:11462–11467. 143–154. 44. Yang F, et al. (2006) Laser microdissection and microarray analysis of breast tumors 19. Rhodes DR, et al. (2007) Oncomine 3.0: Genes, pathways, and networks in a collection reveal ER-alpha related genes and pathways. Oncogene 25:1413–1419. of 18,000 cancer gene expression profiles. Neoplasia 9:166–180. 45. Bhat-Nakshatri P, et al. (2004) Tumour necrosis factor and PI3-kinase control 20. Rhodes DR, et al. (2004) ONCOMINE: A cancer microarray database and integrated data-mining platform. Neoplasia 6:1–6. oestrogen receptor alpha protein level and its transrepression function. Br J Cancer – 21. Munday AD, et al. (1999) The inositol polyphosphate 4-phosphatase forms a complex 90:853 859. with phosphatidylinositol 3-kinase in human platelet cytosol. Proc Natl Acad Sci USA 46. Harvell DME, Richer JK, Allred DC, Sartorius CA, Horwitz KB (2006) Estradiol regulates 96:3640–3645. different genes in human breast tumor xenografts compared with the identical cells 22. Giles GG, English DR (2002) The Melbourne Collaborative Cohort Study. IARC Sci Publ in culture. Endocrinology 147:700–713. 156:69–70. 47. Lin H-JL, et al. (2008) Breast cancer-associated fibroblasts confer AKT1-mediated 23. Blows FM, et al. (2010) Subtyping of breast cancer by immunohistochemistry to epigenetic silencing of Cystatin M in epithelial cells. Cancer Res 68:10257–10266. investigate a relationship between subtype and short and long term survival: A 48. Nielsen TO, et al. (2004) Immunohistochemical and clinical characterization of the collaborative analysis of data for 10,159 cases from 12 studies. PLoS Med 7:e1000279. basal-like subtype of invasive breast carcinoma. Clin Cancer Res 10:5367–5374.

22236 | www.pnas.org/cgi/doi/10.1073/pnas.1015245107 Fedele et al. DOI:10.1158/0008-5472.CAN-10-3738

Cancer Tumor and Stem Cell Biology Research

Hedgehog Overexpression Is Associated with Stromal Interactions and Predicts for Poor Outcome in Breast Cancer

Sandra A. O'Toole1,3,5,6, Dorothy A. Machalek1, Robert F. Shearer1, Ewan K.A. Millar1,8,9,10, Radhika Nair1, Peter Schofield2,3, Duncan McLeod10,11, Caroline L. Cooper5, Catriona M. McNeil1,7, Andrea McFarland1, Akira Nguyen1, Christopher J. Ormandy1,3, Min Ru Qiu4, Brian Rabinovich13, Luciano G. Martelotto12, Duc Vu12, Gregory E. Hannigan12, Elizabeth A. Musgrove1,3, Daniel Christ2,3, Robert L. Sutherland1,3, D. Neil Watkins12, and Alexander Swarbrick1,3

Abstract Hedgehog (Hh) signaling plays an important role in several malignancies but its clinical significance in breast cancer is unclear. In a cohort of 279 patients with invasive ductal carcinoma of the breast, expression of Hh ligand was significantly associated with increased risk of metastasis, breast cancer-specific death, and a basal- like phenotype. A paracrine signature, encompassing high epithelial Hh ligand and high stromal Gli1, was an independent predictor for overall survival in multivariate analysis. In 2 independent histological progression series (n ¼ 301), Hh expression increased with atypia. Hh ligand overexpression in a mouse model of basal breast cancer increased growth, induced a poorly differentiated phenotype, accelerated metastasis, and reduced survival. A stromal requirement for these effects was supported by the lack of similar Hh-mediated changes in vitro, and by stromal-specific expression of Hh target genes in vivo. Furthermore, inhibition of Hh ligand with a monoclonal antibody (5E1) inhibited tumor growth and metastasis. These data suggest that epithelial–stromal Hh signaling, driven by ligand expression in carcinoma cells, promotes breast cancer growth and metastasis. Blockade of Hh signaling to peritumoral stromal cells may represent a novel therapeutic approach in some basal- like breast cancers. Cancer Res; 71(11); 4002–14. 2011 AACR.

Introduction receptor positive disease (1). Recent data show that therapeutic targeting of the HER2 receptor is also making an The decrease in deaths from breast cancer over the last 2 impact in breast cancer mortality (2). However, in women decades reflects improvements in early detection, and the with basal-like breast cancer, an aggressive subtype lacking success of targeted therapies such as tamoxifen in hormone expression of the estrogen receptor (ER), progesterone receptor (PR), and HER2, the prognosis remains poor (3). At present, there is a shortage of targetable signaling pathways in basal- Authors' Affiliations: 1Cancer Research and 2Immunology Programs, like breast cancer. Garvan Institute of Medical Research; 3St Vincent's Clinical School, Several studies have suggested a role for hedgehog (Hh) Faculty of Medicine, University of New South Wales; and 4Department – of Anatomical Pathology, SydPath, St Vincent's Hospital, Darlinghurst; signaling in breast cancer (4 6). This pathway is a highly 5Department of Tissue Pathology and Diagnostic Oncology, Royal Prince conserved developmental signaling system essential for Alfred Hospital; 6Central Clinical School, University of Sydney; and epithelial to mesenchymal signaling in development (7). 7Department of Medical Oncology, Sydney Cancer Centre, Royal Prince Alfred Hospital, Camperdown; 8Department of Anatomical Pathology, Deregulation of Hh signaling has been implicated in the South Eastern Area Laboratory Service, St George Hospital, Kogarah; pathogenesis of carcinoma, in part through the promotion 9 School of Medical Sciences, University of New South Wales, Kensington; of epithelial–stromal interactions (8–10). 10School of Medicine, University of Western Sydney, Campbelltown; and 11Department of Tissue Pathology, Institute of Clinical Pathology and The Hh ligands, Sonic (Shh), Indian (Ihh), and Desert Medical Research, Westmead, NSW; and 12Monash Institute of Medical (Dhh) hedgehogs bind to and inactivate the transmembrane Research, Monash University, Clayton, Victoria, Australia; and 13Experi- mental Diagnostic Imaging, UT MD Anderson Cancer Center, Houston, receptor Patched (Ptch). Ptch is a constitutive inhibitor of Texas Smoothened (Smo), a transmembrane protein required for Note: Supplementary data for this article are available at Cancer Research all Hh signaling. In its inactive state, Smo permits the Online (http://cancerres.aacrjournals.org/). formation of a multiprotein complex that constitutively Corresponding Author: Alexander Swarbrick, Cancer Research Program, processes the Gli proteins (Gli1–3) to short, transcriptionally Garvan Institute of Medical Research, 384 Victoria St, Darlinghurst, NSW repressive forms. Activation of Smo decouples this complex 2010, Australia. Phone: 61-2-9295-8366; Fax: 61-2-9295-8321; E-mail: [email protected] or D. Neil Watkins, Monash Institute of Medical from microtubule domains and leads to stabilization of full Research, Monash University, 27-31 Wright St, Clayton, Victoria 3186, length, transactivating Gli proteins that initiate transcrip- Australia. E-mail: [email protected] tion of Hh target genes, including Ptch, Gli1,andHedgehog doi: 10.1158/0008-5472.CAN-10-3738 interacting protein (Hhip; ref. 7). Expression of these tran- 2011 American Association for Cancer Research. scripts can be used as an indirect measure of canonical

Hedgehog in Breast Cancer

Hh signaling (10), although Gli1 can be driven by non- for Hh pathway antibodies was carried out on a DAKO Hh–dependent mechanisms (11). autostainer using the following antigen retrieval protocols Early reports suggesting Hh signaling may contribute to after dewaxing and rehydration: Hh ligand H-160 sc-9024, breast carcinogenesis came through the studies of Lewis and Santa Cruz, 1:80, 20 minutes in a boiling waterbath in Dako colleagues in mice with heterozygous disruption of Ptch1 retrieval solution S2367; Ptch1 ab27529, Abcam, 1:50 antigen which showed marked abnormalities in mammary ductal retrieval for 20 minutes boiling waterbath in Dako retrieval structures resulting in hyperplasias and dysplasias similar to solution S2367 and Gli1, sc-20687, Santa Cruz, 1:100, antigen human breast lesions (12). More recently members of the same retrieval 30 seconds at maximum temperature and pressure in group (13) studying mice with constitutive activation of human a DAKO pressure cooker in DAKO solution s1699. Full details Smo under control of the mouse mammary tumor virus of the protocols for immunohistochemistry are shown in (MMTV) promoter, found that mammary ductal cells showed Supplementary Table S1. These antibodies were rigorously increased proliferation, altered differentiation, and developed validated with robust controls as shown in Supplementary ductal dysplasias. This group had also previously shown that Figures S1–3. mammary ducts of mice with loss of Gli2 had a range of For each marker, 2 specialist breast pathologists (S.A. histological alterations similar to micropapillary ductal hyper- O’Toole and either D. McLeod, E.A.K. Millar, M.R. Qiu, or plasia in the human breast (14). Tissue recombination studies C.L. Cooper) independently calculated the percentage staining showed that the role of Gli2 is particular to the stroma, as these and the predominant intensity on a predetermined scale of 0: duct changes were not seen when epithelium with Gli2 deleted no staining to 3: strong staining in both the epithelial cells of was transplanted into wild-type mouse stroma, supporting a the lesions and the adjacent stroma for each Hh pathway critical compartmentalization of Hh signaling in development protein (and for Gli1 both nuclear and cytoplasmic expressions and proliferative mammary ductal lesions. were scored). Each core had an H score that was generated by Although several studies suggest that deregulation of Hh multiplying these scores (17–19, 21). signaling might be important in breast cancer (4, 6, 12–16), the Intrinsic breast cancer subtypes were assessed immuno- clinical and functional significance of these findings and their histochemically using criteria similar to those recently potential therapeutic impact is unclear. Therefore, we carried described by Cheang and colleagues (22) but using FISH to out a detailed immunohistochemical study using rigorously determine HER-2 status (19). Details for antibodies, immuno- validated antibodies against Hh ligand (the initiating signal), histochemistry, in situ hybridization, and scoring for these Ptch1 (the pathway receptor), and Gli1 (a widely accepted markers, have been previously reported in the invasive ductal readout of active canonical signaling) in large, well-character- carcinoma cohort (17–19, 21). ized cohorts of invasive ductal carcinoma and premalignant and proliferative breast lesions. We also employed a mouse Immunofluorescence model of mammary carcinogenesis to show that Hh ligand Immunofluorescence (IF) was carried out on 7-mm-thick overexpression promotes stromal-dependent tumor growth. mouse tumor samples fixed in 4% (w/v) paraformaldehyde, Finally, we show that use of a Hh-ligand blocking antibody blocked for 90 minutes with 2% (v/v) horse serum, and incu- reduces tumor growth and pulmonary metastases, supporting bated with primary antibody (Hhip rabbit H-280 Santa Cruz sc- a functional role of the Hedgehog pathway in breast cancer 25465, 1:200) and vimentin (chicken, 50-264, Pro Sci, 1:200). development and progression. Secondary antibodies Alexa555 Cy3 (red) fluorophore conjugated anti-rabbit 1:300 (A-31572, Molecular Probes) or Materials and Methods Alexa488 Cy2 (green) fluorophore conjugated anti-guinea pig (A-11073, Molecular Probes) 1:500 were applied to the sections Patients and incubated for 1 hour in the dark at room temperature. Three cohorts of patients were used as follows: first, the Coverslips were mounted onto slides using Vectashield mount- Garvan/St Vincent's Hospital outcome series (17–19) of 292 ing medium with 6-diamidino-2-phenylindole (DAPI) counter- patients with invasive ductal carcinoma, and 2 independent stain (H-1500, Vector Laboratories). Slides were imaged cohorts of a histological progression series; the Garvan Insti- immediately using a Zeiss Axioplan upright fluorescence tute/St Vincent's Hospital progression series (GSVH-PS) con- microscope with Zeiss Axiocam MRm digital camera. Digital sisting of a subset of 79 patients for whom tissue was still images were captured using Axiovision V 4.8.1.0 software. available, from a larger series originally (20) and the Garvan Institute/Royal Prince Alfred Hospital with 222 patients diag- Animals nosed with invasive ductal carcinoma (IDC) or ductal carci- All experiments involving mice were carried out under the noma in situ (DCIS) described in an earlier report (21). Ethics supervision and in accordance with the regulations of the approval was granted for the use of pathology specimens and Garvan/St Vincent's Animal Ethics Committee (Approval 08/ cognate clinicopathologic data (HREC SVH H94/080, SVH 41). C3(1)/SV40 large T transgenic mice inbred on FVBN H00036, and RPAH X05-0115). background was obtained from Dr. Jeff Green (NIH). Inbred mice were obtained from the Animal Resource Centre, Wes- Antibodies and immunohistochemistry tern Australia (BALB/c, Rag / ). To ensure the specificity of Hh pathway antibodies, rigorous Mammary gland transplantation. For the 4T1 model antibody validation was carried out. Immunohistochemistry the fourth fat pad of 28-day-week old female BALB/c mice was

www.aacrjournals.org Cancer Res; 71(11) June 1, 2011 4003 Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

O'Toole et al.

surgically visualized and 1 106 cells (10 mL) injected. For the ACA TT, Gli1 forward GGA CCC ACT CCA ATG AGA AG, Gli1 M6-SHH/Vector transplants, 0.75 106 cells (10 mL) were reverse CAT GCA CTG TCT TCA CGT GTT, Hhip forward GTG surgically injected via direct visualization into the fourth fat TTC GGA GAT CGC AAT G, Hhip reverse TTT TCT TGC CAT pad of 21-day-old Rag / mice. TGC TTG GT, PECAM1 forward AGC CAG TAG CAT CAT GGT In vivo imaging. Animals were imaged twice weekly. CA, PECAM1 reverse AGC AGG ACA GGT CCA ACA AC, Briefly mice were first injected intraperitoneally with 200 b-actin forward GGA TGC AGA AGG AGA TTA CTG C, b-actin mL of 30% D-luciferin (diluted from 10 mg/mL in PBS, 6160- reverse CCA CCG ATC CAC ACA GAG TA. 80-1 GoldBio) and imaged under anesthesia in a Xenogen IVIS Further primer details are shown in Supplementary 200 biophotonic imager. Luminescence is expressed as Table S2. The programs used are shown in Supplementary photons/sec/ROI (region of interest) minus background lumi- Table S3. Data were analyzed using the comparative CT DD nescence for a similarly sized region. method ( CT).

Cell lines Expression constructs M6 mouse mammary carcinoma cells derived from the C3 The full-length SHH cDNA (2,716 bp) was subcloned from (1)/SV40 Tag mouse model (23; kindly provided by Jeff Green, pRK5-SHH (24) into pMSCV-puro retroviral expression vector NIH) were grown in DMEM high glucose medium (11995, (634401, Clontech). The pLV4301G-enhanced luciferase third Invitrogen), supplemented with 10% (v/v) FBS and penicillin/ generation lentiviral construct was used for in vivo imaging streptomycin. 4T1 cells (2939, ATCC) were maintained in and expresses green fluorescent protein (GFP) and luciferase RPMI 1640, 10% FBS, 10 mmol/L HEPES, 100 mmol/L sodium was used for in vivo imaging (25). To generate pLV4301G viral pyruvate and 0.25% glucose. Phoenix Ecotropic packaging cell stocks, 293T lentiviral packaging cells were transfected as line (3444, ATCC) and Lenti-X293T packaging cells (632180, described for the retrovirus production below. 4T1 and M6 Clontech) were maintained in DMEM supplemented with 100 cells were infected with the construct and sorted for GFP mmol/L sodium pyruvate, 200 mmol/L L-glutamine and non- before expanding for subsequent experiments. For retrovirus essential amino acids. Cell lines were authenticated by short production, phoenix-eco cells (26) were seeded at 1.8 106 tandem repeat (STR), single-nucleotide polymorphism (SNP), cells/well in 60-mm dishes and transfected with 2 mgof and fingerprint analyses and passaged for less than 6 months. plasmid (SHH or vector alone) and 1 mg of pCL-Eco (12371, The "Light II" NIH/3T3 cell line stably transfected with a Gli- Addgene) using Effectene transfection reagent (301425, Qia- responsive firefly luciferase reporter and Renilla-luciferase gen) as directed. Media was changed 24 hours after transfec- expression vector (CRL-2795, ATCC) was grown in DMEM tion and a further 24 hours were allowed for virus production. supplemented with 10% FCS containing 0.1 mg/mL Zeocin, Forty-eight hours posttransfection diluted viral supernatant and 0.4 mg/mL Geneticin. (1:10 in M6 media containing 8 mg/mL polybrene) was used to Cell viability assay (MTS assay) was carried out using the infect M6 pLV4301G sorted cells seeded at 0.8 105 cells/well m CellTiter 96 AQueous Cell Proliferation Assay (G5421, Promega) in 6-well plates. Transduced cells were selected with 5 g/mL according to the manufacturers recommendation. of puromycin 48 hours after infection. Selection pressure was maintained throughout subsequent passages. Flow cytometry Tumors were processed into single cell suspensions before Neutralizing antibody staining and fluorescence-activated cell sorting (FACS). Before Hybridomas for hedgehog (clone 5E1) or control antigalac- staining, samples were blocked and then stained with a tosidase (clone 40-1a) IgG1 monoclonal antibodies were R-Phycoerythrin (R-PE)–Conjugated Rat anti-Mouse CD24 obtained from the Developmental Studies Hybridoma Bank monoclonal antibody (553262, BD Pharmingen) or the PE (UIOWA, developed under the auspices of the NICHD and Rat IgG2b negative control (122-116-072 Jackson ImmunoR- maintained by The University of Iowa). Ultrapure antibodies esearch) for 30 minutes on ice. Samples were sorted on the (0.23EU endotoxin/mg of protein) were generated and purified FACS Vantage SETM Cell Sorter (with FACSDiVa Option; BD by the Recombinant Products Facility at the University of New Biosciences) running BD FACSDiVa software version 5.0.3 (BD South Wales. Mice were treated twice a week by intraperito- Biosciences). neal (i.p.) injection with either 0.5 mg 5E1 or IgG1 control antibody. Quantitative real-time PCR cDNA was synthesized using an avian myeloblastosis virus Statistical analyses (AMV) reverse transcriptase system (A3500 Promega) as per Statistical evaluation was carried out using Statview 5.0 protocol. Quantitative real-time PCRs (qRT-PCR) were carried Software (Abacus Systems). A value of P < 0.05 was accepted as out using the Roche Universal Probe Library System on a statistically significant. Baseline characteristics of the cohort Roche LightCycler480, 384 well platform, with primers as were defined using simple frequency distributions. Cutoffs for follows: SHH forward CAA ATT ACA ACC CCG ACA TC, expression of the biological markers examined were deter- SHH reverse GCA TTT AAC TTG TCT TTG CAC CT, Ptch1 mined using an optimal cut-point technique (27). forward GGC CTG GCA GAG GAC TTA C, Ptch1 reverse GGA Multivariate Cox proportional hazards analyses used AGC ACC TTT TGA GTG GA, Ptch2 forward GTC CAC CTA "backwards" modeling to generate models predictive of GTG CTC CCA AC, Ptch2 reverse CTC AGC TCC TGA GCC outcome. Spearman–Rank correlation was used to explore

4004 Cancer Res; 71(11) June 1, 2011 Cancer Research Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

Hedgehog in Breast Cancer

the relationship between 2 continuous variables in the 16% of cases (44/282) and was independently prognostic in human Hh protein expression studies. Simple unpaired t multivariate analysis for overall survival (HR1.7, P ¼ 0.04, 95% tests were used to compare 2 groups in the mouse studies. 1.0–2.8; Table 2; Fig. 1B) and approached significance for breast cancer–specific death (P ¼ 0.09) in the resolved model. Results Aberrant expression of Hh ligand is an early event in Hh ligand is a marker of poor prognosis in invasive mammary carcinogenesis ductal carcinoma In view of our findings of the prognostic significance of Hh We examined our cohort of 292 patients with invasive ligand expression in invasive ductal carcinoma, we were ductal carcinoma of the breast using rigorously validated interested to identify at what stage of breast cancer develop- antibodies to Hh ligand, Ptch, and Gli1. Examples of the ment Hh ligand expression was first upregulated. We exam- expression of Hh ligand and Gli1 are shown in Figure 1A. ined expression of Hh ligand using immunohistochemistry in Cytoplasmic expression of Hh ligand could be assessed in 279 2 cohorts of preinvasive and malignant breast lesions; a test patients and was present in virtually all cases (275 of 279, 98%) cohort and a validation cohort both comprising a histological in both the epithelium and the stroma (279 of 279, 100%). progression series of increasing architectural atypia and However, only 34% (96 of 279) of cases showed high intensity malignancy including in situ and invasive ductal carcinoma. Hh staining in carcinoma cells (Fig. 1A). Kaplan–Meier survi- We observed a striking, progressive increase in the expres- val analysis showed that those patients had a poorer outcome sion of epithelial Hh ligand in hyperplasia, atypia, and in situ in terms of breast cancer metastasis (P ¼ 0.0004, HR 1.95, 95% malignancy (DCIS) in both the test and validation cohorts CI 1.2–3.1) and breast cancer–specific death (P ¼ 0.002, HR 2.3, (Fig. 1C and D). This was seen in the earliest lesions in the 95% CI 1.3–4.0, Fig. 1B). High Hh ligand expression was also progression series, with greater expression of Hh ligand in associated with grade 3, larger (>20 mm), and more prolif- histologically normal ducts adjacent to invasive carcinoma erative (high Ki67) tumors, PR negative status (all P < 0.05, compared with normal ducts from reduction mammoplasty Table 1), and was strongly correlated with the basal-like patients in the test cohort (P < 0.05), with further significant subtype of breast cancer (P ¼ 0.001). There was no association increases in ductal hyperplasia and then DCIS. with any other subtypes. High Hh ligand expression was not There were also incremental increases in Hh ligand expres- independently prognostic in multivariate analysis. There was a sion from low- to intermediate- to high-grade DCIS in the strong association between Hh ligand expression and prolif- validation cohort. In the test cohort, there was no significant erative cell-cycle proteins such as cyclin A (Spearman–Rank change in the expression of Hh between grade 3 DCIS and correlation Rho ¼ 0.344, P < 0.0001), cyclin B1 (Rho ¼ 0.354, invasive carcinoma. P < 0.0001) as well as cyclin E1 (Rho ¼ 0.405, P < 0.0001). There In view of our findings in human tissue, we explored was no significant association of stromal Hh expression with similarities with a mouse model of basal-like mammary survival. carcinoma, in which we could study functional effects of Ptch1 was assessable in 197 cases and showed epithelial Hh pathway manipulation. We selected the C3(1)/Tag model cytoplasmic expression in 93% of cases (184/197) and as it has a well-defined premalignant sequence and gives rise stromal cytoplasmic expression in 89%. Univariate analysis to basal-like tumors (28). In keeping with our human data, we for breast cancer–specific death showed no prognostic sig- found that lesions of the C3 (1)/Tag model also showed nificance for cytoplasmic epithelial Ptch1 expression (P ¼ 0.8) increased expression of Hh ligand in early proliferative lesions and only a borderline association for stromal Ptch1 (P ¼ 0.05). (hyperplasia) such that Hh ligand is significantly higher in Both nuclear and cytoplasmic Gli1 were assessed in the proliferative and malignant lesions than normal mammary cohort but only cytoplasmic Gli1 showed any association with ductal epithelium (Supplementary Fig. S4), supporting the use survival. High cytoplasmic stromal Gli1 expression (example of this model in subsequent studies of Hh function in mam- shown in Fig. 1A) was determined by optimal cut-point mary carcinoma. determination as more than 20% of stromal cells expressing Gli1. These Gli1 positive stromal cells had the appearance of Functional effects of Hh overexpression fibroblasts based on morphological assessment by a specialist In view of our data in human breast cancer showing that pathologist. There were 83 (31%) stromal Gli1 positive cases, high level Hh ligand expression is a marker of poor prognosis, which were associated with breast cancer–specific death (P ¼ we investigated underlying mechanisms that might contribute 0.004, HR 2.4, 95% CI 1.3–4.1) on univariate analysis (Fig. 1B). to this finding. We elected to use the M6 cell line derived from There was no prognostic significance to nuclear Gli1 expres- the C3 (1)/Tag model as a particularly suitable model for in sion in either the stroma (P ¼ 0.8) or the epithelium (P ¼ 0.4) vivo and in vitro studies. We first determined the effects of Hh by Kaplan– Meier survival analysis. overexpression in a transplant model in which M6 cells stably In view of the known paracrine mechanism of Hh pathway expressing Hh ligand (M6-HH), or controls, were transplanted signaling in development in many organs and in a number of into the fat pad of immunodeficient Rag / mice. M6 cells malignancies, we developed a "paracrine" signature defined as stably expressing Hh formed significantly larger tumors with a cases with both high Hh expression in the epithelium (Hh 4-fold increase in mean tumor volume (P ¼ 0.0006) and a 3- intensity score 3) and high Gli1 in the stroma (>20% of stromal fold increase in weight compared with controls at endpoint cells expressing Gli1). This "paracrine" signature identified (P ¼ 0.005, Fig. 2A). This was confirmed by live imaging studies

www.aacrjournals.org Cancer Res; 71(11) June 1, 2011 4005 Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

O'Toole et al.

A Strong Hh ligand expression Strong stromal Gli1 expression

B i Breast cancer–specific death ii Breast cancer metastasis 1 1 .8 .8 .6 .6 .4 High Hh ligand .4 High Hh ligand n = 96 n = 96 .2 HR 2.3 .2 HR 1.95 P = 0.002 P = 0.0004 Cumulative survival 0 Cumulative survival 0 0 25 50 75 100 125150 175 200 0 25 50 75 100 125 150 175 200 Time (months) Time (months) iiiBreast cancer–specific death iv Breast cancer–specific death 1 1 .8 .8 .6 .6 High stromal Gli1 Paracrine signature .4 .4 n = 83 n = 44 .2 HR 2.4 .2 HR 2.7

Cumulative survival P = 0.004 P = 0.001 0 Cumulative survival 0 0 25 50 75 100 125 150 175 200 0 25 50 75 100 125 150 175 200 Time (months) Time (months) C D Test cohort

250 * 200

150 ** 100 Normal duct UDH * 50 Hh ligand H score 0 Benign Cancer UDH DCIS DCIS DCIS IDC duct assoc G1 G2 G3 N duct lesion Validation cohort 250 ** ** LG DCIS IG DCIS 200 ** 150 ** 100 50 Hh ligand H score 0 Benign Cancer UDH DCIS DCIS DCIS IDC HG DCIS IDC duct assoc G1 G2 G3 N duct lesion

Figure 1. High Hh ligand is associated with a poor prognosis in invasive ductal carcinoma and is an early event in breast cancer progression. A, high Hh ligand expression (400) and high stromal Gli1 expression (1,000). B, Kaplan–Meier curves for (i) breast cancer recurrence and (ii) breast cancer–specific death for high Hh, (iii) breast cancer–specific death for high stromal Gli1, and (iv) breast cancer–specific death for the "paracrine" signature. C, increasing intensity of Hh ligand expression in lesions with greater cytological and architectural atypia (all images 200). UDH, usual ductal hyperplasia; LG, low grade; IG, intermediate grade; HG, high grade; IDC, invasive ductal carcinoma. D, quantitation of Hh staining intensity in 2 independent cohorts of lesions.

4006 Cancer Res; 71(11) June 1, 2011 Cancer Research Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

Hedgehog in Breast Cancer

Table 1. Clinicopathologic associations of Hh overexpression

Clinicopathologic parameter Total Hh high Hh low Pa N ¼ 279 N ¼ 96 (34%) N ¼ 183 (66%) n (%) n (%)

Median, y (range): 55 (24–87) 0.03 Age 55 y 127 53 (42) 74 (58) Age <55 y 152 43 (28) 109 (72) Median size, mm (range): 21 (0.9–80) 0.01 Size 20 mm 115 50 (43) 65 (57) Size <20 mm 164 46 (28) 118 (72) Grade <0.0001 3 126 61 (48) 65 (52) 1 and 2 153 35 (23) 118 (77) Lymph node status 0.01 Positive 121 52 (43) 69 (57) Negative 155 44 (28) 111 (72) Estrogen receptor 0.07 Positive 187 58 (31) 129 (69) Negative 88 38 (43) 50 (57) Progesterone receptor 0.03 Positive 158 46 (29) 112 (71) Negative 118 50 (42) 68 (58) HER2 status 0.99 Positive 51 17 (33) 34 (67) Negative 217 75 (35) 142 (65) Ki67 > median 0.0001 High 121 58 (47) 63 (53) Low 127 31 (24) 96 (75)

ac2 analysis P value.

using luciferase expression which showed greater signal in the Table 2. Univariate and multivariate analysis of Hh overexpressing tumors compared with controls (Fig. 2B). clinicopathologic variables for overall survival Ninety days after transplantation, no mice carrying Hh-overexpressing tumors (N ¼ 8) survived, compared with 88% Variable HR (95% CI) P survival for mice bearing control tumors (N ¼ 8; P < A, Univariate analysis 0.0004). There were also significant differences in the local Histological grade 3 3.8 (2.1–7.0) <0.0001 invasion of the 2 groups. Control tumors were adherent to the Size > 20 mm 2.4 (1.6–3.7) <0.0001 overlying skin but showed well defined edges. In contrast, the Lymph nodes > 0 3.3 (1.9–6.0) <0.0001 Hh overexpressing tumors were locally very aggressive, invad- ER positive 0.4 (0.3–0.6) <0.0001 ing through the abdominal musculature, and penetrating the PR positive 0.3 (0.2–0.6) <0.0001 peritoneal cavity (Fig. 2C). HER2 positive 2.3 (1.8–5.8) <0.0001 Histological analysis showed that 4 of 5 (80%) of Hh over- Epithelial Hh high 2.3 (1.3–4.0) 0.002 expressing tumors showed invasion of lymphatic spaces by Stromal Gli1 high 2.4 (1.3–4.1) 0.004 tumor cells, confirmed by LYVE-1 immunohistochemistry Paracrine Hh signature 2.7 (1.5–4.8) 0.001 (Fig. 2D), compared with only 1 of 5 in the control group. B, Multivariate analysis, We further explored this model to investigate the pheno- resolved model type of M6-HH and M6-control tumors matched for size Lymph nodes > 0 2.2 (1.3–3.6) 0.001 before they became large and necrotic. Although there was PR positive 0.3 (0.2–0.7) 0.0002 no difference in the incidence of peritumoral lymphatic inva- HER2 positive 1.9 (1.1–3.2) 0.02 sion (data not shown) when tumors were of an equivalent size, Paracrine Hh signature 1.7 (1.01–2.8) 0.04 Hh overexpressing tumors were denser and were histologically less well differentiated, with no glandular structures observed

www.aacrjournals.org Cancer Res; 71(11) June 1, 2011 4007 Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

O'Toole et al.

A i ii

500 1.2 ** P = 0.0045 400 Vector 1.0

) HH 3 0.8 300 0.6 200 Weight (g)

Volume (mm 0.4 100 0.2

0 0.0 22 26 27 29 30 34 37 41 43 HH Days Vector Image Min = –2398 Image Max = 46182 Min = –1215 B i ii Max = 24898

30000 15000

20000 10000

10000 5000

Counts Counts Color Bar Color Bar Min = 1245 Min = 2309 Max = 17803 Max = 35873 C i ii

Figure 2. Hh overexpression promotes tumor growth, local aggressiveness, and lymphatic invasion and is associated with poor survival. A, M6-HH allografts show increased tumor volume and a 3-fold increase in tumor weight at endpoint. B, luciferase activity of M6 tumors in vivo. M6-HH tumors (left) show larger signals than controls (right). C, control tumors show limited local invasion and are confined to the overlying skin (left, tumor highlighted by black circle), whereas M6-HH tumors (right) show marked local invasion, growing through the abdominal wall and penetrating the peritoneal cavity. D, M6-HH tumors show increased peritumoral lymphatic invasion (LYVE-1 immunohistochemistry, 100).

4008 Cancer Res; 71(11) June 1, 2011 Cancer Research Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

Hedgehog in Breast Cancer

A M6-control tumor M6-HH tumor

Figure 3. Hh overexpression results in more poorly differentiated and more proliferative tumors. A, control M6-HH tumor M6-control tumor tumors of equivalent volume to B M6-HH tumors show better differentiation with well developed glandular structures and more stroma. In contrast M6-HH tumors (right) show no glandular differentiation and very little stroma, consisting of sheets of poorly differentiated malignant cells (H&E 200). B, phospho- histone H3 immunohistochemistry in M6-HH tumors compared with control tumors (200). C, M6-HH C *P = 0.0417 *P = 0.0182 tumors have higher mitotic counts 50 400 (left) and phospho-histone H3 40 counts (right). 300 30 200 20 Mitotic count 100 10 Phospho-Histone count 0 0

HH HH Vector Vector

in any Hh tumors compared with glandular structure in the in vivo studies. Based on our observation of increased observed in all of vector controls (Fig. 3A). We observed no proliferation in M6-HH tumors, we hypothesized that M6-HH difference in tumor infiltration by inflammatory cells (data not would proliferate faster than control cells. However, in vitro shown, assessed by IF for CD45 and F4/80), microvessel MTS assays showed no difference in the growth curves density (data not shown, assessed by CD31 IF) or in apoptosis (Fig. 4B). We also speculated that in view of its known role (assessed by caspase-3 IF, data not shown). However, there in stem cell regulation (5), Hh might promote increased self- was a significant increase in tumor cell proliferative fraction renewal capacity. However, modified "mammosphere" assays with a 2-fold higher proportion of phospho-histone H3-posi- showed no difference in sphere forming capacity between tive cells per hpf in the Hh expressing tumors compared with control or M6-HH cells in primary, secondary, or tertiary vector controls of the same size (Fig. 3B and C). Mitotic counts cultures (data not shown). Furthermore, inhibition of Hh (per 10 hpf) were also 50% higher in the Hh-positive tumors signaling with 5E1 antibody also had no effect on the devel- compared with vector controls of the same size (Fig. 3C). opment of primary or secondary mammospheres between the 2 groups (data not shown). Finally, we found no difference in A stromal requirement for Hh-mediated tumor growth migration of M6-HH cells in vitro compared with vector We confirmed M6 cells with Hh produced functional ligand control M6 cells (data not shown). using a Gli1 luciferase reporter, the "light II assay." This activity These findings suggest that there may be a critical stromal could be blocked by 5E1 an anti-Hh blocking antibody that interaction to account for the dramatic difference in tumor inhibits binding of all 3 ligands to the Ptch receptor, but not by growth seen with Hh overexpression. We further explored this control antibody (Fig. 4A). Using these cells, we then con- by carrying out quantitative PCR (qPCR) for key Hh pathway ducted detailed in vitro studies to investigate possible components, including readouts of canonical signaling; Hhip, mechanisms for the differences in tumor growth observed Ptch1, Ptch2, and Gli1. There was no change in the expression

www.aacrjournals.org Cancer Res; 71(11) June 1, 2011 4009 Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

O'Toole et al.

A B 4 14 Vector 12 3 HH 10

8 2 6 Absorbance 4 1

Relative reporter activity 2

0 0 12347

recHH M6-HH Days M6-vector Media alone M6-HH + 5E1

M6-HH + IgG control C D

Vector 10,000 Vector ** 10,000 HH HH 1,000 * 1,000 ** 100 100 ** 10 * 10 Relative expression Relative expression 1 1

0.1 0.1

SHH Gli1 Gli2 Hhip Gli1 Gli2 Ptch1 Ptch2 SHH Ptch1 Ptch2 Hhip Pecam1 E F 1,000 Stroma Epithelium Vimentin 100

Tumor epithelium 10 Hhip

1 Relative expression

0.1

Gli1 Gli2 Hhip SHH Ptch1 Ptch2 Pecam1

Figure 4. Hh-mediated growth requires stromal interaction. A, light II luciferase assay showing M6-HH cells produce functional Hh ligand which can be blocked by 5E1, but not by the control antibody. Recombinant Hh ligand is used as a positive control. B, M6 cells show no differences in growth measured by MTS assay with and without HH overexpression. C, qPCR for HH genes showing no change in pathway genes compared with vector controls except for SHH in the HH cells in culture. D, qPCR for HH genes in whole tumors showing increased HH responsive genes. E, qPCR for HH genes in separated epithelium and stroma in M6-HH tumors. SHH as expected is seen at highest levels in the epithelium, whereas HH responsive genes induced by HH overexpression are observed at highest levels in the stroma. F, dual immunofluorescence for vimentin (green) and Hhip (red). Nuclei are shown in blue (DAPI) with densely packed nuclei within the epithelial compartment. Hhip expression is not seen in the tumor epithelium but only in the peritumoral fibroblasts highlighted by vimentin expression (200). *, P < 0.05; **, P < 0.01.

4010 Cancer Res; 71(11) June 1, 2011 Cancer Research Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

Hedgehog in Breast Cancer

of any hedgehog pathway genes in the M6-HH cells in culture was no apparent growth or histological differences in the compared with controls except for SHH as expected (Fig. 4C). primary tumors (data not shown). However, the size of However, there was a significant increase in Hhip (P ¼ 0.014), metastatic deposits was smaller in the 5E1 treatment group Ptch1 (P ¼ 0.023), Ptch2 (P ¼ 0.009), and Gli1 (P ¼ 0.009) mRNA (P ¼ 0.02, Fig. 5E), suggesting that Hh controls the prolifera- in Hh overexpressing tumors compared with control (Fig. 4D). tion of primary and metastatic tumors. To identify in which compartment these changes occurred we separated the stroma and epithelium of the M6-HH tumor Discussion using FACS sorting for CD24 (a marker of epithelium; ref. 29). These studies revealed upregulation of Ptch1, Ptch2, Gli1, and The importance of Hh signaling in a subset of common Hhip expression exclusively in the stromal compartment human cancers is thought to rest on its ability to recreate compared with the epithelium (Fig. 4E). The epithelial com- developmental epithelial–mesenchymal paracrine signaling partment showed only upregulation of Shh as expected. These (30) where the signal from Hh ligand expressed by tumor findings were further validated using indirect double immu- cells is received by the receptor Ptch in the adjacent stroma nofluorescence on the tumor samples. Hhip expression was (8–10, 31, 32). Our study represents the first detailed descrip- confined to peritumoral stromal fibroblasts (confirmed by tion of the localization of key Hh pathway components in a vimentin expression, Fig. 4F) with no expression in the large, well-characterized breast cancer cohort using rigorously epithelial cells (data not shown, keratin immunostaining) validated antibodies. For the first time, we also report that the or inflammatory cells (data not shown, CD45 immunostain- expression of Hh ligand in the epithelial cells of breast cancer ing) and was not seen in the control tumors (data not shown). is associated with increased risk of metastasis, breast cancer– These data confirm that there is upregulation of downstream specific death, and a more proliferative, aggressive, basal-like Hh pathway components predominantly in the stroma in phenotype and that Hh ligand expression increases during response to Hh produced by the epithelium, supporting a progression of premalignant breast epithelial lesions. Taken paracrine signaling mechanism in our model of mammary together, these data strongly suggest the Hh ligands play a role carcinoma. in the progression and invasiveness of a subset of breast cancers. Hh blockade inhibits tumor growth and metastasis Our data support a paracrine mode of canonical Hh signal- In view of our data showing that Hh overexpressing tumors ing in breast cancer, although we cannot exclude a cell- were associated with larger, more poorly differentiated tumors autonomous role for Hh signaling in a small subset of tumors, with an increased rate of lymphatic invasion, we investigated or within a small population of tumor cells within a given the functional requirements for Hh ligand in the development tumor. In keeping with previously reported smaller studies (4, of tumor growth and metastatic dissemination. Using the M6- 13, 16), we found expression of Ptch and Gli1 in breast cancer HH allograft model, we treated mice with twice weekly specimens. Our data show that a combination of high Hh intraperitoneal injections of either 5E1 or a control antibody. ligand in the epithelium and Gli1 in the stroma, a "paracrine" The ability of 5E1 to inhibit Hh signaling was confirmed by pattern, is independently prognostic for overall survival and qPCR in tumor samples showing marked inhibition of Hhip approaches significance for breast cancer–specific death. The expression in the treatment group (Fig. 5A). The 5E1 treated importance of this "paracrine" Hh signature is supported by mice (N ¼ 10) showed a markedly slower rate of tumor growth evidence that Hh target gene expression is limited to the (Fig. 5B) and also had statistically significant longer median stromal compartment in mouse models, and that Hh ligand survival times (>110 days compared with 78 days, P ¼ 0.0002) expression produces no detectable cell-autonomous effect in than the control antibody treated group (N ¼ 5; Fig. 5C). In mammary carcinoma cells in vitro. Recent studies show that addition, ex vivo imaging and histological analysis of harvested Hh signaling may be important in cellular responses in organs showed a difference in the pattern of metastatic endothelial and mesenchymal cells (33, 34), adding weight disease between control and treated groups. Although all to the idea that Hh ligand can drive the formation of an mice in both groups developed lung metastases equivalently, optimal stromal environment in some solid tumors. Our data 100% of control mice developed metastases to the liver and also show that inhibition of the stromal Hh response, rather pancreas (Fig. 5D), but only 25% of 5E1 treated mice developed than direct targeting of neoplastic cells, is a new potential liver metastases, and none developed pancreatic metastases. therapeutic approach. These data suggested that Hh played a role in the development By manipulating the expression and/or activity of Hh ligand of metastatic disease. in breast cancer cells in vivo, we have also shown the func- We next sought to independently validate these findings in tional significance of this pathway in a model that closely the well-characterized 4T1 allograft model (25), which rapidly resembles basal-like breast cancer. The marked effect of Hh develops spontaneous metastases. qPCR showed that HH expression on tumor growth, grade, histology, and metastatic pathway genes, including all 3 Hh ligands were expressed potential in these models are remarkably consistent with our in untreated 4T1 cells (data not shown), confirming it is a good data in human breast cancer. Interestingly, blockade of Hh model in which to investigate the effects of HH inhibition. We signaling in vivo altered not only the size of metastases but transplanted the same number of 4T1 cells into the fat pads of also their tissue profile, suggesting that Hh signaling may also immunocompetent BALB/c mice followed by twice weekly act to promote organ-specific growth based on selection for treatment with intraperitoneal 5E1 or control antibody. There more favorable microenvironments.

www.aacrjournals.org Cancer Res; 71(11) June 1, 2011 4011 Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

O'Toole et al.

ABIgG control 5E1 antibody 4 600 2 * * IgG control 1 5E1 antibody ) 3 0.5 400 0.25 0.125

0.0625 200

Relative expression 0.03125 Tumor volume (mm 0.15625 0.0078125 0 30 40 50 60 70 Gli1 Gli2 Hhip SHH Ptch1 Ptch2 Days Pecam1 C D 100

80 IgG control 5E1 treatment ***P = 0.0002 60

40 Percent survival 20

0 0 50 100 Days E 1.0 × 1006 P = 0.0225 ) 2 8.0 × 1005

6.0 × 1005

4.0 × 1005

× 05

Average of lung met (um 2.0 10

5E1

IgG control

Figure 5. Hh inhibition decreases tumor growth and reduces the size of lung metastases. A, qPCR for HH genes showing inhibition of key readouts of Hh pathway activity with 5E1 treatment. B, M6-HH tumors show reduced growth rates in vivo with 5E1 treatment in comparison to the control antibody. C, M6-HH tumors treated with 5E1 show significantly longer survival compared with those treated with control antibody. D, representative pancreatic metastasis in M6-HH mouse (H&E, 200). E, Hh inhibition with 5E1 decreases the size of lung metastatic deposits in 4T1 allografts compared with control antibody treated shown graphically (left; *, P < 0.05) with an example of a lung metastasis in this model shown at right. The clinical implications of these findings for breast cancer treatment for basal-like breast cancer for which there is no include (i) a biological and functional connection between Hh effective targeted therapy, and (iii) the importance of epithelial ligand expression and basal-like breast cancer, (ii) a potential interactions as potential therapeutic targets. The development

4012 Cancer Res; 71(11) June 1, 2011 Cancer Research Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

Hedgehog in Breast Cancer

of therapeutics directed at stromal Hh signaling may represent Grant Support a novel approach to the treatment of locally advanced or metastatic breast cancer. Australia Breast Cancer Tissue Bank is generously supported by the National Health and Medical Research Council of Australia, The Cancer Institute NSW, and the National Breast Cancer Foundation. This work was supported by the Disclosure of Potential Conflicts of Interest following grants: the National Health and Medical Research Council (grant nos. 481378 (S.A. O’Toole), 427601 (R.L. Sutherland) and 535947 (S.A. O’Toole, A. No potential conflicts of interest were disclosed. Swarbrick, D.N. Watkins), the Cancer Institute NSW [grant nos. 08/ECF/1–12 (A. Swarbrick), 07/CRF/1–06 (S.A. O’Toole), and 07-CDF-1/28 (E.A. Musgrove)], the Acknowledgments St Vincent's Clinic Foundation, the National Breast Cancer Foundation, the Australian Cancer Research Foundation, the RT Hall Trust, and the Petre Foundation. 5E1 hybridomas were obtained from the University of Iowa Hybridoma Bank. The costs of publication of this article were defrayed in part by the payment Some tissues were also received from the Australia Breast Cancer Tissue Bank. of page charges. This article must therefore be hereby marked advertisement in The authors thank the assistance of Ms. Alice Boulghourjian and Ms. Sarah accordance with 18 U.S.C. Section 1734 solely to indicate this fact. Eggleton with immunohistochemistry, Ms. Anne Holliday for data management, Dr. Samantha Oakes for providing mouse tissue, and Mrs. Gillian Lehrbach with cell culture. The authors also thank Prof. Roger Daly and Dr. Elizabeth Caldon Received October 14, 2010; revised February 4, 2011; accepted March 8, 2011; for their advice on the manuscript. published online June 1, 2011.

References 1. Peto R, Boreham J, Clarke M, Davies C, Beral V. UK and USA breast methylation and reduced expression of PTCH in breast cancer. Breast cancer deaths down 25% in year 2000 at ages 20–69 years. Lancet Cancer Res Treat 2007;105:139–55. 2000;355:1822. 17. Lopez-Knowles E, O’Toole SA, McNeil CM, Millar EK, Qiu MR, Crea P, 2. Viani GA, Afonso SL, Stefano EJ, De Fendi LI, Soares FV. Adjuvant et al. PI3K pathway activation in breast cancer is associated with the trastuzumab in the treatment of her-2-positive early breast cancer: a basal-like phenotype and cancer-specific mortality. Int J Cancer; meta-analysis of published randomized trials. BMC Cancer 2007;7:153. 126:1121–31. 3. Rakha EA, Reis-Filho JS, Ellis IO. Basal-like breast cancer: a critical 18. Lopez-Knowles E, Zardawi SJ, McNeil CM, Millar EK, Crea P, Mus- review. J Clin Oncol 2008;26:2568–81. grove EA, et al. Cytoplasmic localization of beta-catenin is a marker of 4. Kubo M, Nakamura M, Tasaki A, Yamanaka N, Nakashima H, Nomura poor outcome in breast cancer patients. Cancer Epidemiol Biomar- M, et al. Hedgehog signaling pathway is a new therapeutic target for kers Prev; 19:301–9. patients with breast cancer. Cancer Res 2004;64:6071–4. 19. Millar EK, Anderson LR, McNeil CM, O'Toole SA, Pinese M, Crea P, 5. Liu S, Dontu G, Mantle ID, Patel S, Ahn NS, Jackson KW, et al. et al. BAG-1 predicts patient outcome and tamoxifen responsiveness Hedgehog signaling and Bmi-1 regulate self-renewal of normal and in ER-positive invasive ductal carcinoma of the breast. Br J Cancer malignant human mammary stem cells. Cancer Res 2006;66:6063–71. 2009;100:123–33. 6. Fiaschi M, Rozell B, Bergstrom A, Toftgard R. Development of mam- 20. Alle KM, Henshall SM, Field AS, Sutherland RL. Cyclin D1 protein is mary tumors by conditional expression of GLI1. Cancer Res overexpressed in hyperplasia and intraductal carcinoma of the breast. 2009;69:4810–7. Clin Cancer Res 1998;4:847–54. 7. Ingham PW, McMahon AP. Hedgehog signaling in animal develop- 21. Zardawi SJ, Zardawi I, McNeil C, Millar EK, McLeod D, Morey AL, et al. ment: paradigms and principles. Genes Dev 2001;15:3059–87. High Notch1 protein expression is an early event in breast cancer 8. Shaw A, Gipp J, Bushman W. The Sonic Hedgehog pathway stimu- development and is associated with the HER-2 molecular subtype. lates prostate tumor growth by paracrine signaling and recapitulates Histopathology 2010;56:286–96. embryonic gene expression in tumor myofibroblasts. Oncogene 22. Cheang MC, Voduc D, Bajdik C, Leung S, McKinney S, Chia SK, et al. 2009;28:4480–90. Basal-like breast cancer defined by five biomarkers has superior 9. Tian H, Callahan CA, DuPree KJ, Darbonne WC, Ahn CP, Scales SJ, prognostic value than triple-negative phenotype. Clin Cancer Res et al. Hedgehog signaling is restricted to the stromal compartment 2008;14:1368–76. during pancreatic carcinogenesis. Proc Natl Acad Sci U S A 2009; 23. Holzer RG, MacDougall C, Cortright G, Atwood K, Green JE, Jorcyk 106:4254–9. CL. Development and characterization of a progressive series of 10. Yauch RL, Gould SE, Scales SJ, Tang T, Tian H, Ahn CP, et al. A mammary adenocarcinoma cell lines derived from the C3(1)/SV40 paracrine requirement for hedgehog signalling in cancer. Nature Large T-antigen transgenic mouse model. Breast Cancer Res Treat 2008;455:406–10. 2003;77:65–76. 11. Lauth M, Toftgard R. Non-canonical activation of GLI transcription 24. Maity T, Fuse N, Beachy PA. Molecular mechanisms of Sonic hedge- factors: implications for targeted anti-cancer therapy. Cell Cycle hog mutant effects in holoprosencephaly. Proc Natl Acad Sci U S A 2007;6:2458–63. 2005;102:17026–31. 12. Lewis MT, Ross S, Strickland P, Sugnet CW, Jimenez E, Scott MP, 25. Tao K, Fang M, Alroy J, Sahagian GG. Imagable 4T1 model for the et al. Defects in mouse mammary gland development caused by study of late stage breast cancer. BMC Cancer 2008;8:228. conditional haploinsufficiency of Patched-1. Development 1999;126: 26. Pear WS, Nolan GP, Scott ML, Baltimore D. Production of high-titer 5181–93. helper-free retroviruses by transient transfection. Proc Natl Acad Sci 13. Moraes RC, Zhang X, Harrington N, Fung JY, Wu MF, Hilsenbeck SG, U S A 1993;90:8392–6. et al. Constitutive activation of smoothened (SMO) in mammary 27. Harvey JM, Clark GM, Osborne CK, Allred DC. Estrogen receptor glands of transgenic mice leads to increased proliferation, altered status by immunohistochemistry is superior to the ligand-binding differentiation and ductal dysplasia. Development 2007;134:1231–42. assay for predicting response to adjuvant endocrine therapy in breast 14. Lewis MT, Ross S, Strickland PA, Sugnet CW, Jimenez E, Hui C, et al. cancer. J Clin Oncol 1999;17:1474–81. The Gli2 transcription factor is required for normal mouse mammary 28. Herschkowitz JI, Simin K, Weigman VJ, Mikaelian I, Usary J, Hu Z, gland development. Dev Biol 2001;238:133–44. et al. Identification of conserved gene expression features between 15. Mukherjee S, Frolova N, Sadlonova A, Novak Z, Steg A, Page GP, murine mammary carcinoma models and human breast tumors. et al. Hedgehog signaling and response to cyclopamine differ in Genome Biol 2007;8:R76. epithelial and stromal cells in benign breast and breast cancer. Cancer 29. Sleeman KE, Kendrick H, Ashworth A, Isacke CM, Smalley MJ. CD24 Biol Ther 2006;5:674–83. staining of mouse mammary gland cells defines luminal epithelial, 16. Wolf I, Bose S, Desmond JC, Lin BT, Williamson EA, Karlan BY, et al. myoepithelial/basal and non-epithelial cells. Breast Cancer Res Unmasking of epigenetically silenced genes reveals DNA promoter 2006;8:R7.

www.aacrjournals.org Cancer Res; 71(11) June 1, 2011 4013 Downloaded from cancerres.aacrjournals.org on June 1, 2011 Copyright © 2011 American Association for Cancer Research DOI:10.1158/0008-5472.CAN-10-3738

O'Toole et al.

30. Litingtung Y, Lei L, Westphal H, Chiang C. Sonic hedgehog is 33. Nakamura K, Sasajima J, Mizukami Y, Sugiyama Y, Yamazaki M, Fujii essential to foregut development. Nat Genet 1998;20:58–61. R, et al. Hedgehog promotes neovascularization in pancreatic cancers 31. Fan L, Pepicelli CV, Dibble CC, Catbagan W, Zarycki JL, Laciak R, by regulating Ang-1 and IGF-1 expression in bone-marrow derived et al. Hedgehog signaling promotes prostate xenograft tumor growth. pro-angiogenic cells. PLoS One 5:e8824. Endocrinology 2004;145:3961–70. 34. Yamazaki M, Nakamura K, Mizukami Y, Ii M, Sasajima J, Sugiyama Y, 32. Zhang J, Lipinski RJ, Gipp JJ, Shaw AK, Bushman W. Hedgehog et al. Sonic hedgehog derived from human pancreatic cancer cells pathway responsiveness correlates with the presence of primary cilia augments angiogenic function of endothelial progenitor cells. Cancer on prostate stromal cells. BMC Dev Biol 2009;9:50. Sci 2008;99:1131–8.

Immunocytochemistry was performed on adherent cells grown overnight on coverslips, fixed in 4% (w/v) paraformaldehyde, blocked for 90 min with 2% (v/v) horse serum and incubated with primary antibody for 1 h at room temperature.

Alexa555 Cy3 fluorophore conjugated anti-rabbit (A-31572, Invitrogen, Victoria,

AUS) was diluted at 1:500 and incubated for 1 h in the dark at room temprature.

Coverslips were mounted onto slides using Vectashield mounting medium with 6- diamidino-2-phenylindole (DAPI) counterstain (H-1500, Vector Laboratories,

California, USA). Slides were stored at -20 oC. IF was performed as described for immunocytochemistry, however frozen tissue sections were cut on Superfrost coated slides.

Western Blotting

Cell pellets were re-suspended in RIPA buffer (50 mM Tris, 150 mM NaCl, 0.1%

SDS, 0.5% Na.Deoxychlorate, 1% Triton 100, 1mM PMSF, Complete Mini, EDTA-

Free protease inhibitor tablets(11836170001, Roche, Basel, Switzerland)) and incubated on ice for 10 min after which the lysates were cleared by centrifugation at top speed for 10 min. Protein quantitation was carried out using the Bio-Rad Protein

Assay Reagent (500-0006, Bio-Rad California, USA). 40ug of protein was denatured in LDS sample buffer (NP0007, Invitrogen) as per protocol. Proteins were separated by SDS-PAGE on a 4-12% Bis-Tris Gel (NP0323, Invitrogen) in MOPS running buffer (1 M MOPS, 0.5 M EDTA, 2 M Sodium Acetate) then transferred to PVDF membrane (IPVH00010, Millipore, NSW, AUS). Non-specific binding was blocked in 5% skim milk powder TBS-.Tween (0.1%), followed by incubation with primary antibody overnight at 4 C. All subsequent washes were made in TBS-T(0.1%). Following appropriate horseradish peroxidase-conjugated secondary antibody

(NA931V/NA934V GE Healthcare NSW, AUS) incubation the membrane was washed before detection with enhanced chemiluminescence (ECL; NEL104/105

PerkinElmer, MA USA).

Mammosphere assays

Briefly M6 cells with and without SHH transduction were trypsinized and filtered to obtain a single cell suspension. Cells were thoroughly washed and resuspended in mammosphere media. A minimum of 1000 cells were counted. The number of single cells was >99% in all experiments. Cells were grown in a serum-free M6 growth medium, supplemented with B27 (17504-044, Invitrogen) and 20 ng/mL bFGF

(354060 BD Biosciences, CA, USA), and 4 ug/mL heparin (H0777 Sigma-Aldrich).

Single cells were plated in ultralow attachment 6 well plates (Corning) at a density of 20,000 viable cells/well in primary culture and 1000 cells/well in subsequent passages.

To generate secondary mammospheres and subsequent tertiary mammospheres, primary mammospheres were collected by gentle centrifugation (800 g, 10 seconds) after 10 days. Cells were centrifuged and the pellet gently resuspended in the residual medium to prevent dissociation of mammospheres. After washing, primary mammospheres were non-enzymatically dissociated then neutralized with mammosphere medium. The cells obtained from dissociation were sieved through a

40-um sieve and analyzed microscopically for single cellularity. Single cells were then plated in ultralow attachment 6 well plates (Corning) at a 1000 cells/well for secondary mammospheres.

Supplementary Table 1. Antibodies used for immunohistochemistry.

Antibody Species Dilution Antigen Retrieval Hh Ligand H-160 Rabbit 1:80 20 minutes boiling (sc-9024, Santa Cruz) Polyclonal waterbath (Dako retrieval solution S2367) PTCH Rabbit 1:50 20 minutes boiling (ab27529, Abcam) Polyclonal waterbath (Dako retrieval solution S2367) GLI1 H-300 Rabbit 1:100 30 s at maximum (sc-20687, Santa Cruz) Polyclonal temperature and pressure in a pressure cooker (DAKO) in DAKO pH 6.1 solution (s1699) LYVE-1 (BAF2125, Goat 1:100 20 minutes boiling R&D) Polyclonal waterbath (Dako retieval solution S1699) Phospo-Histone H3 Rabbit 1:100 2 minutes at maximum (Ser10) (Cell signaling polyclonal temperature and pressure in 9701) a pressure cooker in DAKO pH 6.1 solution (s1699) Ki67 (SP6), Rabbit 1:200 Leica/Vision Biosystem Neomarkers, USA; monoclonal Bondmax automated

system using ER2 (high pH) antigen retrieval solution Cyclin A (6E6) Mouse 1:100 2 minutes at maximum monoclonal temperature and pressure in

Novocastra, UK a pressure cooker in DAKO pH 6.1 solution (s1699) Cyclin B1 (7A9) Mouse 1:40 2 minutes at maximum monoclonal temperature and pressure in

Novocastra, UK a pressure cooker in DAKO pH 6.1 solution (s2367) Cyclin E (13A3), Mouse 1:40 2 minutes at maximum Novocastra, Newcastle, monoclonal temperature and pressure in

UK a pressure cooker in DAKO pH 6.1 solution (s1699)

Primer/Probe sets

Supplementary table 2: primers used for Roche LightCycler480

Gene Primer Fwd 5’ Primer Rev 5’ UPL Cat No SHH CAA ATT ACA ACC CCG ACA GCA TTT AAC TTG TCT TTG CAC TC CT 32 04687655001

Ptch1 GGC CTG GCA GAG GAC TTA C GGA AGC ACC TTT TGA GTG GA 10 04685091001

Ptch2 GTC CAC CTA GTG CTC CCA AC CTC AGC TCC TGA GCC ACA TT 40 04687990001

Gli1 GGA CCC ACT CCA ATG AGA AG CAT GCA CTG TCT TCA CGT GTT 33 04687663001

Gli2 AGA ACC GCA CTC ACT CCA AT AGC TGG GGT CTG TGT ACC TC 64 04688635001

Hhip GTG TTC GGA GAT CGC AAT G TTT TCT TGC CAT TGC TTG GT 67 04688660001

Pecam AGC CAG TAG CAT CAT GGT CA AGC AGG ACA GGT CCA ACA AC 25 04686993001 1 -actin GGA TGC AGA AGG AGA TTA CTG C CCA CCG ATC CAC ACA GAG TA 63 04688627001 Supplementary table 3: program used for Roche LightCycler480

Target Acquisition Hold Ramp rate Sec Target Step size temp mode ( C/s) (per C) ( C) Pre-Incubation 94 None 7 min 4.8 0 0 Amplification Analysis mode: Quantification Cycles: 45 Target Acquisition Hold Ramp rate Sec Target Step size temp mode ( C/s) (per C) ( C) 94 None 15 sec 4.8 0 0 60 None 30 sec 2.5 50 0.5 72 Single 15 sec 1.5 0 0 Cooling 40 30 sec 2.5 0 0 0

Normal duct Hyperplasia

Mammary intraepithelial neoplasia Invasive carcinoma

C3 (1)/Tag lesions B

250 **

200 n=15 n=15

n=9 150

100 n=10

50 Hh ligand H score

0 normal hyperplasia MIN Ca Supplementary Figure 4. (A) IHC for HH in mammary glands from C(#)-Tag transgenic mice. (B) Quantitation of Hh IHC. Supplementary Figure 3. (A) Western blot for GLI1 (left) and Flag epitope (right) in HCT-116 cells transfected with Gli1Flag construct (transfection performed in triplicate). HCT-116 are endogenously negative for Gli1 expression. Flag +ve known to contain a Flag tagged protein of 60 kDa. (B) Immunofluorescence of M6 cells for GLI1 antigen. M6 cells show GLI1 in both the cytoplasm and in the nucleus. All images 400x, bar = 20 Ʌm. (C) Immunohistochemistry of FFPE mouse embryo (E14,5) for GLI1 with matched concentration IgG negative control. Perichondrium (arrow) is positive for nuclear GLI1. Images at 200x, bar = 50 Ʌm. Supplementary Figure 2. (A) Western blot for PTCH in HCT-116 cells transfected with mouse Ptch (transfection performed in triplicate). HCT-116 cells are endogenously negative for Ptch expression. (B) Immunofluorescence of mouse embryo (E13.5) and M6 mouse mammary carcinoma cell line for PTCH antigen. Image of developing bone around growth plate (transverse section) shows cells of the perichondrium (arrow) positive for PTCH. Image of cerebellar anlage shows external granular cells (asterix) positive for PTCH. M6 cells show correct subcellular localisation of PTCH at the plasma membrane. All images at 400x, bar = 20 Ʌm. (C) Immunohistochemistry of FFPE mouse embryo (E13.5) for PTCH antigen with matched concentration of IgG negative control. Images show developing bone around growth plate (transverse section) shows cells of the perichondrium (arrow) positive for PTCH. Images at 200x, bar = 50 Ʌm. Supplementary Figure 1. (A) Western blot for Hh ligand in M6 cells transfected with a siRNA directed against Hh ligand for 48 h or 72 h (samples transfected in triplicate). M6 cells endogenously express Hh ligand. A549 cells do not endogenously express Hh ligand and served as a negative control. (B) Immunofuorescence of M6 cells and M6 cells infected to stably express Hh ligand. Images 400x, bar = 20 Ʌm. (C) Immunofuorescence (top) and immunohistochemistry (bottom) of mouse neural tube for Hh ligand. Both show Hh ligand around cells of the floor plate (arrows), with matched concentration IgG negative control for the immunohistochemistry analysis. All images of ventral portion of developing neural tube (transverse sections). Immunofuorescent images 400x, bar = 20 Ʌm, Immunohistochemical images 200x, bar = 50 Ʌm. Histopathology 2010, 56, 286–296. DOI: 10.1111/j.1365-2559.2009.03475.x

High Notch1 protein expression is an early event in breast cancer development and is associated with the HER-2 molecular subtype

Sarah J Zardawi,1 Ibrahim Zardawi,2 Catriona M McNeil,1,3 Ewan K A Millar,1,4 Duncan McLeod,5 Adrienne L Morey,6 Paul Crea,7 Niamh C Murphy,1 Mark Pinese,1 Elena Lopez-Knowles,1 Samantha R Oakes,1 Christopher J Ormandy,1 Min Ru Qiu,1,6 Anne Hamilton,8 Andrew Spillane,8,9 Cheok Soon Lee,5,10 Robert L Sutherland,1,11 Elizabeth A Musgrove1,11 & Sandra A O’Toole1,5,11 1Cancer Research Program, Garvan Institute of Medical Research, St Vincent’s Hospital, Darlinghurst, 2Department of Anatomical Pathology, HAPS, John Hunter Hospital, New Lambton, 3Department of Medical Oncology, University of Sydney, Westmead Hospital, Westmead, 4Department of Anatomical Pathology, South Eastern Area Laboratory Service, St George Hospital, Kogarah, 5Department of Anatomical Pathology, Royal Prince Alfred Hospital, Camperdown, 6Department of Anatomical Pathology, Sydpath and 7Department of Surgery, St Vincent’s Hosptial, Darlinghurst, 8Sydney Cancer Centre, Royal Prince Alfred Hospital, Camperdown and University of Sydney, 9Mater Hospital, North Sydney, 10Bosch Institute, University of Sydney and School of Medicine, University of Western Sydney, and 11Faculty of Medicine, St Vincent’s Clinical School, University of NSW, Sydney, NSW, Australia

Date of submission 16 March 2009 Accepted for publication 11 May 2009

Zardawi S J, Zardawi I, McNeil C M, Millar E K A, McLeod D, Morey A L, Crea P, Murphy N C, Pinese M, Lopez- Knowles E, Oakes S R, Ormandy C J, Qiu M R, Hamilton A, Spillane A, Soon Lee C, Sutherland R L, Musgrove E A & O’Toole S A (2010) Histopathology 56, 286–296 High Notch1 protein expression is an early event in breast cancer development and is associated with the HER-2 molecular subtype

Aims: Activation of Notch signalling results in hyper- Notch1 protein was an early event in both murine and plasia and tumorigenesis in murine mammary epithe- human breast cancer development with progressive lium. However, there is little information regarding the increases in expression with the development of expression of Notch1 in premalignant lesions and early hyperplasia and malignancy. High Notch1 was not breast cancer. We investigated expression of Notch1 in prognostic in the outcome cohort. There was, however, breast cancer development and its association with a highly signiﬁcant association of high Notch1 protein molecular subtypes. with the HER-2 molecular subtype of breast cancer Methods and results: Immunohistochemical expression (P = 0.008). of Notch1 was determined in a murine model of Conclusions: These data demonstrate that aberrant mammary carcinogenesis and in breast tissue from two Notch regulation is an early event in mammary cohorts of breast cancer patients, the ﬁrst (n = 222) carcinogenesis and is associated with the HER-2 comprising a histological progression series and the molecular subtype of breast cancer, and suggest the second an outcome series of 228 patients with operable Notch signalling pathway may be a potential thera- invasive ductal carcinoma. Enhanced expression of peutic target worthy of further investigation. Keywords: breast cancer, developmental signalling pathways, Notch1

Address for correspondence: Robert L Sutherland, PhD, Cancer Research Program, Garvan Institute of Medical Research, 384 Victoria St, Darlinghurst, NSW 2010, Australia. e-mail: [email protected]

2010 The Authors. Journal compilation 2010 Blackwell Publishing Limited. Notch is an early event in breast cancer development 287

Abbreviations: ADH, atypical ductal hyperplasia; CK, cytokeratin; DAPT, N-[N-(3,5-difluorophenacetyl)-l-alanyl]- S-phenylglycine t-butyl ester; DCIS, ductal carcinoma in situ; EGFR, epidermal growth factor receptor; ER, oestrogen receptor; FFPE, formalin-fixed paraffin-embedded; FISH, fluorescence in situ hybridization; IDC, invasive ductal carcinoma; LOESS, Locally weighted scatterplot smoothing; MIN, mammary intraepithelial neoplasia; N1ICD, Notch1 intracellular domain; PR, progesterone receptor; TMA, tissue microarray; UDH, usual ductal hyperplasia

carcinogenesis; Farnie et al.11 have reported that Introduction primary culture of human ductal carcinoma in situ There is increasing evidence that dysregulation of (DCIS) mammospheres had higher levels of Notch developmental signalling pathways such as Notch, activation than normal duct-derived mammospheres, Wnt and Hedgehog is associated with the development and increased Notch signalling in DCIS was associated and progression of malignancy in a number of organs. with an increased risk of recurrence. These developmental signalling pathways regulate self- Gene expression profiling has identified intrinsic 12 renewal and maintenance of normal stem cells in most molecular phenotypes of breast cancer. There is tissues and are proposed to contribute to the mainte- speculation that the breast cancer molecular subtypes, nance of cancer stem cells.1 i.e. luminal, basal-like, HER-2 and normal breast-like Notch signalling is critical in mammalian embryonic cancers observed in expression profiles, may reflect 13 development, particularly in neurogenesis, angiogene- distinct cells of origin, although this is not universally 14 sis and in development of the breast, heart and accepted. However, there is relatively little evidence to lymphoid systems.2 More recently, Bouras et al.3 have date to link breast cancer molecular subtypes with described a key role for Notch signalling in the particular developmental pathways. For example, there regulation of mammary progenitor cells. are conflicting data regarding the association of Notch Notch signalling occurs through two separate path- signalling pathway components with specific breast 8 ways, CSL-dependent signalling (CBP ⁄ RBP-jj in verte- cancer subtypes. Reedijk et al. observed that breast brates, Suppressor of Hairless in Drosophila and Lag-1 cancers with high levels of Jagged1 ligand, Notch1 and in Caenorhabditis elegans) and Deltex protein signalling, Notch3 receptors were almost exclusively triple nega- with the majority of signalling occurring through the tive [oestrogen receptor (ER)-negative, progesterone CSL-dependent pathway.4 In humans, Notch signalling receptor (PR)-negative, HER-2-negative] and that high is mediated through the Notch transmembrane recep- levels of Jagged1 were associated with expression of 15 tors (Notch 1–4), which are activated by the Delta-like basal cytokeratins (CK). Dontu et al. reported that (Delta-like 1, 3, 4) and Jagged (Jagged 1, 2) ligands. Notch pathway activity drove early progenitor cells Binding of ligand causes cleavage of the transmem- towards the development of myoepithelial cells, while 3 brane receptor by an ADAM metalloproteinase and Bouras et al. have recently demonstrated that Notch c-secretase, leading to release of the intracellular signalling regulates luminal cell-fate commitment. In domain (NICD)5 and translocation to the nucleus contrast, others have reported that Notch activity is where it forms a trimeric activation complex with linked with HER-2 overexpression, suggesting both 16,17 18 CSL and Mastermind to induce transcription activation positive and negative relationships between these 19 of the HES6 and HERP5 gene families. oncogenes. Most recently, Magnifico et al. identified There is increasing evidence that aberrations in the that Notch1 directly regulates HER-2. Thus, there is a Notch pathway, in particular the Notch1 receptor, are need to clarify further the role of aberrant Notch implicated in breast carcinogenesis. Constitutive acti- signalling in the development of breast cancer and any vation of Notch1 causes mammary hyperplasia3 and association with particular molecular subtypes of the carcinogenesis in mice,7 and Notch1 overexpression disease. Here, we investigated the potential associations has been associated with a significantly poorer prog- of immunohistochemical expression of Notch1 and nosis in human breast cancer.8 Activation of Notch clinicopathological parameters and disease outcome in signalling in mouse mammary epithelial cells is asso- tissue from two series of breast cancer patients. ciated with disorganized and invasive growth in matrigel and collagen assays,9 and with transformation Materials and methods of human breast epithelial cell lines through loss of study populations E-cadherin expression and resistance to apoptosis.10 There is relatively little published data regarding Two separate cohorts of breast cancer patient tissues the role of Notch signalling in early human breast were used in these studies. The first cohort comprised

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. 288 S J Zardawi et al.

a histological ‘progression’ series of 222 patients Table 1. Clinicopathological features of the invasive ductal diagnosed with invasive ductal carcinoma (IDC) or DCIS carcinoma cohort (n = 228) between 1996 and 2005 at Royal Prince Alfred Hospital Clinicopathological features n (Sydney, NSW, Australia) retrospectively identified for this study. This series was developed to include lesions Median length of 68 (range 0–159) associated with increasing risk of developing breast follow-up (months) cancer, rather than implying a direct neoplastic Median age (years) 54 (range 24–87) progression. A range of preinvasive and invasive lesions was identified in the associated formalin-fixed paraffin- Tumour size (mm) embedded (FFPE) resected breast tissue using standard 0–10 38 ⁄ 228 (17%) diagnostic criteria by experienced breast pathologists 11–20 96 ⁄ 228 (42%) (S.A.O’T., E.K.A.M. or D.M.). This series comprised 724 independent lesions that were suitable for interpretation 21–50 85 ⁄ 228 (37%) of Notch1 expression: 66 morphologically normal ducts >50 9 ⁄ 228 (4%) adjacent to invasive carcinoma, four atypical ductal hyperplasias (ADH), 15 usual ductal hyperplasias Tumour grade (UDH), 511 cases of DCIS and 101 IDCs. ADH was I39⁄ 228 (17%) excluded from the analysis due to the small number of ⁄ cases. The use of clinical specimens and associated data II 89 228 (39%) was approved by the Ethics Committee of the Royal III 100 ⁄ 228 (44%) Prince Alfred Hospital (X05-0115). The second cohort consisted of 292 patients with Lymph node metastasis operable IDC treated by a single surgeon (P.C.) between 0 128 ⁄ 228 (56%) February 1992 and August 2002 and has been previously reported.20 The clinicopathological features 1–3 68 ⁄ 228 (30%) of the subset of 228 patients suitable for assessment of 4–10 22 ⁄ 228 (10%) Notch1 staining from this cohort are summarized in Table 1. Prior approval for this study was obtained >10 10 ⁄ 228 (4%) from the Human Research Ethics Committee of St Oestrogen receptor positive 156 ⁄ 227 (69%) Vincent’s Hospital, Sydney (HREC SVH H94 ⁄ 080, HREC 06336 SVH H00036). Progesterone receptor positive 133 ⁄ 228 (58%)

HER-2+ (FISH) 38 ⁄ 218 (17%) tissue microarrays and immunohistochemistry Total recurrences 54 ⁄ 228 (24%) Two separate sets of tissue microarrays (TMAs) of FFPE Breast cancer deaths 36 ⁄ 228 (16%) tumour tissue blocks were constructed for each cohort. Five-year recurrence-free survival 79% For the IDC cohort, each patient was represented by two to six 1-mm cores. For the progression series four Five-year breast cancer-speciﬁc 85% representative cores from each donor block were taken survival and deposited in the recipient array block for sampling Notch1 positive 88 ⁄ 228 (39%) of DCIS. Three representative cores were taken when sampling IDC. FISH, ﬂuorescence in situ hybridization. TMAs were produced using the MTA-1 Manual Tissue Arrayer (Beecher Instruments, Sun Prairie, WI, USA). boiling water bath for 40 min, followed by cooling for Haematoxylin and eosin slides for all arrays were 15 min in a running waterbath. Following a thorough prepared and individual cores were reviewed by an wash in distilled water, endogenous peroxidase activity experienced breast pathologist (S.A.O’T. or E.K.A.M.). was eliminated with 3% hydrogen peroxide for 5 min. A small number of whole tissue sections were stained for Slides were incubated for 1 h with the Notch1 H-131 Notch1 and compared visually with TMAs and the antibody diluted 1:100 (Santa Cruz Biotechnology, pattern of immunoreactivity was found to be concordant. Santa Cruz, CA, USA). Following a buffer wash, Antigen retrieval was performed using Dako solution detection employed Dako Envision+ rabbit secondary pH 9.0, (s2367; Dako, Carpinteria, CA, USA) in a reagent (Dako) for 30 min at room temperature,

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. Notch is an early event in breast cancer development 289

followed by Dako diaminobenzidine (DAB)+ chromo- cohort comprised 135 patients, with a median tumour gen for 10 min. This antibody was selected because it size of 18 mm and a median follow-up of 122 months has previously been shown to be highly specific for with 70% ER positivity. Data were presented from Notch1 expression in a study of transgenic mice with Agilent Human 1A arrays which were available as raw inducible inactivation of Notch1.21 In our study, scanner data files and sourced from Array Express endothelial cells22 were used as the positive tissue (http://www.ebi.ac.uk/microarray-as/ae/) accession control and a matched concentration isotype IgG was E-UCON-1. Using the limma R package,26 background- used as the negative control. subtracted data were normalized by the global locally ER, PR, CK5 ⁄ 6 and epidermal growth factor receptor weighted scatterplot smoothing (LOESS) technique (EGFR) were also stained using the following antibodies: applied to non-control spots only. To combine informa- ER 1:100 (clone 6F11; Dako) and PR 1:200 (clone PgR tion from duplicate dye swap arrays, a linear model was 636; Dako), CK5 ⁄ 6 1:80 (clone MAB1602; Chemicon fitted to the normalized data using limma. Model fit Int., Temecula, CA, USA), and EGFR 1:100 (clone H11; coefficients for each sample were then used as final Dako). HER-2 fluorescence in situ hybridization (FISH) expression estimates, expressed relative to a pooled was assessed in the Australian National Reference reference RNA. The NOTCH1 probe was analysed for Laboratory. (Department of Pathology, St Vincent’s frequency distribution of mRNA and its associations Hospital, Sydney) using the Vysis PathVysion HER-2 with patient outcome using Cox proportional hazards DNA dual colour probe kit. A HER-2:chromosome 17 for univariate analysis. The mean score was used to ratio >2.2 was classified as HER-2 amplification. dichotomize the data to generate Kaplan–Meier curves. Cytoplasmic immunoreactivity for Notch1 was The dataset of Wang et al.25 comprised 286 patients assessed by two independent observers (S.J.Z. and with lymph node-negative primary breast cancer with S.A.O’T.), including an experienced breast pathologist a mean age of 52 years, and 73% ER positivity. The (S.A.O’T.) and described in terms of the intensity data presented were from the Affymetrix (Santa Clara, (0 = negative, 1+ = weak, 2+ = moderate and 3+ = CA, USA) oligonucleotide microarray U133a GeneChip, strong) and percentage of immunopositive cells. From which were scanned using standard protocols. Expres- these indices a simplified ‘H score’ (i.e. intensity · sion values were calculated by use of the Affymetrix percentage of positive cells) was calculated for each GeneChip analysis software MAS 5.0. For chip nor- core and a mean and median score for each index malization, probe sets were scaled to a target intensity calculated for each tumour.20 ER and PR were assessed of 600. as positive if they had an H score of >10. CK5 ⁄ 6 and EGFR were assessed as positive if there was any positive mouse mammary gland tissues cytoplasmic or membranous immunoreactivity present at any intensity. The immunohistochemical expression of Notch1 was also examined in a mouse model of mammary carcino- ⁄ definition of intrinsic molecular phenotype genesis. The C3 SV40T mouse model directs transgenic of breast cancer expression of the SV40 large T antigen to the distal mammary duct and prostatic ductal epithelium, result- This was approximated using a surrogate ‘signature’ ing in multistage ductal dysplasia and ultimately similar to the immunohistochemical signature recently invasive cancer.27 Whole mammary gland tissue sec- described by Cheang et al.,23 but also using FISH to tions from these mice at various stages of carcinogenesis determine HER-2 status.20 Five different subgroups were harvested, and routinely prepared as previously were defined: luminal A = ER+ and ⁄ or PR+, HER-2); described.28 All experiments involving mice were per- luminal B = ER+ and ⁄ or PR+, HER-2+; HER-2 = ER) formed under the supervision of and in accordance with and PR), HER-2+; basal-like = ER),PR), HER-2),CK the regulations of the Garvan ⁄ St Vincent’s Animal 5 ⁄ 6+ and ⁄ or EGFR+; unclassified = negative for all Experimentation Committee. five markers. Essentially the same immunohistochemical protocol for Notch1 was used as described above, with minor modification due to the significantly shorter fixation notch pathway genes mrna expression and outcome period of the mouse tissues; antigen retrieval was Gene expression data from two publically available performed in Dako pH 9 target retrieval solution for datasets, the Naderi24 and Wang25 studies, were 15 min in a boiling water bath. Expression of Notch1 employed to establish relationships between NOTCH1 was assessed as described for the human tissues by an mRNA levels and breast cancer outcome. The Naderi experienced breast pathologist (S.A.O’T.).

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. 290 S J Zardawi et al.

statistical analysis Notch1 expression was observed in the cytoplasm of epithelial cells ranging from weak (1+) to moderate Statistical evaluation was performed using Statview (2+) and strong (3+) (Figure 1B–D, respectively). The 5.0 Software (Abacus Systems, Berkeley, CA, USA). A specificity of immunoreactivity was confirmed by the P-value of <0.05 was accepted as statistically signifi- presence of immunopositivity in endothelial cells, with cant. Baseline characteristics of the cohort were defined no expression in smooth muscle cells (Figure 1A). using simple frequency distributions. There was a range of expression of Notch1 observed in Univariate analyses using the Kaplan–Meier and normal, premalignant and invasive breast lesions; Cox proportional hazards model were applied to the using the mean H score of 85 (derived from the cohort for established clinicopathological variables invasive carcinoma cohort) to dichotomize the data in using recurrence, distant relapse and breast cancer- the ‘progression series’, high Notch1 was seen in only related death as end-points. nine of 66 cases (13.6%) of morphologically normal Descriptive statistics for Notch1 protein expression in ducts with a progressive increase in the proportion of the invasive carcinoma cohort were determined and the lesions showing high relative Notch1 protein expres- frequency distribution of the scores assessed. No appar- sion with 33% of UDH, 30% of nuclear grade 1 DCIS, ent break points were observed to suggest an obvious 57% of nuclear grade 2 DCIS and 59% of nuclear grade cut-point for dichotomization. The optimal cut-point 3 DCIS with high Notch1 (Figure 2Bii). Intriguingly, determination technique using serial survival analysis there was a small but statistically significant (log rank P-value in Kaplan–Meier analysis) was then (P = 0.005) decline in Notch1 expression in IDC used to search for a biologically significant cut-point, but when compared with DCIS, with 39% of IDCs in the none was identified. Therefore the mean H score progression series showing high Notch1 (Figure 2Bi). (H > 85) was used to dichotomize the data for univariate Photomicrographs showing the pattern of expression survival analysis. This cut-point of H ‡85 was subse- in the human breast progression cohort are also quently also applied to the ‘progression’ cohort. presented in Figure 2A with a graphical representation Subgroup analysis, to determine any prognostic of the anova of Notch1 H score in each of the human significance of Notch1 expression, was also performed breast lesions shown in Figure 2Bi. There was also an where possible. However, the smaller subgroups of increase in Notch1 protein expression between different HER-2 and basal-like breast cancers lacked sufficient nuclear grades of DCIS (Figure 2Bii) with a progressive statistical power for analysis. Clinicopathological vari- increase between nuclear grades 1 and 2 DCIS (P < ables were dichotomized using standard parameters. 0.001) and between DCIS grades 2 and 3 (P = 0.03). Correlations between the determined cut-points for We next investigated the expression of Notch1 in a each variable and clinicopathological variables were murine model of mammary carcinogenesis and tested for statistical significance using contingency progression, the C3 ⁄ SV40T mouse model, and found tables, applying the v2 test. The relationship between an almost identical pattern to that seen in the human Notch1 expression as a continuous variable and breast breast lesion progression series (Figure 2C,D). These cancer intrinsic phenotypes was explored using an mice develop atypia of the mammary ductal epithelium analysis of variance (anova) with Fisher’s post hoc test at about 8 weeks of age, progressing to mammary of significance. H score was also used as a continuous intraepithelial neoplasia (MIN; analogous to human variable in anova of the progression series to charac- DCIS) at about 12 weeks of age with the development terize the changing levels of Notch1 protein expression of invasive carcinomas at about 16 weeks of age with increasing architectural and cytological atypia in 100% of female mice. The carcinomas have a over the whole cohort of lesions. similar morphology to human infiltrating ductal carcinomas.27 Mammary glands from 15 C3 ⁄ SV40T transgenic Results mice were evaluated and the expression of Notch was significantly elevated in hyperplastic lesions (n = 14), notch1 overexpression is an early event in the development and progression of breast cancer MIN (a DCIS-like lesion, n = 12) and invasive carcinoma (n = 7) compared with morphologically normal Although Notch pathway activation has been reported ducts (n = 15) (P < 0.05, Figure 2C,D). Furthermore, to induce murine mammary gland hyperplasia and ducts from syngeneic wild-type mice showed no neoplasia, there are relatively few data reporting the expression of Notch1 (Figure 2Ci). These data suggest expression of Notch pathway components in prolifer- that aberrant expression of Notch1 is an early event in ative and premalignant tissues of the human breast. both human and murine mammary neoplasia and

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. Notch is an early event in breast cancer development 291

Figure 1. Range of expression of Notch1 in invasive ductal carcinoma. A, Notch1– IgG control. B, Weak Notch1 expression, intensity score 1 (matching section from IgG control in A). C, Moderate Notch1 expression, intensity score 2. D, Strong Notch1 expression, intensity score 3 (H&E). progressively increases with the development of hyper- and 13% showed strong expression (3+) (Figure 1). The plasia and malignancy. mean H score in the IDC cohort was 85. In comparison, the mean Notch1 H score in morphologically normal ducts was 33.9, showing there is significantly higher high relative notch1 protein expression is a common event in breast cancer but is not relative expression of Notch1 in invasive carcinoma. We prognostic found a very similar mean Notch1 H score of 82.7 in cases of IDC within an independent cohort, the ‘progres- There have been a handful of reports describing the sion’ series. Using the mean H score to dichotomize the expression of Notch1 in human breast cancer.8,29,30 Notch1 data, 39% of the IDC cohort showed high relative However, to date the association with molecular Notch1 protein expression. subtypes of breast cancer has only been rarely explored. There was no prognostic significance of Notch1 We investigated the expression of Notch1 by immuno- expression, either on univariate survival analysis, histochemistry in 292 patients from a well-character- using Cox proportional hazards model or on Kaplan– ized, single-surgeon IDC cohort (n = 292). Once Meier survival analysis, in the cohort as a whole or in uninformative cores (folded or missing) were excluded, subgroup analysis, where possible. The Kaplan–Meier expression of Notch1 was assessable in 228 patients. curves for breast cancer recurrence and breast cancer- The clinicopathological features of this group are specific death are shown in Figure 3A,B. shown in Table 1. v2 analysis was performed to determine any associ- There was a range of Notch1 expression in this cohort; ation between high relative Notch1 protein expres- 4% of cases showed no expression of Notch1, whereas sion and clinicopathological characteristics of the 46% showed weak (1+), 37% showed moderate (2+), patients. There was no association between Notch1

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. 292 S J Zardawi et al.

Figure 2. Immunohistochemistry for Notch1 in human and murine mammary cancer development. A, Photomicrographs of (i) normal duct, (ii) usual ductal hyperplasia, (iii) ductal carcinoma in situ (DCIS) and (iv) invasive ductal carcinoma. B, (i) anova of Notch1 expression in human breast lesions and (ii) specifically in grades 1–3 DCIS. C, Photomicrographs of (i) normal ducts, (ii) hyperplasia, (iii) mammary intraepithelial neoplasia and (iv) invasive carcinoma in the C3 ⁄ SV40T mouse. D, anova of Notch1 expression in C3 ⁄ SV40T mouse. *P < 0.05; **P < 0.01. overexpression and tumour grade, tumour size, patient publically available breast cancer gene expression array age, lymph node status, hormone receptor status or datasets.24,25 The probes for NOTCH1 in either dataset Her-2 immunohistochemistry or HER-2 FISH (P all showed no prognostic significance (P = 0.58 for breast >0.05). However, ER) patients who were HER-2+ by cancer recurrence in the Wang cohort and P = 0.51 for FISH or immunohistochemistry showed a significant breast cancer-related death in the Naderi dataset, association with high Notch1 protein expression Figure 3C,D). (P = 0.02 FISH and P = 0.049 immunohistochemistry). In contrast, ER+, HER-2 FISH+ cases showed no notch1 is associated with the her2 subtype association with high Notch1 (P = 0.33 FISH and P = 0.99 immunohistochemistry). In view of our hypothesis that dysregulation of devel- In view of previous studies reporting that Notch1 opmental signalling pathways may be associated with overexpression is associated with a poor prognosis,8,29,30 specific breast cancer subtypes, the association we investigated the expression of NOTCH1 mRNA in two of Notch1 with each molecular subtype was then

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. Notch is an early event in breast cancer development 293

Figure 3. Relationship between expression of Notch1 and breast cancer patient outcome. Kaplan–Meier plots illustrating the relationship between Notch1 protein expression and A, breast cancer recurrence or B, breast cancer-specific death and between NOTCH1 mRNA expression and C, breast cancer recurrence in the Wang dataset23 and D, breast cancer-related death in the Naderi cohort.22 determined. v2 analysis demonstrated that high relative we report that high expression of Notch1 protein is a Notch1 protein expression was significantly associated very early event in both human and murine mammary with the HER-2 subtype (defined as ER) and PR), carcinogenesis. Elevated expression of Notch1 was HER-2+ FISH); P = 0.008. There was no association noted first in morphologically normal ducts in breast between Notch1 and any other subtype (Table 2). tissue containing invasive carcinoma and showed a Analysis of variance was then used to explore the progressive increase in lesions of UDH and DCIS. We relationship between the Notch1 H score (as a contin- also noted an increase in the expression of Notch1 from uous variable) and the breast cancer subtypes, using DCIS grade 1 to grade 2, with a further increased level our modified definition of the molecular subtypes. noted in grade 3 DCIS. A similar pattern was also Figure 4 shows that there is enrichment of high observed in preinvasive lesions from the C3 ⁄ SV40T relative Notch1 protein expression in the HER-2 mouse model of mammary carcinogenesis, which subtype, which is significantly elevated in this group shares significant homology with similar lesions in compared with all other subtypes (P < 0.05). the human breast. Interestingly, ducts from syngeneic wild-type mice showed no expression of Notch1, in Discussion contrast to that detected in morphologically normal ducts from the C3 ⁄ SV40T mice, suggesting that There has been considerable recent interest in the role increased Notch1 expression accompanies very early of Notch signalling in breast cancer development and events in the development of mammary carcinoma. progression. Bouras et al.3 have recently reported that There are relatively little previously published data Notch signalling regulates mammary stem cell function regarding a potential role for Notch early in breast and that constitutive activation of Notch1 in luminal cancer development. Although aberrant activation of precursor cells results in the development of hyper- the Notch signalling pathway results in the develop- plastic nodules and subsequently invasive cancer. Here ment of adenocarcinomas in murine models, and

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. 294 S J Zardawi et al.

Table 2. Associations of high relative Notch1 protein expres- to these reports postulating a significant role for Notch sion with breast cancer subtypes signalling in early breast cancer development and progression. Breast cancer Number (%) cases 8 subtype with high Notch1 v2 P-value In contrast to the earlier findings of Reedijk et al., who reported that high levels of NOTCH1 mRNA were Luminal A 54 ⁄ 141 (38) 0.40 associated with a poorer prognosis, which had even Luminal B 5 ⁄ 21 (23) 0.24 greater prognostic significance when combined with high levels of JAG1 mRNA, our data failed to demon- HER-2 12 ⁄ 17 (70) 0.008 strate any prognostic significance for high Notch1 protein expression. There are a number of possible Basal-like 11 ⁄ 31 (35) 0.84 reasons for these differences. Although the cohorts in Unclassified 2 ⁄ 14 (14) 0.09 both these studies were of comparable size, (n = 184 and 228, respectively), there were significant differences in cohort composition as evidenced by the clinicopathological variables. For example, the patients in one of the Reedijk cohorts had larger tumours (average size 37 mm compared with 21 mm in our cohort) as well as a higher proportion of grade 3 tumours (63% versus 44%). Furthermore, it has been demonstrated for JAG1 that there is a poor correlation between mRNA levels and protein expression,31 which may also be true for Notch1. To test this possibility, we performed in silico analysis of the prognostic significance of relative mRNA levels of NOTCH1 in two publically available gene expression array data for patient cohorts with similar clinicopathological characteristics to our cohort. In keeping with our immunohistochemical findings, we found no prognostic significance of high relative NOTCH1 mRNA expression in either cohort. Figure 4. Notch1 protein expression in breast cancer subtypes. Another novel finding of our study is a significant Analysis of variance demonstrates a statistically significant increase association between elevated expression of Notch1 and in Notch1 expression in the HER-2 subtype of breast cancer, the HER-2 subtype of breast carcinoma. There is some compared with all other molecular subtypes (basal-like, luminal A, experimental evidence to support a role of Notch luminal B and unclassified). *P < 0.05; **P < 0.01. pathway signalling in HER-2-expressing tumours. Chen et al.32 identified that a palindromic binding site within overexpression of activated forms of the Notch1 recep- the promoter region of HER-2 (ERBB-2) was identical to tor can transform normal human epithelial cells in vitro, that for RBPJj, which transports the intracellular only one report has previously supported a role in domain of Notch receptors to the nucleus, where it acts preinvasive human disease.11 In this study Farnie and as a site-specific DNA binding partner leading to tran- coworkers developed a novel method of primary scription of Hes1. Binding of RBPJj to the promoter culture of DCIS from human lesions, using techniques region of HER-2 resulted in increased transcriptional similar to the mammosphere assay. Elevated levels of activity, which was further augmented by addition of the Notch1 intracellular domain (N1ICD) were noted in Notch-IC. In the light of these data, it is possible that these mammospheres, when compared with those elevated Notch1 receptor expression may be linked derived from normal breast tissue. The growth of mechanistically to augmented HER-2 transcription. DCIS mammospheres could be inhibited by N-[N-(3, However, there are conflicting data regarding an 5-difluorophenacetyl)-l-alanyl]-S-phenylglycine t-butyl association between Notch1 and HER-2. Osipo et al.18 ester (DAPT), a c-secretase inhibitor specific to Notch in studies in human breast cancer cell lines demon- pathway activity. These authors also examined expres- strated an inverse relationship between levels of HER-2 sion of the N1ICD in a cohort of 50 patients with DCIS and Notch transcriptional activity, such that inhibition and found that elevated N1ICD was associated with a of HER-2 resulted in elevation of Notch activity, which higher risk of recurrence. Our data add further support was abrogated by a Notch inhibitor, c-secretase. It is

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. Notch is an early event in breast cancer development 295

interesting to note that this group also reported that Reedijk’s group defined the HER-2 subgroup as any inhibition of Notch activity (either by a c-secretase patients with positive HER-2 expression, basal-like inhibitor or by a Notch1 siRNA) increased sensitivity of tumours as those lacking ER and expressing CK5, cell lines to the growth-inhibitory effects of trast- whereas luminal cancers were simply defined as those uzumab, and suggested that inhibition of Notch expressing ER. These definitions are less stringent than signalling may be a promising therapeutic approach the surrogate molecular signatures as described by in cases of trastuzumab-resistant HER-2+ cancers. This Livasy,35 and more recently by Cheang et al.,23 and disparity with our findings is not entirely surprising, as have had a significant influence on the proportion of there are likely to be significant differences between patients assigned to each phenotypic subgroup. For Notch1 transcriptional activity as measured by a example, Reedijk’s definition of luminal excludes lumi- luciferase reporter gene in three breast cancer cell nal B patients (ER+ and HER-2+), whereas the HER-2 lines18 and Notch1 protein levels as assessed by group will include luminal B patients. A further immunohistochemistry in human breast tissue samples. confounder is the method of assessment of HER-2 This is also supported by a recent report by Rizzo et al.,33 status in Reedijk’s cohort, in that three different who found a poor correlation between Notch-induced methods were used to assess HER-2 status–polymerase transcriptional activity and Notch receptor levels. It is chain reaction, Southern blot and slot blot, whereas also interesting that a recent abstract presented by the HER-2 status in our cohort was performed using FISH, Osipo group18 reports that constitutively active Notch1 currently regarded as best practice of assessment of increased tyrosine phosphorylated ErbB-2 (HER-2) and HER-2 status.36 Validation of these findings in a larger reversed the inhibition of a c-secretase inhibitor,16 cohort would add substantially to our understanding of showing a direct positive relationship between Notch1 this relationship. and HER-2, in contrast to their recent paper,18 and In summary, these data indicate that there is an these data may provide evidence of a negative feedback association between aberrant expression of Notch1 and loop between Notch and HER-2. HER-2, and that high relative Notch1 protein expres- Yamaguchi et al.34 reported no evidence of growth sion is an early event in both human and murine inhibition of ERBB2 (HER-2)-positive cell lines follow- mammary development and progression. Although it ing treatment with siRNA targeting NOTCH1, remains to be determined whether high Notch1 recep- although they found that a NOTCH3 siRNA inhibited tor protein expression correlates with Notch pathway the growth of some ERBB2– cell lines. In contrast, a activity, our data combined with other recent positive association between Notch2 and HER-2 reports16,18 suggest that Notch signalling may present protein expression was reported by Florena et al.17 in a novel therapeutic target for the future management a small cohort of 98 invasive breast cancers. of breast cancer, particularly in the poor prognosis Our data are also supported by the recent findings of HER-2 subtype. Magnifico et al.,19 who have identified that Notch1 signalling directly regulates HER-2 expression levels in Acknowledgements tumour initiating cells, so that inhibition of Notch1 resulted in a reduction of HER-2 cell surface levels. The Cancer Institute NSW, National Health and There are also extensive data linking Notch signal- Medical Research Council of Australia, the RT Hall ling to the basal-like subtype of breast carcinoma. Trust, the Sydney Breast Cancer Foundation and the Notch pathway activity in a mammosphere assay, an Petre Foundation supported this work. The authors in vitro model of stem cell and early mammary wish to thank Ms Alice Boulhhourjian, who provided development, increased the differentiation of progenitor assistance with immunohistochemistry, Ms Anne cells into a myoepithelial lineage,15 while Rizzo et al.33 Holliday for data management and Ms Joanne Scorer demonstrated that in a series of breast cancer cell lines, for her assistance in editing of the manuscript. Notch receptor levels are highest in ER), HER-2 non- overexpressing cells, analogous to clinical triple-nega- References tive phenotype, a group that includes most basal-like breast cancers. The apparent discrepancies between 1. Zardawi SJ, O’Toole SA, Sutherland RL, Musgrove EA. these data and those reported here may be attributable Dysregulation of Hedgehog, Wnt and Notch signalling to examination of different components of the pathway, pathways in breast cancer. Histol. Histopathol. 2009; 24; 385–398. different parameters of expression, i.e. protein com- 2. Artavanis-Tsakonas S, Rand MD, Lake RJ. Notch signaling: cell pared with mRNA, and the different criteria employed fate control and signal integration in development. Science 1999; to assign patients to the different molecular subtypes. 284; 770–776.

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. 296 S J Zardawi et al.

3. Bouras T, Pal B, Vaillant F et al. Notch signaling regulates 21. Croquelois A, Blindenbacher A, Terracciano L et al. Inducible mammary stem cell function and luminal cell-fate commitment. inactivation of Notch1 causes nodular regenerative hyperplasia Cell Stem Cell 2008; 3; 429–441. in mice. Hepatology 2005; 41; 487–496. 4. Brennan K, Brown AM. Is there a role for Notch signalling in 22. Takeshita K, Satoh M, Ii M et al. Critical role of endothelial human breast cancer? Breast Cancer Res. 2003; 5; 69–75. Notch1 signaling in postnatal angiogenesis. Circ. Res. 2007; 5. Bray SJ. Notch signalling: a simple pathway becomes complex. 100; 70–78. Nat. Rev. Mol. Cell Biol. 2006; 7; 678–689. 23. Cheang MC, Voduc D, Bajdik C et al. Basal-like breast cancer 6. Iso T, Kedes L, Hamamori Y. HES and HERP families: multiple defined by five biomarkers has superior prognostic value than effectors of the Notch signaling pathway. J. Cell. Physiol. 2003; triple-negative phenotype. Clin. Cancer Res. 2008; 14; 1368– 194; 237–255. 1376. 7. Hu C, Dievart A, Lupien M, Calvo E, Tremblay G, Jolicoeur P. 24. Naderi A, Teschendorff AE, Barbosa-Morais NL et al. A gene- Overexpression of activated murine Notch1 and Notch3 in expression signature to predict survival in breast cancer across transgenic mice blocks mammary gland development and independent data sets. Oncogene 2007; 26; 1507–1516. induces mammary tumors. Am. J. Pathol. 2006; 168; 973– 25. Wang Y, Klijn JG, Zhang Y et al. Gene-expression profiles to 990. predict distant metastasis of lymph-node-negative primary breast 8. Reedijk M, Odorcic S, Chang L et al. High-level coexpression of cancer. Lancet 2005; 365; 671–679. JAG1 and NOTCH1 is observed in human breast cancer and is 26. Limma SG. Linear models for microarray data. In Gentlemen V, associated with poor overall survival. Cancer Res. 2005; 65; Carey S, Dudoit R, Irizarry R, Huber W eds. Bioinformatics and 8530–8537. computational biology solutions using R and Bioconductor. New 9. Soriano JV, Uyttendaele H, Kitajewski J, Montesano R. Expression York: Springer, 2005; 397–420. of an activated Notch4(int-3) oncoprotein disrupts morphogen- 27. Green JE, Shibata MA, Yoshidome K et al. The C3(1) T-⁄ SV40 esis and induces an invasive phenotype in mammary epithelial antigen transgenic mouse model of mammary cancer: ductal cells in vitro. Int. J. Cancer 2000; 86; 652–659. epithelial cell targeting with multistage progression to carci- 10. Stylianou S, Clarke RB, Brennan K. Aberrant activation of notch noma. Oncogene 2000; 19; 1020–1027. signaling in human breast cancer. Cancer Res. 2006; 66; 1517– 28. Oakes SR, Robertson FG, Kench JG et al. Loss of mammary 1525. epithelial prolactin receptor delays tumor formation by reducing 11. Farnie G, Clarke RB, Spence K et al. Novel cell culture technique cell proliferation in low-grade preinvasive lesions. Oncogene for primary ductal carcinoma in situ: role of Notch and epidermal 2007; 26; 543–553. growth factor receptor signaling pathways. J. Natl. Cancer Inst. 29. Dickson BC, Mulligan AM, Zhang H et al. High-level JAG1 mRNA 2007; 99; 616–627. and protein predict poor outcome in breast cancer. Mod. Pathol. 12. Perou CM, Sorlie T, Eisen MB et al. Molecular portraits of human 2007; 20; 685–693. breast tumours. Nature 2000; 406; 747–752. 30. Parr C, Watkins G, Jiang WG. The possible correlation of Notch-1 13. Sims AH, Howell A, Howell SJ, Clarke RB. Origins of breast and Notch-2 with clinical outcome and tumour clinicopatholog- cancer subtypes and therapeutic implications. Nat. Clin. Pract. ical parameters in human breast cancer. Int. J. Mol. Med. 2004; Oncol. 2007; 4; 516–525. 14; 779–786. 14. Gusterson B. Do ‘basal-like’ breast cancers really exist? Nat. Rev. 31. Reedijk M, Pinnaduwage D, Dickson BC et al. JAG1 expression is Cancer 2009; 9; 128–134. associated with a basal phenotype and recurrence in lymph 15. Dontu G, Abdallah WM, Foley JM et al. In vitro propagation and node-negative breast cancer. Breast Cancer Res. Treat. 2008; 111; transcriptional profiling of human mammary stem ⁄ progenitor 439–448. cells. Genes Dev. 2003; 17; 1253–1270. 32. Chen Y, Fischer WH, Gill GN. Regulation of the ERBB-2 promoter 16. Clementz AG, Osipo C. Notch-1 activates ErbB-2 through a by RBPJkappa and NOTCH. J. Biol. Chem. 1997; 272; 14110– PEA3-dependent mechanism. Cancer Res. 2009; 69; 362s. 14114. 17. Florena AM, Tripodo C, Guarnotta C et al. Associations between 33. Rizzo P, Miao H, D’Souza G et al. Cross-talk between notch and Notch-2, Akt-1 and HER2 ⁄ neu expression in invasive human the estrogen receptor in breast cancer suggests novel therapeutic breast cancer: a tissue microarray immunophenotypic analysis approaches. Cancer Res. 2008; 68; 5226–5235. on 98 patients. Pathobiology 2007; 74; 317–322. 34. Yamaguchi N, Oyama T, Ito E et al. NOTCH3 signaling pathway 18. Osipo C, Patel P, Rizzo P et al. ErbB-2 inhibition activates Notch- plays crucial roles in the proliferation of ErbB2-negative human 1 and sensitizes breast cancer cells to a gamma-secretase breast cancer cells. Cancer Res. 2008; 68; 1881–1888. inhibitor. Oncogene 2008; 27; 5019–5032. 35. Livasy CA, Karaca G, Nanda R et al. Phenotypic evaluation of the 19. Magnifico A, Albano L, Campaner S et al. Tumor-initiating cells basal-like subtype of invasive breast carcinoma. Mod. Pathol. of HER2-positive carcinoma cell lines express the highest 2006; 19; 264–271. oncoprotein levels and are sensitive to trastuzumab. Clin. Cancer 36. Wolff AC, Hammond ME, Schwartz JN et al. American Society Res. 2009; 15; 2010–2021. of Clinical Oncology ⁄ College of American Pathologists guide- 20. Millar EK, Anderson LR, McNeil CM et al. BAG-1 predicts patient line recommendations for human epidermal growth factor outcome and tamoxifen responsiveness in ER-positive invasive receptor 2 testing in breast cancer. J. Clin. Oncol. 2007; 25; ductal carcinoma of the breast. Br. J. Cancer 2009; 100; 123–133. 118–145.

2010 The Authors. Journal compilation 2010 Blackwell Publishing Ltd, Histopathology, 56, 286–296. Cancer Research Article Epidemiology, Biomarkers & Prevention Cytoplasmic Localization of β-Catenin is a Marker of Poor Outcome in Breast Cancer Patients

Elena López-Knowles1, Sarah J. Zardawi1, Catriona M. McNeil1,3,4,5, Ewan K.A. Millar1,6,7,8, Paul Crea10, Elizabeth A. Musgrove1,9, Robert L. Sutherland1,9, and Sandra A. O'Toole1,2,9,11

Abstract β-catenin is involved in cell adhesion through catenin-cadherin complexes and as a transcriptional regulator in the Wnt signaling pathway. Its deregulation is important in the genesis of a number of human malignancies, particularly colorectal cancer. A range of studies has been undertaken in breast cancer, with contradictory associations reported among β-catenin expression, clinicopathologic variables, and disease outcome. We undertook an immunohistochemical study measuring the levels and subcellular localization of β- catenin in 292 invasive ductal breast cancers with known treatment and outcome. No association with breast cancer–specific death was observed for cytoplasmic or membrane expression alone; however, a continuous score representing both locations (membrane minus cytoplasmic expression: MTC score) was associated with a worse outcome in univariate analysis (P = 0.004), and approached significance in a multivariate analysis model that included lymph node, progesterone receptor (PR), and HER2 status (P = 0.054). Therefore, the MTC score was used for further statistical analyses due to the importance of both the subcellular location and the levels of expression of β-catenin. An association was identified between high cytoplasmic expression (low MTC score), and high tumor grade (P = 0.004), positive Ki67 (P = 0.005), negative estrogen receptor (ER) (P = 0.005), positive HER2 (P = 0.04) status, and an active phosphoinositide 3-kinase pathway (P = 0.005), measured as PIK3CA mutations (P = 0.05) or PTEN loss (P = 0.05). Low cytoplasmic expression (high MTC score) was associated with the luminal A subtype (P = 0.004). In conclusion, a low β-catenin MTC score is associated with an adverse outcome in breast cancer, which may be of mechanistic significance in the disease process. Cancer Epidemiol Biomakers Prev; 19(1); 301–9. ©2010 AACR.

Introduction status, are useful in guiding therapeutic decision making, but there is a pressing need to develop new biomarkers Breast cancer is the most common invasive cancer in and therapeutic strategies to combat the disease. women in industrialized nations and is second only to lung Numerous prognostic factors have been evaluated in cancer as the leading cause of cancer-related death in breast cancer patients to predict clinical outcome, and women. Current treatments, which are largely determined there are multiple lines of evidence that suggest an impor- by the estrogen receptor, HER2 status, or clinicopathologic tant role for β-catenin in breast cancer. β-catenin is an on- variables, such as tumor size and grade, and lymph node cogene, and its dysregulation or mutational activation can lead to cancer (1). In breast cancer, mutations are rare (2), Authors' Affiliations: 1Cancer Research Program, Garvan Institute of Medical Research; 2Discipline of Pathology, Faculty of Medicine, unlike in colon or hepatocellular cancers, but the level of University of Sydney; 3Department of Medical Oncology, Westmead expression or activation of β-catenin may be associated Hospital, Sydney, New South Wales, Australia; 4Breast Cancer Institute with breast cancer progression. Previous data, however, of New South Wales and 5Western Clinical School, University of Sydney, Westmead Hospital, Westmead, New South Wales, Australia; have presented mixed results about the subcellular local- 6Department of Anatomical Pathology, South Eastern Area Laboratory ization of β-catenin, and its relationship with clinicopath- Service, St George Hospital, Kogarah, New South Wales, Australia; 7Department of Pathology, School of Medicine, University of Western ologic variables and disease outcome. Sydney, Campbelltown, Australia; 8School of Medical Sciences, Faculty β-catenin is a critical component of cadherin-based of Medicine and 9St Vincent's Clinical School, Faculty of Medicine, cell-cell adhesion, has a central role in transcriptional University of New South Wales, Kensington, New South Wales, Australia; 10Department of Surgical Oncology, St Vincent's Hospital, Darlinghurst, regulation in the Wnt signaling pathway, and is an impor- New South Wales, Australia; and 11Department of Anatomical Pathology, tant intermediate in many other signal transduction Royal Prince Alfred Hospital, Camperdown, New South Wales, Australia pathways, such as the phosphoinositide 3-kinase(PI3K)/ Corresponding Author: Sandra A. O'Toole, Cancer Research Program, AKT pathway. β-catenin is located at the cell membrane, Garvan Institute of Medical Research, 384 Victoria Street, Darlinghurst, New South Wales 2010, Australia. Phone: 612-9295-8338; Fax: 612- and the cytoplasm and/or nucleus. At the cell membrane, it 9295-8321. E-mail: [email protected] isboundtothecytoplasmicdomainoftypeIcadherins,and doi: 10.1158/1055-9965.EPI-09-0741 is essential for the structural organization and function ©2010 American Association for Cancer Research. of cadherins by linking through α-catenin to the actin

www.aacrjournals.org 301 López-Knowles et al.

cytoskeleton (3). This complex is stabilized by serine/ but only breast cancer–related deaths (18%) were con- threonine phosphorylation of β-catenin (4). Other phos- sidered for survival analyses. phorylation events, however, such as tyrosine phosphorylation of β-catenin by epidermal growth factor Tissue Microarray Construction receptors or Src, among others, lead to its dissociation A total of 18 tissue microarrays containing two cores from the adherens complex and transfer to the cyto- of each tumor sample were constructed from the for- plasm (5). Cytosolic β-catenin is subsequently degraded malin-fixed, paraffin-embedded tumor material from each patient in the cohort, as previously reported. or translocated to the nucleus. In the absence of Wnt The tissue microarrays were produced with the use of ligands, β-catenin binds to a complex formed by glyco- the MTA-1 manual tissue arrayer (Beecher Instruments, gen synthase kinase 3β (GSK3β), adenomatous polypo- Woodland, CA). siscoli,andaxin(6).Thiscomplexfacilitatesthe β phosphorylation of the NH2 terminus of -catenin by Immunohistochemistry β GSK3 or other kinases, targeting it for degradation by Three micron sections of each tissue microarray were the ubiquitin-proteasome pathway (7). Activation of cut, deparaffinized, and used for immunostaining. The Wnt signaling is initiated by Wnt ligands binding to β-catenin mouse monoclonal antibody (BD Transduction two receptor molecules, frizzled proteins and lipoprotein Laboratories, Franklin Lakes, NJ) was used at a 1:200 di- receptor–related proteins 5 and 6 (LRP-5/6). This leads to lution for 60 min at room temperature. Antigen retrieval GSK3β inactivation and β-catenin accumulation in the was done with the use of Dako solution (pH 6.0; S1699, cytoplasm with translocation to the nucleus. There it in- Dako, Carpentaria, CA) for 30 min in a water bath. A Dako teracts with TCF/LEF to control transcription of down- autostainer was used for immunostaining (Dako). Reac- stream target genes, such as cyclin D1 and c-Myc. tions were developed with the use of diaminobenzidine, β-catenin stability is also regulated by numerous Wnt- and sections were hematoxylin counterstained. Colon can- independent mechanisms, and can be modulated by cer tissues were used as a positive control and invasive ErbB2, p53, and Pin1 (8), among others. lobular breast carcinoma as a negative control. Immuno- The aim of this study was to evaluate β-catenin expres- histochemistry protocols, and scoring and analysis for sion in a cohort of breast cancer patients, and investigate p27Kip1,p21Cip1/Waf1, cyclin E1, cyclin D1, and Ki67 are the link between subcellular localization and levels of described elsewhere (9). Protocols, scoring, and analysis β-catenin expression with known clinicopathologic vari- of the PI3K/AKT pathway are described in Lopez- ables and outcome, as well as the PI3K/AKT signal Knowles et al. (11), and of Notch 1 in Zardawi et al. (12). transduction pathway and Notch1 as a Wnt pathway downstream marker. Immunohistochemical Scoring Scoring was completed by a specialist breast pathologist (S.A.O'T.) and a scientist (S.J.Z.) blinded to the clin- Materials and Methods ical and pathologic information; in cases of discrepancy a consensus was reached by conferencing. All tumors Patients and Tumors were assessed for both intensity (1+ to 3+), and the pro- Cases were drawn from the St Vincent's Campus portion of cells staining positive in the nucleus, cyto- Outcome Cohort, which comprised 292 invasive ductal plasm, and cell membrane. A histoscore (H) was calculated by multiplying the percentage of positively carcinomas of the breast from patients treated by a sin- stained cells with each category of staining intensity. gle surgeon (P.C.) between February 1992 and August β-catenin expression was found predominantly in the 2002 at St Vincent's Hospital, Sydney, Australia. Ethics membrane and cytoplasm of cells, with only three cases approval was granted for the use of pathology speci- showing detectable nuclear expression; hence, we de- mens and cognate clinicopathologic data (Human Re- scribed a histoscore value for the membrane expression search Ethics Committee of St Vincent's Hospital, (H1) and the cytoplasmic expression (H2) indepen- Sydney, Australia). A more detailed description of the dently, and then derived a third histoscore (H3), which clinicopathologic characteristics of the cohort is pub- represents both categories of expression as one contin- lished elsewhere (9). Forty percent of tumors were uous variable (membrane minus cytoplasmic histoscore) >20 mm, 45% were grade >2, 43% were lymph node as a measure of subcellular localization, the "membrane positive, 68% were ER positive, 57% were PR positive, in situ to cytoplasmic score" (MTC). When this continuous value and 18% were HER2 positive by fluorescence hy- is positive, the tissue shows predominantly membranous bridization (>2.2 ratio of HER2:chromosome 17 centro- expression, and when the value is negative, the tissue some; ref. 10). Median age was 54 y, and patients were shows mainly cytoplasmic expression (examples shown treated with endocrine therapy (49%), chemotherapy in Fig. 1). (38%), or both (24%). Cases were prospectively followed-up for a median of 64 mo, and the outcome Statistical Analyses events measured were recurrence (local or distant; To assess the independence of two categorical vari- 25%) and metastasis (23%); all deaths were recorded ables, the χ2 test was applied when the variables

302 Cancer Epidemiol Biomarkers Prev; 19(1) January 2010 Cancer Epidemiology, Biomarkers & Prevention β-Catenin in Breast Cancer

Figure 1. Patterns of expression of β-catenin expression in human tissues. A. Normal colonic crypts show a membranous pattern of expression. B. Invasive colon adenocarcinoma showing predominantly nuclear expression. C. Invasive lobular breast carcinoma negative control lacks expression of β-catenin. D. Normal breast duct with membranous expression in the basal cell layer. E. Weak predominantly membranous expression in invasive ductal carcinoma. F. Strong predominantly membranous expression in invasive ductal carcinoma. G. Predominantly cytoplasmic expression in invasive ductal carcinoma. H. Strong membranous and cytoplasmic expression in invasive ductal carcinoma. All images, ×400 magnification; hematoxylin counterstained.

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 19(1) January 2010 303 López-Knowles et al.

Figure 2. Kaplan-Meier curves illustrating the relationship between β-catenin expression, and disease progression and treatment. A. Recurrence, metastasis, and cancer-specific death stratified by a cytoplasmic (C; β-catenin MTC score expression <0) or membrane (M; β-catenin MTC score expression >0) expression. The number of patients at risk, and the HRs, 95% CIs, and P-value of the association by log rank testing are shown. B. Breast cancer–specific death in patients treated with adjuvant endocrine (tamoxifen) therapy or chemotherapy, and in untreated patients, stratified by cytoplasmic or membrane staining.

were dichotomous, the Mann-Whitney test was applied Because membranous expression did not show a normal when one variable was dichotomous and the other distribution and no optimal cut point was observed, the continuous, and the Spearman rank correlation was median was used as a threshold for survival analysis. applied when both variables were continuous. Kaplan- The average cytoplasmic H2 score was 104 and the me- Meier survival curves and Cox proportional hazard dian was 80. Again, cytoplasmic expression did not ratios (HR) were estimated to obtain risks of recurrence, show a normal distribution and no optimal cut point metastasis, and death from breast cancer after adjusting was identified; therefore, the median was used as a for other confounding variables. Results were consid- threshold for survival analysis. When one continuous ered significant at the two-sided P < 0.05 level. Statview H3 score was calculated for β-catenin expression by version 5.0 was used for the analysis (Abacus Systems, subtracting the cytoplasmic H2 score from the membra- Berkeley, CA). nous H1 score (the MTC score), the average expression of β-catenin was 21 and the median was 10. This vari- Results able had a normal distribution across the whole cohort of 276 breast cancers. To study the association with out- β-Catenin Expression come, we dichotomized the data with a value <0 of the β Normal colon and adenocarcinoma of the colon were continuous -catenin expression; a score <0 would be used as controls for β-catenin expression. Normal colonic mainly cytoplasmic expression, and one >0 would rep- mucosa showed membranous expression of β-catenin resent predominantly membranous expression. To test (Fig. 1A), whereas adenocarcinoma showed strong nu- other means of obtaining one continuous variable that clear and cytoplasmic expression (Fig. 1B). Invasive lob- represented both the cytoplasmic and membrane stain- ular carcinoma of the breast was used as a negative ing observed, two other variables were calculated. First, control, showing complete absence of expression the ratio dividing the membrane and the cytoplasm (Fig. 1C). Of the original 292 invasive ductal carcino- scores was calculated, and the scores dichotomized at mas, 276 tumors were able to be evaluated once unin- the median. Second, the membrane plus the cytoplasmic formative cores (e.g. folded or missing) were excluded. scores were added, dichotomizing the data again at the β-catenin showed moderate membranous expression median score. Both these scores showed no significant as- in the myoepithelial/basal cell layer of normal ducts sociation with outcome (data not shown). (Fig.1D).Incontrast,β-catenin expression was observed in both the cytoplasm and the membrane com- β-Catenin Expression, Disease Outcome, and partments of carcinoma cells. Four percent of cases Treatment (12 of 276) showed only membranous expression Membranous and cytoplasmic expression of β-catenin (Fig. 1E), 5% of cases (15 of 276) showed only cytoplas- alone did not show an association with breast cancer– mic expression, and the remainder showed expression specific death [HR, 0.88; 95% confidence interval (95% in both subcellular compartments (Fig. 1F-H). Nuclear CI), 0.507-1.554; P = 0.67 and HR, 1.493; 95% CI, 0.853- expression was seen in only three cases. The average 2.613; P = 0.16, respectively], and only cytoplasmic ex- membrane H1 score was 125 and the median was 120. pression showed a weak association with recurrence

304 Cancer Epidemiol Biomarkers Prev; 19(1) January 2010 Cancer Epidemiology, Biomarkers & Prevention β-Catenin in Breast Cancer

P (HR, 1.79; 95% CI, 1.122-2.854; = 0.01) and metastasis Table 2. Clinicopathologic association of β- P (HR, 1.79; 95% CI, 1.097-2.932; = 0.02). However, in a catenin expression univariate analysis a shift to high cytoplasmic expression, measured as a negative MTC score, was associated with Membrane Cytoplasmic MTC recurrence (HR, 1.98; 95% CI, 1.25-3.14; P = 0.003), metas- H1 H2 H3 P tasis (HR, 2.16; 95% CI, 1.33-3.51; = 0.002), and breast P P P cancer–specific death (HR, 2.28; 95% CI, 1.31-3.98; P = Tumor size 0.004; Fig. 2A). In a multivariate analysis incorporating <20 mm 164 all clinicopathologic variables reported in Table 1, β-cate- >20 mm* 112 0.0054* 0.0009* 0.7355 nin approached independent predictor status for breast Tumor grade cancer–related death in a model that included lymph ≤2 150 node status, PR, and HER2 status (HR, 1.77; 95% CI, >2* 126 0.0153* <0.0001* 0.0042* 0.98-3.18; P = 0.054). LN status Next, patients were distributed into three treatment 0 152 subgroups (i.e., endocrine therapy, chemotherapy, and >0* 121 0.0235* 0.0348* 0.9403 no adjuvant treatment), and associations between Age, y β-catenin and breast cancer–specific death were assessed. <50 102 In patients treated with endocrine therapy (n = 140), β- >50 174 0.1166 0.9882 0.1783 catenin shift to cytoplasmic expression was associated ER with breast cancer–specific death (HR, 3.11; 95% CI, Negative*86 0.0456* <0.0001* 0.0052* 1.41-6.86; P = 0.005). There was no such association, how- Positive 188 ever, in the chemotherapy-treated patients (n = 107; (HR, PR 1.56; 95% CI, 0.73-3.34; P = 0.245) or in the patients who Negative* 117 0.0116* 0.0005* 0.5411 did not receive adjuvant treatment (n = 99; HR, 2.76; 95% Positive 158 CI, 0.87-8.77; P = 0.08; Fig. 2B). HER2 Negative 217 β -Catenin Expression and Clinicopathologic Positive* 51 0.5167 0.1511 0.0412* Variables Ki67 β The association between -catenin expression in the Negative 128 cytoplasm, membrane, or both, and the clinicopathologic Positive* 120 0.4825 <0.0001* 0.0054* variables was evaluated (Table 2). High membranous expression was associated with a large tumor size >20 mm *Significantly associated with β-catenin expression. (P = 0.005), a high tumor grade >2 (P = 0.01), and positive

Table 1. Univariate and multivariate outcome lymph node status (P = 0.02). It was also associated with analysis for clinicopathologic variables, ER-negative (P = 0.04) and PR-negative (P = 0.01) status. hormone receptor status, and β-catenin High cytoplasmic expression was also associated with a large tumor size >20 mm (P = 0.0009), a high tumor grade Variable HR 95% CI P >2 (P < 0.0001), and positive lymph node status (P = 0.03). It was also associated with ER-negative (P < Univariate Grade >2 3.520 1.930-6.418 <0.0001 0.0001) and PR-negative (P = 0.0005) status, and a high Size >20 mm 2.468 1.415-4.304 0.0015 proliferation rate, as assessed by Ki67 (P < 0.0001). The Lymph 3.691 2.025-6.729 <0.0001 MTC score showed an association between cytoplasmic node status >0 expression, and higher grade (P = 0.004), ER-negative Age >50 y 1.427 0.799-2.551 0.2297 (P = 0.005) and HER2-positive (P = 0.04) status, and pos- ER+ 0.300 0.172-0.524 <0.0001 itive Ki67 (P = 0.005). PR+ 0.170 0.087-0.333 <0.0001 HER2 FISH+ 3.491 1.956-6.229 <0.0001 β MTC H3 <0 2.28 1.309-3.980 0.0036 -Catenin Expression and Cell Cycle Markers β Multivariate Lymph node 3.295 1.781-6.098 0.0001 -catenin expression increases with cell cycle progres- status >0 sion, and is located in the cytoplasm and nucleus during β PR+ 0.213 0.107-0.422 <0.0001 the cell cycle. Because a role for -catenin in cell prolifer- HER2 FISH+ 2.429 1.293-4.564 0.0058 ation and apoptosis has been previously identified (13), MTC H3 <0 1.775 0.989-3.187 0.0545 and expression was associated with Ki67 in our cohort, we investigated potential relationships between β-cate- Abbreviation: FISH, fluorescence in situ hybridization. nin expression and several cell cycle proteins of importance in breast cancer (Table 3).

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 19(1) January 2010 305 López-Knowles et al.

Table 3. Association of β-catenin expression and molecular markers

Membrane H1 Cytoplasmic H2 MTC H3 N. PPP

Cell cycle markers Cyclin D1 Rho −0.113 Rho −0.154 Rho 0.019 0.0625 0.0108* 0.7568 Cyclin E1 Rho 0.344 Rho 0.364 Rho 0.217 <0.0001* <0.0001* 0.0003* p21 Rho 0.041 Rho −0.018 Rho 0.086 0.4971 0.7604 0.1515 p27 Rho −0.231 Rho −0.363 Rho 0.124 0.0001* <0.0001* 0.0407* Notch1 <67% 133 >67%*930.0077* <0.0001* 0.1731 PI3K/AKT pathway PIK3CA Wt 146 Mut* 12 0.1806 0.4805 0.0528* PIK3CA No change 174 Amp 28 0.9375 0.1885 0.1342 PTEN H >0 181 H = 0*700.0232* <0.0001* 0.0599* pAKT H <22 191 H >22 60 0.5274 0.8769 0.7144 PI3K/AKT Wt 53 Altered* 134 0.2798 0.0989 0.0057*

*Significantly associated with β-catenin expression.

An association was identified between membranous Cross-talk between the Wnt and Notch pathways expression of β-catenin and high cyclin E1 (P < can occur through a number of mechanisms: through 0.0001) and low p27Kip1 (P = 0.0001). An association the physical binding of Notch1 to β-catenin (14); was also identified between cytoplasmic expression of through their association with common cofactors (15); β-catenin, and low cyclin D1 (P = 0.01), high cyclin E1 through GSK3β phosphorylation of Notch, modulating (P < 0.0001), and low p27Kip1 (P < 0.0001) expression its transcriptional activity (16); and through β-catenin– levels. When the MTC score was analyzed, an associa- mediated transcriptional activation of the Notch ligand tion was confirmed between predominantly membra- Jagged1 (17). We identified an association between nous expression and high cyclin E1 (P = 0.0003), and high cytoplasmic (P < 0.0001) and high membrane predominantly cytoplasmic expression and loss of (P = 0.007) β-catenin expression and high Notch1 p27Kip1 (P = 0.04). expression. PIK3CA mutations and copy number, PTEN expres- β-Catenin Expression and Signal Transduction sion, a negative regulator of the PI3K pathway, and Pathways pAKT expression were previously characterized in β-catenin is involved in Wnt signaling and is down- this cohort of patients (11). A positive association was stream of GSK3β, which is also involved in the PI3K/ identified between PIK3CA mutations and cytoplasmic AKT pathway. We therefore next assessed the association location of β-catenin, measured by the MTC score (P = of Notch, as an interactor of the Wnt signaling pathway, 0.05). A positive association was also identified between and pAKT, PTEN, and PIK3CA,ascomponentsof loss of PTEN expression, and high cytoplasmic expression the PI3K/AKT pathway, with β-catenin expression (P < 0.0001) and high membranous expression (P = 0.02) (Table 3). of β-catenin. The MTC score confirmed this association

306 Cancer Epidemiol Biomarkers Prev; 19(1) January 2010 Cancer Epidemiology, Biomarkers & Prevention β-Catenin in Breast Cancer

between PTEN loss and cytoplasmic location of β-catenin variables, outcome, and treatment. A shift to high cytoplas- (P = 0.05). Finally, a surrogate variable for an active PI3K/ mic β-catenin expression was associated with high-grade AKT pathway, measured as an alteration of at least one tumors, high proliferation rate, ER-negative and HER2- member of the PI3K/AKT pathway studied, was asso- positive status, and worse outcome. The latter was signifi- ciated with cytoplasmic location of β-catenin, as mea- cant in univariate analysis, and approached independence sured by the MTC score (P = 0.005). as a prognostic marker of breast cancer–specific death (P = Therefore, an active PI3K/AKT pathway, measured 0.054) in a model including lymph node, PR, and HER2 by PIK3CA mutations, PTEN loss, or an alteration status. in any of the three variables studied in the pathway, β-catenin expression is localized in the nucleus in is associated with β-catenin protein expression located tissues like colorectal cancer (19); however, in the breast predominantly in the cytoplasmic compartment of the only two studies have shown nuclear localization tissue. (20, 21), whereas most publications report a cytoplasmic or membranous pattern of expression in breast cancer cells. β-Catenin Expression and Breast Cancer Subtypes Previously, β-catenin cytoplasmic expression was reported Breast cancer patients were distributed into five molec- to be associated with worse outcome in three studies (20, ular phenotypes or subtypes with the use of criteria 22, 23), whereas others have identified no such relationship similar to those recently described by Cheang (18), but (24-26). Two studies have reported an association between with the use of fluorescence in situ hybridization to deter- cytoplasmic expression and ER-positive status (23, 27), mine HER2 status (9, 10). The five phenotypes were lumi- whereas others have found no association (25, 28). Contra- nal A (ER+ and/or PR+, HER2−), luminal B (ER+ and/or dictory results are also reported for its associations with PR+, HER2+), HER2 (ER− and PR−, HER2+), “basal-like” tumor size and grade, patient age, and lymph node status. (ER−,PR−,HER2−, CK5/6+ and/or EGFR+), and un- Confirming the findings of most previous studies, classified (negative for all markers). β-catenin expression was identified only in the membrane When cases were distributed into these five pheno- and cytoplasm of the breast cancer tissues analyzed. The types, a high membranous β-catenin H1 score was asso- molecular mechanisms that allow the transfer of mem- ciated with the basal-like (P = 0.008) subtype. A high brane bound to cytoplasmic localization is largely un- cytoplasmic β-catenin H2 score was also associated with known, but could be due to changes in adenomatous the basal-like (P < 0.0001) and HER2 (P = 0.01) subtypes, polyposis coli (APC) or axin, which maintains high β- and a low cytoplasmic expression (P < 0.0001) was catenin levels in the cytoplasm (29), APC shuttling associated with the luminal A subtype. A high MTC (30), or galectin-3 recruiting β-catenin from cell adhe- score was also associated with the luminal A subtype sion to signaling pathways (31). Low APC expression (P = 0.004; Table 4). is associated with β-catenin overexpression and localization in the cytoplasm in breast cancer cells (32). In Discussion breast cancer, APC (33) and Axin1 mutations are rare (32), but other events, such as methylation, as described β-catenin has previously been shown to be a potential in the APC promoter of invasive breast cancer patients prognostic marker in breast cancer; however, the associations (34), could regulate APC levels and thus β-catenin ac- with clinicopathologic variables and outcome measures cumulation. Also, APC protein truncation leading to reported in different studies are discrepant. We therefore an increase in cytosolic levels of β-catenin have been evaluated β-catenin expression in a cohort of invasive duc- identified in breast cancer cell lines (35). tal breast cancers with known patient clinicopathologic When analyzing downstream markers, such as c-Myc and cyclin D1, no association was identified. Two studies β β have shown an association between cyclin D1 and - Table 4. -catenin expression and breast catenin expression (20, 32), although two further stud- cancer subtypes ies have not identified this association (25, 36). This lack of consensus could be due to cytoplasmic rather Membrane Cytoplasmic MTC than nuclear localization of the protein. H1 H2 H3 One critical issue related to previous studies is the PPP selection of either membrane or cytoplasmic expression for scoring and statistical analysis. A review of previous Basal-like 31 0.0079 (+) <0.0001 (+) 0.1796 literature shows marked discrepancies in the selection of HER2 24 0.3973 0.0135 (+) 0.1155 a threshold, and the scoring of either of the two locations Luminal A 169 0.1037 <0.0001 (−) 0.0045 (−) to measure β-catenin activation or aberrant expression. Luminal B 27 0.1085 0.6455 0.0974 We have calculated a new score, which evaluates the Unclassified 16 0.4595 0.7109 0.9654 expression in both locations without defining arbitrary NOTE: (+) indicates a positive correlation and (−) a negative thresholds. Both locations are then taken into consider- correlation. ation by calculating a MTC histoscore. This serves as a surrogate marker for β-catenin localization, with low

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 19(1) January 2010 307 López-Knowles et al.

MTC expression measuring a decrease in membrane ex- whole. This trend supports preliminary data by Hiscox pression and an increase in cytoplasmic expression. et al., which showed that tamoxifen-resistant breast PIK3CA mutations, PTEN loss, and the overall PI3K cancer cells had increased cytoplasmic accumulation of active pathway lead to a decrease in GSK3β and thus β-catenin (39). Taken together, these data suggest a po- an increase in β-catenin cytoplasmic accumulation. The tential role for β-catenin in endocrine resistance, but this association of PIK3CA mutations, PTEN loss, and overall requires further investigation, particularly in an appro- PI3K pathway activation with a shift of expression to the priate clinical trial cohort. cytoplasm, and the lack of association between the In conclusion, our study shows that a shift in subcellular activated pathway and either membrane or cytoplasmic localization of β-catenin is associated with alterations in expression alone, highlight the strength of this combined cell cycle regulatory proteins and signaling transduction variable to characterize the localization of β-catenin pathways in clinical breast cancer samples. Furthermore, expression. Recently, an important role for PI3K/AKT/ cytoplasmic accumulation of β-catenin is associated with β catenin pathway has been identified in the regulation poor prognosis in invasive ductal carcinoma. It is possible of malignant stem/progenitor cell populations, suggest- that in a larger cohort, true independence may be estab- ing that inhibiting this pathway could be an effective lished, and further efforts are needed to validate these method to target tumorigenic breast cancer cells (37). findings in independent breast cancer cohorts. No association with breast cancer–specific death was observed with either membrane or cytoplasmic expression Disclosure of Potential Conflicts of Interest alone; however, a shift to cytoplasmic expression, as defined by a MTC score <0, is significantly associated with No potential conflicts of interest were disclosed. recurrence, metastasis, and breast cancer–specific death, highlighting the importance of β-catenin localization Acknowledgments and the pathways that converge on this protein in breast cancer. A shift to high cytoplasmic expression also shows a We thank Alice Boulghourjian for assistance with pro- strong trend toward being an independent prognostic cessing the tissue, Anne Holliday for data management, marker of breast cancer–specific death (P = 0.054). and Assoc. Prof. Adrienne Morey for HER2 fluorescence These observations point to the importance of the mem- in situ hybridization analysis. brane to the cytoplasm shift score. This measures a decrease in β-catenin at the adherens junctions involved in Grant Support cell adhesion, which could have an effect on cell invasion, tumor metastasis, and the increase in β-catenin cytoplas- National Health and Medical Research Council of Aus- mic accumulation due to the activation of pathways con- tralia, the Cancer Institute NSW, the Petre Foundation, verging on this protein (such as the PI3K/AKT pathway). and the R.T. Hall Trust. This accumulation can lead to activation of numerous tar- The costs of publication of this article were defrayed in get genes that play essential roles in normal development part by the payment of page charges. This article must and in human cancers (38). therefore be hereby marked advertisement in accordance Interestingly, we observed an association between β- with 18 U.S.C. Section 1734 solely to indicate this fact. catenin cytoplasmic expression and poor outcome in pa- Received 7/26/09; revised 9/29/09; accepted 10/28/09; tients treated with endocrine therapy in the cohort as a published online 1/7/10.

References 1. Polakis P. Wnt signaling and cancer. Genes Dev 2000;14:1837–51. 9. Millar EK, Dean JL, McNeil CM, et al. Cyclin D1b protein expression 2. Candidus S, Bischoff P, Becker KF, Hofler H. No evidence for muta- in breast cancer is independent of cyclin D1a and associated with tions in the α- and β-catenin genes in human gastric and breast car- poor disease outcome. Oncogene 2009;28:1812–20. cinomas. Cancer Res 1996;56:49–52. 10. Wolff AC, Hammond ME, Schwartz JN, et al. American Society of 3. Jamora C, Fuchs E. Intercellular adhesion, signalling and the cyto- Clinical Oncology/College of American Pathologists guideline re- skeleton. Nat Cell Biol 2002;4:E101–8. commendations for human epidermal growth factor receptor 2 test- 4. Bek S, Kemler R. Protein kinase CKII regulates the interaction of β- ing in breast cancer. J Clin Oncol 2007;25:118–45. catenin with α-catenin and its protein stability. J Cell Sci 2002;115: 11. Lopez-Knowles E, O'Toole SA, McNeil CM, et al. PI3K pathway ac- 4743–53. tivation in breast cancer is associated with the basal-like phenotype 5. Nelson WJ, Nusse R. Convergence of Wnt, β-catenin, and cadherin and cancer-specific mortality. Int J Cancer 2009. pathways. Science (New York, NY) 2004;303:1483–7. 12. Zardawi SJ, Zardawi I, McNeil C, et al. High Notch1 protein expres- 6. Rubinfeld B, Albert I, Porfiri E, et al. Binding of GSK3β to the APC-β- sion is an early event in breast cancer development and is associ- catenin complex and regulation of complex assembly. Science (New ated with the HER-2 molecular subtype. Histopathol. In press. York, NY) 1996;272:1023–6. 13. Olmeda D, Castel S, Vilaro S, Cano A. β-Catenin regulation during 7. Aberle H, Bauer A, Stappert J, Kispert A, Kemler R. β-Catenin is a target the cell cycle: implications in G2/M and apoptosis. Mol Biol Cell for the ubiquitin-proteasome pathway. EMBO J 1997;16:3797–804. 2003;14:2844–60. 8. Hatsell S, Rowlands T, Hiremath M, Cowin P. β-Catenin and Tcfs in 14. Hayward P, Balayo T, Martinez Arias A. Notch synergizes with axin to mammary development and cancer. J Mammary Gland Biol Neopla- regulate the activity of armadillo in Drosophila. Dev Dyn 2006;235: sia 2003;8:145–58. 2656–66.

308 Cancer Epidemiol Biomarkers Prev; 19(1) January 2010 Cancer Epidemiology, Biomarkers & Prevention β-Catenin in Breast Cancer

15. Alves-Guerra MC, Ronchini C, Capobianco AJ. Mastermind-like 1 is a 27. Nakopoulou L, Gakiopoulou H, Keramopoulos A, et al. c-met tyro- specific coactivator of β-catenin transcription activation and is essen- sine kinase receptor expression is associated with abnormal β-cate- tial for colon carcinoma cell survival. Cancer Res 2007;67:8690–8. nin expression and favourable prognostic factors in invasive breast 16. Espinosa L, Ingles-Esteve J, Aguilera C, Bigas A. Phosphorylation by carcinoma. Histopathology 2000;36:313–25. glycogen synthase kinase-3 β down-regulates Notch activity, a link 28. Karayiannakis AJ, Nakopoulou L, Gakiopoulou H, et al. Expression for Notch and Wnt pathways. J Biol Chem 2003;278:32227–35. patterns of β-catenin in in situ and invasive breast cancer. Eur J Surg 17. Rodilla V, Villanueva A, Obrador-Hevia A, et al. Jagged1 is the path- Oncol 2001;27:31–6. ological link between Wnt and Notch pathways in colorectal cancer. 29. Krieghoff E, Behrens J, Mayr B. Nucleo-cytoplasmic distribution of β- Proc Natl Acad Sci U S A 2009;106:6315–20. catenin is regulated by retention. J Cell Sci 2006;119:1453–63. 18. Cheang MC, van de Rijn M, Nielsen TO. Gene expression profiling of 30. Neufeld KL, Zhang F, Cullen BR, White RL. APC-mediated downre- breast cancer. Annu Rev Pathol 2008;3:67–97. gulation of β-catenin activity involves nuclear sequestration and nu- 19. Wong SC, Lo ES, Lee KC, Chan JK, Hsiao WL. Prognostic and di- clear export. EMBO Rep 2000;1:519–23. agnostic significance of β-catenin nuclear immunostaining in colo- 31. Shimura T, Takenaka Y, Tsutsumi S, et al. Galectin-3, a novel binding rectal cancer. Clin Cancer Res 2004;10:1401–8. partner of β-catenin. Cancer Res 2004;64:6363–7. 20. Lin SY, Xia W, Wang JC, et al. β-Catenin, a novel prognostic marker 32. Ozaki S, Ikeda S, Ishizaki Y, et al. Alterations and correlations of the for breast cancer: its roles in cyclin D1 expression and cancer pro- components in the Wnt signaling pathway and its target genes in gression. Proc Natl Acad Sci U S A 2000;97:4262–6. breast cancer. Oncol Rep 2005;14:1437–43. 21. Prasad CP, Gupta SD, Rath G, Ralhan R. Wnt signaling pathway 33. Jin Z, Tamura G, Tsuchiya T, et al. Adenomatous polyposis coli in invasive ductal carcinoma of the breast: relationship between (APC) gene promoter hypermethylation in primary breast cancers. β-catenin, dishevelled and cyclin D1 expression. Oncology 2007; Br J Cancer 2001;85:69–73. 73:112–7. 34. Prasad CP, Mirza S, Sharma G, et al. Epigenetic alterations of 22. Dolled-Filhart M, McCabe A, Giltnane J, et al. Quantitative in situ CDH1 and APC genes: relationship with activation of Wnt/β-catenin analysis of β-catenin expression in breast cancer shows decreased pathway in invasive ductal carcinoma of breast. Life Sci 2008;83: expression is associated with poor outcome. Cancer Res 2006;66: 318–25. 5487–94. 35. Schlosshauer PW, Brown SA, Eisinger K, et al. APC truncation and 23. Fanelli MA, Montt-Guevara M, Diblasi AM, et al. P-cadherin and increased β-catenin levels in a human breast cancer cell line. Carci- β-catenin are useful prognostic markers in breast cancer patients; nogenesis 2000;21:1453–6. β-catenin interacts with heat shock protein Hsp27. Cell Stress Cha- 36. Wong SC, Lo SF, Lee KC, et al. Expression of frizzled-related protein perones 2008;13:207–20. and Wnt-signalling molecules in invasive human breast tumours. J 24. Bankfalvi A, Terpe HJ, Breukelmann D, et al. Immunophenotypic and Pathol 2002;196:145–53. prognostic analysis of E-cadherin and β-catenin expression during 37. Korkaya H, Paulson A, Charafe-Jauffret E, et al. Regulation of mam- breast carcinogenesis and tumour progression: a comparative study mary stem/progenitor cells by PTEN/Akt/β-catenin signaling. PLoS with CD44. Histopathology 1999;34:25–34. Biol 2009;7:e1000121. 25. Chung GG, Zerkowski MP, Ocal IT, et al. β-Catenin and p53 anal- 38. Fuchs SY, Ougolkov AV, Spiegelman VS, Minamoto T. Oncogenic β- yses of a breast carcinoma tissue microarray. Cancer 2004;100: catenin signaling networks in colorectal cancer. Cell Cycle 2005;4: 2084–92. 1522–39. 26. Dillon DA, D'Aquila T, Reynolds AB, Fearon ER, Rimm DL. The ex- 39. Hiscox S, Jiang WG, Obermeier K, et al. Tamoxifen resistance in pression of p120ctn protein in breast cancer is independent of α- and MCF7 cells promotes EMT-like behaviour and involves modulation β-catenin and E-cadherin. Am J Pathol 1998;152:75–82. of β-catenin phosphorylation. Int J Cancer 2006;118:290–301.

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 19(1) January 2010 309 IJC International Journal of Cancer

Loss of STARD10 expression identiﬁes a group of poor prognosis breast cancers independent of HER2/Neu and triple negative status

Niamh C. Murphy1, Andrew V. Biankin1,2,3, Ewan K.A. Millar1,4,5,6, Catriona M. McNeil1,7,8, Sandra A. O’Toole1,8,9, Davendra Segara1,10, Paul Crea10, Monilola A. Olayioye11, C. Soon Lee5,12, Stephen B. Fox12, Adrienne L. Morey13, Michael Christie14, Elizabeth A. Musgrove1,2, Roger J. Daly1,2, Geoffrey J. Lindeman11,14, Susan M. Henshall1,2, Jane E. Visvader11 and Robert L. Sutherland1,2

1 Cancer Research Program, Garvan Institute of Medical Research, Darlinghurst, Sydney, NSW, Australia 2 St. Vincent’s Clinical School, Faculty of Medicine, University of New South Wales, NSW Australia 3 Division of Surgery, Bankstown Hospital, Bankstown, Sydney, NSW, Australia 4 Department of Anatomical Pathology, South Eastern Area Laboratory Service, St. George Hospital, Kogarah, NSW, Australia 5 Department of Pathology, School of Medicine, University of Western Sydney, NSW, Australia 6 School of Medical Sciences, Faculty of Medicine, University of New South Wales, NSW, Australia 7 Department of Medical Oncology, Westmead Hospital, Sydney, NSW, Australia 8 Departments of Medicine and Pathology, University of Sydney, NSW, Australia 9 Department of Anatomical Pathology, Sydney Cancer Centre, Royal Prince Alfred Hospital Camperdown, Sydney, NSW, Australia 10 Department of Surgical Oncology, St Vincent’s Clinic, Darlinghurst, Sydney, NSW, Australia 11 VBCRC Laboratory, The Walter and Eliza Hall Institute of Medical Research, Parkville, VIC, Australia 12 Department of Pathology, University of Melbourne, Peter MacCallum Cancer Centre, East Melbourne, VIC, Australia 13 Department of Pathology (Syd Path), St Vincent’s Hospital, Darlinghurst, Sydney, NSW, Australia 14 Department of Anatomical Pathology and Medical Oncology, The Royal Melbourne Hospital, Parkville, VIC, Australia

The phospholipid transfer protein STARD10 cooperates with c-erbB signaling and is overexpressed in Neu/ErbB2 breast cancers. We investigated if STARD10 expression provides additional prognostic information to HER2/neu status in primary breast cancer. A published gene expression dataset was used to determine relationships between STARD10 and HER2 mRNA levels and patient outcome. The central findings were independently validated by immunohistochemistry in a retrospective cohort of 222 patients with breast cancer with a median follow-up of 64 months. Kaplan–Meier and Cox proportional hazards analyses were used for univariate and multivariate analyses. Patients with low STARD10 or high HER2 tumor mRNA levels formed discrete groups each associated with a poor disease-specific survival (p 5 0.0001 and p 5 0.0058, respectively). In the immunohistochemical study low/absent STARD10 expression i.e. 10% positive cells was observed in 24 of 222 (11%) tumors. In a univariate model, low/absent STARD10 expression was significantly associated with decreased patient survival (p 5 0.0008). In multivariate analyses incorporating tumor size, tumor grade, lymph node status, ER, PR and HER2 status, low STARD10 expression was an independent predictor of death from breast cancer (HR: 2.56 (95% CI: 1.27–5.18), p 5 0.0086). Furthermore, low/absent STARD10 expression, HER2 amplification and triple negative status were independent prognostic variables. Loss of STARD10 expression may provide an additional marker of poor outcome in breast cancer identifying a subgroup of patients with a particularly adverse prognosis, which is independent of HER2 amplification and the triple negative phenotype. Early Detection and Diagnosis

Key words: breast cancer, STARD10, HER2/Neu, triple negative breast cancer Niamh C. Murphy, Andrew V. Biankin and Ewan K.A. Millar contributed equally to this work Grant sponsors: National Health and Medical Research Council of Australia, Victorian Breast Cancer Research Consortium Inc. (Australia), the Cancer Institute of NSW, the Petre Foundation, the RT Hall Trust DOI: 10.1002/ijc.24826 History: Received 1 Jul 2009; Accepted 3 Aug 2009; Online 12 Aug 2009 Monilola A. Olayioye’s current address is: Institute of Cell Biology and Immunology, University of Stuttgart, Stuttgart, Germany Correspondence to: Robert L. Sutherland, Cancer Research Program, Garvan, Institute of Medical Research, 314 Victoria Street, Darlinghurst, Sydney, NSW 2010, Australia., Fax: þ61-2-9295-8321, E-mail: [email protected]

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC 1446 Loss of STARD10 in poor prognosis breast cancers

Breast cancer is a complex, heterogeneous disease, encom- prognostic information to HER2 status in primary breast passing a range of pathologies and patterns of gene expres- cancer. sion resulting in disparate clinical behavior and response to therapy.1,2 Axillary lymph node status, tumor size and grade, Material and Methods patient age and lymphovascular invasion are among the clini- Gene expression data and analysis copathological prognostic indicators currently employed to Published gene expression data17 were downloaded from 3 predict an individual’s risk of metastasis. Axillary lymph http://microarray-pubs.stanford.edu/wound_NKI/explore.html node status is the best clinicopathological prognostic marker and employed to establish relationships between STARD10 available, but it is an imperfect predictor. Approximately 25% and HER2 mRNA levels and breast cancer outcome. Criteria of node-negative patients harbor micrometastases and are employed for the choice of this dataset included: the presence destined to relapse. Conversely, up to 50% of node-positive of probesets for STARD10 on the oligonucleotide array, the patients do not recur after many years of follow-up even size of the cohort (n ¼ 295) and length of follow-up (median 4 without adjuvant treatment. Thus, there is a critical need to 7.8 years, range 0.05–18.3) and the ability of classical clinico- identify more accurate prognostic markers that can identify a pathological parameters (tumor size, grade and lymph node patient’s risk of recurrence and to accurately select which ad- status), ER and HER2 to predict outcome. juvant therapeutic regimens are most likely to benefit individual patients. Patient cohort Currently, the only recommended predictive biomarkers Following approval from the Human Ethics Committee of St. in breast cancer are the estrogen (ER) and progesterone Vincent’s Hospital, a total of 222 patients diagnosed with inva- receptors (PR) for selecting endocrine-sensitive disease3,5 and sive ductal breast carcinoma (IDC) and treated by a single sur- the human epidermal growth factor receptor 2, HER2 (c- geon (PC) between February 1992 and August 2002 was erbB2/neu), for identifying patients who may benefit from included in this study. The clinicopathological variables, Cox trastuzumab, a therapeutic antibody targeting the c-erbB2 re- univariate analyses, event rates and median time to event are ceptor.5,6 HER2 status also indicates an enhanced sensitivity recorded in Table 1. Formalin-fixed, paraffin embedded tissue to high dose anthracycline-based chemotherapeutic regi- was available from the pathology archives of St. Vincent’s Public mens,7 while HER2 overexpression is associated with a lower Hospital (Sydpath) and St. Vincent’s Private Hospital (Douglas probability of response to tamoxifen therapy in both early Hanly Moir), Sydney, Australia. All tumors were classified as and advanced breast cancer.8,9 Advances in expression micro- IDC according to the World Health Organization (WHO) array technology allow classification of intrinsic subtypes of schema18 and were graded using the Nottingham combined his- 19 breast cancer2 which identify basal-like or triple negative phe- tologic grading scheme. Follow-up intervals were calculated notype breast cancers (i.e. ER, PR and HER2 negative, TNP) from the date of definitive procedure (biopsy/lumpectomy/mas- as poor prognosis groups.10,11 Other gene ‘‘signatures’’ that tectomy) to the date of last recorded follow-up (mean: 67 also predict disease outcome include the 70 gene signature,12 months, median: 64 months, range: 1–154 months). These data V 21-gene ‘‘signature’’ (Oncotype DX R RT-PCR assay)13 and were obtained from annual review of patient charts or tumor the genomic grade index.14 However, these advances have yet registry data for patients alive at the previous annual review. to be translated into widespread clinical practice. Thus, there Patients less than 50 years of age with node-positive, ER remains a need for the identification of further single gene or tumors or tumors larger than 3 cm received adjuvant chemo- multiple prognostic markers of breast cancer, which are ame- therapy (cyclophosphamide, methotrexate and 5-fluorouracil or nable to clinical practice, either alone or in combination. adriamycin and cyclophosphamide (AC)). Patients with ERþ STARD10, a recently described member of the START tumors who were more than 50 years of age received 5 years of (steroidogenic acute regulatory proteins) lipid transfer pro- tamoxifen therapy. Tissue microarrays (TMAs) were con- Early Detection and Diagnosis teins,15,16 is overexpressed in mammary tumors from Neu/ structed from formalin-fixed, paraffin embedded tumor mate- ErbB2 transgenic mice and in 35% of primary human breast rial; each patient was represented by 2–6, 1 mm cores. cancers.15 Furthermore, increased expression of STARD10 is correlated with c-erbB2/HER2 status in human breast cancer Immunohistochemistry cell lines and a series of breast carcinomas while coexpression Four-micron sections were cut from each TMA, mounted on V with the epidermal growth factor receptor (EGFR/c-erbB1)in SuperFrost R Plus glass slides, dewaxed by passage through murine fibroblasts enhances anchorage-independent cell xylene (two 5-min washes), cleared and rehydrated in graded growth15 Together, these data imply potential functional alcohol (100%, 95% and 70%) ending in a distilled water wash. cooperation between STARD10 and ErbB receptor signaling Endogenous peroxidases and proteins were blocked with 3% in breast oncogenesis. hydrogen peroxide and protein block, serum free (DAKO Cor- This study was undertaken to further define STARD10 poration Carpenteria, CA). Sections, without antigen retrieval, expression in a large cohort of patients with breast cancer, were incubated with a 1:100 dilution of an affinity purified rab- identify any relationship with known prognostic markers bit anti-STARD10 antibody,15 and detection was performed and determine if STARD10 status could provide additional with EnvisionþRabbit (DAKO) for 30 min at room

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC Murphy et al. 1447

Table 1. Cox univariate analysis, event rates and survival Median time to Hazard ratio (95% CI) p value Event rate event (months) Univariate analysis Grade 3 4.250 (2.152–8.395) <0.0001 34/103 33.2 Size > 20 mm 2.901 (1.562–5.417) 0.0008 30/97 33.9 Lymph node þ 2.877 (1.509–5.486) 0.0013 32/111 35.4 ERþ 0.282 (0.155–0.514) <0.0001 20/152 49.3 PRþ 0.171 (0.084–0.347) <0.0001 10/119 62.7 HER2 IHC 3þ 4.592 (1.767–6.288) <0.0001 12/25 33.9 HER2 amp (FISH) 3.334 (1.767–6.288) 0.0002 16/43 35.7 Triple negative 2.941 (1.549–5.587) 0.001 14/34 35.7 Low StarD10 3.065 (1.549–6.066) 0.0013 11/24 20.3 Survival Mean follow-up 67 months (0.1–154) Alive 165 (74%) Death due to breast cancer 45 (20%) Death due to other causes 12 (6%) Disease speciﬁc survival 5 years 189 (85%) 10 years 175 (79%)

ER, estrogen receptor; PR, progesterone receptor; FISH, fluorescent in-situ hybridization. temperature followed by visualization using DABþ (DAKO) versus positive (3þ). Triple negative was defined as ER, PR for 5 min. Slides were also stained using the following mono- and HER-2 negative. clonal antibodies: ER 1:100 (6F11, DAKO), PR 1:200 (PgR- 636, DAKO). HER2/neu immunohistochemistry was under- Statistical analyses taken using standard procedures,20 and HER2 FISH was Statistical analyses were performed using Statview 5.0 Soft- assessed in The Australian National HER-2 Reference labora- ware (Abacus Systems, Berkeley, CA). A p-value of <0.05 tory (SydPath, St Vincent’s Hospital Sydney). A HER2:chro- was accepted as statistically significant. Correlations between mosome 17 ratio >2.2 was classified as HER2 amplification. STARD10 expression and clinicopathological variables were tested by applying cross tables and the Exact Fisher’s Test. Disease-specific survival, defined as date of definitive proce- Immunohistochemical scoring dure to date of death due to breast cancer, was selected as For STARD10, 2–6 individual cores were assessed independ- the primary endpoint for survival analysis. Kaplan–Meier and ently by 3 pathologists (EKAM, SAO’T, MC) who were the Cox proportional hazards model were used for univariate blinded to each others results and to patient outcome. Scores and multivariate analyses, respectively. Those factors that were assigned as a percentage of cells with positive cytoplas- were prognostic on univariate analysis were then assessed in Early Detection and Diagnosis mic and/or nuclear staining and the absolute intensity of a multivariable model to identify factors that were independ- staining determined on a scale of 0–3 where 0 represents no ently prognostic and those that were the result of staining, 1 represents mild staining, 2 represents moderate confounding. staining and 3 represents strong staining. Criteria to achieve a positive score for STARD10 expression were cytoplasmic Results staining intensity >1in>10% of cells. Where scoring varied STARD10/HER2 mRNA levels and association between individual pathologists, a final score was resolved by with patient outcome conferencing and consensus. Scoring of ER/PR/HER2/neu An initial characterization of STARD10 gene expression in immunohistochemical staining was performed by a single breast cancer and its potential relationship to patient out- breast pathologist (EKAM) according to routine clinical prac- come was undertaken by interrogating the data of Chang et tice guidelines. Assessment of HER2/neu positive immuno- al.17 When modeled as a continuous variable, higher expres- staining was restricted to areas demonstrating a membranous sion levels of STARD10 mRNA were associated with longer staining pattern. HER2/neu was classified as negative (2þ) survival (HR 0.25, 95%CI: 0.09–0.64, p ¼ 0.004). As loss of

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC 1448 Loss of STARD10 in poor prognosis breast cancers

tastases (p ¼ 0.22) but was associated with ER or PR negative tumors (p < 0.0001 for each). As a preliminary study had revealed that 4 of 8 cancers with strong membranous staining of HER2 were also STARD10 positive,15 we determined the relationship between STARD10 and HER2 mRNA. Regression analysis failed to demonstrate any significant relationship between HER2 and STARD10 mRNAs as continuous variables (R ¼ 0.17). When we selected the upper 35% of STARD10 mRNA expression, in line with the earlier estimate of STARD10 overexpression by immunohistochemistry,15 we again failed to demonstrate a significant association with high HER2 mRNA expression. Most notably, however, only 3 of 295 cancers (1%) had both high HER2/neu mRNA and low STARD10 mRNA expression, while 53 of 295 (18%) tumors were classified as having high HER2/neu mRNA; a proportion comparable to the level of HER2/neu amplification observed in this (Table 1) and other studies.20 At this cut-point high HER2/neu mRNA was associated with poor patient survival (p ¼ 0.0058; Fig. 1b). Hence, 46 of a possible 49 cancers with low STARD10 mRNA expression were HER2/neu mRNA negative, and 50 of a possible 53 HER2/neu mRNA positive cancers had high STARD10 mRNA expression. The identification of these 2 almost mutually exclusive subgroups of patients (i.e. HER2 positive and STARD10 negative) with particularly adverse outcomes suggested that they were potentially biologically distinct. To assess if STARD10 mRNA was an independent prognostic factor, Cox proportional hazard models were constructed with stepwise removal of redundant variables until resolution. The resolved model (Table 2, Model 1) demonstrates that low STARD10 mRNA expression was an independent prognostic factor in this cohort when modelled with other clinicopathological markers (Model 1) or HER2 alone (Model 2). Kaplan-Meier analysis based on this model, where HER2/neu status is stratified by STARD10 expression demonstrated that STARD10 positive/HER2/neu negative cancers had the best prognosis, STARD10 negative/HER2/neu positive cancers the worst and STARD10 positive HER2/neu positive cancers characterized an intermediate prognostic group (Fig. 1c). The influence of low STARD10 expression in HER2/neu Early Detection and Diagnosis Figure 1. Disease-specific survival according to STARD10 and HER2 positive cancers was not assessable since this occurred in mRNA levels. only 3 cases.

Immunohistochemical analysis of STARD10 and association with patient outcome STARD10 was apparent in approximately 15% of breast can- We next performed immunohistochemistry on TMAs from cer samples using immunohistochemistry in a preliminary an independent cohort of 222 carcinomas to assess if analysis, a cutpoint was identified to dichotomize these data STARD10 status was also capable of identifying patient out- and define a similar proportion. This cutpoint (absolute value come. Tumor size, tumor grade, lymph node status, ER and –0.210) identified a group of 49 patients (17%) with low PR status, HER2/neu immunohistochemical status (i.e. 25 of expression of STARD10 mRNA that was associated with a 216, 11.6%) positive and HER2/neu amplification by FISH poor disease specific survival (logrank p < 0.0001; Fig. 1a). (i.e. 43 of 212, 20.2%) positive, were all significantly associ- Differential STARD10 expression was not associated with tu- ated with an adverse patient outcome on univariate analysis mor size >2 cm (Chi Square p ¼ 0.53) or lymph node me- (Table 1). The higher rate of HER-2 positivity as assessed by

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC Murphy et al. 1449

FISH includes cases which were equivocal (i.e. 2þ) by IHC tumor size (p¼ 0.2527), lymph node status (p ¼ 0.1949), as well as IHC 3þ. Assignment of STARD10 positivity HER2/neu IHC status (p > 0.9999) or HER2/neu FISH (p ¼ required an intensity >1in>10% of cancer cells. Using these 0.6720). STARD10 was negatively associated with the triple criteria, positive STARD10 immunostaining was observed in negative phenotype (i.e. 24 of 34 TNP cancers were 198/222 (89%) of cancers. Staining was typically cytoplasmic STARD10 positive, p < 0.0001). and nuclear (Fig. 2). Positive STARD10 expression was significantly associated with ER (p 0.0016) and PR positivity (p 0.0001) but not with histological grade (p ¼ 0.4189), Survival analysis There was a significant association between low/absent STARD10 expression and breast cancer recurrence, metastasis Table 2. Multivariate analysis for clinicopathological parameters, to distant sites and disease-specific survival on Kaplan–Meier hormone receptor status and STARD10 and HER2 mRNA expression analysis (Fig. 3.). Within this group of low expressors (n ¼ Variable (wound cohort) Hazard ratio (95% CI) p value 24) there was no significant difference in survival between Model 11 those tumors which had 0 staining (n ¼ 9) and those with Tumor grade 3 2.22 (1.36–3.62) 0.0014 1–10% (n ¼ 15, p ¼ 0.507). As previously reported both Tumor size >20 mm 1.94 (1.12–3.37) 0.0180 HER2/neu overexpression by immunohistochemistry and HER2/neu amplification were associated with an adverse dis- PR positive 0.55 (0.32–0.93) 0.0253 ease-specific survival (p < 0.0001 in both cases). When HER2 positive 1.76 (1.02–3.05) 0.0434 patients were stratified into 4 groups based on STARD10 Low STARD10 1.83 (1.04–3.22) 0.0361 immunohistochemistry and HER2 amplification similar Model 22 results to those derived from the mRNA data were evident. HER2 positive 2.35 (1.41–3.91) 0.0010 The majority of patients (n ¼ 153) were in the STARD10þ/ Low STARD10 2.92 (1.79–4.76) <0.0001 HER2 group and had the best prognosis (Fig. 3d). HER2 amplification was associated with poor survival in both the Wound signature cohort (Ref. 12). STARD10þ (n ¼ 38) and STARD10 (n ¼ 5) groups. How- 1The resolved multivariate model shows that low STARD10 expression is an independent poor prognostic factor. 2The resolved multivariate ever, loss of STARD10 expression in the absence of HER2 model shows that low STARD10 expression is independent of HER2 amplification (n ¼ 16) conferred a poor prognosis similar to amplification. that of HER2 amplification. HER2 FISH data was only Early Detection and Diagnosis

Figure 2. Representative images of STARD10 immunohistochemistry showing no (0), low (1þ), medium (2þ) and high (3þ) intensity staining.

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC 1450 Loss of STARD10 in poor prognosis breast cancers

Figure 3. Relationship between STARD10 expression as assessed by immunohistochemistry and (a) breast cancer recurrence, (b) metastasis and (c) disease speciﬁc survival alone or in (d) combination with HER2 ampliﬁcation status.

available on 212 patients; hence, results are not for the full STARD10 and HER2/neu status both provide important clin- cohort of 222 patients. ical prognostic information but identify independent high- Cox univariate analysis confirmed that loss/absence of risk groups. STARD10 expression was significantly associated with To assess the influence on outcome of a triple negative decreased breast cancer-specific survival (HR: 3.065, 95%CI: phenotype in this cohort, we replaced STARD10 with triple 1.549–6.066, p ¼ 0.0013). Tumor size and grade, lymph node negative phenotype in the resolved model and confirmed the Early Detection and Diagnosis status, ER, PR and HER2 status were also highly significant independent prognostic nature of this phenotype (Table 3, prognostic markers in this cohort. In a multivariate analysis Model 4). However, when STARD10 and triple negative phe- incorporating these markers (Table 3, Model 1), step-wise notype were assessed together with the other significant vari- elimination of noncontributory/redundant variables was ables (Table 3, Model 5), triple negative phenotype was no employed until resolution. Using this approach, lymph node longer independently prognostic but when modeled together involvement, PR status, HER2 amplification and low low STARD10 expression and triple negative phenotype STARD10 remained independent predictors of death from remained independent prognostic factors (Table 3, Model 6). breast cancer (Table 3, Model 2). Thus, STARD10 status by Thus, low STARD10 status identifies a poor prognosis group immunohistochemistry is an independent predictor of death independent of both HER2 amplification and triple negative from breast cancer in this cohort validating the mRNA data phenotype. from the independent cohort presented in Table 2. In a multivariate model incorporating only STARD10 status and Discussion HER2/neu amplification, both variables remain independent STARD10 was originally identified in mammary tumors from prognostic factors (Table 3, Model 3). This implies that Neu/ErbB2 transgenic mice, subsequently isolated from a c-

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC Murphy et al. 1451

Table 3. Multivariate analysis for breast cancer-specific death for signaling may cooperate in breast cancer development and clinicopathological parameters, STARD10 and HER2 status progression. Variable (Garvan cohort) Hazard ratio (95% CI) p value To test this concept we first assessed relationships between Model 11 STARD10 and HER2 mRNA levels in a cohort of 295 Tumor grade 3 1.964 (0.830–4.647) 0.1243 patients where the oligonucleotide expression array and clinicopathological data were in the public domain.17 This dem- Tumor size >20 mm 1.789 (0.930–3.445) 0.0816 onstrated that STARD10 and HER2/neu mRNA were overex- Lymph node positive 2.129 (1.055–4.295) 0.0349 pressed in a significant proportion of breast cancers but their ER positive 1.039 (0.468–2.307) 0.9251 expression levels were not correlated. When each set of gene PR positive 0.304 (0.127–0.731) 0.0078 expression data was dichotomized for high and low mRNA HER2 amplified 2.173 (1.103–4.280) 0.0249 expression, low STARD10 and high HER2/neu formed 2 Low StarD10 2.843 (1.330–6.075) 0.0070 almost mutually exclusive subgroups each of which had a poor disease outcome. Model 22 To validate these findings and assess the potential clinical Lymph node positive 2.33 (1.21–4.51) 0.0119 utility of a simple immunohistochemical test for STARD10, PR positive 0.22 (0.11–0.46) <0.0001 we analyzed the expression of STARD10 in an independent HER2 amplified 2.67 (1.39–5.11) 0.0031 cohort of 222 primary breast cancers. Employing a cut-off Low StarD10 2.56 (1.27–5.18) 0.0086 for positive STARD10 staining as a staining intensity >1in > Model 33 10% of cells we identified expression in 198 of 222 (89%) tumors; 24 cases (11%) were classified as low/absent expres- HER2 amplified 3.19 (1.69–6.02) 0.0003 sion. In agreement with the mRNA data, STARD10 expres- Low StarD10 3.53 (1.77–7.09) 0.0003 sion by immunohistochemistry was significantly associated 4 Model 4 with ER and PR status but not with tumor size, lymph node Lymph node positive 2.57 (1.31–5.02) 0.0059 status or HER2/neu status. Univariate analysis identified PR positive 0.32 (0.14–0.73) 0.0066 absent/low STARD10 expression as a significant predictor of HER2 amplified 4.59 (2.05–10.3) 0.0002 an adverse patient outcome. In a multivariate analysis incorporating STARD10 status with established prognostic factors, Triple negative 2.59 (1.07–6.22) 0.0340 STARD10 status remained an independent prognostic factor Model 55 of disease recurrence, distant metastasis and breast cancer- Lymph node positive 2.57 (1.31–5.02) 0.0059 specific death. When modeled together with HER2/neu status, PR positive 0.31 (0.14–0.70) 0.0051 absent/low STARD10 expression remained a significant prog- HER2 amplification 3.93 (1.73–8.92) 0.0011 nostic factor. This was also true when modeled against the Low StarD10 2.18 (1.06–4.50) 0.0348 triple negative phenotype which is a well-documented poor prognosis group.10,11 Thus, this study demonstrates for the Triple negative 2.13 (0.86–5.27) 0.1037 first time that immunohistochemical assessment of STARD10 Model 66 provides clinically important prognostic information, which Low StarD10 2.84 (1.38–5.81) 0.0044 is independent of other established prognostic factors. Recent Triple negative 2.37 (1.22–4.64) 0.0113 gene expression profiling studies have identified molecular subtypes of breast cancer with markedly different prognoses. 1Multivariate analysis of all significant factors associated with outcome identified using univariate analysis shows that low STARD10 is an Established poor prognosis groups include those with HER2 independent prognostic factor. 2The resolved multivariate model, amplification, the HER2 subtype and the basal-like subtype.2 following elimination of noncontributory variables (grade 3, tumor size, Early Detection and Diagnosis 3 The triple negative phenotype was a proposed surrogate ER). Low STARD10 is still prognostic and independent of HER2 10 amplification. 4Triple negative status is an independent prognostic marker of the basal subtype, but it is now not thought to factor. 5Triple negative status is not independent when STARD10 is be synonymous with this group.11 Interestingly, the low/ 6 added to the model. A bivariate model shows that STARD10 and triple absent STARD10 group appears to be independent of these negative status are independent of each other. molecular subtypes and studies are underway to determine if this group also has a unique molecular signature. ERBB2 amplified, breast cancer cell line15 and demonstrated The known function of c-erbB2 in mammary carcinogene- to be preferentially expressed in breast cancer cells overex- sis21 and breast cancer and the enrichment of tumors with pressing c-erbB2. Furthermore, in 79 invasive breast cancers BRCA1 and p53 mutations and features of stem cells with 28 (35%) demonstrated moderate to intense staining while the basal subtype10 provide some potential mechanistic basis normal breast tissue was negative. Together with in vitro stud- for these different phenotypes. In contrast, little is known of ies, demonstrating that coexpression of STARD10 and the the normal physiological function of STARD10 in mammary EGFR/c-erbB1 enhanced anchorage-independent growth,15 epithelium and breast cancer and whether or not it might be these data led to the hypothesis that STARD10 and c-erbB mechanistically involved in the disease process or is merely a

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC 1452 Loss of STARD10 in poor prognosis breast cancers

surrogate marker of a more relevant underlying biological STARD10’s apparent involvement in breast tumorigenesis process. STARD10 is a member of the START domain pro- may be related to its role as an estrogen target gene.27 In this teins, it was initially discovered on the basis of cross reactiv- regard, a recent study on the molecular response to aroma- ity with a phosphoserine-specific antibody in mammary tase inhibitors in a neoadjuvant treatment paradigm identi- tumors from neu/ErbB2 transgenic mice and was subse- fied STARD10 as one of the most significantly down-regu- quently isolated from SKBR3 human breast carcinoma cells.15 lated genes following treatment.28 Thus, loss of STARD10 Recently, STARD10 has been found to function as a dual expression may reflect a loss of ER or ER function in agree- specificity phospholipid transfer protein for phophatidylcho- ment with data presented here on the relationship with ER line (PC) and phosphatidylethanolamine (PE).22 By binding status, and/or a loss of steroid-regulated differentiated cell PC and PE, STARD10 may function in the transfer of these function i.e. dedifferentiation which is a known facet of can- lipids between subcellular compartments and the regulation cer progression. However, STARD10 immunohistochemical of cellular signaling events. Deregulation may result in aber- expression was not correlated with tumor grade and rant lipid signaling contributing to cellular transformation. remained independently prognostic in a model incorporating STARD10 is widely expressed and synthesized constitutively PR status and thus does not merely reflect morphological dif- in many organs, including the liver, where a putative func- ferentiation or hormone-responsiveness. It is interesting to tion is the export of phospholipids into bile. STARD10 note, however, that the prognostic significance of loss of expression is also regulated during development in the testes STARD10 in breast cancer has parallels with loss of AZGP1 and mammary glands.23 In the mammary gland, STARD10 in prostate cancer.29 Both are steroid-regulated genes in the expression is developmentally regulated with highest expres- more differentiated, hormone-responsive phases of the dis- sion occurring during gestation and lactation.15 This suggests ease30 but when expression is lost, identify patients with a a role for STARD10 in the enrichment of lipids in milk and high probability of relapse, metastasis and death. The molec- as a potential marker of differentiation. In addition, CK2 ular basis of these interactions remains to be defined. (casein kinase 2) a gene involved in Wnt/b-catenin signal- In conclusion, loss of STARD10 expression in breast can- ing24 and overexpressed in breast cancer25 downregulates the cer strongly predicts an aggressive disease course independent lipid transfer activity of STARD10 by phosphorylation of ser- of HER2/neu amplification and the triple negative phenotype. ine 284.26 This may explain a possible biological mechanism These data raise the possibility that loss of expression of responsible for the loss of STARD10 expression in breast STARD10 may provide an additional simple immunohisto- cancer and suggests the need for further investigation of its chemical marker of disease outcome in breast cancer by iden- potential role in tumorigeneis that may be independent of its tifying a subgroup of patients with a particularly adverse established lipid-binding transport function.22 Alternatively, outcome.

References

1. Ellis IO, Schnitt SJ, Sastre-Garau X, 5. Harris L, Fritsche H, Mennel R, Norton L, 9. Houston SJ, Plunkett TA, Barnes DM, Bussolati G, Tavassoli FA, Eusebi V, Ravdin P, Taube S, Somerfield MR, Hayes Smith P, Rubens RD, Miles Peterse JL, Mukai K, Tabar L, Jacquemier DF, Bast RC, Jr.American Society of DW.Overexpression of c-erbB2 is an J, Cornelisse CJ, Sasco AJ, et al. Invasive Clinical Oncology 2007 update of independent marker of resistance to breast carcinomas. In: Tavassoli FA, recommendations for the use of tumor endocrine therapy in advanced breast Devilee P, eds. WHO classification of markers in breast cancer. J Clin Oncol cancer. Br J Cancer 1999;79:1220–6. breast tumors and tumors of the female 2007;25:5287–312. 10. Cleator S, Heller W, Coombes RC.Triple- genital organs: pathology and genetics. 6. Viani GA, Afonso SL, Stefano EJ, De Fendi negative breast cancer: therapeutic options. Lyon: IARC Press, 2003. 13. LI, Soares FV.Adjuvant trastuzumab in the Lancet Oncol 2007;8:235–44. Early Detection and Diagnosis 2. Sorlie T, Perou CM, Tibshirani R, Aas T, treatment of her-2-positive early breast 11. Rakha EA, Reis-Filho JS, Ellis IO.Basal-like Geisler S, Johnsen H, Hastie T, Eisen MB, cancer: a meta-analysis of published breast cancer: a critical review. J Clin Oncol van de Rijn M, Jeffrey SS, Thorsen T, randomized trials. BMC Cancer 2007;7:153. 2008;26:2568–81. Quist H, et al. Gene expression patterns of 7. Pritchard KI, Shepherd LE, O’Malley FP, 12. van’tVeerLJ,DaiH,vandeVijverMJ,He breast carcinomas distinguish tumor Andrulis IL, Tu D, Bramwell VH, Levine YD,HartAAM,MaoM,PeterseHL,van subclasses with clinical implications. Proc MN.HER2 and responsiveness of breast der Kooy K, Marton MJ, Witteveen AT, Natl Acad Sci USA 2001;98:10869–74. cancer to adjuvant chemotherapy. N Engl J Schreiber GJ, Kerkhoven RM, et al. Gene 3. Goldhirsch A, Glick JH, Gelber RD, Coates Med 2006;354:2103–11. expression profiling predicts clinical outcome AS, Thurlimann B, Senn HJ.Meeting 8. Carlomagno C, Perrone F, Gallo C, De of breast cancer. Nature 2002;415:530–6. highlights: international expert consensus Laurentiis M, Lauria R, Morabito A, 13. Paik S, Shak S, Tang G, Kim C, Baker J, on the primary therapy of early breast Pettinato G, Panico L, D’Antonio A, Cronin M, Baehner FL, Walker MG, cancer 2005. Ann Oncol 2005;16:1569–83. Bianco AR, De Placido S.c-erb B2 Watson D, Park T, Hiller W, Fisher ER, 4. Early Breast Cancer Trialists’ Collaborative overexpression decreases the benefit of Wickerham DL, et al. A multigene assay to Group.Polychemotherapy for early breast adjuvant tamoxifen in early-stage breast predict recurrence of tamoxifen treated, cancer: an overview of the randomised cancer without axillary lymph node node-negative breast cancer. N Engl J Med trials. Lancet 1998;352:930–42. metastases. J Clin Oncol 1996;14:2702–8. 2004;351:2817–26.

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC Murphy et al. 1453

14. Sotiriou C, Wirapati P, Loi S, Harris A, 20. Wolff AC, Hammond ME, Schwartz JN, Seldin DS.Protein kinase CK2 in mammary Fox S, Smeds J, Nordgren H, Farmer P, Hagerty KL, Allred DC, Cote RJ, Dowsett gland tumorigenesis. Oncogene 2001;20: Praz V, Haibe-Kains B, Desmedt C, M, Fitzgibbons PL, Hanna WM, Langer A, 3247–57. Larsimont D, et al. Gene expression McShane LM, Paik S, et al. American 26. Olayioye MA, Buchholz M, Schmid S, profiling in breast cancer: understanding Society of Clinical Oncology/College of Schoffler P, Hoffmann P, Pomorski the molecular basis of histologic grade to American Pathologists guideline T.Phosphorylation of StarD10 on serine improve prognosis. J Natl Cancer Inst recommendations for human epidermal 284 by casein kinase II modulates its lipid 2006;98:262–72. growth factor receptor 2 testing in breast transfer activity. J Biol Chem 2007;282: 15. Olayioye MA, Hoffmann P, Pomorski T, cancer. J Clin Oncol 2007;25:118–45. 22492–8. Armes J, Simpson RJ, Kemp BE, Lindeman 21. Ursini-Siegel J, Schade B, Cardiff RD, 27. Musgrove E, Sergio CM, Loi S, Inman CK, GJ, Visvader JE.The phosphoprotein Muller WJ.Insights from transgenic Anderson LR, Alles MC, Pinese M, Caldon STARD10 is overexpressed in breast cancer mouse models of ERBB2-induced CE, Schutte J, Gardiner-Garden M, and cooperates with ErbB receptors in breast cancer. Nat Rev Cancer 2007;7: Ormandy CJ, McArthur G, et al. cellular transformation. Cancer Res 2004; 389–97. Identification of functional networks of 64:3538–44. 22. Olayioye MA, Vehring S, Muller P, estrogen- and c-Myc-responsive genes and 16. Alpy F, Tomasetto C.Give lipids a START: Herrmann A, Schiller J, Thiele C, their relationship to response to tamoxifen the StAR-related lipid transfer (START) Lindeman GJ, Visvader JE, Pomorski therapy in breast cancer. PLoS One 2008;3: domain in mammals. J Cell Sci 2005;118: T.STARD10, a START domain protein e2987. 2791–801. overexpressed in breast cancer, functions as 28. Mackay A, Urruticoechea A, Dixon JM, 17. Chang HY, Nuyten DS, Sneddon JB, Hastie a phospholipid transfer protein. J Biol Dexter T, Fenwick K, Ashworth A, Drury T, Tibshirani R, Sorlie T, Dai H, He YD, Chem 2005;280:27436–42. S, Larionov A, Young O, White S, Miller van’t Veer LJ, Bartelink H, van de Rijn M, 23. Yamanaka M, Koga M, Tanaka H, WR, Evans DB, et al. Molecular response Brown PO, et al. Robustness, scalability, Nakamura Y, Ohta H, Yomogida K, to aromatase inhibitor treatment in and integration of a wound-response gene Tsuchida J, Iguchi N, Nojima H, primary breast cancer. Breast Cancer Res expression signature in predicting breast Nozaki M, Matsumiya K, Okuyama A, 2007;9:R37. cancer survival. Proc Natl Acad Sci USA et al. Molecular cloning and 29. Henshall SM, Horvath LG, Quinn DI, 2005;102:3738–43. characterization of phosphatidylcholine Eggleton SA, Grygiel JJ, Stricker PD, 18. Tavassoli F, Devilee P, eds. WHO transfer protein-like protein gene Biankin AV, Kench JG, Sutherland classification of tumours: pathology and expressed in murine haploid germ cells. RL.Zinc-alpha2-glycoprotein expression as genetics tumours of the breast and female Biol Reprod 2000;62:1694–701. a predictor of metastatic prostate cancer genital organs. Lyon: IARC Press,2003. 24. Seldin DS, Landesman-Bollag E, Farago M, following radical prostatectomy. J Natl 19. Elston CW, Ellis IO.Pathological prognostic Currier N, Lou D, Dominguez I.CK2 as a Cancer Inst 2006;98:1420–4. factors in breast cancer. I. The value of positive regulator of Wnt signaling and 30. Nelson PS, Clegg N, Arnold H, Ferguson histological grade in breast cancer: tumourigenesis. Mol Cell Biochem 2005; C, Bonham M, White J, Hood L, Lin B.The experience from a large study with long- 274:63–7. program of androgen-responsive genes in term follow-up. Histopathology 1991;19: 25. Landesman-Bollag E, Lle Romieu-Mourez neoplastic prostate epithelium. Proc Natl 403–10. RE, Song DH, Sonenshein GE, Cardiff RD, Acad Sci USA 2002; 99:11890–5. Early Detection and Diagnosis

Int. J. Cancer: 126, 1445–1453 (2010) VC 2009 UICC IJC International Journal of Cancer

Overexpression of the oncogenic signal transducer Gab2 occurs early in breast cancer development

Emmy D.G. Fleuren1, Sandra O’Toole1,2, Ewan K. Millar1, Catriona McNeil1, Elena Lopez-Knowles1, Alice Boulghourjian1, David R. Croucher1, Daniel Schramek3, Tilman Brummer1, Josef M. Penninger3, Robert L. Sutherland1,2 and Roger J. Daly1,2

1 Cancer Research Program, Garvan Institute of Medical Research, Sydney, NSW, Australia 2 St. Vincent’s Clinical School, University of New South Wales, Australia 3 Institute of Molecular Biotechnology of the Austrian Academy of Sciences, Vienna, Austria

Gab2, a docking-type signaling protein with demonstrated oncogenic potential, is overexpressed in breast cancer, but its prognostic significance and role in disease evolution remain unclear. Immunohistochemical detection of Gab2 in a large cohort of primary human breast cancers of known outcome revealed that while Gab2 expression was positively correlated with increased tumor grade, it did not correlate with disease recurrence or breast cancer-related death in the total cohort or in patients stratified according to lymph node, estrogen receptor (ER) or HER2 status. Interestingly, analysis of a ‘‘progression series’’ that included premalignant and preinvasive breast lesions as well as samples of metastatic disease revealed that Gab2 expression was significantly enhanced in the earliest lesion examined, usual ductal hyperplasia, with a further increase detected in ductal carcinoma in situ (DCIS). Furthermore, expression was less in invasive cancers and lymph node metastases than in DCIS, but still higher than in normal breast. These findings indicate that while Gab2 expression is not prognostic in breast cancer, its role in early disease evolution warrants further analysis, as Gab2 and its effectors may provide targets for novel strategies aimed at preventing breast cancer development.

Gab2 is a docking-type signaling protein that recruits the ty- and MCF-10A mammary epithelial cells.3 In addition, Gab2 rosine phosphatase Shp2 and the p85 subunit of PI3K and overexpression has been detected in breast and gastric can- thereby positively regulates the Ras/Erk and PI3K pathways cers, metastatic melanoma and acute myeloid leukemia in response to activation of a variety of receptor and recep- (AML)/myelodysplastic syndrome (MDS).4–8 One mechanism tor-associated tyrosine kinases.1 Strong evidence has emerged underpinning Gab2 overexpression in breast cancer, mela- supporting a functional role for Gab2 in development of sev- noma and AML/MDS is amplification of the GAB2 gene eral human malignancies. When uncoupled from negative at chromosome 11q14.1, which also occurs in ovarian can- feedback regulation, Gab2 is transforming in both fibroblasts2 cer.6–11 Importantly, use of transgenic/knock-out approaches has identified a critical role for Gab2 in particular mouse models of tumorigenesis. Gab2 ablation in mice prevents Key words: tyrosine kinase, signal transduction, ductal carcinoma transformation of myeloid cells by Bcr-Abl or mutant in situ Shp2.12,13 In addition, Gab2 overexpression in the mammary Abbreviations: AML: acute myeloid leukemia; Cdk: cyclin- gland accelerates tumorigenesis induced by active erbB2,9 dependent kinase; DCIS: ductal carcinoma in situ; ER: estrogen whereas Gab2 deficiency either attenuates erbB2-induced receptor; Erk: extracellular signal regulated kinase; Gab2: Grb2- mammary tumorigenesis9 or suppresses metastatic spread to associated binder 2; HER2: human epidermal growth factor receptor the lungs without major effects on tumor growth,14 possibly 2; IDCa: invasive ductal carcinoma; IHC: immunohistochemistry; reflecting the use of different Gab2 gene-knockout alleles MDS: myelodysplastic syndrome; MEK: mitogen activated protein and/or oncogenic forms of erbB2. kinase/Erk kinase; PI3K: phosphatidyinositol 3-kinase; Shp2: src Although Gab2 is implicated in breast cancer develop- homology 2-containing phosphatase 2; UDH: usual ductal ment, the only studies that have analyzed Gab2 protein 4,9

Short Report hyperplasia expression in primary breast cancers have lacked the power Grant sponsors: National Health and Medical Research Council of to accurately determine the frequency of Gab2 upregulation Australia, the Petre Foundation, the RT Hall Trust, Cancer Institute and its association with clinicopathological characteristics NSW Clinical Research, Early Career Development Fellowships and patient outcome. In addition, its expression during early DOI: 10.1002/ijc.25172 breast cancer evolution and disease progression is uncharac- History: Received 13 Jul 2009; Accepted 7 Jan 2010; Online 19 Jan terized. To address these issues, we characterized Gab2 2010 expression in a large and well-deﬁned cohort of primary Correspondence to: Roger J. Daly, Cancer Research Program, Garvan breast cancer specimens, as well as a separate cohort that Institute of Medical Research, Sydney, NSW 2010, Australia, Tel.: included premalignant and preinvasive breast lesions as well 61-2-9295-8333, Fax: 61-2-9295-8321, E-mail: [email protected] as regional lymph node metastases.

Int. J. Cancer: 127, 1486–1492 (2010) VC 2010 UICC Fleuren et al. 1487

Material and Methods therapy (38%) or both (24%). Cases were prospectively Generation of anti-Gab2 polyclonal antibody, Western followed-up for a median of 64 months and outcome events blotting and immunohistochemistry measured were: recurrence (local or distant) (25%), metasta- An affinity-purified sheep polyclonal antibody was generated sis (23%) and all deaths, but only breast cancer-related deaths against the C-terminal Gab2 peptide sequence WTDV (18%) were considered for survival analyses. RQSSEPSKGAKL in association with Symansis (Auckland, To complement the data on expression of Gab2 from the New Zealand). Although the peptide sequence exhibits simi- outcome series, a ‘‘progression series’’ was assembled consist- larity to the Gab1 C-terminus, lack of crossreactivity of the ing of tissue microarrays (TMAs) containing usual ductal antibody against Gab1 was confirmed by Western blot and hyperplasia (UDH; n ¼ 39), ductal carcinoma in situ (DCIS) immunohistochemistry (IHC; see Results). The peptide anti- nuclear grades 1 (n ¼ 12), 2 (n ¼ 54), and 3 (n ¼ 34) and gen sequence is poorly conserved in Gab3. Western blot anal- regional lymph node metastases (n ¼ 80), all derived from ysis was undertaken as previously described.4,15 The anti-b- randomly selected primary breast cancer resections of actin antibody (clone AC-15) was from Sigma (Castle Hill, patients with either pure DCIS (n ¼ 8), DCIS with invasive Australia). Before immunohistochemical staining, formalin- ductal carcinoma (IDC; n ¼ 37) or IDC alone (n ¼ 49). fixed, paraffin-embedded breast cancer cell line pellets and Because 1 patient can have several different types of lesion human or mouse tissue sections were dewaxed in xylene and (e.g., ductal hyperplasia adjacent to DCIS, or multiple lymph subsequently rehydrated. Antigen retrieval was performed in node metastases), cores were scored per lesion and not per a low pH target retrieval solution (S1699, DAKO Corporation, patient. Normal ducts (n ¼ 43) derived from patients who Carpinteria, CA) in a pressure cooker for 10 sec at target pres- had undergone reduction mammoplasty were also included sure. IHC was performed using a Dako autostainer (DAKO in this series, none of which contained premalignant breast Corporation). Nonspecific background staining was prevented lesions or breast cancer within the surgical specimen. This by blocking with hydrogen peroxide (3% v/v). Sections were ‘‘progression series’’ (excluding the IDC cases for which the incubated with the anti-Gab2 antibody (1:100) for 60 min at clinicopathological features have already been described) had room temperature. Substitution of the primary antibody by a mean age of 54.3 years and a mean tumor size of 25.4 mm. normal sheep serum or diluent served as an additional negative Twelve percent of the invasive carcinomas were grade 1, 36% control. Sections were then incubated with rabbit-anti-sheep were grade 2 and 52% were grade 3. Hormone and HER2 re- antibody (Vector Laboratories, Burlingame, CA); (1:200 in dilu- ceptor status or additional clinical information was not avail- ent) for 15 min, followed by the addition of Envisionþ Rabbit able on these cases. (DAKO Corporation) for 15 min. Gab2 staining was then Prior approval for this study was obtained from the visualized by incubation of the slides with DAKO DABþ Human Research Ethics Committee of St. Vincent’s Hospital, chromogen (DAKO Corporation) for 10 min. Slides were coun- Sydney (HREC SVH H94/080, HREC 06338 SVH H00 036). terstained with haematoxylin, dehydrated and coverslipped. Slides were coscored for Gab2 expression by 2 independent Statistical analyses observers that included a breast pathologist (S.O’T.). For each Statistical evaluation was performed using Statview 5.0 Soft- tissue microarray section, the level of Gab2 staining was ware (Abacus Systems, Berkeley, CA). A p-value of <0.05 assessed with respect to both the percentage of cells expressing was accepted as statistically significant. Baseline characteris- the protein (0–100%) and the intensity of staining (0–3). The tics of the cohort were defined using simple frequency distri- scoring system for the staining intensity was as follows: 0, no butions. Patient outcomes were analyzed using Kaplan-Meier cellsstainingpositive;1,weakstainingintensity;2,moderate analysis, and a one way analysis of variance (ANOVA) with staining intensity; 3, high staining intensity. Finally, an H-score Fisher’s least significant difference post-hoc test was used to was designated to each core, which was generated by multiply- determine the relationship between Gab2 expression levels ing the percentage of cells stained by the predominant intensity and clinicopathological and molecular markers. score. All the scores were based on cytoplasmic staining. The H-score was used for further analysis. Results Our initial study on Gab2 in breast cancer used a commer- 4

cially available polyclonal Gab2 antibody for IHC, however, Short Report Patient characteristics subsequent batches of this antibody failed to exhibit appro- The cohort of primary human breast cancers of known out- priate selectivity for this application. Therefore, a panel of come has been described previously.16 Brieﬂy, 40% of tumors polyclonal antisera against Gab2 was generated in sheep and were >20 mm with a mean size of 19.6 mm, 45% were grade tested for their sensitivity and speciﬁcity in detecting Gab2. >2, 43% were lymph node positive, 68% were estrogen recep- AC-terminal antibody detected marked Gab2 expression in tor (ER) positive, 57% were progesterone receptor (PR) posi- MDA-MB-134, 175 and 468, and low levels in HMEC- tive and 18% were HER2 FISH positive (>2.2 ratio of 219-4 and BT-549 cells, as previously reported (Fig. 1a).4 HER2:chr.17 centrosome). Median age was 54 years and Prolonged exposure of the blot indicated that the antibody patients were treated with endocrine therapy (49%), chemo- was highly selective for Gab2. In particular, it did not detect

Int. J. Cancer: 127, 1486–1492 (2010) VC 2010 UICC 1488 Overexpression of the signal transducer Gab2

Figure 2. Immunohistochemistry of human breast cancer specimens. Representative specimens demonstrating negative, weak, moderate and high staining intensity are shown. All images are 200 magniﬁcation, haematoxylin counterstain. The higher power inset highlights cytoplasmic localization of Gab2.

expressing cell lines (data for BT-549 is shown in Fig. 1b). Normal human tissue controls were also utilized, with testis as a positive control and skeletal muscle as a negative control, based on Northern Blot results.17 Gab2 staining was strong in the seminiferous tubules, whereas no Gab2 immunopositivity was detected in skeletal muscle (Fig. 1b). In addition, we utilized specific tissues from wild-type and Gab2 gene knock-out mice18 to test the specificity of our antibody. As with human testis tissue, wild-type mouse testis exhibited Figure 1. Characterization of anti-Gab2 polyclonal antibody. (a) strong staining in the seminiferous tubules, but this was Testing of antibody by Western blot. Lysates from the normal absent in the gene knock-out, confirming its dependency on human mammary epithelial cell strain HMEC-219-4, the indicated Gab2 expression (Fig. 1b). Wild-type mammary gland exhib- breast cancer cell lines, and control or Gab2-overexpressing MCF- ited weak staining that was reduced to background in corre- 10A cells19 were subjected to Western blot analysis as indicated. sponding knock-out tissue (data not shown). (b) Validation of the antibody for immunohistochemistry. We then analyzed Gab2 protein expression in a previously 16 Immunohistochemistry for Gab2 was performed on formalin-fixed, characterized breast cancer cohort. Gab2 immunopositivity paraffin-embedded breast cancer cell line pellets, human testis (defined as H-score > 0, see Fig. 2) was observed in 94% (n and skeletal muscle and wild-type mouse testis and Gab2 gene ¼ 236) of 251 patients and the average H-score of the whole knockout testis. Cell pellet images are 400 magnification, all cohort was 136. The distribution of staining intensity over tissues are 200 magnification. The higher power image in the the cohort was: Gab2 low (H-score 100), n ¼ 140; medium inset highlights localization of Gab2 at the plasma membrane and (H-score 101–200), n ¼ 78; and high (H-score > 200), n ¼ 33. Examples of breast cancer specimens negative for Gab2, Short Report in the cytoplasm. or exhibiting low, medium and high expression, are shown in Figure 2. When present, Gab2 expression was cytoplasmic, as Gab1, which has a lower gel mobility and is expressed in all expected (Fig. 2). As reported previously,4 normal breast epi- of these cell lines.4 thelium was either negative for Gab2 or exhibited weak stain- All Gab2-positive cell lines demonstrated cytoplasmic ing. In a recent study, GAB2 gene amplification at 11q14.1 expression by IHC and a higher staining intensity at the was detected in 8–15% of primary breast cancers.11 However, plasma membrane (Fig. 1b and data not shown). Notably, our data indicate that Gab2 overexpression is a common the intensity of expression corresponded with the Western event in this disease. Consequently, while gene amplifica- Blot data, with little or no staining observed in the low- tion represents 1 mechanism leading to deregulated Gab2

Int. J. Cancer: 127, 1486–1492 (2010) VC 2010 UICC Fleuren et al. 1489

Figure 3. Relationship between Gab2 expression, clinicopathological parameters and patient outcome. (a) Relationship between Gab2 and tumor grade. (b–e) Kaplan-Meier curves for breast cancer-specific death. There was a progressive increase in Gab2 H-score between grades Short Report 1 to 2 and 2 to 3 of breast cancer (a). The number of samples per group was: Grade 1, 38 cases; Grade 2, 92 cases; Grade 3, 121 cases. Error bars indicate 95% CI. *p < 0.05, **p < 0.01. There was no prognostic significance of high Gab2 expression in the cohort as a whole (b), or in ER positive (c), ER negative (d), or HER2 amplified (e) subgroups, with all analyses exhibiting p > 0.05. expression in breast cancer,9,11 aberrant transcriptional or ated with tumor size in this cohort. In addition, although post-transcriptional regulation must also contribute. previous research demonstrated a potential link between Although Gab2 enhances cell proliferation in in vitro Gab2 and metastasis,14 and Gab2 is present in a gene expres- models of mammary tumorigenesis,9,19 Gab2 was not associ- sion signature associated with lymph node metastasis in

Int. J. Cancer: 127, 1486–1492 (2010) VC 2010 UICC 1490 Overexpression of the signal transducer Gab2

breast cancer patients,20 no correlation was found with lymph node status. Previous research using breast cancer cell lines demonstrated a trend toward increased Gab2 expression in ER-positive versus ER-negative lines, although this did not reach significance.4 Therefore, it was of interest to examine whether a similar correlation could be demonstrated in a large cohort of primary breast cancers. However, such a relationship was not found. Also, Gab2 expression did not correlate with progesterone receptor- or HER2-positivity. However, significantly higher Gab2 H-scores were associated with increases in tumor grade from 1 to 3 (Fig. 3a). To identify whether there was an association between Gab2 protein expression levels and patient outcome for the whole cohort, Kaplan-Meier survival analysis (log-rank test) was performed. The frequency distribution for Gab2 was analyzed to determine if any biologically significant cut points were observed. None were seen and optimal cut point determination by serial survival analysis similarly revealed no cut point. Therefore, the median H-score for Gab2 expression (H > 100) within the IDC cohort was used to dichotomize the cohort into Gab2-high and -low groups. Kaplan-Meier analysis revealed no significant differences in disease recurrence (data not shown) or breast cancer-specific death (Fig. 3b). Subgroup analysis was then performed to determine if Gab2 was prognostic within specific patient groups. Although Gab2 is estrogen-regulated,4 there was no association between Gab2 expression and outcome in ER-positive patients (Fig. 3c). Gab2 functions downstream of the epidermal growth factor receptor (EGFR) and erbB2,9,19 and both of these receptors are expressed at higher levels in ER-negative versus ER- positive breast cancers.21 Therefore, we investigated the relationship between Gab2 levels and patient outcome in the ER- negative subgroup. However, Gab2 expression was not related to patient outcome in these patients (Fig. 3d). Furthermore, although Gab2 enhances erbB2-induced tumorigenesis in a transgenic mouse model,9 Gab2 was not associated with poor prognosis in HER2-positive breast cancer patients (Fig. 3e). Finally, no association with outcome was detected in the Figure 4. Gab2 expression during breast cancer evolution and lymph node-negative or-positive subgroups (data not shown). progression. (a) Photomicrographs of Gab2 expression in a Next, we determined at which stage of breast cancer de- histological progression series. ‘‘N’’ denotes a normal duct. Note velopment Gab2 overexpression is first observed, and low levels of Gab2 in normal ducts and increasing expression in whether further alterations in expression are associated with lesions with greater architectural and cytological atypia. All images metastatic progression. To address this, a cohort that are 200 magnifications, haematoxylin counterstain. (b) Increased included premalignant lesions including UDH and DCIS Gab2 expression occurs early in breast cancer evolution. Box plot (grades 1, 2, and 3), as well as regional lymph node metasta- of mean Gab2 H-score in each lesion. UDH, usual ductal hyperplasia; DCIS, ductal carcinoma in situ; IDC, invasive ductal

Short Report ses, was subjected to Gab2 IHC. Notably, a significant increase in Gab2 expression was detected in the earliest carcinoma. *p < 0.01 and **p < 0.005. (c) Photomicrograph lesion examined, UDH, and all of the different types of showing a decrease in Gab2 expression in invasive ductal lesions exhibited significantly higher Gab2 expression when carcinoma, compared to the adjacent DCIS. compared with the normal samples (Figs. 4a and b). Also, although there was no difference in Gab2 expression between DCIS grade 2 or 3 were compared to IDC or lymph node UDH and DCIS grade 1, there was a significant increase in metastases, and expression in the latter 2 types of lesion was Gab2 expression between UDH and DCIS grade 2, and similar. Furthermore, for 12 patients, DCIS and IDC were between DCIS grade 1 and DCIS grade 3 (p ¼ 0.011). Inter- present in the same core and 11/12 (92%) exhibited a estingly, there was a significant decrease in expression when reduction in Gab2 expression in the IDC compared to the

Int. J. Cancer: 127, 1486–1492 (2010) VC 2010 UICC Fleuren et al. 1491 associated DCIS (Fig. 4c). Therefore, Gab2 overexpression is early breast lesions is deregulation of cell cycle control mechan early event in breast cancer development, but it does not anisms. The Retinoblastoma (Rb) tumor suppressor is a criti- increase further in the progression to invasive or metastatic cal regulator of cell cycle entry via control of E2F family disease. transcription factors, and Rb is itself regulated by specific cyclin/Cdk complexes, including cyclin D1/Cdk4.23 Aberra- Discussion tions in the Rb pathway are known to occur early in breast In this manuscript, we report that Gab2 expression was not cancer evolution, as cyclin D1 protein levels increase in the associated with outcome in a large breast cancer cohort or its transition from normal mammary epithelium to UDH, and constituent subgroups. This was a surprising result, given from the latter pathology to DCIS.24 Furthermore, E2F-1 strong evidence from cell culture and mouse models linking expression is enhanced in DCIS compared to the normal Gab2 with mammary tumorigenesis and metastatic progres- breast.25 Importantly, Gab2 is a direct E2F target gene that sion.9,14,19,22 However, in a recently published study on breast mediates Akt activation, thereby attenuating the proapoptotic cancer patients, amplification of the GAB2 locus at 11q14.1 effects of the E2F-induced transcriptional program.26 Indeed, was not associated with overall survival,11 supporting our MCF-10A cells overexpressing Gab2 exhibit enhanced Akt findings. Consequently, while Gab2 possesses properties char- activity.19 Therefore, we hypothesize that increased E2F activ- acteristic of a mammary oncogene, these data indicate that ity early in breast cancer development leads to increased effects of increased Gab2 protein expression or gene copy Gab2 expression, and Gab2 contributes to tumorigenesis by number on disease course are not discernable for established providing both proliferative and survival signals. Also, it breast cancers. remains possible that Gab2 promotes progression from In addition, we detected a marked increase in Gab2 in situ carcinoma to invasive cancer, although as expression expression early in breast cancer evolution. The latter finding of Gab2 is lower in invasive disease versus DCIS (but still is potentially highly significant, as a functional role for Gab2 higher than in normal tissue; Fig. 4), the effect must be medi- in this setting is supported by published in vitro studies. In ated via cooperation with an additional oncogene. In support 3D-culture, MCF-10A cells expressing Gab2 at levels compa- of the latter hypothesis, Gab2 cooperates with erbB29 and rable to those detected in breast cancer cell lines generate active Src22 to promote invasive behavior in the MCF-10A acini (spheroids) with enhanced diameter as a result of 3D culture system. These potential roles of Gab2 early in increased cellular proliferation, and further enhancement of breast cancer evolution warrant further investigation, particu- Gab2 expression leads to defects in luminal clearance.19 In larly as targeting of its downstream effector pathways using addition, Gab2 cooperates with antiapoptotic proteins such as Shp2-, MEK- or PI3K-directed small molecule inhibitors 9 Bcl-XL to induce luminal filling in this model. Notably, such could provide a basis for breast cancer-preventative strategies. aberrant morphogenesis of MCF-10A cells is comparable to the histological phenotype of DCIS, which is also character- Acknowledgements ized by partial or complete luminal filling. A potential mech- Sandra O’Toole and David Croucher are recipients of Cancer Institute NSW anism that may underpin the increase in Gab2 expression in Clinical Research and Early Career Development Fellowships, respectively.

References

1. Wohrle FU, Daly RJ, Brummer T. 5. Lee SH, Jeong EG, Nam SW, Lee JY, Yoo Classification of Melanoma. Clin Cancer Function, regulation and pathological roles NJ, Lee SH. Increased expression of Gab2, Res 2009;15:4288–91. of the Gab/DOS docking proteins. Cell a scaffolding adaptor of the tyrosine kinase 9. Bentires-Alj M, Gil SG, Chan R, Commun Signal 2009;7:22. signalling, in gastric carcinomas. Pathology Wang ZC, Wang Y, Imanaka N, 2. Lynch DK, Daly RJ. PKB-mediated 2007;39:326–9. Harris LN, Richardson A, Neel BG, negative feedback tightly regulates 6. Zatkova A, Schoch C, Speleman F, Poppe Gu H. A role for the scaffolding adapter mitogenic signalling via Gab2. EMBO J B, Mannhalter C, Fonatsch C, Wimmer K. GAB2 in breast cancer. Nat Med 2006;12: 2002;21:72–82. GAB2 is a novel target of 11q 114–21. 3. Brummer T, Larance M, Abreu MTH, amplification in AML/MDS. Genes 10. Brown LA, Kalloger SE, Miller MA, Shih Ie Lyons RJ, Timpson P, Emmerich CH, Chromosomes Cancer 2006;45: M, McKinney SE, Santos JL, Swenerton K, Fleuren ED, Lehrbach GM, Schramek D, 798–807. Spellman PT, Gray J, Gilks CB, Huntsman Short Report Guilhaus M, James DE, Daly RJ. 7. Horst B, Gruvberger-Saal SK, Hopkins BD, DG. Amplification of 11q13 in ovarian Phosphorylation-dependent binding of 14– Bordone L, Yang Y, Chernoff KA, Uzoma carcinoma. Genes Chromosomes Cancer 3–3 terminates signalling by the Gab2 I, Schwipper V, Liebau J, Nowak NJ, 2008;47:481–9. docking protein. EMBO J 2008;27:2305–16. Brunner G, Owens D, et al. Gab2-mediated 11. Bocanegra M, Bergamaschi A, Kim YH, 4. Daly RJ, Gu H, Parmar J, Malaney S, signaling promotes melanoma metastasis. Miller MA, Rajput AB, Kao J, Langerod A, Lyons RJ, Kairouz R, Head DR, Henshall Am J Pathol 2009;174:1524–33. Han W, Noh DY, Jeffrey SS, Huntsman SM, Neel BG, Sutherland RL. The docking 8. Chernoff KA, Bordone L, Horst B, Simon DG, Borresen-Dale AL, et al. Focal protein Gab2 is overexpressed and estrogen K, Twadell W, Lee K, Cohen JA, Wang S, amplification and oncogene dependency of regulated in human breast cancer. Silvers DN, Brunner G, Celebi JT. GAB2 GAB2 in breast cancer. Oncogene 2009; Oncogene 2002;21:5175–81. Amplifications Refine Molecular doi:10.1038/onc.2009.364.

Int. J. Cancer: 127, 1486–1492 (2010) VC 2010 UICC 1492 Overexpression of the signal transducer Gab2

12. Sattler M, Mohi MG, Pride YB, Quinnan patient outcome and tamoxifen cancer outcomes. Lancet 2003;361: LR, Malouf NA, Podar K, Gesbert F, responsiveness in ER positive invasive 1590–6. Iwasaki H, Li S, Van Etten RA, Gu H, ductal carcinoma of the breast. Br J Cancer 21. Nicolini A, Carpi A, Tarro G. Biomolecular Grifﬁn JD, et al. Critical role for Gab2 in 2009;100:123–33. markers of breast cancer. Front Biosci transformation by BCR/ABL. Cancer Cell 17. Nishida K, Yoshida Y, Itoh M, Fukada T, 2006;11:1818–43. 2002;1:479–92. Ohtani T, Shirogane T, Atsumi T, 22. Bennett HL, Brummer T, Jeanes A, Yap 13. Mohi MG, Williams IR, Dearolf CR, Chan Takahashi-Tezuka M, Ishihara K, Hibi M, AS, Daly RJ. Gab2 and Src co-operate in G, Kutok JL, Cohen S, Morgan K, Boulton Hirano T. Gab-family adapter proteins act human mammary epithelial cells to C, Shigematsu H, Keilhack H, Akashi K, downstream of cytokine and growth factor promote growth factor independence and Gilliland DG, et al. Prognostic, therapeutic, receptors and T- and B-cell antigen disruption of acinar morphogenesis. and mechanistic implications of a mouse receptors. Blood 1999;93:1809–16. Oncogene 2008;27:2693–704. model of leukemia evoked by Shp2 18. Wada T, Nakashima T, Oliveira-dos-Santos 23. Bosco EE, Knudsen ES. RB in breast (PTPN11) mutations. Cancer Cell 2005;7: AJ, Gasser J, Hara H, Schett G, Penninger cancer: at the crossroads of tumorigenesis 179–91. JM. The molecular scaffold Gab2 is a and treatment. Cell Cycle 2007;6:667–71. 14. Ke Y, Wu D, Princen F, Nguyen T, Pang crucial component of RANK signaling and 24. Alle KM, Henshall SM, Field AS, Y, Lesperance J, Muller WJ, Oshima RG, osteoclastogenesis. Nat Med 2005;11: Sutherland RL. Cyclin D1 protein is Feng GS. Role of Gab2 in mammary 394–9. overexpressed in hyperplasia and tumorigenesis and metastasis. Oncogene 19. Brummer T, Schramek D, Hayes VM, intraductal carcinoma of the breast. Clin 2007;26:4951–60. Bennett HL, Caldon CE, Musgrove EA, Cancer Res 1998;4:847–54. 15. Janes PW, Daly RJ, deFazio A, Sutherland Daly RJ. Increased proliferation and altered 25. Zhang SY, Liu SC, Al-Saleem LF, RL. Activation of the Ras signalling growth factor dependence of human Holloran D, Babb J, Guo X, Klein-Szanto pathway in human breast cancer cells mammary epithelial cells overexpressing AJ. E2F-1: a proliferative marker of breast overexpressing erbB-2. Oncogene 1994;9: the Gab2 docking protein. J Biol Chem neoplasia. Cancer Epidemiol Biomarkers 3601–8. 2006;281:626–37. Prev 2000;9:395–401. 16. Millar EKA, Anderson LR, McNeil CM, 20. Huang E, Cheng SH, Dressman H, Pittman 26. Chaussepied M, Ginsberg D. O’Toole SA, Pinese M, Crea P, Morey A, J, Tsou MH, Horng CF, Bild A, Iversen ES, Transcriptional regulation of AKT Biankin AV, Henshall SM, Musgrove EA, Liao M, Chen CM, West M, Nevins JR, activation by E2F. Mol Cell 2004; Sutherland RL, Butt AJ. BAG-1 predicts et al. Gene expression predictors of breast 16:831–7. Short Report

The key hypoxia regulated gene CAIX is upregulated in basal-like breast tumours and is associated with resistance to chemotherapy

1 2,3 1 4 2,3 1 2,3 1 1 EY Tan , M Yan , L Campo , C Han , E Takano , H Turley , I Candiloro , F Pezzella , KC Gatter , 5,6 5,7 5,8 9 5,9 5 4 ,2,3 EKA Millar , SA O’Toole , CM McNeil , P Crea , D Segara , RL Sutherland , AL Harris and SB Fox*

1Nuffield Department of Clinical Laboratory Sciences, John Radcliffe Hospital, Oxford OX3 9DU, UK; 2Pathology, Peter MacCallum Cancer Centre,

St Andrews Place, East Melbourne 3002, Australia; 3Pathology, University of Melbourne, Melbourne, Victoria, Australia; 4Cancer Research UK Molecular

Oncology Laboratory, Weatherall Institute of Molecular Medicine, John Radcliffe Hospital, Oxford OX3 9DU, UK; 5Cancer Research Program, Garvan 6 Institute of Medical Research, 384 Victoria Street, Darlinghurst, NSW 2010, Australia; Department of Anatomical Pathology, South Eastern Area 7 Laboratory Service, St George Hospital, Kogarah, NSW 2217, Australia; Department of Pathology (Sydpath), St Vincent’s Hospital, Darlinghurst, NSW 8 9 2010, Australia; Department of Medical Oncology, Westmead Hospital, Sydney, NSW, Australia; Department of Surgical Oncology, St Vincent’s Clinic,

Darlinghurst, NSW, Australia

Basal-like tumours account for 15% of invasive breast carcinomas and are associated with a poorer prognosis and resistance to

therapy. We hypothesised that this aggressive phenotype is because of an intrinsically elevated hypoxic response. Microarrayed a tumours from 188 patients were stained for hypoxia-inducible factor (HIF)-1 , prolyl hydroxylase (PHD)1, PHD2, PHD3 and factor

inhibiting HIF (FIH)-1, and carbonic anhydrase (CA) IX stained in 456 breast tumours. Tumour subtypes were correlated with

standard clincopathological parameters as well as hypoxic markers. Out of 456 tumours 62 (14%) tumours were basal-like. These

tumours were positively correlated with high tumour grade (Po0.001) and were associated with a significantly worse disease-free

survival compared with luminal tumours (Po0.001). Fifty percent of basal-like tumours expressed HIF-1a, and more than half

expressed at least one of the PHD enzymes and FIH-1. Basal-like tumours were nine times more likely to be associated with CAIX

expression (Po0.001) in a multivariate analysis. Carbonic anhydrase IX expression was positively correlated with tumour size

(P ¼ 0.005), tumour grade (Po0.001) and oestrogen receptor (ER) negativity (Po0.001). Patients with any CAIX-positive breast

tumour phenotype and in the basal tumour group had a significantly worse prognosis than CAIX-negative tumours when treated with

chemotherapy (Po0.001 and P ¼ 0.03, respectively). The association between basal phenotype and CAIX suggests that the more

aggressive behaviour of these tumours is partly due to an enhanced hypoxic response. Further, the association with chemoresistance

in CAIX-positive breast tumours and basal-like tumours in particular raises the possibility that targeted therapy against HIF pathway or

downstream genes such as CAs may be an approach to investigate for these patients.

British Journal of Cancer (2009) 100, 405 – 411. doi:10.1038/sj.bjc.6604844 www.bjcancer.com & 2009 Cancer Research UK

Keywords: breast; hypoxia; carbonic anhydrase; predictive; basal

Current prognostic and predictive markers of invasive breast HER2 overexpressing and basal (Sorlie et al, 2001, 2003), the latter cancer rely largely on histological indicators such as tumour size, having a particularly poor clinical outcome (Sorlie et al, 2003; Abd grade, nodal status and receptor status, which define broad patient El-Rehim et al, 2004; Nielsen et al, 2004). groups for different treatment regimens. However, it is recognised Basal-like breast cancers derive their name from their charac- that there are many patients within such groups who are teristic expression of basal and myoepithelial markers such as overtreated or undertreated with conventional therapies, strongly cytokeratins (CK) 5, 14 and 17 (Jarasch et al, 1988; Wetzels et al, suggesting that current stratification does not fully account for the 1991). These tumours account for up to 15% of all invasive breast molecular heterogeneity and diverse biological behaviour of breast cancers (van de Rijn et al, 2002; Abd El-Rehim et al, 2004; Nielsen cancer. Recently cDNA microarray studies have classified breast et al, 2004; Matos et al, 2005). Basal-like tumours are more cancers broadly into five subtypes, based on their different frequently observed in patients with BRCA1-related cancers expression profiles: normal breast-like, luminal A (oestrogen (Foulkes et al, 2004). In contrast to the normal breast-like and Genetics and Genomics receptor (ER) positive), luminal B (ER positive and proliferative), luminal subtypes, basal-like breast cancers have a particularly aggressive phenotype, being more likely to be high-grade, ER negative (Malzahn et al, 1998; Korsching et al, 2002; Nielsen et al, *Correspondence: Professor SB Fox, Pathology, Peter MacCallum 2004), and are associated with a higher proliferative index Cancer Centre, St Andrews Place, East Melbourne 3002, Australia; (Korsching et al, 2002; Matos et al, 2005). Expression of basal E-mail: [email protected] CKs is also associated with loco-regional recurrence, distant Received 4 September 2008; revised 31 October 2008; accepted 11 metastasis and a poor overall survival in node-positive and node- November 2008 negative patients (Malzahn et al, 1998; van de Rijn et al, 2002). In CAIX in basal breast cancer EY Tan et al 406 addition, basal-like tumours differ from normal breast-like and PHD1, PHD2, PHD3, FIH-1 and HIF-1a were used as positive luminal tumours in their response to chemotherapy. Although controls. In each run a tumour is known to be positive for CAIX basal-like tumours do not respond well to anthracycline-based was included. therapy (Banerjee et al, 2006), they show an impressive Scoring was carried out by two observers simultaneously. ER, pathological response to paclitaxel followed by a combination of HER2, EGFR and CK5/6 staining was used to classify tumours into 5-fluorouracil, doxorubicin and cyclophosphamide (Rouzier et al, four subgroups as per Neilson et al. basal group (ER negative, 2005). They also have a BRCA1 phenotype and therefore are HER2 negative, CK5/6 and/or EGFR positive), luminal group (ER sensitive to platinum-based therapies (Kennedy et al, 2004). positive, HER2 negative), HER2 group (HER2 positive) and a In addition to being mostly high-grade tumours, a so-called negative group (ER negative, HER2 negative, CK5/6 negative and triple negative phenotype (ER, progesterone receptor (PR) and EGFR negative) (Nielsen et al, 2004). Tumours with any nuclear HER2 negative), basal-type breast cancers may also have a staining for HIF-1a were considered positive in the analysis. The characteristic central acellular or necrotic zone (Tsuda et al, levels of staining for PHD1, PHD2 and PHD3 were scored with 1999; Fulford et al, 2006). Tumour necrosis is a consequence of respect to the percentage of cells and the intensity of staining in hypoxia. Hypoxia is a predictor of poor disease-free and overall the cytoplasm. The scoring system for intensity was: 0 ¼ no survival in breast cancer (Fisher et al, 1993; Hockel and Vaupel, staining, 1 ¼ weak staining, 2 ¼ moderate staining, 3 ¼ strong 2001), and is associated with resistance to radiation and certain staining; and the scoring system for percentage was: 0 ¼ no cells chemotherapeutic regimes (Teicher et al, 1990). As the hypoxia- staining positive, 1 ¼ p10% cells staining positive, 2 ¼ 11–50% inducible factor (HIF) pathway has been shown to be pivotal in the positive cells, 3 ¼ 51–80% positive cells, 4 ¼ X80% positive cells. hypoxic tumour response and to mediate many of the above Tumours with a cytoplasmic intensity score of X2 and a adverse effects, and in view of the link between basal-like breast percentage score of X1 were considered positive for PHD1, cancer, hypoxia and an aggressive phenotype, we have examined PHD2 and PHD3 in the analysis, in accordance with their site of the HIF pathway, including the expression of HIF-1a, its regulatory action. Only tumours showing a strong membranous staining in enzymes the prolyl hydroxylase domain enzymes PHD-1, PHD-2, X10% or more cells were considered positive for CAIX (Tan et al, PHD-3, factor inhibiting HIF-1 (FIH-1) and carbonic anhydrase 2007). Tumours with a nuclear intensity score of X2 and a (CA) IX, a downstream target of HIF-1a in a series of percentage score of X1 were considered positive for FIH-1 in the breast cancers, and compared the expression of each marker analysis (Tan et al, 2007). Owing to loss of tumour cores, only 456 by correlating it with various standard clinicopathological out of 621 patients with full phenotype of ER, HER2, EGFR and parameters. CK5/6, which are required to classify tumours were available. Similarly, HIF-1a and the PHDs were available in only 188 out of 456 tumours. MATERIALS AND METHODS Statistical methods Patient characteristics Correlations were evaluated using either the one-way ANOVA or Tumour tissue microarray cores (1 mm cores) were collected from w2 test where appropriate. Kaplan–Meier survival curves were 621 invasive breast carcinomas collected from patients who had calculated using tumour recurrence (defined as the first re- undergone surgery at the John Radcliffe Hospital, Oxford, UK and appearance of tumour at any site after definitive treatment) and from the Garvan Institute, Sydney, Australia. This study has cancer-related death as the endpoints and compared using a log- Ethical Committee approvals (number JR C02.216 and HREC SVH rank test. Binary logistic regression was used for multivariate H94/080, HREC SVH 06336 H00036). All patients had operable analyses and the Cox proportional hazard regression model was breast carcinomas and were not diagnosed with metastatic disease used to identify independent prognostic factors for disease-free at the time of presentation. Information regarding patient and overall survival. The analyses were carried out with SPSS 16.0 characteristics, including age, tumour size, grade, histology, nodal (SPSS Inc., 233 South Wacker Drive, IL, USA). A two-tailed P-value status, ER and HER2 status, were collected from the clinical and test was used in all analyses and a P-value of o0.05 was considered pathological records. The median age of patients included in this statistically significant. study was 55 years (range 24–87 years). Eighty-eight percent of eeisadGenomics and Genetics tumours were invasive ductal of no specific type, 3% were invasive lobular carcinoma and 3% were of other histological classifications RESULTS (data was unavailable for the remaining 6%). Median tumour size was 20 mm and the median tumour grade was two. Forty-four Expression of ER, HER2, CK5/6 and EGFR in invasive percent of patients had nodal disease. Sixty-eight percent of breast cancer tumours were ER positive and 16% were HER2 positive. Patients less than 50 years of age with node-positive, ER-negative tumours Out of 456 tumours 310 (68%) were ER positive; of these, 289 or tumours larger than 3 cm received adjuvant chemotherapy tumours were also HER2 negative and were therefore considered to (cyclophosphamide, methotrexate and 5-fluorouracil (CMF) or be of the luminal group (Table 1). Seventy-five tumours (16%) adriamycin and cyclophosphamide (AC). Patients with hormone- were HER2 positive, regardless of their ER, CK5/6 and EGFR responsive tumours who were more than 50 years of age received 5 status, and were considered to be of the HER2 group. Of the 169 years of endocrine therapy. Patients were followed up for a median ER-negative tumours, 62 (14%) were also HER2 negative but period of 131.9 months. During this time, 137 patients developed CK5/6 and/or EGFR positive; these were considered to be of the recurrence (30.0%) and 99 deaths (21.7%) were considered as basal-like group. Thirty tumours were negative for ER, HER2, CK5/ breast-cancer related. 6 and EGFR and were considered to be of the negative group (Table 1) (Supplementary material Figure 1). Immunohistochemistry Expression of HIF-1a and its regulatory enzymes PHD1, Formalin-fixed paraffin-embedded tissue sections were stained for PHD2, PHD3 and FIH the various hypoxic markers (Supplementary Table 1). Substitu- tion of the primary antibody with phosphate-buffered saline Ninety-four out of 188 tumours (50%) stained positive for HIF-1a served as a negative control. Transfected COS-1 cells expressing (Table 1) (Supplementary material Figure 1). Of these, 64 tumours

British Journal of Cancer (2009) 100(2), 405 – 411 & 2009 Cancer Research UK CAIX in basal breast cancer EY Tan et al 407 Table 1 Correlation analysis of 456 invasive breast carcinomas with Table 2 Correlation analysis of CAIX expression and clinicopathological clinicopathological parameters and CA9 parameters

Luminal HER2 Basal Negative CAIX positive CAIX negative (n ¼ 289) (n ¼ 75) (n ¼ 62) (n ¼ 30) P-value (n ¼ 59) (n ¼ 348) P-value

Patient age 0.64 Patient age 0.96 Median (years) 56.0 55.2 54.0 52.7 Median (years) 55.0 57.0

Tumour size 0.40 Tumour size 0.005 Median (mm) 18.0 22.0 22.5 20.5 Median (mm) 25.0 19.0

Tumour grade o0.001 Tumour grade o 0.001 171424 1096 2 156 16 12 12 2 19 189 358544714 3 56 154

Nodal status 0.75 Nodal status 0.168 Negative 162 40 34 17 Negative 34 242 Positive 125 34 28 13 Positive 40 195

CAIX o0.001 ER status o0.001 Negative 242 27 57 22 Negative 54 110 Positive 14 26 16 3 Positive 19 326

Nuclear HIF1a 0.53 HER2 status 0.065 Negative 54 10 13 1 Negative 43 291 Positive 64 12 13 5 Positive 16 57

Cytoplasmic PHD1 0.76 Nuclear HIF-1a 0.13 Negative 68 10 15 3 Negative 6 54 Positive 51 12 11 3 Positive 15 62

Cytoplasmic PHD2 0.41 CAIX ¼ carbonic anhydrase IX; PHD ¼ prolyl hydroxylase; HIF ¼ hypoxia-inducible Negative 71 9 16 4 factor. no456 because data are not available. Positive 47 11 7 2

Cytoplasmic PHD3 0.87 Negative 70 13 14 3 Table 3 Multivariate analysis binary logistic regression model of the Positive 43 10 11 3 effect of CAIX expression on tumour subtype, using luminal tumours as a baseline (n ¼ 406) Nuclear FIH-1 0.51 Negative 45 6 12 3 Odds ratio 95% CI P-value Positive 80 17 15 3 Tumour size 1.00 0.98– 1.03 0.79 CAIX ¼ carbonic anhydrase IX; PHD ¼ prolyl hydroxylase. no456 because data are Grade 2.7 1.45– 5.00 0.002 not available. Basal tumours 8.9 3.86– 20.29 o0.001 HER2 tumours 2.7 1.16– 6.21 0.02 Negative tumours 1.6 0.40– 6.09 0.52 (68%) were of the luminal group, 13 tumours (14%) were of the basal group, 12 (13%) were of the HER2 group and 5 CAIX ¼ carbonic anhydrase IX. (5%) were of the negative group. Although only 8, 10 and 12% of luminal, basal-like and HER2 tumours, respectively, expressed all three PHD enzymes, the majority expressed at least one PHD Correlation of breast tumours stratified by intrinsic enzyme (63, 52 and 72, respectively) (Table 1). None of the six subtype with CAIX, HIF-1a and its regulatory enzymes negative tumours expressed all three PHD enzymes although all expressed at least one PHD enzyme. More than half of all the Basal-like tumours were 8.9 times more likely to express CAIX (Po0.001, 95% CI 3.86–20.29) than luminal tumours (Table 3). tumours expressed FIH-1 (63% of luminal tumours, 52% of basal o tumours, 68% of HER2 tumours and 50% of negative tumours) This association is significant ( 0.001) even after controlling for (Table 1). the effect of tumour size and grade. There was no significant correlation between breast tumour subtypes and HIF-1a or its regulatory enzymes (P40.05). Correlation of CAIX expression in invasive breast carcinoma with clinicopathological parameters Correlation of breast tumours stratified by intrinsic subtype with clinicopathological parameters and survival Genetics and Genomics CAIX was positive in 59 out of 407 tumours (14%) (Table 2). The outcome majority of these tumours (26 out of 59 (44%)) were of the basal- like group. Carbonic anhydrase IX expression was significantly Basal-like tumours were associated with high tumour grade correlated with a larger tumour size (P ¼ 0.005), high tumour (Po0.001), but were not correlated with patient age, tumour size grade (Po0.001) and ER negativity (Po0.001) (Table 2). There or nodal status (P40.05). Basal-like tumours and HER2 tumours was no correlation with patient age, nodal status or HER2 status shared a similar poor disease-free survival outcome (Supplemen- (P40.05) (Table 2). CAIX expression showed no significant tary Figure 2) and were associated with a significantly shorter correlation with HIF-1a expression (Table 2). disease-free survival compared with luminal tumours (Po0.001,

& 2009 Cancer Research UK British Journal of Cancer (2009) 100(2), 405 – 411 CAIX in basal breast cancer EY Tan et al 408 OR 2.58, 95% CI 1.82–3.65) (Negative tumours were excluded DISCUSSION from the analysis because of the small sample size but are included the figure). mRNA expression profile studies have confirmed the long- standing histological observation that breast cancer is a heterogeneous disease (Perou et al, 2000; Sorlie et al, 2001, CAIX expression and survival in breast cancer patients 2003). These studies have broadly classified breast cancers into four categories: luminal A (strongly ER positive), luminal B (also Patients with tumours expressing CAIX had a significantly worse ER positive but proliferative), HER2 and basal-like groups. disease-free survival (Po0.001) (Figure 1A). There was a Although these categories were identified using unsupervised significant difference in survival in patients with any type of clustering, the groupings also conferred prognostic information, breast carcinoma treated with chemotherapy stratified by CAIX with carcinomas of the basal-like group having a shorter survival (Po0.001) (Figure 1B). Patients with CAIX-positive basal-like than tumours of the luminal group (Sorlie et al, 2001) and being breast cancers who were treated with chemotherapy had a resistant to anthracycline-based chemotherapy (Banerjee et al, significantly shorter overall survival (P ¼ 0.03) than treated 2006). Expression of the hypoxia-inducible factor-1a is also patients with CAIX-negative basal-like breast cancers associated with a poor prognosis and resistance to chemotherapy (Figure 1C). In the absence of chemotherapy treatment, there (Teicher et al, 1981; Schindl et al, 2002; Bos et al, 2003), and in was no significant difference in overall survival among the view of the association between basal-like breast cancer, necrosis, intrinsic subtypes (P ¼ 0.07), or when basal-like carcinomas were their aggressive behaviour and poor response to chemotherapy we stratified according to CAIX expression (P ¼ 0.27). There was no hypothesised that these tumours may have an intrinsically significant difference in overall survival in patients not treated enhanced hypoxic response. We therefore examined the hypoxic with chemotherapy with any type of breast carcinoma (P ¼ 0.07) or pathway in a series of breast cancers. basal-like carcinoma stratified by CAIX (P ¼ 0.27). In a multi- Although there is no generally recognised definition of the variate Cox regression model patients who had tumours that were categories for stratification into intrinsic subgroups (different CAIX positive and who were treated with chemotherapy had a studies using various markers such as CK5, CK14, CK17, EGFR, significantly shorter overall survival (Po0.001), whereas CAIX had KIT, p63 and smooth muscle actin (Nagle et al, 1986; Wetzels et al, no effect on survival in patients not treated with chemotherapy 1991; Malzahn et al, 1998; Tsuda et al, 1999; van de Rijn et al, 2002; (P ¼ 0.48) (Table 4A and B). To test for treatment interactions, Abd El-Rehim et al, 2004; Matos et al, 2005; Fulford et al, 2006) to CAIX, chemotherapy and their interaction variables were entered identify basal-like tumours), in this study we have used the into the Cox regression model as a second block. The interaction of classification developed by Nielsen et al (2004). This uses a chemotherapy with CAIX was significant in the final model combination of ER, HER2, CK5 and EGFR to generate four groups, (P ¼ 0.04) (Table 4C). which in this study gives a frequency of basal-like cancers in this

AB1.0 1.0 CAIX CAIX negative negative 0.8 positive 0.8 positive

0.6 0.6

0.4 0.4

0.2 0.2 Overall survival probability Overall Disease-free survival probability 0.0 0.0 eeisadGenomics and Genetics 02468 101214 0 2 4 6 8101214 Time in years Time in years

C 1.0

CAIX 0.8 negative (n=15) positive (n=11)

0.6

0.4

0.2 Overall survival probability Overall

0.0

02468 101214 Time in years Figure 1 (A) Kaplan–Meier disease-free survival curves stratifying patients by CAIX expression (Po0.001) (n ¼ 407). (B) Kaplan–Meier overall survival curves stratifying all tumours treated with chemotherapy by expression of CAIX (Po0.001) (n ¼ 427). (C) Kaplan–Meier overall survival curves stratifying patients with basal-like tumours treated with chemotherapy by expression of CAIX (P ¼ 0.03) (n ¼ 26).

British Journal of Cancer (2009) 100(2), 405 – 411 & 2009 Cancer Research UK CAIX in basal breast cancer EY Tan et al 409 Table 4A Cox regression model, overall survival, all breast cancers Nevertheless, despite a strong correlation with CAIX, we did not treated with chemotherapy (n ¼ 182) observe a correlation between basal tumours and HIF-1a or between CAIX and HIF-1a in this study. This is unlikely to be due Odds ratio 95% CI P-value to spatial reasons because TMA are ideal for investigating relationships between biomarkers as the same area of tumour is Nodal status 1.77 0.62–5.05 0.29 being examined, but is probably due to the difference in half-lives Grade 2.38 1.36–4.164 0.002 of HIF-1a and CAIX. Hypoxia-inducible factor-1a is rapidly CAIX 3.20 1.79–5.701 o0.001 Size 1.04 1.02–1.058 o0.001 degraded within minutes of reoxygenation (Jiang et al, 1996), whereas CAIX has a half-life of 2–3 days (Turner et al, 2002; CAIX ¼ carbonic anhydrase IX Rafajova et al, 2004). This absence of an association is in agreement with reports (Tomes et al, 2003; Sobhanifar et al, 2005) of the presence of CAIX without HIF-1a expression in Table 4B Cox regression model, overall survival, all breast cancers perinecrotic regions in solid tumours (Sobhanifar et al, 2005). treated with chemotherapy (n ¼ 262) CAIX expression, like a basal-like phenotype, is associated with an increased risk of recurrence and poorer overall survival (Chia et al, Odds ratio 95% CI P-value 2001). In this series, those patients with tumours of either a basal- like phenotype or expressing CAIX were associated with an adverse Nodal status 2.27 1.04– 4.96 0.04 outcome. Nevertheless, in this study and in others reported earlier, Grade 1.90 1.12– 3.25 0.02 CAIX expression is also positively correlated with high tumour grade CAIX 1.33 0.60– 2.95 0.48 Size 1.04 1.02– 1.06 o0.001 and size (Chia et al, 2001; Span et al, 2003), the former feature also characteristic of basal tumours. Thus, hypoxic stress may not be part CAIX ¼ carbonic anhydrase IX of the basal intrinsic subtype but might be due to an association with grade. Although this cannot be completely discounted, our data show that both basal-like and HER2 cancers are significantly Table 4C Cox regression model, overall survival, all breast cancers with associated with high tumour grade (as expected) but the basal group CAIX and chemotherapy as interaction variables (n ¼ 441) was more likely to be CAIX positive, whereas the HER2 subtype was less likely to be CAIX positive. Furthermore, in a multivariate Odds ratio 95% CI P-value analysis including tumour size and grade, basal-like phenotype was still associated with CAIX expression. Thus, the data are strongly Grade 1.98 1.34– 2.92 0.001 suggestive of an enhanced hypoxic drive in this tumour type. This is Age 1.00 0.99– 1.02 0.65 Size 1.04 1.03– 1.05 o0.001 further supported by interrogation of expression data sets (Onco- Oestrogen receptor 0.74 0.46– 1.20 0.22 mine, www.oncomine.org), where there is significant upregulation of Nodal status 2.22 1.15– 4.26 0.02 not only HIF-1a and CAIX but also other HIF-1a-regulated genes such as vascular endothelial growth factor-A, BNIP3 and Glut-1. Additions to model Moreover, patients with basal-like tumours who have not received CAIX 1.09 0.49– 2.42 0.83 chemotherapy have a similar prognosis irrespective of CAIX Chemotherapy 0.75 0.38– 1.49 0.41 expression, whereas patients with basal-like tumours that are CAIX CAIX chemotherapy 2.65 1.03– 6.82 0.04 positive and who have received chemotherapy have a significantly CAIX ¼ carbonic anhydrase IX. shorter overall survival than patients with basal-like tumours that are CAIX negative. This suggests that CAIX-positive basal-like tumours are particularly resistant to chemotherapy, in contrast to CAIX- cohort similar to that described earlier (van de Rijn et al, 2002; negative basal-like tumours, which are able to respond to Abd El-Rehim et al, 2004; Nielsen et al, 2004; Matos et al, 2005). chemotherapy. A potential mechanism for this action is through Central to the hypoxic response is the transcription factor HIF. trapping of chemotherapeutic agents outside the cell when the We have examined the role of the key pathway members regulating extracellular environment is acidic (Mahoney et al, 2003). HIF-mediated transcription, including HIF-1a, the PHDs, FIH and The resistance effect of CAIX/hypoxia seems to hold irrespective CAIX in a series of breast cancers classified into intrinsic subtypes of tumour subtype as, unlike patients not treated with chemo- as described above (Wykoff et al, 2000). We observed that basal- therapy, treated patients with any tumour phenotype positive for like tumours were many times more likely to express CAIX. CAIX also had a significantly worse survival than patients with Carbonic anhydrase IX is a transmembrane protein involved in tumours negative for CAIX in the multivariate analyses. This is in maintaining a low pericellular pH through its reversible conver- keeping with our studies in the neoadjuvant context, in which sion of carbon dioxide and water into carbonic acid (Pastorek et al, CAIX was associated with a shorter relapse-free survival in patients 1994; Opavsky et al, 1996), and has been shown to maintain treated with CMF (Generali et al, 2006a, b), and in patients treated survival of breast tumour cells under hypoxic conditions (Potter in the adjuvant setting, in which CAIX measured by RT–PCR was and Harris, 2003). Carbonic anhydrase IX is not present in normal associated with chemoresistance (Span et al, 2003). Thus, these breast tissues (Bartosova et al, 2002), but is upregulated in invasive data suggest that CAIX is predictive rather than prognostic. tumours with increasing intensity, with increasing distance from Molecular genetic studies of basal-like breast carcinomas have tumour vessels (Span et al, 2003) tightly under the regulation of shown that tumours with a basal-like phenotype have a higher HIF-1a (Grabmaier et al, 2004). Carbonic anhydrase IX expression number of unbalanced chromosomal changes and rate of loss of has been shown to correlate with the distribution of pimonidazole, heterozygosity than luminal-type breast tumours (Korsching et al, Genetics and Genomics a chemical marker of hypoxia (Opavsky et al, 1996; Wykoff et al, 2002; Jones et al, 2004). In addition, basal-like tumours are a 2000) and with hypoxia measured by Eppendorf microelectrode in distinctive feature of BRCA1-associated breast cancers. Although advanced cervical cancer (Loncaster et al, 2001). On the basis of BRCA1 is involved in a large number of cellular processes, it is the these data it has been suggested that CAIX immunohistochemistry maintenance of genomic stability that has been proposed to be is a surrogate marker for detecting hypoxia in tumour samples. the principal factor underlying cancer predisposition in mutation Thus, the association between basal phenotype and CAIX supports carriers (Baldeyron et al, 2002). Thus, it is also of interest that the notion that these tumours have an enhanced hypoxic response hypoxia is recognised to induce genomic instability (Cheng and (Wykoff et al, 2000; Lal et al, 2001). Loeb, 1993; Sutherland, 1998; Semenza, 2000; Chan et al, 2008),

& 2009 Cancer Research UK British Journal of Cancer (2009) 100(2), 405 – 411 CAIX in basal breast cancer EY Tan et al 410 and it is interesting to speculate whether the genomic alterations ACKNOWLEDGEMENTS observed in basal-like breast carcinomas may partly be the result of the enhanced hypoxic response in this subtype. Nevertheless, This study has been supported by the National Medical Research these genomic studies have been carried out on a small number of Council (NMRC), Singapore, the Victorian Breast Cancer Research tumours (Jones et al, 2001, 2004) and additional data are required Consortium, Australia, the Victorian Cancer Biobank, the Euro- to support this thesis. pean Union 6th Framework Euroxy, NHS Biomedical Research In summary, the more aggressive phenotype seen in basal-like Centre Oxford, Cancer Research UK, the National Health and tumours may be at least partly due to the intrinsically enhanced Medical Research Council of Australia (NHMRC), the Cancer hypoxic properties of basal-like breast cancers. It suggests that as Institute NSW, the RT Hall Trust and the Petre Foundation. We this tumour type is particularly difficult to treat, because they are would also like to thank Dr Kingsley Micklem for help with the both hormonal non-responsive and resistant to chemotherapy, photomicrographs. targeting HIF with small molecule inhibitors, HRE-regulated gene therapies, carbonic anhydrase inhibitors or the use of bio-reductive drugs may be of particular use in treating this Supplementary Information accompanies the paper on British aggressive cancer (Supuran, 2008). Journal of Cancer website (http://www.nature.com/bjc)

REFERENCES

Abd El-Rehim DM, Pinder SE, Paish CE, Bell J, Blamey RW, Robertson JF, Grabmaier K, MC AdW, Verhaegh GW, Schalken JA, Oosterwijk E (2004) Nicholson RI, Ellis IO (2004) Expression of luminal and basal Strict regulation of CAIX(G250/MN) by HIF-1alpha in clear cell renal cell cytokeratins in human breast carcinoma. J Pathol 203: 661–671 carcinoma. Oncogene 23: 5624–5631 Baldeyron C, Jacquemin E, Smith J, Jacquemont C, De Oliveira I, Gad S, Hockel M, Vaupel P (2001) Tumor hypoxia: definitions and current clinical, Feunteun J, Stoppa-Lyonnet D, Papadopoulo D (2002) A single mutated biologic, and molecular aspects. J Natl Cancer Inst 93: 266–276 BRCA1 allele leads to impaired fidelity of double strand break end- Jarasch ED, Nagle RB, Kaufmann M, Maurer C, Bocker WJ (1988) joining. Oncogene 21: 1401–1410 Differential diagnosis of benign epithelial proliferations and carcinomas Banerjee S, Reis-Filho JS, Ashley S, Steele D, Ashworth A, Lakhani SR, of the breast using antibodies to cytokeratins. Hum Pathol 19: 276–289 Smith IE (2006) Basal-like breast carcinomas: clinical outcome and Jiang BH, Semenza GL, Bauer C, Marti HH (1996) Hypoxia-inducible factor response to chemotherapy. J Clin Pathol 59: 729–735 1 levels vary exponentially over a physiologically relevant range of O2 Bartosova M, Parkkila S, Pohlodek K, Karttunen TJ, Galbavy S, Mucha V, tension. Am J Physiol 271: C1172–C1180 Harris AL, Pastorek J, Pastorekova S (2002) Expression of carbonic Jones C, Ford E, Gillett C, Ryder K, Merrett S, Reis-Filho JS, Fulford LG, anhydrase IX in breast is associated with malignant tissues and is related Hanby A, Lakhani SR (2004) Molecular cytogenetic identification of to overexpression of c-erbB2. J Pathol 197: 314– 321 subgroups of grade III invasive ductal breast carcinomas with different Bos R, van der Groep P, Greijer AE, Shvarts A, Meijer S, Pinedo HM, clinical outcomes. Clin Cancer Res 10: 5988–5997 Semenza GL, van Diest PJ, van der Wall E (2003) Levels of hypoxia- Jones C, Nonni AV, Fulford L, Merrett S, Chaggar R, Eusebi V, Lakhani SR inducible factor-1alpha independently predict prognosis in patients with (2001) CGH analysis of ductal carcinoma of the breast with basaloid/ lymph node negative breast carcinoma. Cancer 97: 1573–1581 myoepithelial cell differentiation. Br J Cancer 85: 422–427 Chan N, Koritzinsky M, Zhao H, Bindra R, Glazer PM, Powell S, Belmaaza Kennedy R, Quinn J, Mullan P, Johnston P, Harkin D (2004) The role of A, Wouters B, Bristow RG (2008) Chronic hypoxia decreases synthesis of BRCA1 in the cellular response to chemotherapy. J Natl Cancer Inst 96: homologous recombination proteins to offset chemoresistance and 1659– 1668 radioresistance. Cancer Res 68: 605–614 Korsching E, Packeisen J, Agelopoulos K, Eisenacher M, Voss R, Isola J, Cheng KC, Loeb LA (1993) Genomic instability and tumor progression: van Diest PJ, Brandt B, Boecker W, Buerger H (2002) Cytogenetic mechanistic considerations. Adv Cancer Res 60: 121–156 alterations and cytokeratin expression patterns in breast cancer: Chia SK, Wykoff CC, Watson PH, Han C, Leek RD, Pastorek J, Gatter KC, integrating a new model of breast differentiation into cytogenetic Ratcliffe P, Harris AL (2001) Prognostic significance of a novel hypoxia- pathways of breast carcinogenesis. Lab Invest 82: 1525–1533 regulated marker, carbonic anhydrase IX, in invasive breast carcinoma. Lal A, Peters H, St Croix B, Haroon ZA, Dewhirst MW, Strausberg RL, J Clin Oncol 19: 3660– 3668 Kaanders JH, van der Kogel AJ, Riggins GJ (2001) Transcriptional eeisadGenomics and Genetics Fisher ER, Anderson S, Redmond C, Fisher B (1993) Pathologic findings response to hypoxia in human tumors. J Natl Cancer Inst 93: 1337– 1343 from the National Surgical Adjuvant Breast Project protocol B-06. Loncaster JA, Harris AL, Davidson SE, Logue JP, Hunter RD, Wycoff CC, 10-year pathologic and clinical prognostic discriminants. Cancer 71: Pastorek J, Ratcliffe PJ, Stratford IJ, West CM (2001) Carbonic anhydrase 2507– 2514 (CA IX) expression, a potential new intrinsic marker of hypoxia: Foulkes WD, Brunet JS, Stefansson IM, Straume O, Chappuis PO, Begin LR, correlations with tumor oxygen measurements and prognosis in locally Hamel N, Goffin JR, Wong N, Trudel M, Kapusta L, Porter P, Akslen LA advanced carcinoma of the cervix. Cancer Res 61: 6394–6399 (2004) The prognostic implication of the basal-like (cyclin E high/p27 Mahoney BP, Raghunand N, Baggett B, Gillies RJ (2003) Tumor acidity, ion low/p53+/glomeruloid-microvascular-proliferation+) phenotype of trapping and chemotherapeutics. I. Acid pH affects the distribution of BRCA1-related breast cancer. Cancer Res 64: 830–835 chemotherapeutic agents in vitro. Biochem Pharmacol 66: 1207– 1218 Fulford LG, Easton DF, Reis-Filho JS, Sofronis A, Gillett CE, Lakhani SR, Malzahn K, Mitze M, Thoenes M, Moll R (1998) Biological and prognostic Hanby A (2006) Specific morphological features predictive for the basal significance of stratified epithelial cytokeratins in infiltrating ductal phenotype in grade 3 invasive ductal carcinoma of breast. Histopathology breast carcinomas. Virchows Arch 433: 119–129 49: 22–34 Matos I, Dufloth R, Alvarenga M, Zeferino LC, Schmitt F (2005) p63, Generali D, Berruti A, Brizzi MP, Campo L, Bonardi S, Wigfield S, Bersiga cytokeratin 5, and P-cadherin: three molecular markers to distinguish A, Allevi G, Milani M, Aguggini S, Gandolfi V, Dogliotti L, Bottini A, basal phenotype in breast carcinomas. Virchows Arch 447: 688–694 Harris AL, Fox SB (2006a) Hypoxia-inducible factor-1{alpha} expression Nagle RB, Bocker W, Davis JR, Heid HW, Kaufmann M, Lucas DO, Jarasch predicts a poor response to primary chemoendocrine therapy and ED (1986) Characterization of breast carcinomas by two monoclonal disease-free survival in primary human breast cancer. Clin Cancer Res antibodies distinguishing myoepithelial from luminal epithelial cells. 12: 4562–4568 J Histochem Cytochem 34: 869–881 Generali D, Fox SB, Berruti A, Brizzi MP, Campo L, Bonardi S, Wigfield SM, Nielsen TO, Hsu FD, Jensen K, Cheang M, Karaca G, Hu Z, Bruzzi P, Bersiga A, Allevi G, Milani M, Aguggini S, Dogliotti L, Bottini A, Hernandez-Boussard T, Livasy C, Cowan D, Dressler L, Akslen LA, Harris AL (2006b) Role of carbonic anhydrase IX expression in Ragaz J, Gown AM, Gilks CB, van de Rijn M, Perou CM (2004) prediction of the efficacy and outcome of primary epirubicin/tamoxifen Immunohistochemical and clinical characterization of the basal-like therapy for breast cancer. Endocr Relat Cancer 13: 921–930 subtype of invasive breast carcinoma. Clin Cancer Res 10: 5367–5374

British Journal of Cancer (2009) 100(2), 405 – 411 & 2009 Cancer Research UK CAIX in basal breast cancer EY Tan et al 411 Opavsky R, Pastorekova S, Zelnik V, Gibadulinova A, Stanbridge EJ, Zavada tumor subtypes in independent gene expression data sets. Proc Natl Acad J, Kettmann R, Pastorek J (1996) Human MN/CA9 gene, a novel member Sci USA 100: 8418–8423 of the carbonic anhydrase family: structure and exon to protein domain Span PN, Bussink J, Manders P, Beex LV, Sweep CG (2003) Carbonic relationships. Genomics 33: 480–487 anhydrase-9 expression levels and prognosis in human breast cancer: Pastorek J, Pastorekova S, Callebaut I, Mornon JP, Zelnik V, Opavsky R, association with treatment outcome. Br J Cancer 89: 271–276 Zat’ovicova M, Liao S, Portetelle D, Stanbridge EJ, Zavada J, Burny A, Supuran CT (2008) Development of small molecule carbonic anhydrase IX Kettmann R (1994) Cloning and characterization of MN, a human tumor- inhibitors. BJU Int 101(Suppl 4): 39–40 associated protein with a domain homologous to carbonic anhydrase and a Sutherland RM (1998) Tumor hypoxia and gene expression–implications putative helix-loop-helix DNA binding segment. Oncogene 9: 2877–2888 for malignant progression and therapy. Acta Oncol 37: 567–574 Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack Tan E, Campo L, Han C, Turley H, Pezzella F, Gatter K, Harris A, Fox S JR, Ross DT, Johnsen H, Akslen LA, Fluge O, Pergamenschikov A, (2007) Cytoplasmic location of factor inhibiting-HIF (FIH)-1 is Williams C, Zhu SX, Lonning PE, Borresen-Dale AL, Brown PO, Botstein associated with an enhanced hypoxic response and a shorter survival D (2000) Molecular portraits of human breast tumours. Nature 406: in invasive breast cancer. Breast Cancer Res 9: R89, doi:10.1186/bcr1838 747–752 Teicher BA, Holden SA, al-Achi A, Herman TS (1990) Classification of Potter CP, Harris AL (2003) Diagnostic, prognostic and therapeutic antineoplastic treatments by their differential toxicity toward putative implications of carbonic anhydrases in cancer. Br J Cancer 89: 2–7 oxygenated and hypoxic tumor subpopulations in vivo in the FSaIIC Rafajova M, Zatovicova M, Kettmann R, Pastorek J, Pastorekova S (2004) murine fibrosarcoma. Cancer Res 50: 3339–3344 Induction by hypoxia combined with low glucose or low bicarbonate and Teicher BA, Lazo JS, Sartorelli AC (1981) Classification of antineoplastic high posttranslational stability upon reoxygenation contribute to carbonic agents by their selective toxicities toward oxygenated and hypoxic tumor anhydrase IX expression in cancer cells. Int J Oncol 24: 995–1004 cells. Cancer Res 41: 73– 81 Rouzier R, Perou CM, Symmans WF, Ibrahim N, Cristofanilli M, Anderson Tomes L, Emberley E, Niu Y, Troup S, Pastorek J, Strange K, Harris A, K, Hess KR, Stec J, Ayers M, Wagner P, Morandi P, Fan C, Rabiul I, Ross Watson PH (2003) Necrosis and hypoxia in invasive breast carcinoma. JS, Hortobagyi GN, Pusztai L (2005) Breast cancer molecular subtypes Breast Cancer Res Treat 81: 61–69 respond differently to preoperative chemotherapy. Clin Cancer Res 11: Tsuda H, Takarabe T, Hasegawa T, Murata T, Hirohashi S (1999) 5678– 5685 Myoepithelial differentiation in high-grade invasive ductal carcinomas Schindl M, Schoppmann SF, Samonigg H, Hausmaninger H, Kwasny W, with large central acellular zones. Hum Pathol 30: 1134– 1139 Gnant M, Jakesz R, Kubista E, Birner P, Oberhuber G (2002) Turner KJ, Crew JP, Wykoff CC, Watson PH, Poulsom R, Pastorek J, Overexpression of hypoxia-inducible factor 1alpha is associated with Ratcliffe PJ, Cranston D, Harris AL (2002) The hypoxia-inducible genes an unfavorable prognosis in lymph node-positive breast cancer. Clin VEGF and CA9 are differentially regulated in superficial vs invasive Cancer Res 8: 1831–1837 bladder cancer. Br J Cancer 86: 1276–1282 Semenza GL (2000) Hypoxia, clonal selection, and the role of HIF-1 in van de Rijn M, Perou CM, Tibshirani R, Haas P, Kallioniemi O, Kononen J, tumor progression. Crit Rev Biochem Mol Biol 35: 71–103 Torhorst J, Sauter G, Zuber M, Kochli OR, Mross F, Dieterich H, Seitz R, Sobhanifar S, Aquino-Parsons C, Stanbridge EJ, Olive P (2005) Reduced Ross D, Botstein D, Brown P (2002) Expression of cytokeratins 17 and 5 expression of hypoxia-inducible factor-1alpha in perinecrotic regions of identifies a group of breast carcinomas with poor clinical outcome. Am J solid tumors. Cancer Res 65: 7259– 7266 Pathol 161: 1991–1996 Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie T, Wetzels RH, Kuijpers HJ, Lane EB, Leigh IM, Troyanovsky SM, Holland R, Eisen MB, van de Rijn M, Jeffrey SS, Thorsen T, Quist H, Matese JC, van Haelst UJ, Ramaekers FC (1991) Basal cell-specific and hyperpro- Brown PO, Botstein D, Eystein Lonning P, Borresen-Dale AL (2001) Gene liferation-related keratins in human breast cancer. Am J Pathol 138: expression patterns of breast carcinomas distinguish tumor subclasses 751–763 with clinical implications. Proc Natl Acad Sci USA 98: 10869–10874 Wykoff CC, Beasley NJ, Watson PH, Turner KJ, Pastorek J, Sibtain A, Sorlie T, Tibshirani R, Parker J, Hastie T, Marron JS, Nobel A, Deng S, Wilson GD, Turley H, Talks KL, Maxwell PH, Pugh CW, Ratcliffe PJ, Johnsen H, Pesich R, Geisler S, Demeter J, Perou CM, Lonning PE, Brown Harris AL (2000) Hypoxia-inducible expression of tumor-associated PO, Borresen-Dale AL, Botstein D (2003) Repeated observation of breast carbonic anhydrases. Cancer Res 60: 7075–7083 Genetics and Genomics

& 2009 Cancer Research UK British Journal of Cancer (2009) 100(2), 405 – 411 Yan et al. Breast Cancer Research 2011, 13:R47 http://breast-cancer-research.com/content/13/2/R47

RESEARCHARTICLE Open Access Recruitment of regulatory T cells is correlated with hypoxia-induced CXCR4 expression, and is associated with poor prognosis in basal-like breast cancers Max Yan1,2,3*, Nicholas Jene1, David Byrne1, Ewan KA Millar2,3,4,5, Sandra A O’Toole2,6, Catriona M McNeil2,7, Gaynor J Bates8, Adrian L Harris9, Alison H Banham8, Robert L Sutherland2 and Stephen B Fox1

Abstract Introduction: Basal-like breast cancers behave more aggressively despite the presence of a dense lymphoid infiltrate. We hypothesised that immune suppression in this subtype may be due to T regulatory cells (Treg) recruitment driven by hypoxia-induced up-regulation of CXCR4 in Treg. Methods: Immunoperoxidase staining for FOXP3 and CXCL12 was performed on tissue microarrays from 491 breast cancers. The hypoxia-associated marker carbonic anhydrase IX (CA9) and double FOXP3/CXCR4 staining were performed on sections from a subset of these cancers including 10 basal-like and 11 luminal cancers matched for tumour grade. Results: High Treg infiltration correlated with tumour CXCL12 positivity (OR 1.89, 95% CI 1.22 to 2.94, P = 0.004) and basal phenotype (OR 3.14, 95% CI 1.08 to 9.17, P = 0.004) in univariate and multivariate analyses. CXCL12 positivity correlated with improved survival (P = 0.005), whereas high Treg correlated with shorter survival for all breast cancers (P = 0.001), luminal cancers (P < 0.001) and basal-like cancers (P = 0.040) that were confirmed in a multivariate analysis (OR 1.61, 95% CI 1.02 to 2.53, P = 0.042). In patients treated with hormone therapy, high Treg were associated with a shorter survival in a multivariate analysis (OR 1.78, 95% CI 1.01 to 3.15, P = 0.040). There was a tendency for luminal cancers to show CXCL12 expression (102/138, 74%) compared to basal-like cancers (16/27, 59%), which verged on statistical significance (P = 0.050). Up-regulation of CXCR4 in Treg correlated with the basal- like phenotype (P = 0.029) and tumour hypoxia, as indicated by CA9 expression (P = 0.049). Conclusions: Our data show that in the setting of hypoxia and CXCR4 up-regulation in Treg, CXCL12 expression may have the negative consequence of enhancing Treg recruitment and suppressing the anti-tumour immune response.

Introduction the repression of autoimmune disorders and transplant Cancer is rarely suppressed by the host immune rejection [3,4]. Although the role of Treg in cancer has response since tumour cells acquire immune tolerance. not been fully elucidated, these cells are likely to be The failure of an anti-cancer immune response may be responsible for maintaining the self-tolerance that may due to a specific subpopulation of regulatory T cells hinder the generation and activity of anti-tumour reac- (Treg) [1]. Treg down-regulate the activation and expan- tive T cells [2]. This is supported by observations that sion of self-reactive lymphocytes [2], and are crucial for depletion of Treg [1,5,6] and transforming growth b secreted by Treg [7,8] correlate with an enhanced immune response to cancer vaccines. Recently we and * Correspondence: [email protected] 1Department of Pathology, Peter MacCallum Cancer Centre, St Andrews others have demonstrated that tumour infiltration by Place, East Melbourne, Melbourne, VIC 3002, Australia Treg, independent of other lymphoid populations, is Full list of author information is available at the end of the article

© 2011 Yan et al.; licensee BioMed Central Ltd This is an open access article distributed under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/2.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited Yan et al. Breast Cancer Research 2011, 13:R47 Page 2 of 10 http://breast-cancer-research.com/content/13/2/R47

associated with a reduced survival in breast and other beneficial for this subset of breast cancers that are less cancers [9-13]. likely to respond to conventional therapies. Breast cancers are heterogeneous and one recognised subgroup, basal-like breast cancers, derive their name Materials and methods from the characteristic expression of basal cytokeratins Patient characteristics (CK) 5, 14 and 17 [14,15]. These tumours account for The flow of patients through the study as per the up to 15% of all invasive breast cancers [16], and are REporting recommendations for tumour MARKer prog- frequently observed in patients with BRCA1-related can- nostic studies (REMARK) criteria [36] is as follows. Six cers [17]. Despite the presence of a dense lymphoid hundred and twenty-one invasive breast carcinomas, infiltrate on histology, which is suggestive of an anti- characterised in a previous study [32], were retrospec- tumour immune response [17], they are associated with tively collected from the John Radcliffe Hospital, Oxford, a more aggressive clinical course characterised by UK and from the Garvan Institute of Medical Research, shorter survival and a higher risk of metastasis [17]. We Sydney, Australia. Of the 621 tumours, 594 tumours hypothesize that this is due, in part, to suppression of had tissue available for tissue microarray (TMA) con- the immune response by Treg. struction, of which 491 tumours were available for In non-neoplastic tissues, Treg are recruited by che- FOXP3 staining due to core drop-out. A subset of these mokines such as CXCL12 secreted by bone marrow, tumours (254 cases) was also stained for CXCL12. lymph node and inflammatory cells [18], a mechanism Arrays contained duplicate cores (1 mm cores). FOXP3/ that is replicated in tumours through chemokine secre- CXCR4 double staining and CA9 staining were per- tion by neoplastic cells [18]. Thus CXCL12, which binds formed on whole stained sections of 10 grade 3 basal- to its cognate receptor CXCR4 expressed by Treg, has like and 11 grade 3 luminal cancers from the Peter Mac- been implicated in the recruitment of Treg in a number Callum Cancer Centre. A waiver for informed consent, of tumours including ovarian cancer [19], adenocarci- for the use of archival material, has been obtained as noma of the lung [20], malignant mesothelioma [21], part of Ethics Committee approvals (JR C02.216, HREC and the myelodysplastic syndromes [22]. CXCR4 expres- SVH H94/080 and PMCC 09/36). All patients had oper- sion is induced under hypoxic stress via activation of able breast carcinomas and were not diagnosed with the HIF pathway in a number of cell types including B metastatic disease at the time of presentation. Median lymphocytes [23], tumour associated monocytes and age of patients included in this study was 55 years endothelial cells [24], microglia [25], multipotent stem (range 24 to 87 years). Median tumour size was 20 mm cells, stromal cells [26,27], cardiac monocytes and fibro- and the median tumour grade was 2. Forty-four percent blasts [28]. Furthermore, the HIF pathway enhances the of patients had nodal disease. Sixty-eight percent of immunosuppressive activity of Treg by promoting the tumours were ER positive and 16% were HER2 positive. expression of their lineage transcriptional regulator A total of 198 patients (31.9%) received adjuvant che- FOXP3 [29]. Given the role of hypoxia in T cell activa- motherapy (cyclophosphamide, methotrexate and tion [30,31], and also specifically in Treg [29], we 5-fluorouracil (CMF) or adriamycin and cyclophospha- hypothesised that Treg recruitment is dependent on mide (AC)). Adjuvant tamoxifen was given to 222 both CXCL12 production by tumour cells and hypoxia- (35.7%) patients. Patients were followed-up for a median induced CXCR4 expression in Treg. We further period of 131.9 months. During this time, 137 patients hypothesize that since basal-like tumours have an developed recurrence (30.0%) and 99 deaths (21.7%) enhanced hypoxic drive [32] this mechanism may be were considered breast-cancer related. prominent in basal-like breast cancer. We, therefore, investigated CXCR4 expression in Treg, Immunohistochemistry and scoring together with the expression of CA9 and CXCL12 in Paraffin embedded tissues were dewaxed. For FOXP3 basal-like and other subtypes of breast cancers. The sig- and CA9 staining, antigen retrieval was performed by nificance of this project lies in the rational design of microwaving in 50 mmol/L Tris/2 mmol/L EDTA (pH tumour vaccine approaches or blocking antibodies [33]. 9.0). Labelling was performed using the mouse monoco- Therapies targeting Treg are entering clinical trials lonal antibodies 236A/E7 (FOXP3, Abcam, Cambridge, [34,35]; therefore, it is important to quantify Treg num- UK) at 10 μg/mL [11] and M75 (CA9) at 1:100 [32]. For bers and to assess factors that may affect their recruit- CXCL12 immunohistochemistry, antigen retrieval was ment to the tumour microenvironment. Thus, should performed using the DAKO PT Link retrieval system findings suggest that hypoxia driven recruitment of Treg (Glostrup, Denmark) at high pH for 20 minute at 100°C. via the CXCL12/CXCR4 axis plays a significant role in Labeling was performed using mouse CXCL12 antibody basal-like tumours, therapies targeting CXCL12/CXCR4 (R&D Systems, Minneapolis, MN, USA, MAB350) [37] and HIF pathways, in addition to targeting Treg may be diluted to 8 μg/mL. Labeling was detected using the Yan et al. Breast Cancer Research 2011, 13:R47 Page 3 of 10 http://breast-cancer-research.com/content/13/2/R47

Dako Envision System. The stained arrays were counter- Statistical methods stained with haematoxylin. Positive and negative staining Group comparisons were made for non-thresholded controls for both antibodies were carried out in parallel data using unpaired t test, and for thresholded data using tonsillar tissue. using the chi square test. Kaplan-Meier survival curves Double immunohistochemical staining on whole sec- were plotted using breast cancer related death as the tions for FOXP3 and CXCR4 was performed using the endpoint and compared using a log rank test. Binary ® Ventana Benchmark ULTRA system (Tucson, Ari- logistic regression was used for multivariate analyses zona, USA). Antigen retrieval was performed using and the Cox proportional hazard regression model was Ventana ULTRA Cell Conditioning 1 (95°C for 36 min- used to identify independent prognostic factors for over- utes). Sections were incubated with FOXP3 antibody all survival. Analyses were performed with SPSS 16.0 (10 μg/mL) for 20 minutes (room temperature), fol- (SPSS Inc., 233 South Wacker Drive, Chicago, IL, 60606, lowed by visualization with Ventana Ultraview Univer- USA). A two-tailed P-value test was used in all analyses sal DAB. This was followed by incubation with and a P-value of less than 0.05 was considered statisti- polyclonal rabbit CXCR4 antibody (Sigma, St Louis, cally significant. MO,USA,C3116)at4μg/mL [38] for 16 minutes (room temperature) and visualization with Ventana Results Ultraview Alkaline Phosphatase Red Detection Kit. Tumour infiltration by Treg correlates with tumour Centroblasts and centrocytes within lymph node germ- CXCL12 expression and basal phenotype inal centres were used as positive and negative controls The number of Treg was correlated with clinicopatholo- respectively [39]. gical parameters (Table 1). In a univariate analysis, Cores were available for 456 tumours to be classified tumours with high Treg (defined as ≥15 Treg per core) into four intrinsic subgroups, based on ER, HER2 in situ were significantly more frequently grade 3 tumours hybridization, EGFR and CK5/6 staining, as per Nielson (115/224, 51%), compared with tumours with low Treg et al. [16] and Millar et al. [40]: basal-like group (ER (80/259, 31%) (P < 0.001). Compared to low Treg negative, HER2 negative, CK5/6 and/or EGFR positive), tumours, high Treg tumours also significantly correlated luminal group (ER positive, HER2 negative), HER2 with negative ER status (73/262 (28%) and 81/216 (38%) group (HER2 positive) and a negative group (ER, HER2, ER negative respectively, P = 0.025), positive HER2 sta- CK5/6 and EGFR negative). Absolute numbers of tus (7/245 (3%) and 20/206 (10%) HER2 positive respec- FOXP3-positive, Treg lymphocytes in assessable 1-mm tively, P = 0.002) and positive CXCL12 expression (74/ diameter invasive tumour cores were counted manually 125 (59%) and 83/110 (75%), P = 0.008). No correlation (Figure 1a). A cut-off of 15 Treg per core was used to was observed between Treg and nodal status, (P = divide the tumours into two groups (as previously 0.062), endocrine therapy (P = 0.255) and chemotherapy defined by Bates et al. [11]) The level of staining for (P = 0.148). Basal-like cancers were more likely to have CXCL12 was scored with respect to the intensity and high Treg (36/55, 75%, median 25.46) compared with percentage of staining in the cytoplasm. The scoring luminal cancers (99/258, 38%, median 7.64) (P <0.001, system for intensity was: 0 = no staining, 1 = weak Table 2). staining, 2 = moderate staining, 3 = strong staining. The Inamultivariateanalysisusingthebinarylogistic percentage of tumour cells stained in the given core regression model, high Treg significantly correlated with scoredas:0%=0;1to10%=1;11to50%=2;51to basal-like phenotype (OR 3.14, 95% CI 1.08 to 9.17, P = 80% = 3; 81 to 100% = 4. The CXCL12 scores for both 0.004) and CXCL12 expression (OR 1.89, 95% CI 1.22 staining intensity and the percentage of positive tumour to 2.94, P = 0.004), but not tumour grade (OR 1.43, 95% cells were added together to give a maximum score of 7. CI 0.88 to 2.32, P = 0.151, Table 3). A CXCL12 cut-off score of 7 was used to divide the tumours into approximately two equal groups (Figures High Treg infiltration is associated with reduced breast 1b, c). For FOXP3/CXCR4 double staining, the number cancer-specific survival in luminal and basal-like cancers of Treg co-expressing FOXP3/CXCR4 was enumerated High tumour infiltration by Treg (≥15) was significantly from four (x40, 0.55 mm diameter) high power fields of associated with reduced breast cancer specific survival lymphoid rich infiltrate (approximately equal to 0.95 (P = 0.001), with divergence of the survival curves occur- mm2) within the tumour (Figures 1d, e). For CA9 stain- ring two years after initial surgical treatment (Figure 2a). ing, positive expression was defined as the presence of The association of high Treg infiltration with poorer sur- strong membranous staining in ≥10% of tumour cells vival in all tumours was confirmed in a multivariate ana- (Figure 1f) [32,41]. lysis including age, lymph node status, grade, size, ER, Yan et al. Breast Cancer Research 2011, 13:R47 Page 4 of 10 http://breast-cancer-research.com/content/13/2/R47

Figure 1 Immunohistochemistry for FOXP3, CXCL12, CXCR4 and CA9. Immunohistochemical staining for (A) FOXP3 positive tumour infiltrating Treg (x20), (B) Tumour cells negative for CXCL12 (x10), (C) Tumour cells showing strong positivity for CXCL12 (x10). (D, E) Double immunohistochemical staining for FOXP3 (brown, nuclear) and CXCR4 (red, cytoplasmic). (D) CXCR4 negative Treg (x40), (E) CXCR4 positive Treg (arrows) (x40), Inset (x100 oil immersion), (F) Tumour cells with positive CA9 staining.

HER2 status, hormone therapy and chemotherapy (OR were observed for patients treated with chemotherapy 1.62, 95% CI 1.02 to 2.55, P = 0.040) (Supplementary when stratified by Treg infiltration (P = 0.565) Table S1a in Additional file 1). Breast cancer-specific survival based on Treg infiltra- CXCL12 expression does not differ between the different tion was analysed for each intrinsic subtype. High Treg intrinsic subtypes, and is associated with longer survival infiltration correlated with poorer survival in luminal (P Although a smaller proportion of basal-like (16/27, 59%) < 0.001) (Figure 2b) and basal-like cancers (P = 0.040) and HER2 tumours (25/47, 53%) was positive for (Figure 2c), but not in HER2 (P = 0.255) or null type CXCL12 expression compared with luminal tumours cancers (P = 0.128). (102/138, 74%), this did not reach statistical significance Treg infiltration was associated with reduced breast can- (P = 0.050) (Table 2). There was no correlation between cer-specific survival in patients treated with hormone ther- CXCL12 expression and tumour size, grade, lymph node apy (P = 0.001) (Figure 2d). This was confirmed in a status, ER, endocrine treatment and chemotherapy (all multivariate analysis where high Treg were an indepen- P > 0.05) (Supplementary Table S2 in Additional file 1). dent predictor of a poor response to hormone therapy On a log rank test, positive CXCL12 expression corre- (OR 1.78, 95% CI 1.01 to 3.15, P = 0.040) (Supplementary lated with longer breast cancer-specific survival (P = Table S1b in Additional file 1). No differences in survival 0.024) (Figure 2e). Yan et al. Breast Cancer Research 2011, 13:R47 Page 5 of 10 http://breast-cancer-research.com/content/13/2/R47

Table 1 Correlation of FOXP3 Treg with Table 3 Multivariate analysis, correlation of high Treg clinicopathological parameters and CXCL12 expression (≥15) with tumour size, grade, CXCL12 and tumour type Treg < 15 Treg ≥15 P-value Odds ratio 95% CI P-value Tumour size (mm) Tumour size 0.56 0.27 to 1.14 0.560 Median 21.2 23.8 0.571 Grade 1.43 0.88 to 2.32 0.151 Tumour grade < 0.001 CXCL12 1.89 1.22 to 2.94 0.004 1 56 (22%) 25 (11%) Tumour type 2 123 (47%) 84 (38%) Luminal (baseline) 1.00 - - 3 80 (31%) 115 (51%) Basal-like 3.14 1.08 to 9.17 0.004 Nodal status 0.062 n = 165 Negative 154 (59%) 109 (51%) Positive 107 (41%) 107 (49%) Treg infiltration correlates with tumour hypoxia ER 0.025 Carbonic anhydrase IX (CA9) is a transmembrane pro- Negative 73 (28%) 81 (38%) tein involved in maintaining a low pericellular pH Positive 189 (72%) 135 (62%) through the conversion of carbon dioxide to carbonic HER2 0.002 acid [42]. Its expression has been shown to correlate Negative 238 (97%) 186 (90%) with hypoxia as measured by Eppendorf microelectrode Positive 7 (3%) 20 (10%) [43] and the distribution of pimonidazole (a chemical CXCL12 0.008 marker of hypoxia) [42,44]. Negative 51 (41%) 27 (25%) CA9 expression was correlated with Treg numbers in Positive 74 (59%) 83 (75%) 448 breast cancers: increased numbers of Treg were Endocrine Rx 0.255 observed in CA9 positive tumours (median Treg = 32, n Negative 99 (37%) 72 (32%) = 66) compared to CA9 negative tumours (median Treg Positive 167 (63%) 151 (68%) = 10, n = 382) (Mann-Whitney U P < 0.001). In order Chemotherapy 0.148 to further explore whether Treg recruitment was asso- Negative 168 (64%) 127 (57%) ciated with hypoxia or other factors expressed by the Positive 96 (35%) 95 (43%) basal subtype, we correlated Treg numbers with CA9 ER, estrogen receptor; Rx, therapy. expression in a subset of 327 non-basal breast cancers.

Table 2 Correlation analysis of intrinsic subtypes with clinicopathological parameters, Treg and CXCL12 expression (n = 456) Luminal (n = 289) HER2 (n = 75) Basal-like (n = 62) Negative (n = 30) P-value Patient age Median (years) 56.0 55.2 54.0 52.7 0.64 Tumour size Median 18.0 22.0 22.5 20.5 0.40 Tumour grade < 0.001 1 71 (25%) 4 (5%) 2 (3%) 4 (13%) 2 156 (55%) 16 (22%) 12 (20%) 12 (40%) 3 58 (20%) 54 (73%) 47 (77%) 14 (47%) Nodal status 0.75 Negative 162 (56%) 40 (54%) 34 (55%) 17 (57%) Positive 125 (44%) 34 (46%) 28 (45%) 13 (43%) Treg < 0.001 < 15 159 (62%) 28 (39%) 19 (25%) 23 (92%) ≥15 99 (38%) 43 (61%) 36 (75%) 2 (8%) median 7.64 33.54 25.46 0.83 CXCL12 0.050 Negative 36 (26%) 22 (47%) 11 (41%) 6 (35%) Positive 102 (74%) 25 (53%) 16 (59%) 11 (65%) Treg, regulatory T cells. Yan et al. Breast Cancer Research 2011, 13:R47 Page 6 of 10 http://breast-cancer-research.com/content/13/2/R47

Figure 2 Kaplan Meier curves, breast cancer specific survival. Kaplan Meier curves, breast cancer specific survival, (A) all cancers stratified by Treg infiltration (n = 479, P = 0.001), (B) luminal cancers stratified by Treg infiltration (n = 258, P < 0.001), (C) basal-like cancers stratified by Treg infiltration (n = 54, P = 0.040), (D) patients treated with hormone therapy (n = 309, P = 0.001), (E) all cancers stratified by CXCL12 expression (n = 236, P = 0.005),

CA9 positive non-basal cancers had higher numbers of Infiltration by CXCR4 positive Treg correlates with basal Treg (median Treg = 21 per 1 mm core, n = 32) com- phenotype and tumour hypoxia pared to CA9 negative non-basal cancers (median Treg Double CXCR4/FOXP3 immunoperoxidase staining was =8per1mmcore,n=21)(Mann-WhitneyUP = performed to evaluate CXCR4 positive Treg infiltration 0.044). These results suggest hypoxia promotes Treg in whole stained sections from 10 grade 3 basal-like and recruitment independent of basal subtype. 11 grade 3 luminal cancers. A higher proportion of Treg Yan et al. Breast Cancer Research 2011, 13:R47 Page 7 of 10 http://breast-cancer-research.com/content/13/2/R47

in basal-like cancers expressed CXCR (median = 18.1%, Discussion interquartile range = 4.9% to 38.1%), when compared to In view of the association among basal-like breast can- luminal cancers (median = 3.2%, interquartile range = cer, its dense lymphoid infiltrate and aggressive beha- 2.1% to 19.9%) (P = 0.029) (Figure 3a). No difference in viour, we hypothesized that basal-like tumours evade CXCR4 positive Treg was observed between the peritu- the anti-tumour immune response via the recruitment moural stroma and the tumour bed (P = 0.337). of Treg. In our study, basal-like tumours were associated Using CA9 as a surrogate marker of hypoxia, in keep- with high Treg and a four-fold increase in the median ing with our previous studies [32], grade 3 basal-like number of Treg compared with luminal tumours. tumours were more likely to be hypoxic (8/10 positive Furthermore, this association was independent of for CA9), compared to grade 3 luminal cancers (0/11 tumour size and grade and conferred a poor survival in positive for CA9) (P < 0.001). Furthermore, in hypoxic this subtype. These results are in keeping with our pre- tumours with positive CA9 expression, a higher propor- vious study in an independent cohort where Treg infil- tion of Treg was positive for CXCR4 (median = 18.1%, tration correlated with high tumour grade, HER2 interquartile range = 9.7% to 40.3%) compared to CA9 positivity, ER negativity and poor survival [11]. and with negative tumours (median = 4.3%, interquartile range = Bohling et al. [45] in their comparison of 26 grade 3 tri- 2.1% to 19.9%) (P = 0.049) (Figure 3b). ple negative cancers with 71 non-grade 3 non-triple negative cancers. Although high Treg were also associated with an adverse outcome in luminal cancers in our cohort of HER positive patients, who were not treated with trastuzumab, accumulation of Treg did not correlate with survival. Although the reason for the latter result is unclear, one potential explanation is Treg immune suppression may have no effect on an immune response which is ineffective in the first instance. This is supported by studies where the anti-tumour immune response against HER2 cancers could be boosted by an infusion of HER2 specific T cells [46]. A number of chemokines have been implicated in the recruitment of Treg in non-neoplastic tissues [18]. Recent studies on breast cancers have shown that two of these chemokines, CCL20 [13] and CCL22 [12], may recruit Treg that express the corresponding chemokine receptors CCR6 and CCR4. Our results demonstrate CXCL12 derived from tumour cells may also be involved in Treg recruitment. While tumour CXCL12 expression correlated with tumour Treg recruitment, this did not appear to account for the increased number of Treg observed in basal-like cancers. Indeed, a lower proportion of basal-like cancers expressed CXCL12 (59%) compared with luminal cancers (74%). Our findings suggest preferential Treg recruitment in basal-like cancers may be in part explained by CXCR4 up-regulation in Treg. This is supported by in vitro cell migration assays where induction of CXCR4 expression in Treg resulted in their migration towards a CXCL12 gradient Figure 3 CXCR4 expression in Treg. (A) Proportion of Treg expressing CXCR4 in basal (median = 18.1%, interquartile range = [19,20], and could be terminated by incubation with an 4.9% to 38.1%) vs. luminal cancers (median = 3.2%, interquartile anti-CXCR4 antibody [19]. Preferential accumulation of range = 2.1% to 19.9%) (P = 0.029). (B) Proportion of Treg CXCR4 positive Treg, and its correlation with tumour expressing CXCR4 in CA9 positive (median = 18.1%, interquartile CXCL12 expression, has also been previously demon- range = 9.7% to 40.3%) vs. CA9 negative cancers (median = 4.3%, strated in adenocarcinomas of the lung [20] and malig- interquartile range = 2.1% to 19.9%) (P = 0.049). nant mesothelioma [21]. Yan et al. Breast Cancer Research 2011, 13:R47 Page 8 of 10 http://breast-cancer-research.com/content/13/2/R47

CXCL12 is a chemokine which exclusively binds to prognosis in non-basal breast cancers (log rank test P = the CXCR4 receptor. It is also the only ligand for the 0.002), but no such correlation is seen for basal cancers CXCR4 receptor [47]. CXCR4 expression is induced by (P = 0.688). In the setting of profound hypoxia and hypoxia [24,48]. Using CA9 as a surrogate marker of CXCR4 up-regulation in Treg, as occurs in basal-like hypoxia [42-44], we demonstrated that hypoxia is asso- breast cancer, CXCL12 may have a negative conse- ciated with the accumulation of Treg and also the sub- quence of enhancing Treg recruitment and suppressing set of CXCR4 positive Treg in breast cancer. This, the anti-tumour immune response. Although CXCL12 together with our previous finding that hypoxia is a fea- may also recruit other T cell subsets, it appears to pre- ture of basal-like breast cancers, suggests increased Treg ferentially recruit Treg, rather than CD8 cytotoxic or infiltration in basal-like cancers may be in part due to CD4 helper T cells [20,53]. hypoxia-induced up-regulation of CXCR4 in Treg. The correlations identified in our study are mechanistically Conclusions in vitro supported by studies, where T cells incubated These findings have important implications, as they sug- under hypoxic conditions show a time-dependent gest basal-like cancers, which are traditionally resistant to a increase in HIF-1 [29-31]. Furthermore, the link targeted therapies and may potentially respond to immu- a between HIF-1 and CXCR4 expression is supported by notherapy targeting Treg. Furthermore, Treg recruitment a number of findings [24] including: a) reduction of by CXCL12/CXCR4 in these cancers may potentially be CXCR4 in VHL mutated cell lines [24], b) increased IL- modulated by treatment directed against the HIF-1a 2 receptor (also implicated in CXCR4 up-regulation), pathway. Thus there is an opportunity for clinical trials and reduced CCR6 (another receptor implicated in Treg based on robust reagents directed against Treg, or antibo- recruitment) expression in T cells stimulated by hypoxia dies blocking CXCR4, to stratify patients for anti-HIF a CXCR4 [31], c) HIF-1 recruitment to the promoter in therapies. Indeed, many potent HIF-1 inhibitors are the hypoxic state [24] and d) hypoxia-induced expres- FDA-approved cancer treatment agents including anthra- sion of CXCR4, but not CCR6, CCR7, CXCR3 or cyclines and topotecan enabling clinical trials to test their a CXCR5 [23]. HIF-1 mayalsoinduceCXCR4expres- effectiveness [54,55]. Side effects should be minimal since sion indirectly by up-regulating the expression of these agents are chronically administered at low dose to FOXP3, which binds regulatory sequences upstream of derive their anti-HIF activity. There are some data to sug- the transcriptional start site of CXCR4 resulting in gest that the anthracycline analogue, mitoxanthrone, low- CXCR4 over-expression [49]. Furthermore, hypoxia ers the number of Treg in tumours [56] and such agents increases the potency of Treg in suppressing the prolif- may be combined with multiple immunotherapy strate- eration of effector CD4+ T cells [29,50]. There are likely gies that also reduce Treg numbers, enabling an also to be other cytokines regulating T cell recruitment improved effector cell response to a vaccine. and a comprehensive analysis of hypoxia-induced cytokines and their cognate receptors would be valuable, but Additional material need to be directed by detailed analysis of pathways regulated in these cells in vitro. The effect of hypoxia on Additional file 1: Supplementary Tables S1-S2. Supplementary Table Treg appears to be independent of other factors S1: (A) Multivariate analysis, Cox regression model, breast cancer specific expressed by the basal subtype as the correlation survival, all tumours (n = 398). (B) Multivariate analysis, Cox regression between hypoxia and Treg was re-duplicated in non- model, breast cancer specific survival, patients given with hormone therapy (n = 253). Supplementary Table S2: Correlation of CXCL12 basal tumours. expression with clinicopathological parameters. Loss of CXCL12 expression, in this study and in previous studies [51], is associated with a poor prognosis. Tumour cells with reduced CXCL12 in their immediate Abbreviations microenvironment may be at an advantage to receive BRCA1: breast cancer 1; early onset; CA9: carbonic anhydrase IX; CK: endocrine CXCL12 signals, promoting their migration cytokeratin; CXCL12: chemokine (C-X-C motif) ligand 12; CXCR4: C-X-C towards ectopic sources of the CXCR4 ligand. This is chemokine receptor type 4; EGFR: epidermal growth factor receptor; ER: estrogen receptor; FOXP3: forkhead box P3; HER2: human epidermal growth supported by mouse models where metastasis of tumour receptor 2; HIF: hypoxia inducible factor; Treg: regulatory T cell; TMA: tissue xenografts to the lung may be inhibited by endogenous microarray CXCL12 expression in the xenografted tumour [52]. Acknowledgements While CXCL12 expression is associated with a favour- This study was supported in part by the Victorian Breast Cancer Research able prognosis in an analysis of all breast cancers, there Consortium, the Victorian Cancer Biobank, Cancer Council Victoria, Cancer may be significant heterogeneity in the impact of Institute of New South Wales, the Australian Cancer Research Foundation, the Petre Foundation, the National Health and Medical Research Council CXCL12 on tumour behaviour between subtypes. For (Project grant 400207 to HX; Project Grants: 535903 and 535947 to EM and example, while CXCL12 is associated with a good RS) and the RT Hall Trust, Australia. Yan et al. Breast Cancer Research 2011, 13:R47 Page 9 of 10 http://breast-cancer-research.com/content/13/2/R47

Author details 14. Wetzels RH, Holland R, van Haelst UJ, Lane EB, Leigh IM, Ramaekers FC: 1Department of Pathology, Peter MacCallum Cancer Centre, St Andrews Detection of basement membrane components and basal cell keratin 14 Place, East Melbourne, Melbourne, VIC 3002, Australia. 2Cancer Research in noninvasive and invasive carcinomas of the breast. Am J Pathol 1989, Program, Garvan Institute of Medical Research, 384 Victoria Street, 134:571-579. Darlinghurst, Sydney, NSW 2010, Australia. 3School of Medical Sciences, 15. Wetzels RH, Kuijpers HJ, Lane EB, Leigh IM, Troyanovsky SM, Holland R, van University of New South Wales, High Street, Kensington, Sydney, NSW 2052, Haelst UJ, Ramaekers FC: Basal cell-specific and hyperproliferation-related Australia. 4South East Area Laboratory Services, St George Hospital, South keratins in human breast cancer. Am J Pathol 1991, 138:751-763. Street, Kogarah, Sydney, 2217, Australia. 5School of Medicine, University of 16. Nielsen TO, Hsu FD, Jensen K, Cheang M, Karaca G, Hu Z, Hernandez- Western Sydney, Narellan Road, Campbelltown, Sydney, NSW 2560, Australia. Boussard T, Livasy C, Cowan D, Dressler L, Akslen LA, Ragaz J, Gown AM, 6Department of Tissue Pathology, Royal Prince Alfred Hospital, Missenden Gilks CB, van de Rijn M, Perou CM: Immunohistochemical and clinical Road, Camperdown, Sydney, NSW 2010, Australia. 7Department of Medical characterization of the basal-like subtype of invasive breast carcinoma. Oncology, Royal Prince Alfred Hospital, Missenden Road, Camperdown, Clin Cancer Res 2004, 10:5367-5374. Sydney, NSW 2010 Australia. 8Nuffield Department of Clinical Laboratory 17. Rakha EA, Reis-Filho JS, Ellis IO: Basal-like breast cancer: a critical review. Sciences and Medical Oncology, University of Oxford, Henry Wellcome J Clin Oncol 2008, 26:2568-2581. Building for Molecular Physiology, Old Road Campus, Headington, Oxford, 18. Wei S, Kryczek I, Zou W: Regulatory T-cell compartmentalization and OX3 7BN, UK. 9Weatherall Institute of Molecular Medicine, University of trafficking. Blood 2006, 108:426-431. Oxford, John Radcliffe Hospital, Oxford, OX3 9DS, UK. 19. Wei S, Kryczek I, Edwards RP, Zou L, Szeliga W, Banerjee M, Cost M, Cheng P, Chang A, Redman B, Herberman RB, Zou W: Interleukin-2 Authors’ contributions administration alters the CD4+FOXP3+ T-cell pool and tumour SF and MY conceived the experiments. Experiments were carried out by MY, trafficking in patients with ovarian carcinoma. Cancer Res 2007, NJ and DB. MY, SF, EM, SO, CM, GB, AH, AB and RS were involved in the 67:7487-7494. collection and analysis of data. All authors were involved in writing the 20. Wald O, Izhar U, Amir G, Avniel S, Bar-Shavit Y, Wald H, Weiss ID, Galun E, paper and had final approval of the submitted version. Peled A: CD4+CXCR4highCD69+ T cells accumulate in lung adenocarcinoma. J Immunol 2006, 177:6983-6990. Competing interests 21. Shimizu Y, Dobashi K, Imai H, Sunaga N, Ono A, Sano T, Hikino T, Shimizu K, The authors declare that they have no competing interests. Tanaka S, Ishizuka T, Utsugi M, Mori M: CXCR4+FOXP3+CD25+ lymphocytes accumulate in CXCL12-expressing malignant pleural Received: 2 November 2010 Revised: 7 March 2011 mesothelioma. Int J Immunopathol Pharmacol 2009, 22:43-51. Accepted: 26 April 2011 Published: 26 April 2011 22. Kotsianidis I, Bouchliou I, Nakou E, Spanoudakis E, Margaritis D, Christophoridou AV, Anastasiades A, Tsigalou C, Bourikas G, Karadimitris A, References Tsatalas C: Kinetics, function and bone marrow trafficking of CD4+CD25 1. Disis ML: Immune regulation of cancer. J Clin Oncol 2010, 28:4531-4538. +FOXP3+ regulatory T cells in myelodysplastic syndromes (MDS). 2. Sakaguchi S: Regulatory T cells: key controllers of immunologic self- Leukemia 2009, 23:510-518. tolerance. Cell 2000, 101:455-458. 23. Piovan E, Tosello V, Indraccolo S, Masiero M, Persano L, Esposito G, 3. Albert MH, Anasetti C, Yu XZ: T regulatory cells as an immunotherapy for Zamarchi R, Ponzoni M, Chieco-Bianchi L, Dalla-Favera R, Amadori A: transplantation. Expert Opin Biol Ther 2006, 6:315-324. Differential regulation of hypoxia-induced CXCR4 triggering during B- 4. Kang HK, Datta SK: Regulatory T cells in lupus. Int Rev Immunol 2006, cell development and lymphomagenesis. Cancer Res 2007, 67:8605-8614. 25:5-25. 24. Schioppa T, Uranchimeg B, Saccani A, Biswas SK, Doni A, Rapisarda A, 5. Dannull J, Su Z, Rizzieri D, Yang BK, Coleman D, Yancey D, Zhang A, Bernasconi S, Saccani S, Nebuloni M, Vago L, Mantovani A, Melillo G, Sica A: Dahm P, Chao N, Gilboa E, Vieweg J: Enhancement of vaccine-mediated Regulation of the chemokine receptor CXCR4 by hypoxia. J Exp Med antitumour immunity in cancer patients after depletion of regulatory T 2003, 198:1391-1402. cells. J Clin Invest 2005, 115:3623-3633. 25. Wang X, Li C, Chen Y, Hao Y, Zhou W, Chen C, Yu Z: Hypoxia enhances 6. Morse MA, Hobeika AC, Osada T, Serra D, Niedzwiecki D, Lyerly HK, Clay TM: CXCR4 expression favoring microglia migration via HIF-1alpha activation. Depletion of human regulatory T cells specifically enhances antigen- Biochem Biophys Res Commun 2008, 371:283-288. specific immune responses to cancer vaccines. Blood 2008, 112:610-618. 26. Ceradini DJ, Kulkarni AR, Callaghan MJ, Tepper OM, Bastidas N, 7. Disis ML: Enhancing cancer vaccine efficacy via modulation of the Kleinman ME, Capla JM, Galiano RD, Levine JP, Gurtner GC: Progenitor cell tumour microenvironment. Clin Cancer Res 2009, 15:6476-6478. trafficking is regulated by hypoxic gradients through HIF-1 induction of 8. Disis ML, Wallace DR, Gooley TA, Dang Y, Slota M, Lu H, Coveler AL, SDF-1. Nat Med 2004, 10:858-864. Childs JS, Higgins DM, Fintak PA, de la Rosa C, Tietje K, Link J, Waisman J, 27. Hung SC, Pochampally RR, Hsu SC, Sanchez C, Chen SC, Spees J, Salazar LG: Concurrent trastuzumab and HER2/neu-specific vaccination in Prockop DJ: Short-term exposure of multipotent stromal cells to low patients with metastatic breast cancer. J Clin Oncol 2009, 27:4685-4692. oxygen increases their expression of CX3CR1 and CXCR4 and their 9. Curiel TJ, Coukos G, Zou L, Alvarez X, Cheng P, Mottram P, Evdemon- engraftment in vivo. PLoS One 2007, 2:e416. Hogan M, Conejo-Garcia JR, Zhang L, Burow M, Zhu Y, Wei S, Kryczek I, 28. Hu X, Dai S, Wu WJ, Tan W, Zhu X, Mu J, Guo Y, Bolli R, Rokosh G: Stromal Daniel B, Gordon A, Myers L, Lackner A, Disis ML, Knutson KL, Chen L, cell derived factor-1 alpha confers protection against myocardial Zou W: Specific recruitment of regulatory T cells in ovarian carcinoma ischemia/reperfusion injury: role of the cardiac stromal cell derived fosters immune privilege and predicts reduced survival. Nat Med 2004, factor-1 alpha CXCR4 axis. Circulation 2007, 116:654-663. 10:942-949. 29. Ben-Shoshan J, Maysel-Auslender S, Mor A, Keren G, George J: Hypoxia 10. Beyer M, Schultze JL: Regulatory T cells in cancer. Blood 2006, 108:804-811. controls CD4+CD25+ regulatory T-cell homeostasis via hypoxia-inducible 11. Bates GJ, Fox SB, Han C, Leek RD, Garcia JF, Harris AL, Banham AH: factor-1alpha. Eur J Immunol 2008, 38:2412-2418. Quantification of regulatory T cells enables the identification of high-risk 30. Larbi A, Zelba H, Goldeck D, Pawelec G: Induction of HIF-1alpha and the breast cancer patients and those at risk of late relapse. J Clin Oncol 2006, glycolytic pathway alters apoptotic and differentiation profiles of 24:5373-5380. activated human T cells. J Leukoc Biol 2010, 87:265-273. 12. Gobert M, Treilleux I, Bendriss-Vermare N, Bachelot T, Goddard-Leon S, 31. Makino Y, Nakamura H, Ikeda E, Ohnuma K, Yamauchi K, Yabe Y, Arfi V, Biota C, Doffin AC, Durand I, Olive D, Perez S, Pasqual N, Faure C, Poellinger L, Okada Y, Morimoto C, Tanaka H: Hypoxia-inducible factor Ray-Coquard I, Puisieux A, Caux C, Blay JY, Menetrier-Caux C: Regulatory T regulates survival of antigen receptor-driven T cells. J Immunol 2003, cells recruited through CCL22/CCR4 are selectively activated in 171:6534-6540. lymphoid infiltrates surrounding primary breast tumours and lead to an 32. Tan EY, Yan M, Campo L, Han C, Takano E, Turley H, Candiloro I, Pezzella F, adverse clinical outcome. Cancer Res 2009, 69:2000-2009. Gatter KC, Millar EK, O’Toole SA, McNeil CM, Crea P, Segara D, 13. Xu L, Xu W, Qiu S, Xiong S: Enrichment of CCR6(+)Foxp3(+) regulatory T Sutherland RL, Harris AL, Fox SB: The key hypoxia regulated gene CAIX is cells in the tumour mass correlates with impaired CD8(+) T cell function upregulated in basal-like breast tumours and is associated with and poor prognosis of breast cancer. Clin Immunol 2010, 135:466-475. resistance to chemotherapy. Br J Cancer 2009, 100:405-411. Yan et al. Breast Cancer Research 2011, 13:R47 Page 10 of 10 http://breast-cancer-research.com/content/13/2/R47

33. Disis ML: Molecular targeting with cancer vaccines. J Clin Oncol 2005, 52. Wendt MK, Cooper AN, Dwinell MB: Epigenetic silencing of CXCL12 23:4840-4841. increases the metastatic potential of mammary carcinoma cells. 34. Emens LA, Reilly RT, Jaffee EM: Breast cancer vaccines: maximizing cancer Oncogene 2008, 27:1461-1471. treatment by tapping into host immunity. Endocr Relat Cancer 2005, 53. Jaafar F, Righi E, Lindstrom V, Linton C, Nohadani M, Van Noorden S, 12:1-17. Lloyd T, Poznansky J, Stamp G, Dina R, Coleman DV, Poznansky MC: 35. Colombo MP, Piconese S: Regulatory T-cell inhibition versus depletion: Correlation of CXCL12 expression and FoxP3+ cell infiltration with the right choice in cancer immunotherapy. Nat Rev Cancer 2007, human papillomavirus infection and clinicopathological progression of 7:880-887. cervical cancer. Am J Pathol 2009, 175:1525-1535. 36. McShane LM, Altman DG, Sauerbrei W, Taube SE, Gion M, Clark GM: 54. Hede K: Low-dose anthracyclines may block HIF-1 and stop tumour Reporting recommendations for tumour marker prognostic studies. J growth. J Natl Cancer Inst 2009, 101:368-370. Clin Oncol 2005, 23:9067-9072. 55. Lee K, Qian DZ, Rey S, Wei H, Liu JO, Semenza GL: Anthracycline 37. Hanna J, Wald O, Goldman-Wohl D, Prus D, Markel G, Gazit R, Katz G, chemotherapy inhibits HIF-1 transcriptional activity and tumour-induced Haimov-Kochman R, Fujii N, Yagel S, Peled A, Mandelboim O: CXCL12 mobilization of circulating angiogenic cells. Proc Natl Acad Sci USA 2009, expression by invasive trophoblasts induces the specific migration of 106:2353-2358. CD16- human natural killer cells. Blood 2003, 102:1569-1577. 56. Nizar S, Copier J, Meyer B, Bodman-Smith M, Galustian C, Kumar D, 38. Barbieri F, Bajetto A, Stumm R, Pattarozzi A, Porcile C, Zona G, Dorcaratto A, Dalgleish A: T-regulatory cell modulation: the future of cancer Ravetti J, Minuto F, Spaziante R, Schettini G, Ferone D, Florio T: immunotherapy? Br J Cancer 2009, 100:1697-1703. Overexpression of stromal cell-derived factor 1 and its receptor CXCR4 induces autocrine/paracrine cell proliferation in human doi:10.1186/bcr2869 pituitaryadenomas. Clin Cancer Res 2008, 14:5022-5032. Cite this article as: Yan et al.: Recruitment of regulatory T cells is 39. Caron G, Le Gallou S, Lamy T, Tarte K, Fest T: CXCR4 expression correlated with hypoxia-induced CXCR4 expression, and is associated functionally discriminates centroblasts versus centrocytes within human with poor prognosis in basal-like breast cancers. Breast Cancer Research germinal center B cells. J Immunol 2009, 182:7595-7602. 2011 13:R47. 40. Millar EK, Graham PH, O’Toole SA, McNeil CM, Browne L, Morey AL, Eggleton S, Beretov J, Theocharous C, Capp A, Nasser E, Kearsley JH, Delaney G, Papadatos G, Fox C, Sutherland RL: Prediction of local recurrence, distant metastases, and death after breast-conserving therapy in early-stage invasive breast cancer using a five-biomarker panel. J Clin Oncol 2009, 27:4701-4708. 41. Tan EY, Campo L, Han C, Turley H, Pezzella F, Gatter KC, Harris AL, Fox SB: Cytoplasmic location of factor-inhibiting hypoxia-inducible factor is associated with an enhanced hypoxic response and a shorter survival in invasive breast cancer. Breast Cancer Res 2007, 9:R89. 42. Opavsky R, Pastorekova S, Zelnik V, Gibadulinova A, Stanbridge EJ, Zavada J, Kettmann R, Pastorek J: Human MN/CA9 gene, a novel member of the carbonic anhydrase family: structure and exon to protein domain relationships. Genomics 1996, 33:480-487. 43. Loncaster JA, Harris AL, Davidson SE, Logue JP, Hunter RD, Wycoff CC, Pastorek J, Ratcliffe PJ, Stratford IJ, West CM: Carbonic anhydrase (CA IX) expression, a potential new intrinsic marker of hypoxia: correlations with tumour oxygen measurements and prognosis in locally advanced carcinoma of the cervix. Cancer Res 2001, 61:6394-6399. 44. Wykoff CC, Beasley NJ, Watson PH, Turner KJ, Pastorek J, Sibtain A, Wilson GD, Turley H, Talks KL, Maxwell PH, Pugh CW, Ratcliffe PJ, Harris AL: Hypoxia-inducible expression of tumour-associated carbonic anhydrases. Cancer Res 2000, 60:7075-7083. 45. Bohling SD, Allison KH: Immunosuppressive regulatory T cells are associated with aggressive breast cancer phenotypes: a potential therapeutic target. Mod Pathol 2008, 21:1527-1532. 46. Disis ML, Salazar LG, Coveler AL, Waisman J, Higgins D, Childs J, Bates N, Dang Y: Phase I study of infusion of HER2/neu (HER2) specific T cells in patients with advanced-stage HER2 overexpressing cancers who have received a HER2 vaccine. 2009 ASCO Annual Meeting Proceedings. J Clin Oncol 2009, 27:15s. 47. Luker KE, Luker GD: Functions of CXCL12 and CXCR4 in breast cancer. Cancer Lett 2006, 238:30-41. 48. Staller P, Sulitkova J, Lisztwan J, Moch H, Oakeley EJ, Krek W: Chemokine receptor CXCR4 downregulated by von Hippel-Lindau tumour suppressor pVHL. Nature 2003, 425:307-311. Submit your next manuscript to BioMed Central 49. Zheng Y, Josefowicz SZ, Kas A, Chu TT, Gavin MA, Rudensky AY: Genome- and take full advantage of: wide analysis of Foxp3 target genes in developing and mature regulatory T cells. Nature 2007, 445:936-940. 50. Sitkovsky MV: T regulatory cells: hypoxia-adenosinergic suppression and • Convenient online submission re-direction of the immune response. Trends Immunol 2009, 30:102-108. • Thorough peer review 51. Zhou W, Jiang Z, Liu N, Xu F, Wen P, Liu Y, Zhong W, Song X, Chang X, • No space constraints or color ﬁgure charges Zhang X, Wei G, Yu J: Down-regulation of CXCL12 mRNA expression by promoter hypermethylation and its association with metastatic • Immediate publication on acceptance progression in human breast carcinomas. J Cancer Res Clin Oncol 2009, • Inclusion in PubMed, CAS, Scopus and Google Scholar 135:91-102. • Research which is freely available for redistribution

Submit your manuscript at www.biomedcentral.com/submit Chan et al. Breast Cancer Research 2011, 13:R19 http://breast-cancer-research.com/content/13/1/R19

RESEARCHARTICLE Open Access The expression of the ubiquitin ligase SIAH2 (seven in absentia homolog 2) is mediated through gene copy number in breast cancer and is associated with a basal-like phenotype and p53 expression Peter Chan1†, Andreas Möller2,3*†, Mira CP Liu2, Jaclyn E Sceneay2, Christina SF Wong2, Nic Waddell4, Katie T Huang1,3, Alexander Dobrovic1,3, Ewan KA Millar5,6,10,11, Sandra A O’Toole7,12, Catriona M McNeil8,9, Robert L Sutherland5, David D Bowtell2, Stephen B Fox1,3*

Abstract Introduction: The seven in absentia homolog 2 (SIAH2) protein plays a significant role in the hypoxic response by regulating the abundance of hypoxia-inducible factor-a; however, its role in breast carcinoma is unclear. We investigated the frequency and expression pattern of SIAH2 in two independent cohorts of sporadic breast cancers. Methods: Immunohistochemical evaluation of SIAH2protein expression was conducted in normal breast tissues and in tissue microarrays comprising ductal carcinoma in situ (DCIS) and a cohort of invasive breast carcinomas. Correlation analysis was performed between SIAH2 and clinicopathological variables and intrinsic breast cancer subgroups and validated in a cohort of 293 invasive ductal carcinomas. Promoter methylation, gene copy number and mRNA expression of SIAH2 were determined in a panel of basal-like tumors and cell lines. Results: There was a significant increase in nuclear SIAH2 expression from normal breast tissues through to DCIS and progression to invasive cancers. A significant inverse correlation was apparent between SIAH2 and estrogen receptor and progesterone receptor and a positive association with tumor grade, HER2, p53 and an intrinsic basal-like subtype. Logistic regression analysis confirmed the significant positive association between SIAH2 expression and the basal-like phenotype. No SIAH2 promoter methylation was identified, yet there was a significant correlation between SIAH2 mRNA and gene copy number. SIAH2-positive tumors were associated with a shorter relapse-free survival in univariate but not multivariate analysis. Conclusions: SIAH2 expression is upregulated in basal-like breast cancers via copy number changes and/or transcriptional activation by p53 and is likely to be partly responsible for the enhanced hypoxic drive through abrogation of the prolyl hydroxylases.

Introduction through binding to hypoxia response elements in the pro- Hypoxia in breast cancer has profound effects on tumor moters of genes, results in expression of proteins involved biology that are reflected in a poor prognosis and resis- in angiogenesis (vascular endothelial growth factor tance to both chemotherapy and radiotherapy in patients (VEGF)), glucose metabolism (glucose transporter 1), [1]. Hypoxia-inducible factor (HIF)-1 is critical to the metastasis (chemokine (C-X-C motif) receptor 4 and stro- hypoxic response, being a transcription factor that, mal cell-derived factor-1), cell survival and proliferation. HIF-1 is a dimer consisting of a constitutively * Correspondence: [email protected]; [email protected] expressed aryl nuclear translocator or HIF-1b and a † Contributed equally hypoxia-inducible HIF-1a. The levels of HIF-1a are 1 Department of Pathology, Peter MacCallum Cancer Centre, Locked Bag 1 tightly regulated by three prolyl hydroxylases. In the pre- A’Beckett Street, Melbourne, Victoria 8006, Australia 2Research Division, Cancer Genetics and Genomics, Peter MacCallum Cancer sence of molecular oxygen, these enzymes hydroxylate Centre, Locked Bag 1 A’Beckett Street, Melbourne, Victoria 8006, Australia the prolyl residues 402 and 564 in the oxygen-dependent Full list of author information is available at the end of the article

© 2011 Chan et al.; licensee BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Chan et al. Breast Cancer Research 2011, 13:R19 Page 2 of 10 http://breast-cancer-research.com/content/13/1/R19

domain of HIF-1a, resulting in conformational change from the John Radcliffe Hospital, Oxford, UK, of which and recognition by the von Hippel-Lindau protein that 54 had DCIS on tissue microarrays (TMAs) for staining leads to its ubiquitination and degradation via the protea- and clinical data available. Ten cases of normal postme- some. In contrast, under hypoxia, the prolyl hydroxylases nopausal breast tissues from mammoplasties were also have limited molecular oxygen and are therefore less collected. This study has Ethics Committee approvals effective, which enables HIF-1a stabilization, transloca- (numbers 00/81, 03/90, 09/36, JRC02.216, HREC SVH tion to the nucleus and initiation of gene transcription H94/080 and H00/36). All patients had operable breast that benefits the tumor. carcinomas and were not diagnosed with distant meta- Seven in absentia homolog 2 (SIAH2) is one of a static disease at the time of presentation. Information family of RING domain proteins which act alone or as regarding patient characteristics, including age, tumor components of ubiquitin ligase complexes that target size, grade, histology and nodal status were collected proteins for proteasomal degradation [2]. Siah proteins from the clinical and pathological records. The median can interact with many intracellular pathways, including age of patients included in this study was 54 years the scaffold proteins, transcriptional repressors and (range, 24 to 87 years). Ninety-three percent of tumors nuclear receptor corepressors and b-catenin. Siah pro- were invasive ductal tumors not otherwise specified teins are also involved in hypoxia signaling via regula- type, 3% were invasive lobular carcinomas and 4% were tion of HIF-1a [3] through the targeted degradation of of other histological types (data were unavailable for two prolyl hydroxylases under hypoxic conditions. Indeed, cases). Median tumor size was 20 mm, and the median SIAH2-knockout mice have a delayed and abrogated tumor grade was 2. Forty-one percent of patients had response to hypoxic conditions that is mediated through nodal disease. Sixty-nine percent of tumors were estro- reduced levels of HIF-1a [3,4]. These data suggest that gen receptor (ER)-positive, and 14% were human epider- Siah proteins may significantly alter HIF signaling mal growth factor receptor 2 (HER2)-positive. Patients through modulation of the prolyl hydroxylases. less than 50 years of age with node-positive, ER-negative Although the role of HIF has been documented in tumors or tumors larger than 3 cm received adjuvant breast cancer [5,6], there are no data on the expression chemotherapy (cyclophosphamide, methotrexate and of SIAH2 in this disease. We have therefore investigated 5-fluorouracil or adriamycin and cyclophosphamide. SIAH2 expression in breast cancer in two independent Patients with hormone-responsive tumors who were cohorts. Our aims were to (1) document the pattern more than 50 years of age received 5 years of endocrine and level of SIAH2 expression in breast cancer, (2) cor- therapy. Patients were followed up for a median period relate expression with conventional clinicopathological of 58.1 months. During this time, in the 100 patients factors, (3) investigate associations of SIAH2 expression from the invasive cohort, two developed recurrence and with intrinsic subtypes of breast cancer and (4) deter- 86 deaths were considered breast cancer-related. mine the effect of SIAH2 expression on relapse-free survival. Immunohistochemistry TMAs were constructed from 1-mm diameter (invasive Materials and methods cancers) or 2-mm cores (DCIS). Sections of 4-μm thick- Patients ness were used for immunostaining. TMA sections were The flow of patients through the study according to the dewaxed, and antigen retrieval was performed in 10 mM Reporting Recommendations for Tumor Marker Prog- sodium citrate, pH 6, in a pressure cooker for 3 min- nostic Studies (REMARK) criteria is listed in Supple- utes. Sections were then treated with 3% H2O2 for mentary Table 1 in Additional file 1 [7] The first cohort 5 minutes to remove endogenous peroxides, washed and was derived from the Department of Pathology, Peter incubated with a SIAH2 antibody (NB110-88113; Novus MacCallum Cancer Centre, Melbourne, Australia, and Biologicals, Littleton, CO, USA) [9,10] at 1:50 dilution comprised 120 cases with full clinicopathological charac- for 90 minutes at room temperature. The peroxidase- teristics but without survival data. The second cohort coupled Mouse ImmPRESS (Vector Laboratories, was from the Garvan Institute, Sydney, Australia, and Burlingame, CA, USA) detection reagent was then used, comprised 293 cases with full clinicopathological charac- and staining was visualized with diaminobenzidine plus teristics including survival data [7]. In total, 439 invasive (DAB+; Dako, Campbellfield, VIC, Australia). Sections cancers with clinicopathological data and follow-up were counterstained with hematoxylin to visualize were available for study. Of these 439 cases, 61 cases nuclei.ToanalyzetheexpressionofSIAH2inbreast were excluded because of inadequate tumor tissue on cancer progression, we assessed expression using a com- the array. The final cohort of invasive cancers comprised bination of both intensity and proportion of cells 378 cases (246 cases with survival data). Eighty cases of expressing SIAH2 [10] Normal breast epithelium pure ductal carcinoma in situ (DCIS) were obtained and tumors were scored for intensity (0 = no staining, Chan et al. Breast Cancer Research 2011, 13:R19 Page 3 of 10 http://breast-cancer-research.com/content/13/1/R19

1 = weak staining, 2 = moderate staining and 3 = strong then cooled to 75°C with temperature rising by 0.2°C per staining) and the percentage of cells (0 = no cells stain- second to 97°C and holding for 1 second after each step- ing positive, 1 = <10% cells staining positive, 2 = 10% to wise increment. Methylated sequences could be identified 50% cells staining positive, 3 = 51% to 80% cells staining by their increased melting temperature [13]. In each assay, positive and 4 = >80% cells staining positive) as pre- fully methylated, peripheral blood DNA (unmethylated), viously reported [12]. The scores for intensity and per- different methylation percentage dilution standards and centage of positive tumor cells were added to give a nontemplate controls were included as controls and stan- maximum score of 7. A cutoff of >2 (median value) was dards. All assays were performed in duplicate. used to define two patient groups of approximately equal size for subsequent statistical analyses. Copy number analysis of SIAH2 ER, HER2, epidermal growth factor receptor (EGFR) A previous study analyzed both gene expression and and cytokeratin (CK)5/6 staining were used to classify copy number variation using the Illumina Human-6 tumors into four intrinsic subgroups: the basal group BeadArray (Illumina, San Diego, CA, 92121, USA) and (ER-negative, HER2-negative, CK5/6-positive and/or the CNV370 SNP array (Illumina) respectively, in a EGFR-positive), the luminal group (ER-positive, HER2- cohort of familial tumors [14]. These familial tumors negative), the HER2 group (HER2-positive) and the were known to be breast cancer 1, early onset gene negative (null) group (ER-negative, HER2-negative, CK5/ (BRCA1), BRCA2 or non-BRCA1 and non-BRCA2 6-negative and EGFR-negative) [13]. tumors, and in this previous study the familial tumors Analysis of SIAH2 Methylation were classified into one of the breast tumor subtypes: DNA from a separate series of 60 breast carcinomas basal-like, luminal A, luminal B, HER2-positive and (John Radcliffe Hospital, Oxford, UK) and five normal normal-like. These data were used to determine the breast tissues (Peter MacCallum Cancer Centre, Mel- expression of SIAH2 and its copy number status in 15 bourne, VIC, Australia), comprising all breast cancer basal-like tumors. The copy number of SIAH2 in each phenotypes (50 ER-positive and 10 ER-negative) (also tumor was inferred from the average logR value of eight see Supplementary Table 2 in Additional file 2), and single-nucleotide polymorphisms (SNPs), which were DNA was also obtained from the breast cancer cell lines within the SIAH2 open reading frame (n =3)orinthe MCF-10A, MCF-7, BT20, SkBr3, Hs578T, T47D, MDA- sequence flanking the gene (n = 5). MB-157, MDA-MB-468, MDA-MB-453, MDA-MB-231, MDA-MB-361, BT483 and ZR75. Bisulfite-modified Results (EpiTect Bisulfite kit; Qiagen, Hilden, Germany) DNA SIAH2 expression in normal breast, in situ and invasive were assessed for SIAH2 methylation using methylation- breast carcinomas sensitive high-resolution melting (MS-HRM) [14]. The SIAH2 expression was identified in the nuclei of occa- MS-HRM primers for SIAH2 were as follows: sional cells within the luminal layer of ducts and acini 5’-TAGAAGCGGGTGGGTTAGGGTTT-3’ (forward) in normal breast tissues in three of 10 patients (30%). and 5’-CTAATACACTCCGCAACCCCC-3’ (reverse) This expression was usually of mild to moderate inten- amplified a region corresponding to GenBank accession sity, and when stratified using the cutoff used for number AC011317.23, nucleotides 105279 to 105409, the tumors, all were considered negative for SIAH2 which contains 17 CpG islands. A polymerase chain (Figure 1A). Expression of SIAH2 was observed in the reaction (PCR) assay was performed in a final volume of nuclei of seven of 54 (13%) DCIS cases. The staining 20 μl. The PCR reaction mixture consisted of 1× PCR buf- was generally of a moderate to strong intensity with a fer (Qiagen), 2.5 mM MgCl2,200μM concentrations of homogeneous distribution (Figures 1B and 1C). There each deoxyribonucleotide triphosphate, a 200 nM concen- was a nonsignificant increase in SIAH2 from the transi- tration of the forward primer, a 200 nM concentration of tion of normal to in situ disease (P = 0.13) and a signifi- the reverse primer, 5 μM SYTO9 intercalating dye (Invi- cant increase in SIAH2 from in situ to invasive breast trogen, Mulgrave, VIC, Australia), 0.5 U of HotStarTaq carcinoma (P = 0.0006) (Table 1 and Figures 1D and 2). DNA Polymerase (Qiagen), and 1 μl (theoretical amount 10 ng) of bisulfite-modified DNA. The PCR amplification Association between SIAH2 protein expression and was performed with an activation step of 15 minutes at clinicopathological characteristics in DCIS 95°C, followed by 50 cycles of 10 seconds at 95°C, 10 sec- There was no significant correlation between SIAH2 onds at an annealing temperature of 66°C, 20 seconds at expression and nuclear grade, presence of necrosis, age, ER, 72°C for extension and one denaturation step of 1 minute progesterone receptor (PR), EGFR or HER2 (all P > 0.05) at 97°C. HRM was directly performed after PCR amplifica- or intrinsic phenotypes in DCIS (P = 0.471) (Supplementary tion. PCR products were denatured at 97°C for 1 minute, Table 3 in Additional file 3). Chan et al. Breast Cancer Research 2011, 13:R19 Page 4 of 10 http://breast-cancer-research.com/content/13/1/R19

Figure 1 Immunohistochemistry of seven in absentia homolog 2 (SIAH2) gene in normal, in situ and invasive breast carcinomas. (A) Occasional nuclear positivity (arrows) in luminal cells in the terminal duct lobular unit. (B) Moderate to strong staining of SIAH2 in the nucleus of a small proportion of the cell in a high nuclear grade ductal carcinoma in situ with comedo necrosis. (C) Occasional weak to moderate SIAH2 (arrows) staining in a luminal type ductal carcinoma. (D) Strong SIAH2 staining in all nuclei in this basal-like breast carcinoma.

Correlation between SIAH2 protein expression with Table 4 in Additional file 4). In the validation cohort, there clinicopathological characteristics and intrinsic subtypes was a significant inverse correlation between SIAH2, ER in invasive cancer (P < 0.0001) and PR (P < 0.0001) and a significant positive In the primary cohort, there was a significant inverse corre- association with tumor grade (P < 0.0001), patient age (P = lation between SIAH2 protein expression and ER (P < 0.009), HER2 (P = 0.007) and intrinsic subtype 0.0001), PR (P = 0.011) and a positive association with (P < 0.0001), but not with tumor size or lymph node status tumor grade (P < 0.0001) and intrinsic subtype (P = 0.028), (P > 0.05) (Supplementary Table 5 in Additional file 5). but there was no association with patient age, tumor size, In the combined cohort, there was a significant inverse lymph node status or HER2 (all P > 0.05) (Supplementary correlation between SIAH2 and ER (P < 0.0001) and PR (P < 0.0001) and a positive correlation with tumor grade (P < 0.0001), HER2 (P = 0.007), p53 (P < 0.001) and intrin- Table 1 c2 tests, SIAH2 expression in normal breast, DCIS sic subtype (P < 0.0001), but not with patient age, tumor and invasive cancersa size or lymph node status (P > 0.05) (Table 2). The signifi- Tissue type Negative, n (%) Positive, n (%) Total, n (%) cant associations between SIAH2 expression grade and Normal 10 (100%) 0 (0%) 10 (100%) intrinsic subgroups were confirmed in a multivariate analy- DCIS 47 (87%) 7 (13%) 54 (100%) sis of tumor phenotype, age, grade and lymph node status Invasive 155 (45%) 194 (55%) 349 (100%) with the basal-like phenotype being more than five Total 212 (51%) 201 (49%) 413 (100%) times more likely to express SIAH2 than luminal tumors P aSIAH2, seven in absentia homolog 2 gene; DCIS, ductal carcinoma in situ; P < (Table 3) ( = 0.015), which was also observed in the 0.0001. combined cohort (P =0.042). Chan et al. Breast Cancer Research 2011, 13:R19 Page 5 of 10 http://breast-cancer-research.com/content/13/1/R19

Relationship between SIAH2 expression and relapse-free and overall survival There was no correlation present between SIAH2 expression and overall relapse-free survival in DCIS-only patients (P = 0.68). Although there was a significantly shorter relapse-free survival in all patients with invasive carcinomas stratified by SIAH2 (P = 0.002) (Figure 4), no significant association with relapse-free survival was observed in univariate analysis in different breast cancer intrinsic groups stratified by SIAH2 (data not shown). There was also no significant association between SIAH2 in invasive carcinomas of all patients and relapse-free survival in multivariate analysis (Table 4).

Discussion Hypoxia is a pivotal driver in breast tumor progression, leading to transcription of several suites of genes involved in angiogenesis, cell survival, cell proliferation and an enhanced metastatic phenotype that are advanta- geous to the neoplastic cells [1]. SIAH2 is part of the ubiquitin ligase complex that target proteins for protea- Figure 2 Semiquantitative seven in absentia homolog 2 (SIAH2) somal degradation and enhances HIF-1a expression by gene expression in normal, ductal carcinoma in situ (DCIS) and invasive carcinoma samples. reducing the abundance of the prolyl hydroxylases [16,17]. These enzymes, in the absence of SIAH2, hydroxylate prolyl residues in the oxygen-dependent domain of HIF-1a, resulting in HIF-1a proteasomal degradation Promoter methylation of SIAH2 in cell lines and tumors and attenuation of the hypoxic response. Since SIAH2 To assess whether SIAH2 expression in tumors is has the potential to profoundly influence the hypoxic modulated by promoter methylation, CpG islands were response, we investigated its expression in normal and identified in the promoter region of SIAH2, and MS- neoplastic breast tissues. HRM primers were designed to cover the CpG-rich area We observed significant upregulation of SIAH2 in the of the promoter region of SIAH2. Five normal breast nucleus in the transition from normal to in situ and tissues, 60 breast carcinomas and 13 breast cancer cell invasive carcinomas in breast cancer, supporting the lines were screened for methylation of SIAH2, but no notion of an important role of SIAH2 in breast cancer promoter methylation was detected in these cancer cell progression. SIAH2 has a nuclear localization signal that lines and samples, as shown by the absence of altered could account for its subcellular pattern of expression methylation profiles (Supplementary Figure 1 in Addi- [18]. The increase in SIAH2 in in situ and invasive car- tional file 6). cinomas correlates with the hypoxia that occurs in neoplasia as the metabolic demand of the tumor exceeds Correlation of gene expression and relative copy number the supply of nutrients and oxygen from the disordered of SIAH2 in basal-like tumors vasculature that is developing. A cohort of familial tumors, which included 15 basal- Correlation analysis showed a significant relationship like tumors, was previously analyzed on the basis of between high levels of breast tumor SIAH2, negative ER gene expression and copy number analysis [15]. The 15 and PR and high HER2. The absence of a positive corre- basal-like tumors showed a significant correlation lation with ER is of interest, since estrogen has been between SIAH2 expression and estimated copy number reported to induce expression of SIAH2 in ER-positive (r = 0.675, P = 0.003) (Figure 3). Two of the 15 basal- breast cancer cell lines [19] and there is a positive rela- like tumors showed a copy number gain (copy number tionship between Siah2 and ER-positive breast tumors of 3) and a further three of 15 basal-like tumors showed but not the basal-like phenotype in six publicly accessi- loss of heterozygosity at this region. In contrast, 15 non- ble data sets [7,20-25] (analysis not shown). This discre- basal-like tumors (10 luminal A, four luminal B and one pancy is likely due to the well-described differences normal-like tumor) did not show any copy number between gene expression and protein abundance or to change in this region. the fact that ER expression may be heterogeneous in Chan et al. Breast Cancer Research 2011, 13:R19 Page 6 of 10 http://breast-cancer-research.com/content/13/1/R19

Table 2 Contingency table of SIAH2 expression in invasive breast carcinomas of the combined cohort with clinicopathological parameters Parameter Negative, N = 169 (45%) Positive, N = 209 (55%) Total, N = 378 (100%) P value Grade <0.0001 Low 40 (78.4%) 11 (21.6%) 51 (100%) Intermediate 77 (60.2%) 51 (39.8%) 128 (100%) High 32 (20.4%) 125 (79.6%) 157 (100%) Age, yr 0.187 <50 48 (39.3%) 74 (60.7%) 122 (100%) >50 107 (47.8%) 117 (52.1%) 224 (100%) Tumor size 0.168 <20 mm 89 (48.6%) 94 (51.4%) 183 (100%) >20 mm 61 (39.9%) 92 (60.1%) 153 (100%) Lymph node status 0.915 Negative 83 (45.4%) 100 (54.6%) 147 (100%) Positive 64 (43.0%) 85 (57.0%) 185 (100%) ER status <0.0001 Negative 21 (20.8%) 80 (79.2%) 101 (100%) Positive 129 (54.4%) 108 (45.6%) 237 (100%) PR status <0.0001 Negative 39 (28.5%) 98 (71.5%) 137 (100%) Positive 111 (55.2%) 90 (44.8%) 201 (100%) HER2 status 0.003 Negative 128 (48.3%) 137 (51.7%) 265 (100%) Positive 22 (29.7%) 52 (70.3%) 74 (100%) Tumor subtype <0.0001 Luminal 114 (57.6%) 84 (42.4%) 198 (100%) Basal-like 4 (9.6%) 37 (90.2%) 41 (100%) Her2 22 (29.3%) 53 (70.7%) 75 (100%) Null 11 (44%) 14 (56%) 25 (100%) P53 Negative 108 (57.8%) 79 (42.2%) 187 (100%) <0.001 Positive 12 (18.2%) 54 (81.8%) 66 (100%) aSIAH2, seven in absentia homolog 2 gene. ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; PR, progesterone receptor. both the pattern and level of expression within tumors. support of this notion is the significant correlation Nevertheless, although there was an inverse relationship between SIAH2 and p53 immunostaining. A further between SIAH2 and ER, approximately half of SIAH2- mechanism in basal-like cancer may also involve p38 positive samples expressed ER, suggesting a complex mitogen-activated protein kinase, which is also upregu- relationship. lated in the basal-like phenotype, as activated p38 The finding that SIAH2 was significantly associated increases the activity of SIAH2 [11]. We also explored with HER2 and basal-like intrinsic breast cancer sub- the role of SIAH2 promoter methylation to assess types, which for basal-like cancers was confirmed in whether protein expression is epigenetically repressed. multivariate analysis, is in accord with our previous We observed no evidence of methylation in any breast report of an enhanced hypoxic drive in basal-like can- carcinoma cell line, normal breast or in a series of 60 cers [26]. In this study, we have demonstrated that breast cancers of variable phenotypes, making this basal-like breast cancers have an intrinsically elevated mechanism of repression highly unlikely in breast tissues SIAH2 level as part of its phenotype that may, partly at in either normal tissue or tumoral tissue. least, explain the mechanism underlying high HIF-1a We then hypothesized that since SIAH2 is located on expression in this tumor subtype. The upregulation of 3q25.1, overexpression might be mediated through gene SIAH2 may be regulated at several levels. We investi- amplification. Indeed, this locus is frequently amplified gated the potential role that p53 may play, since this in basal-like breast cancer [27], and our preliminary gene is frequently mutated in this tumor type [11]. In results showed a significant correlation between DNA Chan et al. Breast Cancer Research 2011, 13:R19 Page 7 of 10 http://breast-cancer-research.com/content/13/1/R19

Table 3 Multivariate analysis in the combined cohort (N = 378) Characteristic P value Hazard 95% CI for hazard ratio ratios Tumor type Luminal 0.09 (reference) Basal 0.02 4.04 1.3 to 12.8 HER2 0.15 1.64 0.8 to 3.2 Null 0.67 1.23 0.5 to 3.2 Grade 1 0.000 2 0.02 2.61 1.2 to 5.8 3 0.0001 12.72 5.2 to 31.4 Lymph node status 0.28 0.75 0.4 to 1.3 Tumor size >20 mm 0.18 0.69 0.4 to 1.2 Multivariate analysis using a binary logistic regression model of the effect of SIAH2 expression on tumor subtype (with luminal tumors as a reference), tumor grade, lymph node status and tumor size. aaSIAH2, seven in absentia homolog 2 gene. 95% CI, 95% confidence interval; HER2, human epidermal growth factor receptor 2. Figure 4 Kaplan-Meier curves stratified by seven in absentia homolog 2 (SIAH2) gene expression for relapse-free survival in copy number and mRNA expression, supporting this (A) all tumors (N = 246) and (B) luminal tumors (n = 144). hypothesis. Specifically, we found that basal-like tumors showed copy number gain of the SIAH2 locus more frequently than luminal tumors and that basal-like tumors Although we observed a significantly shorter relapse- containing copy number gain were associated with high free survival in patients with SIAH2-positive tumors in expression of SIAH2. Nevertheless, using a more sensi- univariate analysis, this was not confirmed in the multi- tive and specific method for quantifying gene copy num- variate analysis model that included conventional prog- ber such as fluorescence in situ hybridization assay, nostic factors such as tumor size, tumor grade and together with SIAH2 protein expression in a validation lymph node status. While SIAH2 was not an indepen- cohort, would be of interest to confirm this finding and dent survival factor in a multivariate analysis model, it assess whether true amplification occurs. was prognostic in the univariate analysis because of its strong association with the basal-like phenotype, which is an independent prognostic factor. Even so, the role of

Table 4 Multivariate analysis using the Cox regression model of relapse-free survival in all breast cancers of validation cohort (N = 245)a Characteristic P value Hazard ratio 95% CI for hazard ratios SIAH2 0.47 1.24 0.7 to 2.2 Grade 1 (reference) 0.54 2 0.70 0.83 0.3 to 2.1 3 0.70 1.22 0.4 to 3.3 Tumor size 0.12 1.37 0.9 to 2.2 Lymph node status 0.001 2.25 1.4 to 3.7 Tumor type Luminal 0.004 Figure 3 Correlation between seven in absentia homolog 2 Basal-like 0.005 3.0 1.4 to 6.4 (SIAH2) gene copy number changes as assessed by the HER2 0.001 2.9 1.6 to 5.7 average logR array value of eight single-nucleotide Null 0.016 2.5 1.2 to 5.1 polymorphisms (SNPs) which located in SIAH2 or within the a flanking region of the gene and normalized expression of SIAH2, seven in absentia homolog 2 gene. SIAH2 in 15 basal-like breast cancers. 95% CI, 95% confidence interval, HER2, human epidermal growth factor receptor 2. Chan et al. Breast Cancer Research 2011, 13:R19 Page 8 of 10 http://breast-cancer-research.com/content/13/1/R19

SIAH2 remains unclear. Thus a report has suggested Additional material that patients with SIAH2-positive, ER-positive tumors have a significantly longer progression-free survival than Additional file 1: Supplementary Table 1. Flow of breast cancer patients with SIAH2-negative tumors and also that patients through the study, according to REMARK criteria [7] Siah2 levels might be a predictive marker of estrogen- Additional file 2: Supplementary Table 2. Tumor phenotype of 60 samples for which seven in absentia homolog 2 (SIAH2) gene responsive disease [28]. The discrepancy between these methylation analysis was performed findings and our own are likely due to the use of mRNA Additional file 3: Supplementary Table 3. Contingency table of ductal levels to measure SIAH2 by Jansen et al. [28], and, carcinoma in situ (DCIS) and available clinicopathological variables despite enriching for neoplastic cells, the stromal com- Additional file 4: Supplementary Table 4. Contingency table of seven partment contributed to the overexpression of SIAH2, in absentia homolog 2 (SIAH2) gene expression in invasive breast carcinomas of the primary cohort with clinicopathological parameters thus confounding the comparison. In support of this Additional file 5: Supplementary Table 5. Contingency table of seven notion, it has been shown by expression microarrays in absentia homolog 2 (SIAH2) gene expression in invasive breast that downregulated SIAH2 in brain metastasis of breast carcinomas of the initial cohort with clinicopathological parameters cancer corresponds with low stromal contamination Additional file 6: Supplementary Figure 1. Seven in absentia homolog [29]. The concept of SIAH2 being a good prognostic 2 (SIAH2) gene methylation in breast carcinoma samples. Methylation- sensitive high-resolution melting (MS-HRM) detects sample methylation and/or predictive parameter is not in accord with the status by melting the amplicons after polymerase chain reaction assay. role of SIAH2 in regulating the hypoxic response or Methylated samples melt later than unmethylated samples, as they have with the observation that inhibition of SIAH2 is asso- cytosines in their sequences rather than the thymines after the bisulfite modification. Two estrogen receptor (ER)-positive and two ER-negative ciated with reduced metastases in animal models [30]. breast carcinomas show no methylation in SIAH2. Standard controls of SIAH2 appears to have several mechanisms of mediating 100%, 50%, 10% and 0% methylation are shown. The curve for each its effect. Some SIAH2 substrates bind directly through sample represents data from duplicate samples. an AXVXP motif, some require adaptor proteins and still others are targeted independently of the above Abbreviations sequence motif [3]. Thus, depending on the cell context, ER: estrogen receptor; HIF: hypoxia-inducible factor; PR: progesterone SIAH2 is likely to have a variety of effects. receptor; Siah: seven in absentia homologue. SIAH2 is critical to the level of the hypoxic response Acknowledgements and therefore is a potential target for anticancer therapy. This study was supported by the Victorian Breast Cancer Research Indeed, since HIF activation results in the regulation of Consortium, the National Health and Medical Research Council (grant 481378 (to SAO), grant 427601 (to RLS) and grant 535947 (SAO), the a large number of genes, interference with this pathway Association for Cancer Research UK (grant AICR 09-0676 (to AM)), Cancer would have broad antineoplastic effects in contrast to Council Victoria (grant to DDB and AM) the Cancer Institute NSW (grant 07/ targeting individual genes, such as VEGF, with bevacizu- CRF/1-06 (to SAO) and grant 07-CDF-1/28 (EAM), the Australian Cancer Research Foundation (ACRF), the RT Hall Trust and the Petre Foundation, mab, which is currently used in the clinic. Indeed, Australia. We thank Professor Adrian L Harris for the kind donation of the menadione, a specific inhibitor of SIAH2, increased breast tumor DNA for methylation analysis. expression of prolyl hydroxylase with a concomitant Author details decrease in levels of HIF-1a. This promising therapeutic 1Department of Pathology, Peter MacCallum Cancer Centre, Locked Bag 1 approach also retarded the growth of melanoma xeno- A’Beckett Street, Melbourne, Victoria 8006, Australia. 2Research Division, grafts [31]. The potential of this approach has also been Cancer Genetics and Genomics, Peter MacCallum Cancer Centre, Locked Bag 1A’Beckett Street, Melbourne, Victoria 8006, Australia. 3Department of investigated using a short protein fragment that compe- Pathology, University of Melbourne, Royal Parade, Parkville, Victoria 3010, titively binds to Siah, resulting in reduced breast cancer Australia. 4Queensland Centre for Medical Genomics, Institute for Molecular growth, which appeared to be mediated through inhibi- Bioscience, University of Queensland, Brisbane, Queensland 4072, Australia. 5Cancer Research Program, Garvan Institute of Medical Research, 384 Victoria tion of the hypoxic response [4]. Street, Darlinghurst, NSW 2010, Australia. 6South East Area Laboratory Services, St George Hospital, Kogarah, NSW 2217, Australia. 7School of Conclusions Medicine, University of Western Sydney, Campbelltown NSW 2751, Australia. 8School of Medical Sciences, Faculty of Medicine, University of New South In summary, we have shown that in situ and invasive Wales, Sydney, NSW 2052, Australia. 9Department of Tissue Pathology and breast carcinomas upregulate SIAH2 and that it is pre- Diagnostic Oncology, Royal Prince Alfred Hospital, Camperdown, NSW 2006, 10 ferentially highly expressed in the basal-like subtype, Australia. Faculty of Medicine, University of Sydney, NSW 2006, Australia. 11Department of Medical Oncology & Breast Cancer Institute of New South which can be accounted for in part by increased gene Wales, University of Sydney, Westmead Hospital, Westmead, NSW 2145, copy number. High levels of SIAH2 may be partly Australia. 12Western Clinical School, University of Sydney, Westmead Hospital, responsible for the enhanced hypoxic drive that under- Westmead, NSW 2145, Australia. lies this tumor type, which is chemotherapy- and radio- Authors’ contributions therapy-resistant. Targeting SIAH2, the most apical PC carried out the scoring of the tissue microarrays (TMAs) and calculated regulator identified in the hypoxic response pathway, statistics. AM conceived and designed the study and drafted the manuscript. MCPL and JES stained the TMAs. CSFW assisted in staining the TMAs and may be a suitable option for anticancer therapy in this drafting the manuscript. NW conducted copy number experiments and breast tumor subtype. analysis. KTH conducted the promoter analysis. AD conducted and Chan et al. Breast Cancer Research 2011, 13:R19 Page 9 of 10 http://breast-cancer-research.com/content/13/1/R19

supervised promoter analysis. EKAM and SAO generated TMAs and scored Flanagan JM, Khanna K, Simpson PT, Lakhani SR, Chenevix-Trench G: correlative markers. CMM conducted clinical follow-up. RLS provided TMAs, Subtypes of familial breast tumours revealed by expression and copy helped in drafting the manuscript and provided critical discussions. DB number profiling. Breast Cancer Res Treat 2010, 123:661-677. helped in designing the study. SBF scored the TMAs, conceived and 16. Fukuba H, Yamashita H, Nagano Y, Jin HG, Hiji M, Ohtsuki T, Takahashi T, designed the study and drafted the manuscript. All authors read and Kohriyama T, Matsumoto M: Siah-1 facilitates ubiquitination and approved the final version of the manuscript. degradation of factor inhibiting HIF-1α (FIH). Biochem Biophys Res Commun 2007, 353:324-329. Competing interests 17. Simon MC: Siah proteins, HIF prolyl hydroxylases, and the physiological The authors declare that they have no competing interests. response to hypoxia. Cell 2004, 117:851-853. 18. Della NG, Senior PV, Bowtell DD: Isolation and characterisation of murine Received: 17 September 2010 Revised: 23 December 2010 homologues of the Drosophila seven in absentia gene (sina). Development Accepted: 9 February 2011 Published: 9 February 2011 1993, 117:1333-1343. 19. Frasor J, Danes JM, Funk CC, Katzenellenbogen BS: Estrogen down- References regulation of the corepressor N-CoR: mechanism and implications for 1. Harris AL: Hypoxia: a key regulatory factor in tumour growth. Nat Rev estrogen derepression of N-CoR-regulated genes. Proc Natl Acad Sci USA Cancer 2002, 2:38-47. 2005, 102:13153-13157. 2. House CM, Moller A, Bowtell DD: Siah proteins: novel drug targets in the 20. Chin K, DeVries S, Fridlyand J, Spellman PT, Roydasgupta R, Kuo WL, Ras and hypoxia pathways. Cancer Res 2009, 69:8835-8838. Lapuk A, Neve RM, Qian Z, Ryder T, Chen F, Feiler H, Tokuyasu T, Kingsley C, 3. Nakayama K, Qi J, Ronai Z: The ubiquitin ligase Siah2 and the hypoxia Dairkee S, Meng Z, Chew K, Pinkel D, Jain A, Ljung BM, Esserman L, response. Mol Cancer Res 2009, 7:443-451. Albertson DG, Waldman FM, Gray JW: Genomic and transcriptional 4. Moller A, House CM, Wong CS, Scanlon DB, Liu MC, Ronai Z, Bowtell DD: aberrations linked to breast cancer pathophysiologies. Cancer Cell 2006, Inhibition of Siah ubiquitin ligase function. Oncogene 2009, 28:289-296. 10:529-541. 5. Bos R, Zhong H, Hanrahan CF, Mommers EC, Semenza GL, Pinedo HM, 21. Desmedt C, Piette F, Loi S, Wang Y, Lallemand F, Haibe-Kains B, Viale G, ’ Abeloff MD, Simons JW, van Diest PJ, van der Wall E: Levels of hypoxia- Delorenzi M, Zhang Y, d Assignies MS, Bergh J, Lidereau R, Ellis P, Harris AL, inducible factor-1α during breast carcinogenesis. J Natl Cancer Inst 2001, Klijn JG, Foekens JA, Cardoso F, Piccart MJ, Buyse M, Sotiriou C: Strong time 93:309-314. dependence of the 76-gene prognostic signature for node-negative 6. Dales JP, Garcia S, Meunier-Carpentier S, Andrac-Meyer L, Haddad O, breast cancer patients in the TRANSBIG multicenter independent Lavaut MN, Allasia C, Bonnier P, Charpin C: Overexpression of hypoxia- validation series. Clin Cancer Res 2007, 13:3207-3214. inducible factor HIF-1α predicts early relapse in breast cancer: 22. Hess KR, Anderson K, Symmans WF, Valero V, Ibrahim N, Mejia JA, Booser D, retrospective study in a series of 745 patients. Int J Cancer 2005, Theriault RL, Buzdar AU, Dempsey PJ, Rouzier R, Sneige N, Ross JS, 116:734-739. Vidaurre T, Gomez HL, Hortobagyi GN, Pusztai L: Pharmacogenomic 7. Tan EY, Yan M, Campo L, Han C, Takano E, Turley H, Candiloro I, Pezzella F, predictor of sensitivity to preoperative chemotherapy with paclitaxel Gatter KC, Millar EK, O’Toole SA, McNeil CM, Crea P, Segara D, and fluorouracil, doxorubicin, and cyclophosphamide in breast cancer. J Sutherland RL, Harris AL, Fox SB: The key hypoxia regulated gene CAIX is Clin Oncol 2006, 24:4236-4244. upregulated in basal-like breast tumours and is associated with 23. Miller LD, Smeds J, George J, Vega VB, Vergara L, Ploner A, Pawitan Y, resistance to chemotherapy. Br J Cancer 2009, 100:405-411. Hall P, Klaar S, Liu ET, Bergh J: An expression signature for p53 status in 8. Millar EK, Anderson LR, McNeil CM, O’Toole SA, Pinese M, Crea P, human breast cancer predicts mutation status, transcriptional effects, Morey AL, Biankin AV, Henshall SM, Musgrove EA, Sutherland RL, Butt AJ: and patient survival. Proc Natl Acad Sci USA 2005, 102:13550-13555. BAG-1 predicts patient outcome and tamoxifen responsiveness in 24. Minn AJ, Gupta GP, Siegel PM, Bos PD, Shu W, Giri DD, Viale A, Olshen AB, ER-positive invasive ductal carcinoma of the breast. Br J Cancer 2009, Gerald WL, Massague J: Genes that mediate breast cancer metastasis to 100:123-133. lung. Nature 2005, 436:518-524. 9. Schmidt RL, Park CH, Ahmed AU, Gundelach JH, Reed NR, Cheng S, 25. Wang Y, Klijn JG, Zhang Y, Sieuwerts AM, Look MP, Yang F, Talantov D, Knudsen BE, Tang AH: Inhibition of RAS-mediated transformation and Timmermans M, Meijer-van Gelder ME, Yu J, Jatkoe T, Berns EM, tumorigenesis by targeting the downstream E3 ubiquitin ligase seven in Atkins D, Foekens JA: Gene-expression profiles to predict distant absentia homologue. Cancer Res 2007, 67:11798-11810. metastasis of lymph-node-negative primary breast cancer. Lancet 10. Ahmed AU, Schmidt RL, Park CH, Reed NR, Hesse SE, Thomas CF, 2005, 365:671-679. Molina JR, Deschamps C, Yang P, Aubry MC, Tang AH: Effect of disrupting 26. Manié E, Vincent-Salomon A, Lehmann-Che J, Pierron G, Turpin E, seven-in-absentia homolog 2 function on lung cancer cell growth. J Natl Warcoin M, Gruel N, Lebigot I, Sastre-Garau X, Lidereau R, Remenieras A, Cancer Inst 2008, 100:1606-1629. Feunteun J, Delattre O, de Thé H, Stoppa-Lyonnet D, Stern MH: High TP53 BRCA1 11. Tan EY, Campo L, Han C, Turley H, Pezzella F, Gatter KC, Harris AL, Fox SB: frequency of mutation in and sporadic basal-like BRCA1 Cytoplasmic location of factor-inhibiting hypoxia-inducible factor is carcinomas but not in luminal breast tumors. Cancer Res 2009, associated with an enhanced hypoxic response and a shorter survival in 69:663-671. invasive breast cancer. Breast Cancer Res 2007, 9:R89. 27. Gatza ML, Lucas JE, Barry WT, Kim JW, Wang Q, Crawford MD, Datto MB, 12. Mirzoeva OK, Das D, Heiser LM, Bhattacharya S, Siwak D, Gendelman R, Kelley M, Mathey-Prevot B, Potti A, Nevins JR: A pathway-based Bayani N, Wang NJ, Neve RM, Guan Y, Hu Z, Knight Z, Feiler HS, Gascard P, classification of human breast cancer. Proc Natl Acad Sci USA 2010, Parvin B, Spellman PT, Shokat KM, Wyrobek AJ, Bissell MJ, McCormick F, 107:6994-6999. Kuo WL, Mills GB, Gray JW, Korn WM: Basal subtype and MAPK/ERK kinase 28. Jansen MP, Ruigrok-Ritstier K, Dorssers LC, van Staveren IL, Look MP, Meijer- (MEK)-phosphoinositide 3-kinase feedback signaling determine van Gelder ME, Sieuwerts AM, Helleman J, Sleijfer S, Klijn JG, Foekens JA, susceptibility of breast cancer cells to MEK inhibition. Cancer Res 2009, Berns EM: Downregulation of SIAH2, an ubiquitin E3 ligase, is associated 69:565-572. with resistance to endocrine therapy in breast cancer. Breast Cancer Res 13. Nielsen TO, Hsu FD, Jensen K, Cheang M, Karaca G, Hu Z, Hernandez- Treat 2009, 116:263-271. Boussard T, Livasy C, Cowan D, Dressler L, Akslen LA, Ragaz J, Gown AM, 29. Palmieri D, Fitzgerald D, Shreeve SM, Hua E, Bronder JL, Weil RJ, Davis S, Gilks CB, van de Rijn M, Perou CM: Immunohistochemical and clinical Stark AM, Merino MJ, Kurek R, Mehdorn HM, Davis G, Steinberg SM, characterization of the basal-like subtype of invasive breast carcinoma. Meltzer PS, Aldape K, Steeg PS: Analyses of resected human brain Clin Cancer Res 2004, 10:5367-5374. metastases of breast cancer reveal the association between up- 14. Wojdacz TK, Dobrovic A: Methylation-sensitive high resolution melting regulation of hexokinase 2 and poor prognosis. Mol Cancer Res 2009, (MS-HRM): a new approach for sensitive and high-throughput 7:1438-1445. assessment of methylation. Nucleic Acids Res 2007, 35:e41. 30. Qi J, Nakayama K, Gaitonde S, Goydos JS, Krajewski S, Eroshkin A, Bar- 15. Waddell N, Arnold J, Cocciardi S, da Silva L, Marsh A, Riley J, Johnstone CN, Sagi D, Bowtell D, Ronai Z: The ubiquitin ligase Siah2 regulates Orloff M, Assie G, Eng C, Reid L, Keith P, Yan M, Fox S, Devilee P, tumorigenesis and metastasis by HIF-dependent and -independent Godwin AK, Hogervorst FB, Couch F, Investigators kConFab, Grimmond S, pathways. Proc Natl Acad Sci USA 2008, 105:16713-16718. Chan et al. Breast Cancer Research 2011, 13:R19 Page 10 of 10 http://breast-cancer-research.com/content/13/1/R19

31. Shah M, Stebbins JL, Dewing A, Qi J, Pellecchia M, Ronai ZA: Inhibition of Siah2 ubiquitin ligase by vitamin K3 (menadione) attenuates hypoxia and MAPK signaling and blocks melanoma tumorigenesis. Pigment Cell Melanoma Res 2009, 22:799-808.

doi:10.1186/bcr2828 Cite this article as: Chan et al.: The expression of the ubiquitin ligase SIAH2 (seven in absentia homolog 2) is mediated through gene copy number in breast cancer and is associated with a basal-like phenotype and p53 expression. Breast Cancer Research 2011 13:R19.

Submit your next manuscript to BioMed Central and take full advantage of:

• Convenient online submission • Thorough peer review • No space constraints or color ﬁgure charges • Immediate publication on acceptance • Inclusion in PubMed, CAS, Scopus and Google Scholar • Research which is freely available for redistribution

Submit your manuscript at www.biomedcentral.com/submit Xu et al. Breast Cancer Research 2011, 13:R9 http://breast-cancer-research.com/content/13/1/R9

RESEARCHARTICLE Open Access Enhanced RAD21 cohesin expression confers poor prognosis and resistance to chemotherapy in high grade luminal, basal and HER2 breast cancers Huiling Xu1,3†, Max Yan2†, Jennifer Patra1, Rachael Natrajan4, Yuqian Yan1, Sigrid Swagemakers5,6, Jonathan M Tomaszewski1,7, Sandra Verschoor1, Ewan KA Millar8,9,10,11, Peter van der Spek5, Jorge S Reis-Filho4, Robert G Ramsay1, Sandra A O’Toole8,12,13,14, Catriona M McNeil8,15,16,17, Robert L Sutherland8,12, Michael J McKay1,18*, Stephen B Fox2*

Abstract Introduction: RAD21 is a component of the cohesin complex, which is essential for chromosome segregation and error-free DNA repair. We assessed its prognostic and predictive power in a cohort of in situ and invasive breast cancers, and its effect on chemosensitivity in vitro. Methods: RAD21 immunohistochemistry was performed on 345 invasive and 60 pure in situ carcinomas. Integrated genomic and transcriptomic analyses were performed on a further 48 grade 3 invasive cancers. Chemosensitivity was assessed in breast cancer cell lines with an engineered spectrum of RAD21 expression. Results: RAD21 expression correlated with early relapse in all patients (hazard ratio (HR) 1.74, 95% confidence interval (CI) 1.06 to 2.86, P = 0.029). This was due to the effect of grade 3 tumors (but not grade 1 or 2) in which RAD21 expression correlated with early relapse in luminal (P = 0.040), basal (P = 0.018) and HER2 (P = 0.039) groups. In patients treated with chemotherapy, RAD21 expression was associated with shorter overall survival (P = 0.020). RAD21 mRNA expression correlated with DNA copy number, with amplification present in 32% (7/22) of luminal, 31% (4/13) of basal and 22% (2/9) of HER2 grade 3 cancers. Variations in RAD21 mRNA expression in the clinical samples were reflected in the gene expression data from 36 breast cancer cell lines. Knockdown of RAD21 in the MDA-MB-231 breast cancer cell line significantly enhanced sensitivity to cyclophosphamide, 5-fluorouracil and etoposide. The findings for the former two drugs recapitulated the clinical findings. Conclusions: RAD21 expression confers poor prognosis and resistance to chemotherapy in high grade luminal, basal and HER2 breast cancers. RAD21 may be a novel therapeutic target.

Introduction fundamental to several key cellular processes, including Cohesin is a multi-protein complex that is highly con- chromosome segregation during mitosis and meiosis, served from yeast to humans. Its primary role is to error-free homologous recombinational repair (HRR) of adhere sister chromatids in close apposition, a mechan- DNA double strand breaks and the regulation of gene ism termed ‘sister chromatid cohesion’ (SCC). SCC is transcription [1-7]. The core cohesin complex consists of four proteins, RAD21 (also known as SCC1 or * Correspondence: [email protected]; [email protected] MCD1), SMC1, SMC3 and SCC3 [8]. † Contributed equally RAD21 is a central component of the cohesin com- 1Research Division, Peter MacCallum Cancer Centre, Locked Bag 1, A’Beckett Street, Melbourne, Vic 8006, Australia plex, both structurally and functionally [8]. Aberrant 2Department of Pathology, Peter MacCallum Cancer Centre, Locked Bag 1, RAD21 expression has been reported in multiple cancers A’Beckett Street, Melbourne, Vic 8006, Australia and cancer cell lines [9-12]. In a mega-scale microarray Full list of author information is available at the end of the article

© 2011 Xu et al.; licensee BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Xu et al. Breast Cancer Research 2011, 13:R9 Page 2 of 10 http://breast-cancer-research.com/content/13/1/R9

analysis of multiple cancers, RAD21 was identified as 1 mM EDTA for three minutes at 125°C in a pressure one of 69 signature genes in undifferentiated cancers cooker (Biocare Decloker, Concord, CA, USA). Sections that had aggressive in vitro or clinical courses and poor were then treated with 3% H2O2 for five minutes to patient outcomes [9]. Further, an intronal single nucleo- remove endogenous peroxides, washed and incubated tide polymorphism (SNP) in the RAD21 gene is strongly with a rabbit polyclonal anti-RAD21 antibody (1:200) associated with increased breast cancer risk [10]. (Abcam, Cambridge, UK), with anti-rabbit IgG as nega- Although these reports support the notion that the tive control, for two hours at room temperature. Horse- abnormal activity of RAD21 may be an important fea- radish peroxidase (HRP)-conjugated secondary antibody, ture of human breast cancer, there are no data available the signal was detected using DAB (3’,3’-diaminobenzi- from clinical breast cancer samples. dine) substrate. Sections were counter-stained with We therefore, evaluated RAD21 expression in a cohort hematoxylin to visualize nuclei. Validation of the anti- of well-characterised human in situ and invasive breast RAD21 antibody was performed using small interference cancers to 1) assess the correlation between RAD21 RNA (siRNA) knockdown of the human RAD21 gene in expression, and conventional and molecular clinico- MCF10A cells on cell blocks (Additional file 2). pathological parameters and patient prognostic data; Nuclear RAD21 expression was assessed for intensity (0 and 2) determine whether aberrant RAD21 expression = no staining, 1 = weak, 2 = moderate, 3 = strong) and might predict therapeutic outcomes. the percentage of positive cells (0 = 0%, 1 ≤10%, 2 = 10% to 50%, 3 = 51% to 80%, 4 ≥80% positive cells) as defined Materials and methods previously [14]. The scores for intensity and percentage Patient clinicopathological variables were added and a cut-off of 7 was used to define two The flow of patients through the study according to the approximately equal size groups of patients for subse- reporting recommendations for tumor marker prognos- quent statistical analyses. tic studies (REMARK) criteria [13] is listed in Additional ER, HER2, EGFR and CK5/6 staining were used to file 1. Four hundred and nine invasive cancers were classify tumors into four intrinsic subgroups: the basal obtained from the Garvan Institute (292 cases with sur- group (ER negative, HER2 negative, CK5/6 and/or EGFR vival and treatment data) and the Peter MacCallum positive), luminal group (ER positive, HER2 negative), Cancer Centre (117 cases without survival data). Sixty- HER2 group (HER2 positive) and the negative (null) four cases were excluded due to lack of tissue available group (ER, HER2, CK5/6 and EGFR negative) [15]. for tissue microarray (TMA) construction or absence of Cell lines tumor on the array. The final cohort of invasive cancers Cell lines (MCF7, MCF10A, MDA-MB-231, MDA-MB- was 345 cases (251 cases with survival data). For ductal 468, SK-BR-3, T47 D and ZR75-1) were obtained from the carcinoma in situ (DCIS), 60 cases of pure DCIS were American Type Culture Collection (ATCC, Rockville, MD, obtained from the John Radcliffe Hospital, UK, and USA). MCF10A was grown and maintained in a 1:1 mix- were assessed on TMAs. This study has Ethics Commit- ture of Dulbecco’s modified Eagle medium (DMEM) F-12 tee approvals (numbers 00/81, 03/90, 09/36, JRC02.216, supplemented with 5% horse serum, 20 ng of epidermal HREC SVH H94/080 and HREC SVH 06336 H00036). growth factor (EGF) per ml, 10 μg/ml of insulin, and 0.5 Patient median age was 54 years (range 24 to 87 years). μg/ml of hydrocortisone. MDA-MB-468, SK-BR-3 and Forty-one percent of patients received adjuvant che- ZR75-1 were grown in MEM with 10% fetal bovine serum motherapy with cyclophosphamide/methotrexate/5- (FBS) and 1% Pen/Strep. Other cell lines were grown in fluorouracil (CMF), or doxorubicin (adriamycin)/cyclo- RPMI-1640 with 10% FBS and 1% Pen/Strep. SVCT was phosphamide (AC). Fifty-two percent received adjuvant obtained from European Collection of Cell Cultures endocrine therapy with tamoxifen. Patient median fol- (ECACC; Salisbury, Wilts, UK) and grown in DMEM with low-up was 58.1 months. During this time, 100 patients 10% FBS, 5 μg/ml hydrocortisone, and 10 μg/ml insulin. developed recurrence (24.4%) and 86 deaths (21.0%) Quantitative real time PCR and Semi-quantitative Western were considered breast-cancer related. blot analysis Exponentially growing cells were harvested and total Immunohistochemistry RNA was extracted using RNeasy kits (Qiagen, Valencia, TMAs were constructed from 1 mm diameter (invasive CA, USA). DNA was removed by on-column treatment cancers) or 2 mm cores (DCIS). Sections of 4 μm thick- with RNase-free DNase. One microgram of total RNA ness were used for immunostaining using the EnVi- was used for cDNA synthesis using a Superscript III kit sion™kit (DAKO, Glostrup, Denmark) following the (Invitrogen, Carlsbad, CA, USA). cDNA was used for manufacturer’s instructions. Briefly, de-paraffinized and quantitative real time polymerase chain reaction (qRT- rehydrated tissue sections were treated for antigen PCR) using primers (5’-AATTTGGCTAGCGGCCCAT- retrieval in 10 mM Tris-HCl (pH 9.0) buffer containing 3’ and 5’-TGTCCGTAATGCCATTTTCACC-3’)which Xu et al. Breast Cancer Research 2011, 13:R9 Page 3 of 10 http://breast-cancer-research.com/content/13/1/R9

span exon 10 and exon 11 of the human RAD21 gene. basal subtype. Two different short-hairpin RNA Phosphoglycerate kinase (PGK) was used as a house- (shRNA) shRNAmir constructs, shRNA 57223 and keeping gene and the relative expression of RAD21 gene shRNA 57224, and shRNAmir vector control (Open Bio- was determined from three independent experiments systems, Huntsville, AL, USA) were introduced into cells using the DeltaDelta CT method [16]. using Lipofectamine 2000 (Invitrogen, Carlsbad, CA, Total protein extraction and Western blots were carried USA) and clones were selected following a two-week out essentially as described previously [17]. For Western culture in 2 μg/ml puromycin. Work involving recombi- blot analysis, blots were incubated with a rabbit polyclonal nant DNA was conducted in approved Physical Contain- anti-RAD21 antibody (Abcam, Cambridge, UK) followed ment level 2 (PC2) facilities. Western blot analysis and by a fluorescence-conjugated secondary antibody, Alexa quantitative real-time PCR were used to verify the level 680 anti-rabbit (Invitrogen, Carlsbad, CA, USA). The sig- of RAD21 protein and mRNA. nal intensity was measured using the Li-Cor Odyssey sys- Clonogenic survival was performed as described [21]. tem. Membranes were then probed with a mouse Exponentially growing cells were seeded in triplicate monoclonal anti-pan actin antibody (Cell Signaling Tech- plates, allowed to adhere for four to six hours. Following nology, Danvers, MA, USA) followed by an IRDye800- 24-hour incubation with individual drugs at graded con- conjugated anti-mouse antibody (Rockland, Gilbertsville, centrations, cells were cultured for two weeks. Surviving PA, USA). The relative level of protein expression was colonies defined as containing more than 50 cells were determined by normalizing to the pan-actin loading con- counted. Three independent experiments were per- trol from a minimum of three independent experiments. formed for each treatment. Survival curves were generated using the linear quadratic model, GraphPad Prism Array CGH and gene expression analysis version 5.01 for Windows, GraphPad Software, San Array comparative genomic hybridization (CGH) and Diego, CA, USA. microarray-based expression profiling were obtained from analysis of 48 microdissected grade 3 invasive duc- Statistical analysis tal carcinomas as described [18]. Tumor subtypes were Correlations were examined using the one-way determined as described above [15]. ANOVA, Students t-test, or chi-square test where appropriate. Kaplan-Meier survival curves were calcu- Gene expression data mining lated using tumor recurrence (relapse free survival) Gene expression profiles of 38 breast cancer cell lines were and breast cancer-related death (overall survival) as obtained by mining a microarray dataset described by Hol- the endpoints and compared using a log rank test. lestelle et al. (2009) [19] at GEO (accession number (GEO: Binary logistic regression was used for multivariate GSE16795)) [20]. Raw intensity values of all samples were analyses and the Cox proportional hazard regression normalized by RMA normalization (Robust Multichip model was used to identify independent prognostic Analysis) (background correction and normalization) factors for disease-free and overall survival. Analyses using Partek version 6.4 (Partek Inc., St. Louis, MO). The were performed with SPSS 16.0 (SPSS Inc., IL, USA). normalized data file was transposed, back transformed to Atwo-tailedP-value test was used in all analyses and normal intensity values and imported into OmniViz ver- a P-value of less than 0.05 was considered statistically sion 6.0.1 (BioWisdom Ltd., Cambridge, UK) for further significant. analysis. For each probe set, the geometric mean of the hybridization intensity of all samples was calculated. The Results level of expression of each probe set was determined rela- RAD21 protein expression in in situ and invasive breast tive to this geometric mean and 2log-transformed. The cancer geometric mean of the hybridization signal of all samples Expression of RAD21 in DCIS ranged from negative was used to ascribe equal weight to gene expression levels (2/60 cases, 3%) to heterogenous staining (30/60 cases, with similar relative distances to the geometric mean. 50%) and homogenous strong staining (28/60 cases, Next a query-by-example numerical query was performed 47%) (Figure 1A). For invasive cancers, RAD21 staining in OmniViz to find records most closely related to RAD21. patterns were similar to DCIS, ranging from heteroge- The top 25 genes that correlated the best with RAD21 neous (50/95 cases, 53%) and homogeneous strong were visualized using a treescape view. staining (42/95 cases, 44%) to negative (3/95 cases, 3%) (Figure 1B, C). No staining or minimal weak staining Generation of stable RAD21 knockdown cell lines and was present in the cytoplasm in all cases. Expression clonogenic survival assays of RAD21 in invasive cancers was significantly lower A single cell clone was derived from the human breast (104/343, 30%) than in in situ cancers (28/60, 47%) cancer cell line, MDA-MB-231. This cell line is of the (P =0.001). Xu et al. Breast Cancer Research 2011, 13:R9 Page 4 of 10 http://breast-cancer-research.com/content/13/1/R9

Table 1 RAD21 expression by DCIS grade (P = 0.428) Grade Negative Positive Total n (%) n (%) n (%) Low 3 (50.0%) 3 (50.0%) 6 (100.0%) Intermediate 9 (60.0%) 6 (40.0%) 15 (100.0%) High 8 (38.1%) 13 (61.9%) 21 (100.0%) Total 20 (47.6%) 22 (52.4%) 42 (100.0%) DCIS, ductal carcinoma in situ.

but not with tumor grade (P = 0.328), age (P = 0.815), HER2 status (P = 0.564) or ER status (P = 0.054) (Table 3). Positive RAD21 expression was seen in 37% (75/201) luminal, 24% (10/42) basal, 22% (9/49) HER2 and 18% (5/28) null, cancers. When compared to luminal cancers, null type cancers, but not basal and HER2 cancers, were significantly more likely to be RAD21 negative (P = 0.043) (Table 4).

RAD21 expression and its correlation with relapse-free survival There was a significant correlation between positive RAD21 expression and shorter relapse-free survival (RFS) (P = 0.009) (Figure 2A). The association with early relapse was confirmed on multivariate analysis (P =0.029,HR= 1.74, 95% CI 1.06 to 2.86) (Table 5). Subset analysis revealed RAD21 expression correlated with relapse in grade 3 (P = 0.023) (Figure 2B) but not in grade 1 or 2 tumors (P = 0.342) (Figure 2C). Further analysis of grade 3 tumors according to subtype showed a significant correlation between RAD21 expression and shorter RFS in the grade 3 luminal (P = 0.040) (Figure 2D), grade 3 basal (P = 0.018) (Figure 2E) and grade 3 HER2 cancers (P =0.039) (Figure 2F), but not null type cancers (P =0.247).

RAD21 expression and its correlation with overall survival Figure 1 RAD21 immunohistochemistry in DCIS and invasive in patients treated with chemotherapy carcinoma. A, Strong nuclear RAD21 staining in DCIS. Scale bar = Among patients not treated with chemotherapy, there was 20 μm. B, Strong nuclear RAD21 staining in an invasive carcinoma, no correlation between RAD21 expression and overall sur- μ luminal type. Scale bar = 20 m. C, Absence of nuclear staining in vival (P = 0.779) (Figure 2G), whereas among patients an invasive carcinoma, basal type. Scale bar = 20 μm. treated with chemotherapy there was a significantly shorter overall survival in patients whose tumors were Association between RAD21 expression and positive for RAD21 expression (P = 0.020) (Figure 2H). clinicopathological characteristics in DCIS This association is also true for patients with grade 3 RAD21 expression did not significantly associate with nuclear grade (P = 0.428) (Table 1) or the intrinsic phenotypes in DCIS (P = 0.471) (Table 2). There was no Table 2 RAD21 expression in DCIS subtypes (P = 0.471) correlation between RAD21 expression in DCIS and risk Subtype Negative Positive Total of relapse (P = 0.834). n (%) n (%) n (%) Luminal 21 (55.3%) 17 (44.7%) 38 (100%) Association between RAD21 expression and Basal 0 (.0%) 1 (100.0%) 1 (100%) clinicopathological characteristics and intrinsic subtypes HER2 7 (43.8%) 9 (56.2%) 16 (100%) in invasive cancer Null 3 (75.0%) 1 (25.0%) 4 (100%) RAD21 expression correlated with larger tumor size Total 31 (52.5%) 28 (47.5%) 59 (100.0%) (P = 0.012) and lymph node involvement (P < 0.001), DCIS, ductal carcinoma in situ. Xu et al. Breast Cancer Research 2011, 13:R9 Page 5 of 10 http://breast-cancer-research.com/content/13/1/R9

Table 3 RAD21 expression in invasive carcinoma by clinicopathological parameters negative positive Total P value n (%) n (%) n (%) Grade P = 0.328 1 42 (17.6%) 16 (15.4%) 58 (16.9%) 2 96 (40.2%) 35 (33.7%) 131 (38.2%) 3 101 (42.3%) 53 (51.0%) 154 (44.9%) Size P = 0.012 <20 mm 147 (61.8%) 50 (47.2) 197 (57.3%) >20 mm 91 (38.2%) 56 (52.8%) 147 (42.7) Lymph node P < 0.001 Negative 148 (63.5%) 43 (42.2%) 191 (57.0%) Positive 85 (36.5%) 59 (57.8%) 144 (43.0%) Age P = 0.815 <50 84 (35.1%) 39 (36.4%) 123 (35.5%) >50 155 (64.9) 68 (63.6%) 223 (64.5%) ER P = 0.054 Negative 63 (34.4%) 14 (21.5%) 77 (31/0%) Positive 120 (65.6%) 51 (78.5%) 171 (69.0%) HER2 P = 0.564 Negative 142 (79.8%) 54 (83.1%) 196 (80.7%) Positive 36 (20.2%) 11 (16.9%) 47 (19.3%) Chemotherapy* P = 0.032 No 117 (62.95) 31 (47.7%) 148 (59.0%) Yes 69 (37.1%) 34 (52.3%) 103 (41.0%) Endocrine therapy P = 0.034 No 97 (52.2%) 24 (36.9%) 121 (48.2%) Yes 89 (47.8%) 41 (63.1%) 130 (51.8%) Figure 2 Kaplan-Meier curves stratified by nuclear RAD21 * Chemotherapy was given either as cylcophosphamide/methotrexate/ expression, relapse-free survival (A-F) and overall survival (G- fluorouracil (CMF), or doxorubicin (adriamycin)/cyclophosphamide (AC). H). Relapse free survival: A, in all tumors, (P = 0.009) (n = 247); B, ER, oestrogen receptor. grade 3 cancers (P = 0.023) (n = 117); C, grade 1 and 2 cancers (P = 0.342) (n = 130); D, grade 3 luminal cancers (P = 0.040) (n = 32); E, tumors (P = 0.021). No significant difference in overall grade 3 basal cancers (P = 0.018) (n = 29); F, grade 3 HER2 cancers (n = 34); Overall survival: G, without chemotherapy (P = 0.779) (n = survival was seen in patients treated with endocrine ther- 148). H, treated with chemotherapy (P = 0.020) (n = 91). apy, when stratified by RAD21 expression (P = 0.231).

RAD21 gene expression correlates with copy number number, in an integrated array CGH and transcriptional alterations, and RAD21 is amplified in a subset of grade 3 dataset generated from 48 microdissected grade 3 inva- luminal, basal and HER2 cancers sive ductal carcinomas of luminal (n = 22), basal-like (n In view of the correlation of RAD21 expression with = 13) and HER2 (n = 13) subtypes [18]. Array CGH and prognosis in grade 3 cancers, we examined RAD21 microarray expression profiling showed RAD21 mRNA mRNA expression for its association with gene copy expression correlated with gene copy number in luminal

Table 5 Multivariate analysis, Cox regression model of Table 4 RAD21 expression in breast cancers by intrinsic relapse-free survival in all breast cancers subtype P value Hazard Ratio 95% CI for hazard ratio Subtype negative positive Total P-value RAD21 0.029 1.74 1.06 to 2.86 n (%) n (%) n (%) Relative to luminal Grade 0.018 1.63 1.09 to 2.44 Luminal 126 (63%) 75 (37%) 201 (100%) - Size>20mm 0.656 1.11 0.71 to 1.73 Basal-like 32 (76%) 10 (24%) 42 (100%) 0.095 Lymph node status 0.023 1.69 1.07 to 2.65 HER2 38 (78%) 11 (22%) 49 (100%) 0.050 ER 0.003 0.46 0.28 to 0.77 Null 23 (82%) 5 (18%) 28 (100%) 0.043 Age>50 0.325 1.28 0.781 to 2.11 Total 239 (70%) 104 (30%) 320 (100%) CI, confidence interval; ER, oestrogen receptor. Xu et al. Breast Cancer Research 2011, 13:R9 Page 6 of 10 http://breast-cancer-research.com/content/13/1/R9

(P = 0.003), basal (P = 0.0086) and HER2 (P = 0.0035) tumors (Pearson correlation, Table 6). RAD21 amplification is present in 32% (7/22) of luminal, 31% (4/13) of basal and 22% (2/9) of HER2 subtypes. These proportions were very similar to our immunohistochemistry analysis of a different sample set described above, where 30% of luminal (14/46), 25% of basal (10/40), and 22% of HER2 (9/41) grade 3 cancers showed positive RAD21 expression. Collectively, these data suggest that positive RAD21 expression observed in a subset of grade 3 tumors may be due to gene amplification.

RAD21 expression in breast cancer cell lines Variations in RAD21 protein expression in clinical samples were reflected by gene expression analysis using qRT-PCR of a panel of breast cancer cell lines (Figure 3A), and by microarray profiling of 36 breast cancer cell lines derived from Hollestelle et al. [19] (Figure 3B). Further analysis revealed that TOP2A which encodes topoisomerase II (a protein also required for sister chromatid separation) and NIPBL (encoding a cohesin loading protein) are among top 25 genes positively correlated with RAD21 expression (Figure 3B).

Knockdown of RAD21 gene expression with short-hairpin RNA in a basal-like breast cancer cell line, MDA-MB-231, results in its enhanced sensitivity to chemotherapeutic Figure 3 Expression of RAD21 in breast cancer cell lines. A. drugs Quantitative real time RT-PCR of RAD21 transcripts in human breast To test the functional significance of our cancer therapy cancer cell lines. The expression level in MCF10A was used as a reference and given an arbitary value of 1. Relative expression of results that RAD21 expression affects sensitivity to che- RAD21 gene was compared with MCF10A. Error bar = standard error motherapeutic drug response, we used a small hairpin mean (SEM). B. OmniViz Treescape showing the hierarchical shRNA-mediated gene-silencing approach to knockdown clustering of the top 25 genes that correlated best with the two the RAD21 gene in MDA-MB-231 breast cancer cell RAD21 gene probes (200607_s_at and 200608_s_at). Gene line. Of several independent cell clones generated using expression levels: red, up-regulation compared with the geometric mean; green, down-regulation compared with the geometric mean. two different shRNA constructs, three clones exhibited The color intensity correlates with the degree of change. Raw RAD21 a reduction in both transcripts and protein dataset (GEO:GSE16795) was sourced from GEO [20]. (Figure 4). The relative levels of RAD21 mRNA in four stable clones, as determined by qRT-PCR analysis, were 56 ± 4% for sh223_sc1 (P = 0.010), 90 ± 13% for Table 6 Correlation of RAD21 gene expression with sh223_sc3 (P = 0.531), 62 ± 13% for sh224_sc4 (P = genomic alterations* 0.111) and 55 ± 2% for sh224_sc5 (P = 0.002), relative Subtype Copy number Gaina Amplificationb to the parental cell line (Figure 4A). No apparent reduc- Pearson Mann Whitney Mann Whitney tion in RAD21 mRNA was detected in the control clone correlation U test U test P P-value P-value P-value transfected with shRNAmir vector (96% ± 8%, = 0.726) (Figure 4A). Further examination of the corre- Luminal (n = 0.0030 0.0169 0.0465 22) sponding RAD21 protein by semi-quantitative Western Basal-like (n 0.0086 – 0.0111 blot analysis revealed a statistically significant reduction = 13) in the levels of RAD21 protein in the three clones HER2 (n = 13) 0.0035 0.0503 0.1025 (sh223_sc1, sh224_sc4 and sh224_sc5), which exhibited * Patients and Methods were as described in Natrajan et al. (2009). RAD21 mRNA reduction (Figure 4A, B). The relative a Gain corresponds to approximately three to five copies of the locus as levels of RAD21 protein were 82% ± 4% for sh223_sc1 defined using a smoothed Log2 ratio of between 0.08 and 0.45. (P = 0.017), 65 ± 4% for sh224_sc4 (P = 0.002) and 60 ± b Amplification was defined as having a Log2 ratio > 0.45, corresponding to more than five copies. 6% for sh224_sc5 (P = 0.007), relative to the parental CGH, comparative genomic hybridization. cell line (Figure 4B). The RAD21 protein levels in the Xu et al. Breast Cancer Research 2011, 13:R9 Page 7 of 10 http://breast-cancer-research.com/content/13/1/R9

and protein levels was confirmed in three stable shRAD21 knockdown clones. We next tested the response of shRAD21 clones to two breast cancer chemotherapeutic drugs, cyclophosphamide and 5-fluorouracil (5-FU). All three RAD21 knockdown clones, sh223_sc1, sh224_sc4 and sh224_sc5, showed increased sensitivity to the drug following treatment with cyclophosphamide, in a manner that directly correlates with the level of RAD21 expression (Figure 5A). In contrast, such enhanced sensitivity was not observed in the clone sh223_sc3 which did not show a reduction in either RAD21 mRNA or protein level (Figure 4). We noted that the sh223_sc1 clone exhibited a more reduced clonogenic survival compared to the other clones, sh224_sc4 and sh224_sc5. This may be due to the difference in targeting sequences between the sh223 and sh224 shRNAs. Similarly, treatment of three shRAD21 clones (sh223_sc1, sh224_sc4 and sh224_sc5) with 5-FU resulted in a significant reduction in the clonogenic survival of all three clones compared to the parental line (Figure 5B). These data recapitulate our findings in patients treated with cyclophosphamide/ methotrexate/5-fluorouracil (CMF) or doxorubicin (adriamycin)/cyclophosphamide (AC) (Figure 2H), providing further evidence that RAD21 expression correlates with cellular sensitivity to chemotherapeutic-drugs. Since our analysis showed that RAD21 expression Figure 4 Validation of RAD21 expression in MDA-MB-231 cells TOP2A carrying shRAD21 knockdown constructs. A. Quantitative real strongly correlates with expression in a number time PCR analysis of RAD21 expression in clones stably expressing of breast cancer cell lines (see Figure 3), we assessed two different shRAD21 constructs and shRNAmir vector, relative to the sensitivity of RAD21 knockdown clones to etoposide, the parental cell line. B. Western blot analysis of RAD21 protein a topoisomerase II inhibitor and commonly used anti-c- level. RAD21 expression in five independently derived clonal cell ancer drug. Etoposide treatment also led to a marked lines with two different stable RAD21 knockdown constructs was compared to the parental cell line, cells transfected with shRNAmir decrease in cell survival in the shRAD21 clones tested, vector and an immortalized human breast epithelial cell line compared to the parental cells (Figure 5C). This result MCF10A. Pan-actin was used as loading control. A reduction in the is consistent with an early report of an enhanced etopo- levels of RAD21 protein in three stable clones (sh223_sc1, sh224_sc4 side-sensitivity following a siRNA-mediated transient and sh224_sc5) was verified by semi-quantitative Western blot RAD21 knockdown in MCF7 breast cancer cell line [10]. analysis (bottom panel). The levels of RAD21 protein were normalized to the pan-actin loading control and expressed as the percentage of the parental line where the RAD21 expression was Discussion given an arbitrary value of 100%. The values represent the mean of This translational study is the first comprehensive analy- four independent experiments except for shRNAmir vector and sis of a novel chromosomal DNA repair protein, RAD21 MCF10A where three independent experiments were performed. cohesin, in breast cancer. Our analyses provide compel- Clones with a significant reduction in either RAD21 mRNA or protein levels were marked by asterisks (* P < 0.05; ** P < 0.005, Student t- ling evidence that RAD21 expression is a novel prognos- test). Error bar = SEM. tic marker in breast cancer and is also highly predictive of anti-cancer therapy outcomes. Tumor RAD21 overexpression strongly correlated with amplification of the sh224_sc4 and sh224_sc5 clones are comparable to that RAD21 gene locus in a significant subset of high grade in an immortalized human mammary epithelial cell line, luminal, basal and HER2 cancers. This suggests that the MCF10A (Figure 4B). No apparent reduction in RAD21 observed RAD21 overexpression resulted from gene protein was detected in sh223_sc2 (99% ± 4%, P = amplification, and provides a plausible explanation for 0.700), sh223_sc3 (97 ± 4%, P = 0.367) and shRNAmir the strong RAD21 prognostic effects observed in these vector (101 ± 9%, P = 0.947), compared to the parental tumors. Our findings in breast cancer with the RAD21 line (Figure 4B). Thus, reduction of both RAD21 mRNA cohesin may also generalize to some other epithelial Xu et al. Breast Cancer Research 2011, 13:R9 Page 8 of 10 http://breast-cancer-research.com/content/13/1/R9

simple, cost-effective and novel way for evaluating HR activity on tissue sections. Alternatively, numerical chromosome (ploidy) alterations as a result of chromosome segregation errors could contribute to the poorer survival outcomes we observed. RAD21 expression is a prognostic and predictive factor that affects the ultimate outcome in many breast cancer sub-histotypes. Prior to clinical use, these data will require validation, for example, in another (larger) clinical dataset, including assessment in a randomised clinical trial. Once validated, RAD21 expression would have potential translational use in the clinical management of patients. Immunohistochemical evaluation of RAD21 protein on breast cancer specimens could be routinely incorporated into standard pathology reporting to estimate levels of RAD21 that might determine drug response of certain drug regimens. Furthermore, our survival analysis revealed that in patients receiving chemotherapy, those patients with tumors positive for RAD21 expression showed a significantly shorter overall survival than patients whose tumors were negative for RAD21, highlighting an exciting potential role for RAD21 expression in predicting cancer therapy response. We further verified the relevance of RAD21 expression to the therapeutic response in vitro. Stable knockdown of RAD21 significantly enhanced, in a Figure 5 Effect of shRNA-mediated RAD21 knockdown on graded fashion, cellular sensitivity to 5-FU, cyclophospha- cellular sensitivity to anti-cancer drugs. Independent cell clones mide and etoposide. The first two of these drugs are with stably reduced RAD21 expression were derived from the breast components of the commonly used FEC (5-FU/epirubi- cancer cell line, MDA-MB-231. Relative levels of RAD21 gene expression when compared to the parental line by qRT-PCR and cin/cyclophosphamide), CMF (cyclophosphamide/metho- semi-quantitative Western blot analysis were shown in Figure 4. trexate/5-fluorouracil) and AC (doxorubicin (adriamycin)/ Error bar = SEM. Clonogenic survival following treatment with: A. cyclophosphamide) regimens for breast cancer. The cyclophosphamide: sh223_sc1 (P = 0.0316), sh223_sc3 (P = 0.175), repair of DNA adducts caused by 5-FU, cyclophospha- sh224_sc4 (P = 0.0187) and sh224_sc5 (P = 0.0563); B. 5-FU: mide and etoposide depends on HR [24-26] although this sh223_sc1 (P = 0.020), sh224_sc4 (P = 0.0257) and sh224_sc5 (P = 0.0242); and C. etoposide: sh223_sc1 (P = 0.020) and sh224_sc5 (P = dependency on HR is only partial for the repair of cyclo- 0.042). phosphamide-induced DNA interstrand crosslinks [27]. Thedecreaseincellsurvivalthatcorrelatedwithlevelsof RAD21 in breast cancer cells after RAD21 knockdown, is, cancer types: consistent with our data, RAD21 at chro- therefore, in keeping with the dependence of breast can- mosomal locus 8q24 is also commonly amplified in cer cells on the HR pathway to repair DNA damage from advanced androgen-resistant prostate cancer [11]. chemotherapy [10]. Consistent with this proposition, we Our immunohistochemical analysis showed that noted that both topoisomerase II and the RAD21 loading RAD21 expression associates with shorter relapse-free protein NIPBL, showed strong coordinately regulated survival in patients with high grade breast cancer. expression with RAD21 in breast cancer cell lines. Based on known RAD21 functions, the adverse out- RAD21 is recruited, in a manner dependent on the cohe- come in breast cancer patients with RAD21 expression sin loading protein NIPBL, to the sites of DNA double could be due to an elevated level of homologous strand breaks such as those generated by topoisomerase recombination (HR) repair activity as a result of II, promoting DNA repair in human cells [28]. Further, RAD21 overexpression. We favor this explanation. topoisomerase II also decatenates chromosomes before Overexpression of other HR proteins (for example, the condensation needed for RAD21 cohesin-mediated RAD51, BRCA1) has also been shown to be associated chromosome segregation. Collectively, ours and other with increased resistance to radio- and chemo-therapy data highlight the likely effects of the DNA repair func- [22,23]. Furthermore, should RAD21 expression prove tions of RAD21 in tumor biology, and their potential to be a surrogate for HR activity, this may provide a importance in cancer treatment, although other aberrant Xu et al. Breast Cancer Research 2011, 13:R9 Page 9 of 10 http://breast-cancer-research.com/content/13/1/R9

functions of RAD21 could theoretically contribute to our Australian Cancer Research Foundation, the Petre Foundation, the National Health and Medical Research Council (Project grant 400207 to HX; Project observed associations and phenotypes. Grants: 535903 and 535947 to EAKM and RLS) and the RT Hall Trust, In conclusion, the predictive function of RAD21 Australia. SS is supported by grants from the Netherlands Genomics expression for chemotherapy outcome suggests that Initiative (NGI)/NWO.

RAD21 expression could be used to guide treatment Author details selection. For example, in high grade tumors with 1Research Division, Peter MacCallum Cancer Centre, Locked Bag 1, A’Beckett enhanced RAD21 expression, consideration could be Street, Melbourne, Vic 8006, Australia. 2Department of Pathology, Peter MacCallum Cancer Centre, Locked Bag 1, A’Beckett Street, Melbourne, Vic given to using chemotherapeutic drugs that are inhibi- 8006, Australia. 3Department of Pathology, Faculty of Medicine and Dental tors of HR repair [25]. Furthermore, it could guide the Sciences, The University of Melbourne, Elizabeth Street, Parkville, Vic 3000, use of alternative strategies, such as drug dosage intensi- Australia. 4The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, 237 Fulhan Road, London SW3 6JB, UK. 5Department of fication, other chemotherapy drugs with activity in Bioinformatics, Erasmus University Medical Centre, Dr. Molewaterplein 50, breast cancer, or chemotherapy in combination with 3015 GE Rotterdam, The Netherlands. 6Department of Genetics, Erasmus radiotherapy (RT) which is a potent inducer of DNA University Medical Centre, and Cancer Genomics Centre, Dr. Molewaterplein 50, 3015 GE Rotterdam, The Netherlands. 7Department of Radiation double stand breaks. Further testing of the sensitivity of Oncology, Peter MacCallum Cancer Centre, Locked Bag 1, A’Beckett Street, RAD21 knockdown clones to other chemotherapeutic Melbourne, Vic 8006, Australia. 8Cancer Research Program, Garvan Institute of agents may be of utility. Because of its effect on prog- Medical Research, 384 Victoria Street, Darlinghurst, NSW 2010, Australia. 9Department of Anatomical Pathology, South Eastern Area Laboratory nosis and therapeutic outcome, enhanced RAD21 Service, St George Hospital, Gary Street, Kogarah, NSW 2217, Australia. expression may also be a novel therapeutic target. Many 10School of Medical Sciences, University of NSW, High Street, Kensington, of the methodologies used to clinically counteract onco- NSW 2052, Australia. 11School of Medicine, University of Western Sydney, Locked Bag 1797, Penrith, NSW 2751, Australia. 12St Vincent’s Clinical School, gene expression, (for example, gene therapy, antisense University of NSW, Victoria Street, Darlinghurst, NSW 2010, Australia. oligonucleotide therapy, specific microRNA expression), 13Department of Tissue Pathology and Diagnostic Oncology, Royal Prince could be entertained to reduce RAD21 levels. Alfred Hospital, Missenden Road, Camperdown, NSW 2050, Australia. 14Central Clinical School, Sydney Medical School, Edward Ford Building A27, The University of Sydney, NSW 2006, Australia. 15Department of Medical Conclusions Oncology, Westmead Hospital, Darcy Road, Westmead, NSW 2145, Australia. In summary, expression of RAD21 in a significant subset 16Breast Cancer Institute, Westmead Hospital, Darcy Road, Westmead, NSW 2145, Australia. 17Western Clinical School, Sydney Medical School, Edward of breast cancers confers poor prognosis in high grade Ford Building A27, The University of Sydney, NSW 2006, Australia. luminal, basal and HER2 breast cancers, and resistance 18Australian National University and Department of Radiation Oncology, to chemotherapy in breast cancer. RAD21 may be a Canberra Hospital, Yamba Drive, Garran, Australian Capital Territory 2605, novel marker of poor prognosis, a predictive factor for Australia. systemic therapy outcomes and a new target for breast Authors’ contributions cancer therapy. HX contributed to the conception and design of the study, performed experiments, data analysis and interpretation, and drafted and revised the manuscript. MY scored and analyzed immunohistochemistry data, performed Additional material the statistical analysis, and drafted and revised the manuscript. JP participated in antibody validation, immuno-staining and the generation of RAD21 knockdown cells. RN contributed to the collection, analysis and Additional file 1: Summary of breast cancer patient samples used interpretation of CGH and microarray data, and revised the manuscript. YY for the study. A pdf file containing a table that summarizes breast participated in immuno-staining, characterization and survival assays of cancer patient samples used for this study. RAD21 knockdown cells. SS and PS carried out the analysis and Additional file 2: Anti-RAD21 antibody validation. A pdf file showing interpretation of microarray data on cell lines, and revised the manuscript. the validation of the anti-RAD21 antibody using siRNA-mediated JMT participated in the gene expression analysis, and revised the knockdown of the human RAD21 gene in MCF10A cells. manuscript. SV performed qRT-PCR analysis and contributed to data analysis. JSRF contributed to CGH and microarray data analysis and interpretation. EAKM, SAO, CMN and RLS contributed to the study materials, patient data collection, and revised the manuscript. RGR contributed to the conception Abbreviations and design, data analysis and interpretation, and revised the manuscript. 5-FU: 5-fluorouracil; AC: doxorubicin (adriamycin)/cyclophosphamide; CGH: MJM contributed to the conception and design of the study, data analysis comparative genomic hybridization; CMF: cyclophosphamide/methotrexate/ and interpretation, and drafted and revised the manuscript. SF contributed 5-fluorouracil; DAB: 3’:3’-diaminobenzidine; DCIS: ductal carcinoma in situ; to the study conception and design, provision of study materials, collection DMEM: Dulbecco’s modified Eagle medium; EGF: epidermal growth factor; and assembly of data, data analysis and interpretation, and drafted and HR: homologous recombination; HRP: horseradish peroxidase; HRR: revised the manuscript. All authors read and approved the final manuscript. homologous recombinational repair; PC2: Physical Containment level 2; PGK: phosphoglycerate kinase; qRT-PCR: quantitative real time PCR; RFS: relapse- Competing interests free survival; RMA: Robust Multichip Analysis; RT: radiotherapy; SCC: sister The authors declare that they have no competing interests. chromatid cohesion; shRNA: short-hairpin RNA; siRNA: small interference RNA; SNP: single nucleotide polymorphism; TMA: tissue microarray. Received: 16 July 2010 Revised: 12 June 2010 Accepted: 21 January 2011 Published: 21 January 2011 Acknowledgements This study was supported in part by the Victorian Breast Cancer Research References Consortium, the Victorian Cancer Biobank, Cancer Council Victoria (Grant 1. McKay MJ, Troelstra C, van der Spek P, Kanaar R, Smit B, Hagemeijer A, 091224 to HX and MJM), Cancer Institute of New South Wales, the Bootsma D, Hoeijmakers JH: Sequence conservation of the rad21 Xu et al. Breast Cancer Research 2011, 13:R9 Page 10 of 10 http://breast-cancer-research.com/content/13/1/R9

Schizosaccharomyces pombe DNA double-strand break repair gene in 22. Richardson C, Stark JM, Ommundsen M, Jasin M: Rad51 overexpression human and mouse. Genomics 1996, 36:305-315. promotes alternative double-strand break repair pathways and genome 2. Nasmyth K: Disseminating the genome: joining, resolving, and separating instability. Oncogene 2004, 23:546-553. sister chromatids during mitosis and meiosis. Annu Rev Genet 2001, 23. Husain A, He G, Venkatraman ES, Spriggs DR: BRCA1 up-regulation is 35:673-745. associated with repair-mediated resistance to cis- 3. Peters JM, Tedeschi A, Schmitz J: The cohesin complex and its roles in diamminedichloroplatinum(II). Cancer Res 1998, 58:1120-1123. chromosome biology. Genes Dev 2008, 22:3089-3114. 24. Matuo R, Sousa FG, Escargueil AE, Soares DG, Grivicich I, Saffi J, Larsen AK, 4. Gause M, Schaaf CA, Dorsett D: Cohesin and CTCF: cooperating to control Henriques JA: DNA repair pathways involved in repair of lesions induced chromosome conformation? Bioessays 2008, 30:715-718. by 5-fluorouracil and its active metabolite FdUMP. Biochem Pharmacol 5. Birkenbihl RP, Subramani S: Cloning and characterization of rad21 an 2010, 79:147-153. essential gene of Schizosaccharomyces pombe involved in DNA double- 25. Helleday T, Petermann E, Lundin C, Hodgson B, Sharma RA: DNA repair strand-break repair. Nucleic Acids Res 1992, 20:6605-6611. pathways as targets for cancer therapy. Nat Rev Cancer 2008, 8:193-204. 6. Parelho V, Hadjur S, Spivakov M, Leleu M, Sauer S, Gregson HC, Jarmuz A, 26. Treszezamsky AD, Kachnic LA, Feng Z, Zhang J, Tokadjian C, Powell SN: Canzonetta C, Webster Z, Nesterova T, Cobb BS, Yokomori K, Dillon N, BRCA1- and BRCA2-deficient cells are sensitive to etoposide-induced Aragon L, Fisher AG, Merkenschlager M: Cohesins functionally associate DNA double-strand breaks via topoisomerase II. Cancer Res 2007, with CTCF on mammalian chromosome arms. Cell 2008, 132:422-433. 67:7078-7081. 7. Wendt KS, Peters JM: How cohesin and CTCF cooperate in regulating 27. Vasquez KM: Targeting and processing of site-specific DNA interstrand gene expression. Chromosome Res 2009, 17:201-214. crosslinks. Environ Mol Mutagen 2010, 51:527-539. 8. Nasmyth K, Haering CH: Cohesin: Its Roles and Mechanisms. Annu Rev 28. Bauerschmidt C, Arrichiello C, Burdak-Rothkamm S, Woodcock M, Hill MA, Genet 2009, 43:525-558. Stevens DL, Rothkamm K: Cohesin promotes the repair of ionizing 9. Rhodes DR, Yu J, Shanker K, Deshpande N, Varambally R, Ghosh D, radiation-induced DNA double-strand breaks in replicated chromatin. Barrette T, Pandey A, Chinnaiyan AM: Large-scale meta-analysis of cancer Nucleic Acids Res 2010, 38:477-487. microarray data identifies common transcriptional profiles of neoplastic transformation and progression. Proc Natl Acad Sci USA 2004, doi:10.1186/bcr2814 101:9309-9314. Cite this article as: Xu et al.: Enhanced RAD21 cohesin expression 10. Atienza JM, Roth RB, Rosette C, Smylie KJ, Kammerer S, Rehbock J, confers poor prognosis and resistance to chemotherapy in high grade Ekblom J, Denissenko MF: Suppression of RAD21 gene expression luminal, basal and HER2 breast cancers. Breast Cancer Research 2011 13: decreases cell growth and enhances cytotoxicity of etoposide and R9. bleomycin in human breast cancer cells. Mol Cancer Ther 2005, 4:361-368. 11. Porkka KP, Tammela TL, Vessella RL, Visakorpi T: RAD21 and KIAA0196 at 8q24 are amplified and overexpressed in prostate cancer. Genes Chromosomes Cancer 2004, 39:1-10. 12. van’t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M, Peterse HL, van der Kooy K, Marton MJ, Witteveen AT, Schreiber GJ, Kerkhoven RM, Roberts C, Linsley PS, Bernards R, Friend SH: Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002, 415:530-536. 13. McShane LM, Altman DG, Sauerbrei W, Taube SE, Gion M, Clark GM: Reporting recommendations for tumor marker prognostic studies (REMARK). J Natl Cancer Inst 2005, 97:1180-1184. 14. Yan M, Rayoo M, Takano EA, Fox SB: BRCA1 tumours correlate with a HIF- 1alpha phenotype and have a poor prognosis through modulation of hydroxylase enzyme profile expression. Br J Cancer 2009, 101:1168-1174. 15. Nielsen TO, Hsu FD, Jensen K, Cheang M, Karaca G, Hu Z, Hernandez- Boussard T, Livasy C, Cowan D, Dressler L, Akslen LA, Ragaz J, Gown AM, Gilks CB, van de Rijn M, Perou CM: Immunohistochemical and clinical characterization of the basal-like subtype of invasive breast carcinoma. Clin Cancer Res 2004, 10:5367-5374. 16. Livak KJ, Schmittgen TD: Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods 2001, 25:402-408. 17. Xu H, Beasley M, Verschoor S, Inselman A, Handel MA, McKay MJ: A new role for the mitotic RAD21/SCC1 cohesin in meiotic chromosome cohesion and segregation in the mouse. EMBO Rep 2004, 5:378-384. 18. Natrajan R, Weigelt B, Mackay A, Geyer FC, Grigoriadis A, Tan DS, Jones C, Lord CJ, Vatcheva R, Rodriguez-Pinilla SM, Palacios J, Ashworth A, Reis- Filho JS: An integrative genomic and transcriptomic analysis reveals molecular pathways and networks regulated by copy number aberrations in basal-like, HER2 and luminal cancers. Breast Cancer Res Treat 2010, 121:575-89. Submit your next manuscript to BioMed Central 19. Hollestelle A, Nagel JH, Smid M, Lam S, Elstrodt F, Wasielewski M, Ng SS, and take full advantage of: French PJ, Peeters JK, Rozendaal MJ, Riaz M, Koopman DG, Ten Hagen TL, de Leeuw BH, Zwarthoff EC, Teunisse A, van der Spek PJ, Klijn JG, Dinjens WN, Ethier SP, Clevers H, Jochemsen AG, den Bakker MA, • Convenient online submission Foekens JA, Martens JW, Schutte M: Distinct gene mutation profiles • Thorough peer review among luminal-type and basal-type breast cancer cell lines. Breast Cancer • No space constraints or color ﬁgure charges Res Treat 2010, 121:53-64. 20. GEO. [http://www.ncbi.nlm.nih.gov/geo/]. • Immediate publication on acceptance 21. Franken NA, Rodermond HM, Stap J, Haveman J, van Bree C: Clonogenic • Inclusion in PubMed, CAS, Scopus and Google Scholar assay of cells in vitro. Nat Protoc 2006, 1:2315-2319. • Research which is freely available for redistribution

Submit your manuscript at www.biomedcentral.com/submit Pathology International 2007; 57: 183–189 doi:10.1111/j.1440-1827.2007.02079.x

Original Article p27KIP-1 cyclin A and cyclin D1 protein expression in ductal carcinoma in situ of the breast: p27KIP-1 correlates with hormone receptor status but not with local recurrence

Ewan K. A. Millar,1,3 Kayla Tran,1 Penny Marr1 and Peter H. Graham2,3 1Department of Anatomical Pathology, South-Eastern Area Laboratory Service and 2Department of Radiation Oncology, Cancer Care Center, St George Hospital, Kogarah and 3University of New South Wales, New South Wales, Australia

Using whole sections of formalin-fixed paraffin-embedded heterogeneous nature of DCIS makes its biological behavior material the expression of p27KIP-1, cyclin A and cyclin D1 ill-defined, and current predictions are based on the appli- was examined in 60 cases of ductal carcinoma in situ (DCIS) cation of histological grading criteria, with nuclear features using routine immunohistochemistry with a median follow currently taking precedence. The reduction in recurrence up of 95 months (range 10–139 months) to identify any rates with local surgical excision and adjuvant radiotherapy association with disease recurrence. Fifty-six patients were treated by local excision and radiotherapy and four by described in the National Surgical Adjuvant Breast Project mastectomy without radiotherapy. There was a highly (NSABP)-17 trial2 has fuelled the trend towards conserva- significant positive association between p27KIP-1 and estro- tive treatment and away from mastectomy over the past gen receptor/progesterone receptor (ER/PR) status (P = decade. Importantly, however, recurrences after conserva- KIP-1 0.002, P = 0.02) and with p27 and cyclin D1 expression tive therapy are associated with invasive disease in up to = (P 0.002). A trend between cyclin A and PR status 50% of cases. Subsequently the NSABP-26 study has (P = 0.08) was also identified. These findings mirror those shown that the addition of the anti-estrogen Tamoxifen described in invasive ductal carcinoma, but there were no associations of any biomarker with histological parameters reduced the rate of non-invasive and invasive recurrences 3 such as nuclear grade or with local recurrence on univariate further, even in the presence of positive margins. There has analysis, which was present in four of the 56 locally excised been much interest in identifying clinically useful tissue- group (7.1%). Further examination of a larger cohort may be based markers, which may provide further predictive infor- worthwhile to explore the possible role as adjunctive pre- mation to help guide therapy. Such biomarker expression dictive markers to aid clinical decision making. can also provide insight into the pathogenesis of the disease. Cell cycle proteins are of interest because they lie Key words: breast cancer, cell cycle, DCIS, hormone receptors, at the center of the control of proliferation and response to prognosis hormonal stimulation. In vitro studies in human breast cancer cell lines have shown that both cyclin A and cyclin Ductal carcinoma in situ (DCIS) is a pre-invasive malig- D1 are induced by estrogens,4 while p27 is downregulated.5 nancy that has a risk of progression to invasive carcinoma p27KIP-1 and cyclin A, however, can both be induced by and its incidence has increased largely through earlier Tamoxifen.5,6 During the cell cycle, progression through G1 detection via breast screening programs, with an incidence and entry into the S phase is dependent largely on the of approximately 20% of all screen-detected cancers.1 The interaction between cyclin D1 with cyclin-dependent kinases (cdk) 4 and 6. These activated complexes along with cyclin Correspondence: Ewan Millar, MB, ChB, FRCPath, FRCPA, Depart- E/cdk2 phosphorylate members of the pocket protein family ment of Anatomical Pathology, South-eastern Area Laboratory including pRb, which lead to reduced inhibition and Service, St George Hospital, Gray Street, Kogarah, NSW 2217, increased levels of the transcription factor E2F, with result- Australia. Email: [email protected] ant entry into S phase and subsequent DNA synthesis. Received 19 September 2006. Accepted for publication 21 November 2006. The transition from S phase into G2 is driven by cyclin KIP-1 © 2007 The Authors A–cdk 2 complexes. The inhibitory protein p27 , which Journal compilation © 2007 Japanese Society of Pathology functions as a tumor suppressor, leads to G1 arrest and 184 E. K. A. Millar et al.

Table 1 Baseline clinical and treatment details for 60 DCIS patients Lumpectomy Mastectomy All n = 56 n = 4 n = 60 n (%) n (%) n (%) Age (years); mean (range) 58 (33–81) 53 (42–63) 58 (33–81) Follow up (months); mean (range) 95 (10–139) 111 (87–137) 95 (10–139) Median 98 110 98 Size (mm); mean (range) 18.9 (2.5–60) 75 (50–130) 23 (2.5–130) Margin (mm); mean (range) 5.5 (0–18) 4 (0–11) 5.4 (0–18) Margin type Focal <2 mm 3 (5) 2 (50) 5 (8) Clear or clear re-excision 31 (55) 2 (50) 33 (55) Negative re-excision 22 (39) 0 22 (37) Surgical excisions 1 13 (23) 0 13 (22) 2–3 43 (77) 4 (100) 47 (88) XRT Nil 5 (9) 4 (100) 9 (15) No boost 13 (23) 0 13 (22) Boost 38 (68) 0 38 (63) XRT, radiotherapy.

ubiquitin-mediated degradation via binding to cdk2–cyclin and three had focally involved margins. In the mastectomy E and also to cdk2–cyclin A. Alterations in expression of group two patients had clear margins and two had focally p27KIP-1, cyclin A and cyclin D1 have been described in asso- involved margins. Of the 56 lumpectomy patients five did not ciation with estrogen receptor (ER) and progesterone recep- receive radiotherapy, 13 had only whole breast radiotherapy tor (PR) status with some prognostic significance, mostly in of 4–50 Gy and 38 also received a boost to the tumor bed of invasive carcinoma with few studies in DCIS. We examined 10–16 Gy for a total dose of 60 Gy. No patients received expression of these proteins and correlated their expression Tamoxifen. Van Nuys prognostic index (VNPI) was generated with histological features, hormonal status and clinical out- as previously published.6 One case with unknown size was comes in a series of 60 cases of DCIS. classified as VNPI score 2.

Pathological details MATERIALS AND METHODS

A total of 60 cases were retrieved from the archives of the Clinical details Department of Anatomical Pathology, South-eastern Area Laboratory Services, St George Hospital, Kogarah. Cases Cases of DCIS were identiﬁed from the database of the were reviewed by two pathologists and assessed for nuclear Department of Radiation Oncology, Cancer Care Center at St grade, pattern, presence or absence of necrosis. Nuclear George Hospital, Kogarah, NSW from 1993 to 1999. Ethical grade was assigned using standardized published criteria.7 approval for the study was granted by South-Eastern Area Health Service Human Research Ethics Committee. Clinical details are summarized in Table 1. Only two patients in complete remission were unavailable for at least 5 years follow Immunohistochemistry up. One died at 10 months of an unrelated cause and one was lost to follow up at 53 months. Follow up of cases without One representative section from each case was cut at 4 mm failure ranged from 10 to 139 months with a median of and mounted onto SuperFrost Plus slides (Menzel-Glaser, 98 months. The age range was 33–81 years with a mean of Brauschweig, Germany). Antigens were retrieved by placing 56 years. A total of 56 patients had lumpectomy with or sections in TRIS/HCl buffer at pH 9.0 and heating at full 1 without radiotherapy and four had mastectomy without radio- pressure for 2 /2 min in a microwave pressure cooker. Anti- therapy. The size of DCIS ranged from 2.5 to 60 mm with an gens were detected using the following dilutions of mono- average of 18.9 mm. In the lumpectomy group 53 patients clonal antibodies: cyclin D1 (clone P2E11F11, Novocastra had clear margins (deﬁned as no DCIS at the inked margin) Laboratories, Newcastle Upon Tyne, UK) at 1:50; cyclin A

(clone 6E6, Novocastra Laboratories) at 1:100; and p27 Table 2 Cyclin D1 versus grade, ER, PR, necrosis and pattern of in (clone 1B4, Novocastra Laboratories) at 1:40; ER (clone situ carcinoma 1D5, Immunon, Pittsburgh, PA, USA; and clone 6F11, Novo- Cyclin D1 castra Laboratories) used combined antibodies at dilutions of – + 1:50 and 1:100, respectively; PR (clone 16, Novocastra n % n % P Laboratories), 1:150. Primary antibodies were incubated at Grade 1467233 room temperature for 60 min and detected using horseradish 2 8 36 14 64 peroxidase (HRP)-labeled streptavidin–biotin (BA-1300 and 3 13411959 SA-5004; Vector Laboratories, Burlingame, CA, USA) for all 25 42 35 58 0.40 antibodies other than ER and PR, for which Envision+ (Dako, Grade 1and212431657 Glostrup, Denmark) was used, then visualized using 3 13421959 3,3,diaminobenzidine (D5237; Sigma, St Louis, MO, USA). 25 42 35 58 0.86 Positive controls used were invasive breast carcinoma for ER cyclin D1, ER and PR; and tonsil for cyclin A and p27KIP-1. – 9 36 16 64 + 16 46 19 54 25 42 35 58 0.45 Scoring PR – 11391761 Two pathologists assessed all immunostains independently. + 14 44 18 56 25 42 35 58 0.73 Cases with discrepant results (>10% variability) were Necrosis reviewed on a double-headed microscope and a consensus – 11391761 score obtained. Only nuclear staining was assessed as posi- + 14 44 18 56 tive for all antibodies. Most cases had variability in the inten- 25 42 35 58 0.73 Pattern sity of staining within even single ducts, which ranged from Usual 20 37 34 63 mild to strong. The percentage of positive nuclei that had any Special type 5 83 1 17 0.29 deﬁnite positive nuclear staining was estimated as a percent- ER, estrogen receptor; PR, progesterone receptor. age of the total number of epithelial cells present in ducts involved by DCIS on each slide. Cases were assessed as positive if there was >5% staining for cyclin D1, as previously grade. A total of 54 cases had the usual solid or cribriform published,8 and >10% for cyclin A, ER and PR. p27KIP-1 was pattern, three cases had features of neuroendocrine/solid considered as high or low expression with an arbitrary cut-off papillary DCIS and three were intracystic papillary. Necrosis of greater or less than 50% staining. The cut-offs were deter- was present in 53% of cases. ER and PR negativity both mined before assessment of the slides and were used fol- correlated with nuclear grade 3 (P < 0.001). lowing review of previously published studies, to maintain comparability with our ﬁndings.

Local recurrence Statistical methods

Five local failures occurred at 31, 37, 39 42 and 86 months, Statistical analyses were performed using STATv9 (Stata, none of which was as invasive carcinoma. Two of these College Station, Texas, USA). Tabulations used the c2 test or, patients with local failure had not received radiotherapy, one if appropriate, Fisher’s exact test. Kaplan–Meier survival mastectomy with a focally involved margin and one lumpec- curves were compared using the log–rank test. Univariate tomy with clear margins. Four deaths from unrelated causes analysis of variables tested for survival used the Cox pro- occurred at 10, 61, 105 and 110 months, all in complete portional hazards model. Multivariate analysis was not per- remission (CR). Two patients were lost to follow up at 53 and formed. Because only six cases were grade 1, these were 87 months, both in CR. combined with grade 2 cases for analysis.

RESULTS Correlation of cyclin D1, cyclin A and p27KIP-1 with Histological grade, pattern of DCIS and hormonal histological features and hormone receptor status status p27KIP-1 high status was present in 33 cases (55%) and 27 Of the total of 60 cases, 10% (six) were low nuclear grade, (45%) had low levels of expression (Tables 2–4). p27KIP-1 37% (22) were intermediate grade and 53% (32) were high status had a statistically signiﬁcant association with ER

Table 3 Cyclin A versus grade, ER, PR, necrosis and pattern of Table 4 p27 versus grade, ER, PR, necrosis and pattern of in situ in situ carcinoma carcinoma Cyclin A p27 – + Hi Low n % n % P n % n % P Grade Grade 1 6 100 0 0 1350350 21464836 21359941 3 19591341 3 17531547 39 65 21 35 0.16 33 55 27 45 0.88 Grade Grade 1 and 2 20 71 8 29 1and216571243 3 19591341 3 17531547 39 65 21 35 0.33 33 55 27 45 0.76 ER ER – 15601040 – 8 32 17 68 + 24 69 11 31 + 25 71 10 29 39 65 21 45 0.49 33 55 27 45 0.002 PR PR – 15541346 – 11391768 + 24 75 8 25 + 22 69 10 29 39 65 21 35 0.08 33 55 27 45 0.02 Necrosis Necrosis –982218–545655 + 30 61 19 39 + 28 57 21 43 39 65 21 35 0.20 33 55 27 45 0.48 Pattern Pattern Usual 36 67 18 33 Usual 31 57 23 43 Special type 3 50 3 50 0.42 Special type 2 33 4 67 0.26 There was an association between cyclin A and PR (P = 0.08) that did Bold, statistically signiﬁcant. not reach statistical signiﬁcance. ER, estrogen receptor; PR, progesterone receptor. ER, estrogen receptor; PR, progesterone receptor.

Correlation of cyclin D1, cyclin A and p27KIP-1 with local recurrence (P = 0.002) and PR status (P = 0.02). Twenty-five of 35 ER+ cases (71%) were p27KIP-1 high and 17 of 25 ER– cases (68%) No significant correlation was found for local recurrence on were p27 low. Twenty-two of 32 PR+ cases had high p27KIP-1 univariate analysis with any clinicopathological parameter, expression and 17 of 25 (68%) of PR– cases were p27KIP-1 cyclin D1, cyclin A2, p27KIP-1 or hormone receptor status low. There was no association with any histological features (Table 5; Fig. 1). such as nuclear grade. Cyclin A positivity was present in 21 cases (35%) and had a trend with PR status, although not statistically significant: 24 of 32 PR+ cases (75%) were cyclin A negative (P = 0.08). DISCUSSION There was no association with ER (P = 0.49) or with histological features. The current trend towards conservative local treatment for Cyclin D1 positivity was present in 35 cases (58%) but had DCIS, which is a pathologically heterogeneous disease, no significant correlation with any histological features or with creates some uncertainty as to its likely success or failure. ER (P = 0.45) or PR (P = 0.73) status. Current predictors are based on nuclear grade, size of lesion, margin status and presence or absence of necrosis, which is often combined to create an index of risk of local failure such as the VNPI. However, these parameters do not always Expression profile of cyclin A, cyclin D1 and p27KIP-1 predict individual outcomes well. Local excision with radiotherapy is an effective treatment and the addition of anti- Low p27KIP-1 expression was associated with low cyclin D1 estrogens further reduces the risk of recurrence. Previous and high p27KIP-1 was associated with high cyclin D1 studies have found an association of grade of DCIS with ER, (P = 0.002). There was no association of p27KIP-1 expression PR and also Her2 status, but there has been no consistent with cyclin A positivity or between cyclin A and cyclin D1 association with outcomes. Correlation of p27KIP-1, cyclin A positivity and cyclin D1 with ER and PR status would be of interest to

Table 5 Univariate local failure rates at 10 years n %10 year control 95%CI P Test All 60 91 71–96 Surgery Lumpectomy 56 92 80–97 Total mastectomy 4 75 13–96 0.26 Log–rank Radiotherapy Nil 9 75 31–93 Adjuvant XRT 51 93 74–97 0.079 Log–rank DCIS size <10mm 20 100 NA 310mm 40 86 69–94 0.096 Log–rank VNPI 1 5 100 NA 2 46 90 75–96 3 9 89 43–98 0.76 Log–rank Grade 1 6 80 20–97 2 22 89 62–97 3 32 94 77–98 0.54 Log–rank Grade dichotomous 1 and 2 28 87 64–96 3 32 94 77–98 0.47 Log–rank Necrosis No 11 90 47–99 Yes 49 91 78–97 0.84 Log–rank ER Negative 25 88 67–96 Positive 35 93 74–98 0.42 Log–rank PR Negative 28 93 74–98 Positive 32 89 68–96 0.68 Log–rank Cyclin D1 Negative 25 92 71–98 Positive 35 90 73–97 0.99 Log–rank Cyclin A Negative 39 88 71–96 Positive 21 95 71–99 0.44 Log–rank p27 Low 27 87 64–96 High 33 94 77–98 0.48 Log–rank HR Age 0.92 0.84–1.00 0.078 Cox Margin (mm) 0.85 0.65–1.10 0.225 Cox No significant correlation was identified with any variable for local recurrence. CI, confidence interval; DCIS, ductal carcinoma in situ; ER, estrogen receptor; PR, progesterone receptor; VNPI, Van Nuys prognostic index; XRT, radiotherapy. identify hormone-associated biomarkers that may aid clinical carcinoma, suggesting early dysregulation in the develop- decision making. We chose to study the expression of two ment of breast cancer.9 Low p27KIP-1 in invasive carcinoma key cyclins that have pro-proliferative activity, are induced by has been associated with poor prognosis and steroid inde- estrogen and expressed in different stages of the cell cycle: pendence and ER negativity in some studies,10 while high cyclin D1 plays a critical role in the G1/S phase transition and levels are associated with ER positivity.11In vitro work has cyclin A, in the S/G2 transition. p27KIP-1 is an important cell shown downregulation of p27KIP-1 by estrogen and its induc- cycle inhibitor and somatic mutations and loss of heterozy- tion by anti-estrogens in breast cancer cell lines, highlighting gosity are rare. Loss of p27KIP-1 expression has therefore its importance in the inhibition of proliferation.5 While p27 has been assumed to be due to post-translational modification inhibitory effects on cdk2–cyclin A and cdk-2–cyclin E com- and increased protein degradation largely due to S phase plexes, it also appears to activate cyclin D1–cdk4 complexes, kinase-associated protein 2 (skp-2). p27KIP-1 expression is being required for their assembly.12 When p27KIP-1 binds to high in normal epithelium and reduces in a progressive cyclin D1–cdk4 complexes it becomes sequestered, allowing manner from epithelial hyperplasia to DCIS and invasive cyclin E–cdk2-driven progression through G1 and into S

(a) (b) studies.17 High p27 status in invasive carcinoma similarly also correlates with ER positivity.11 Low p27KIP-1 status has

1.00 also been suggested as a possible predictor of resistance to anti-estrogen therapy in ER-positive invasive carcinoma.18 Low p27KIP-1 status also has some predictive value in early 0.80 node-negative invasive carcinoma, where it correlates with Her2 status and response to chemotherapy.19 As described in

0.60 previous studies,14 we also demonstrated a strong correlation (c) (d) between p27KIP-1 and cyclin D1 expression (P = 0.002). Cyclin A status in DCIS is not well described and only one

1.00 study has been identiﬁed, although in invasive carcinoma it has been associated with an adverse prognosis.20,21 Cyclin A

0.80 positivity has been described in association with ER/PR negativity with poor metastasis-free survival.22 Similarly we 0.60 0 24 48 72 96 120 0 24 48 72 96 120 found an association between cyclin A positivity and PR Time (months) negativity (P = 0.08), which implied a link between these two markers although not at a statistically significant level, but it is Figure 1 Kaplan–Meier local control estimates. (a) All patients, n = 60; (b) cyclin D1: (––) negative, n = 25; (- - -) positive, n = 35; (c) of interest and merits further investigation. Only one other cyclin A: (––) negative, n = 39; (- - -) positive, n = 21; (d) p27 expres- study has examined cyclin A expression in DCIS. That study sion: (––) high, n = 33; (- - -) low, n = 27. found a significant association with local recurrence if the combined cyclin A and Ki-67 scores (‘global proliferation frac- phase.13 The relative expression of these pro- and anti- tion’) were between the 25th and 75th percentiles.23 These proliferative proteins may then provide insight into the rela- investigators did not assess the presence of an association tionship of these proteins to outcome, which may be of between cyclin A and ER/PR. That study showed an asso- clinical benefit. ciation of comedo DCIS and cyclin A positivity but not for The study group consisted of 60 DCIS patients treated in a non-comedo types. We were unable to find any association single center predominantly with local excision (n = 56) and with nuclear grade. radiotherapy (n = 51) and only a small group with mastec- Cyclin D1 is overexpressed in atypical ductal hyperplasia, tomy (n = 4). None received Tamoxifen. Local failures in the DCIS and invasive carcinoma in a step-wise manner com- conservatively managed group were very few, with only four pared to normal epithelium, and its role in progression is well of 56 patients having an event (7.1%). There was no statis- established.8 It is overexpressed in 40% of DCIS but is ampli- tically significant correlation with clinical outcomes for any of fied in only 15% of cases,24 and its expression can be the markers examined, which reflects the success of treat- induced by estrogen. Several studies have found an asso- ment and the small size of the cohort. This low rate of local ciation between cyclin D1 and ER14,24 but no association with failure is consistent with that described in most large series. grade of DCIS.8,14 One recent study in DCIS found that low The main findings of the current study are the presence of cyclin D1 expression (defined as <40% positivity) was asso- significant positive associations between p27KIP-1 and ER/PR ciated with local recurrence if treated only by local excision status and a trend between cyclin A and PR status, which without radiotherapy.15 The majority of the present patients mirror those described in invasive ductal carcinoma: high received radiotherapy and therefore the present findings are p27KIP-1 expression is significantly associated with ER and not strictly comparable. We were unable to demonstrate any PR positivity and low p27KIP-1 with ER and PR negativity significant association of cyclin D1 with hormonal status, (P = 0.002, P = 0.02, respectively). Some studies have pre- nuclear grade or outcome. One study found an association viously described a correlation of high p27KIP-1 with ER posi- between p27KIP-1, cyclin D1 and ER expression14 but we were tivity in DCIS14 but not others.15 Likewise the association of unable to confirm these findings in the present study. high p27KIP-1 with PR positivity is seen in some but not all studies. We did not find any association between p27KIP-1 and nuclear grade, similarly to other studies,14 although one study CONCLUSION did find an inverse correlation, which may reflect differing scoring methods used.16 No relationship of p27KIP-1 status as In this preliminary study group of conservatively treated DCIS a potential predictive marker for anti-estrogens has been patients we have demonstrated a significant association described in DCIS, but high p27KIP-1 has been found to be of between p27KIP-1, ER and PR status and a trend between predictive value in invasive carcinoma treated with tamoxifen cyclin A expression and PR status, which is similar to those and goserelin, with improved disease-free survival in some described in invasive ductal carcinoma. These markers were

© 2007 The Authors Journal compilation © 2007 Japanese Society of Pathology Cell cycle markers in DCIS of the breast 189 not associated with local recurrence, but because of the 14 Oh YL, Choi JS, Song S-Y et al. Expression of p21, p27 and small cohort and few events, this may simply be a function of cyclin D1 proteins in breast ductal carcinoma in-situ: Relation with clinicopathological characteristics and with p53 expression inadequate power. Further investigation in a larger series of and oestrogen receptor status. Pathol Int 2001; 51: 94–9. cases may be more informative. 15 Jirstrom K, Ringberg A, Ferno M, Anagnostaki L, Landberg G. Tissue microarray analyses of G1/S regulatory proteins in ductal carcinoma in-situ of the breast indicate that low cyclin D1 is REFERENCES associated with local recurrence. Br J Cancer 2003; 89: 1920– 26. 1 Leonard GD, Swain SM. Ductal carcinoma in-situ, complexities 16 Leong ASY, Sormunen RT, Vinyvuvat S, Hamdani RW, and challenges. J Nat Cancer Inst 2004; 96: 906–20. Suthipintawong C. Biological markers in ductal carcinoma 2 Fisher ER, Dignam J, Tan-Ciu E et al. Pathologic findings from in-situ and concurrent infiltrating carcinoma. A comparison of the National Surgical Adjuvant Breast Project (NSABP) eight- eight contemporary grading systems. Am J Pathol 2001; 115: year update of protocol B-17. Cancer 1999; 86: 429–38. 709–18. 3 Fisher B, Dignam J, Wolmark N et al. Tamoxifen in treatment of 17 Pohl G, Rudas M, Dietze O et al. High p27 kip1 expression intraductal breast cancer: National surgical adjuvant breast and predicts superior relapse free and overall survival for premeno- bowel project B-24 randomised control trial. Lancet 1999; 353: pausal women with early stage breast cancer receiving adjuvant 1993–2000. treatment with tamoxifen plus goserelin. J Clin Oncol 2003; 21: 4 Hodges LC, Cook JD, Lobenhofer EK et al. Tamoxifen functions 3594–600. as a molecular agonist inducing cell cycle associated genes in 18 Arteaga CL. CDK inhibitor p27kip-1 and hormone dependence in breast cancer cells. Mol Cancer Res 2003; 1: 300–11. breast cancer. Clin Cancer Res 2004; 10 (Suppl.): 368–71. 5 Cariou S, Donovan JC, Flanagan WM, Milic A, Bhattacharya N, 19 Spataro VJ, Litman H, Viale G et al. Decreased immunoreactiv- Slingerland JM. Down regulation of p21waf-1/cip-1 or p27kip-1 abro- ity for p27 protein in patients with early stage breast carcinoma gates anti-oestrogen mediated cell cycle arrest in human breast is correlated with Her-2/neu over-expression and with benefit cancer cell lines. Proc Natl Acad Sci USA 2000; 97: 9042–6. from one course of perioperative chemotherapy in patients with 6 Silverstein MJ, Lagios MD, Craig PH et al. A prognostic index for negative lymph node status. Results from the International ductal carcinoma in situ. Cancer 1996; 77: 2264–74. Breast Cancer Study Group Trial V. Cancer 2003; 97: 1591– 7 Schwartz GF. Consensus conference on the classification of 600. ductal carcinoma in-situ. Hum Pathol 1997; 80: 1798–802. 20 Michalides R, van Tinteren H, Balkenende A. Cyclin A is a 8 Alle KM, Henshall SM, Field AS, Sutherland RL. Cyclin D1 prognostic indicator in early stage breast cancer with and protein is overexpressed in hyperplasia and intraductal carci- without tamoxifen treatment. Br J Cancer 2002; 86: 402–8. noma of the breast. Clin Cancer Res 1998; 4: 847–54. 21 Poikonen P, Sjostrom J, Amini RM, Villman K, Ahlgren J, Blo- 9 De Paolo F, Vecci AM, Granoto AM et al. p27kip-1 expression qvist C. Cyclin A as a marker for prognosis and chemotherapy in normal epithelium, benign and neoplastic breast lesions. response in advanced breast cancer. Br J Cancer 2005; 93: J Pathol 2002; 196: 26–31. 515–19. 10 Catzavelos C, Bhattacharya N, Ung YC et al. Decreased levels 22 Aaltonen A, Ahlin C, Amini RM et al. Reliability of cyclin A of the cell cycle inhibitor p27kip1 protein: Prognostic implica- assessment on tissue microarrays in breast cancer compared to tions in primary breast cancer. Nat Med 1997; 3: 227–30. conventional histological slides. Br J Cancer 2006; 94: 1697– 11 Barnes A, Pinder SE, Bell JA et al. Expression of p27kip1 in 702. breast cancer and its prognostic significance. J Pathol 2003; 23 Chasle J, Delozier T, Denoux Y, Marnay J, Michels JJ. Immu- 201: 451–9. nohistochemical study of cell cycle regulatory proteins in intra- 12 Chiarle R, Pagano M, Inghirami G. The cyclin dependent kinase ductal breast carcinomas: A preliminary study. Eur J Cancer inhibitor p27 and its prognostic role in breast cancer. Br Cancer 2003; 39: 1363–9. Res 2000; 3: 91–4. 24 Vos CB, Ter Harr NT, Peterse JL, Cornelisse CJ, Van de Vijver 13 Sherr CJ, Roberts JM. CDK inhibitors: Positive and negative MJ. Cyclin D1 gene amplification and overexpression are regulators of G1 phase progression. Genes Dev 1999; 13: present in ductal carcinoma in-situ of the breast. J Pathol 1999; 1501–12. 187: 279–84.

Significance and Assessment of Margin Status in Ductal Carcinoma In Situ of the Breast

Ewan K. A. Millar and Anthony S.-Y. Leong

Division of Anatomical Pathology, Hunter Area Pathology Service, John Hunter Hospital, and Discipline of Anatomical Pathology, University of Newcastle, Newcastle, New SouthWales, Australia

Summary: This article reviews practical and theoretical aspects of margin assessment for surgically excised ductal carcinoma in situ of the breast. Different methods of assessing surgical margins are discussed, including selected tangential sections, the margin shaving, and cavity peel methods. Criteria for margin adequacy and the relevance of margin status in the selection of cases for breast conservation are discussed in the context of other important risk factors for local failure. Key Words: Breast neoplasms—Ductal carcinoma in situ—Breast-conserving therapy—Margins—Local recurrence—Multifocality—Tumor size

Pathologic assessment of the surgical margins of ex- cerning margin status has probably arisen because of the cision has long been a key feature of surgical pathology absence of clear consensus as to what defines an ad- reporting and of central importance in achieving local equate margin of clearance. This information needs to be control in excision biopsies, particularly for ductal car- considered in the context of several other discriminators, cinoma in situ (DCIS) of the breast. Clear margins are a including lesion size, grade, and patient age, before suit- prerequisite for breast-conserving therapy (BCT) and im- ability for BCT can be gauged for each individual. Re- proved cosmesis. As the surgical trend toward this goal currence of DCIS after BCT ranges from 6–19% (9) to has increased in recent years, the onus on the surgical 30% (10) and brings with it a 50% risk of invasive car- pathologist has increased such that adequate and accurate cinoma (10). sampling of the margins of excision of every specimen is As the workload in breast pathology continues to rise, mandatory. Although mastectomy is effectively a cure largely through increasing numbers of cases identified for DCIS, for many women this represents overtreatment via breast screening programs, but also by the require- and there is now greater emphasis on the selection of ment for more refined pathologic reports and minimum appropriate cases for BCT. In the search for the selection datasets that necessitate more extensive sampling, the criteria for such patients, a number of pathologic factors pathologist should be aware of the relevance of the in- have been identified as useful in the prediction of likely formation he or she is providing. success or failure. Close or involved margins have been New information continues to become available on shown to be one such predictor and an independent risk this topic and this review aims to encompass contempo- factor for recurrence by several studies (1–4) but not by rary opinions in surgical pathology, while incorporating others (5–8). This seemingly conflicting position con- the more conventionally held views regarding margin assessment. The discussion will focus on several areas concerning practical aspects of margin assessment, what Address correspondence and reprint requests to Anthony S.-Y. an adequate margin is, the impact of a positive margin, Leong, M.D., Medical Director, Hunter Area Pathology Service, Locked Bag 1, Newcastle Regional Mail Center, Newcastle, NSW and its overall importance compared with other risk fac- 2310, Australia. tors for local failure.

338 MARGIN STATUS IN DUCTAL CARCINOMA IN SITU OF BREAST 339

SPECIMEN HANDLING AND TECHNIQUES the histologic section and the specimen radiograph. If FOR MARGIN ASSESSMENT necessary, superimposition of the tissue section on the radiograph provides excellent microscopic correlation. Typically, assessment of the whole excision specimen Radiologic assessment of margins on its own, not sur- assumes orientation by the surgeon, usually by means of prisingly, is totally unreliable. This has been highlighted three orientating sutures or clips, sometimes containing a by two studies: one showed that 67% of cases that ap- small, superficial ellipse of skin. On receipt of the fresh peared clear on radiograph were positive for DCIS on or fixed specimen, the surgical margins are painted using histology (12). Another study showed a false-negative one of any number of techniques involving colored com- rate of 44% using radiologic assessment of margins (13). mercial dyes, Alcian Blue, or India ink, each of which is Should specimen radiography not be available, a hand- non–radio-opaque and easily identifiable on routine he- drawn diagram of the specimen slices can be made to matoxylin and eosin–stained sections. There can be pen- mark the position of suspicious areas in relation to the etration of the dye into the specimen at this time, if used margins. Blocks may also be selected macroscopically to excess, which may cause problems of interpretation. and subsequently radiographed to confirm the presence Care should be taken to blot off excess dye. Examination of calcification within the selected areas. This method is and sectioning of the fresh tissue specimen can be diffi- probably not as satisfactory as sliced specimen radiog- cult owing to softness and plasticity of the predominantly raphy for correlation with mammographic appearances. fatty tissue. This can be overcome by slicing the speci- The assessment process for the breast specimen is men after approximately 20 minutes in a conventional commonplace in current surgical pathology and most freezer. Cooling the tissue in this way allows it to firm up residents and pathologists soon become proficient in the significantly: serial 3- to 4-mm whole-face slices can be orientation, dissection, and block selection of routine cut from the tissue using this technique, which is in rou- palpable and mammographically detected excision bi- tine use in our own laboratory. Other authors have used opsy specimens. The sampling of margins involves se- 3 to 4 hours of immersion in 10% formal saline before lection of blocks taken perpendicular to the inked surgi- dissection (11). Microwave irradiation of the specimen in cal margins so that the margin sampled is only a fraction normal saline to 70°C until firmness is attained is another of the circumference of the entire margin of the specimen convenient method. The specimen may also be sliced in excised. It also allows the pathologist to measure the the fresh state using a dedicated and specifically built distance in millimeters of DCIS from the resection mar- dissection board with a 4-mm depression that provides gin. This approach relies heavily on the macroscopic support for the tissue and aids sectioning. Using this impression of the dissecting pathologist and sampling technique, full-face slices can be serially sampled paral- should include any pale fibrous areas seen. An alterna- lel to the surface of the cutting board. tive to this method of selected inked margin sampling is Whatever the method employed, the whole of the to examine the shave margins, where 2- to 3-mm–thick specimen is serially sectioned into a series of full-face en face samples are obtained before serial sectioning of slices and laid out in an orientated fashion sequentially as the specimen. This assessment procedure naturally cov- the slices are taken. Careful gross inspection can then be ers a greater surface area of the margin and may be undertaken and combined with specimen radiography, if expected to give a greater degree of confidence in the indicated, for detailed localization of microcalcification. margin assessment. However, a precise measurement of If there is an obvious palpable mass, blocks can be se- distance to the margin cannot be given. One study that lected to incorporate what macroscopically appear to be examined the difference in margin positivity rates be- the closest margins. Blocks can be selected as represen- tween these two techniques (inked surgical margin and tative samples from the circumference, from each of the shave margin) showed that 40% of cases with positive superficial (anterior), deep, medial, and lateral margins shave margins did not have positive inked margin (14). (the superficial margin is often skin and the deep margin This would appear to be explained by the fact that a may extend to the pectoral muscle). However, in the positive shave margin contains DCIS anywhere in the assessment of impalpable mammographically detected sampled block, which in reality means it may lie any- lesions where there may be microcalcification, closer in- where from transection of the margin to within 2 to 3 mm spection and correlation with the specimen mammograph of the margin, depending on the thickness of the block. before and after serial sectioning of the specimen are Thus the implications for recurrence of a positive shave required. In this situation the radiograph of the slices can margin are not as meaningful as for inked margin, where be used as a template from which the blocks taken can be a precise measurement can be made. Shave margin as- easily annotated, allowing improved correlation between sessment has some advantages over inked margin, as it

Advances in Anatomic Pathology, Vol. 8, No. 6, November, 2001 340 E. K. A. MILLAR AND A. S.-Y. LEONG covers a greater area and results in fewer blocks for sensus as to the definition of an adequate margin of assessment. However, multiple levels are probably re- excision. Of the many studies in the literature, most have quired. A similar method of margin assessment is the adopted their own definition of a “clear” margin, which performance of cavity margin biopsies, likened to the can range from anything less than transection of the mar- “peeling of an onion from the outside in” (10,15). In this gin to greater than 20-mm clearance (Table 1). So what way, the surgeon samples further tissue from any number is the most appropriate distance? Studies examining the of cavity margin sites, which he feels to be appropriate to distribution of DCIS in the breast can perhaps provide us guide himself and the pathologist in the adequacy of with most guidance about what is adequate and what is excision. During this procedure, any number of sepa- inadequate. Faverly and Holland’s (22) detailed three- rately labeled specimens of breast and fatty tissue from dimensional stereomicroscopic analysis of 119 mastec- as many as six different locations within the excision tomies has provided invaluable information on the dis- cavity site may be received. These specimens can either tribution of DCIS, which is of key importance in our be unoriented or identified by a suture or surgical clip as understanding and management of this disease. They de- to the side facing the tumor bed, to allow orientation and scribed the growth patterns (continuous or discontinu- inking by the accessioning pathologist. It is normal prac- ous) of DCIS and how it varies depending on DCIS tice to process the whole of the cavity margin specimens, grade (Fig. 1). They found that DCIS as a process is but in some cases the large number of blocks, technical essentially unicentric, (i.e., confined to one “segment” of constraints, and cost-effectiveness may preclude this and the breast), but could be multifocal within the affected some selectivity may be required. A study by Holland et segment. The only exception to this finding was in cases al. (10), however, showed that even with a negative cav- of pure micropapillary DCIS, which have been found in ity margin biopsy, the local recurrence rate was 9.8%. other studies to have a multicentric distribution (23). This result led them to abandon this method of assess- High-grade lesions showed continuous growth in 90% of ment. cases, whereas low-grade lesions were discontinuous in Alternatives have also been sought for more rapid as- 70% of cases. Where discontinuous growth was present sessment of surgical margins. These have involved the in low-grade DCIS, the “gaps” between foci were found cytologic assessment of scrape or dabs taken from the to be less than 5 mm in 63% of cases and less than 10 excision specimen at a number of locations around the mm in 85% of cases. For all grades of DCIS, 92% of specimen, with touch imprints being rapidly stained and cases had gaps of less than 10 mm. The implications of examined (16–19). This procedure is uncommon and these findings are that high-grade DCIS is usually a con- probably not reliable or sensitive enough on which to tinuous process and that margin assessment is more make sound surgical decisions. likely to be reliable than the assessment of low-grade Frozen section assessment of surgical margins is to be DCIS, which may have gaps between foci. This fact also discouraged. It is obviously unreliable to assess one or undermines the view that the presence of normal termi- two blocks of tissue from a large specimen that requires nal duct lobular units or ducts between DCIS and the detailed and careful dissection and does not warrant the margin is a reassuring feature (24,25). This assertion pressure of an intraoperative consultation that places surely can only be of relevance if there is 10 mm of strain on both the pathologist, the technical staff, and the clearance and not when clearance is only 1 to 2 mm. surgeon (20,21). Furthermore, these specimens are often Other important and pertinent observations have been difficult to interpret and may result in loss of tissue on made regarding the distribution of cancerization of lob- facing up of the specimen on re-embedding and cutting ules. Early work by Fechner (26) described the presence in. Added to this are the technical difficulties involved in of cancerization of lobules in 80% of cases with DCIS sectioning of predominantly fatty tissue, which often re- and described it as the “outermost reaches of involve- sults in low-quality sections, again reducing the sensitiv- ment of the terminal duct lobular unit system by malig- ity and quality of the entire exercise. TABLE 1. Definitions of margin status according to published series WHAT IS AN ADEQUATE MARGIN? Author Involved Close Satisfactory Large studies have addressed the issue of margin posi- Arnesson 1989 (45) Transected <5 mm >5 mm tivity and its impact on local recurrence and several have Recht 1994 (31) <10 mm 10–20 mm >20 mm identified it as an independent risk factor for local recur- Fisher 1995 (1) Transected <1 mm >1 mm rence (1–4), while others studies have not (5–8). This Silverstein 1996 (29) Transected <2 mm >10 mm Holland 1998 (10) <10 mm >10 mm perplexing situation appears to stem from a lack of con-

Advances in Anatomic Pathology, Vol. 8, No. 6, November, 2001 MARGIN STATUS IN DUCTAL CARCINOMA IN SITU OF BREAST 341

categories. The National Surgical Adjuvant Breast Proj- ect (NSABP) B-17 trial was the first prospective randomized study to examine the impact of pathologic factors affecting the treatment of DCIS. In this study the margin was considered positive only if transected by DCIS and any more than a fraction of a millimeter was considered clear. These criteria for margins have been viewed by some as inadequate (25,32) but provide interesting information on recurrence rates nonetheless. Of interest is the fact that annual relapse rates of 4% in the NSABP B-17 study (33) are similar to those found in other studies (25,34). The most recent update on the NSABP study showed that although margin positivity was an independent risk factor for recurrence at 5 years, this was not the case at 8 years’ follow-up (33). Forty FIG. 1. Schematic representation of the segmental distribution of percent of recurrences were as invasive carcinoma. Sil- ductal carcinoma in situ (DCIS) in the breast. Upper half indicates the verstein’s (25) data have shown that 43% of cases with distribution of DCIS according to grade: low-grade DCIS tends to be clear margins had residual disease on re-excision. discontinuous and multifocal, whereas high-grade DCIS tends to be continuous and unifocal. The lower half indicates the overlap that exists Another confounding variable is the definition of re- between anatomical segments and therefore the difficulty in excising a currence. Many studies have highlighted the fact that single affected segment. failure patterns tend to be limited to the area of the original excision and therefore “recurrence” is more likely to nancy.” The presence of cancerization in lobules and be residual disease (10). Kurtz et al. (35) found that ducts close to margins is a predictor of recurrence 70–80% of recurrences over 10 years were in the same (27,28). quadrant. But many cases classified as recurrence or re- When surgery is being planned, the surgeon must sidual disease have involved a different segment of the know the size of the area involved by DCIS. This deter- breast and therefore represent new disease rather than mination relies heavily on the radiologic impression. treatment failure per se. Of course, the presence of close However, mammographic evaluation of the extent of margins or focal margin positivity is one part of a for- DCIS has been found to underestimate the true sizeby10 mula that dictates whether local control and breast con- to 20 mm (22). Using this fact as a guide, surgical exci- servation therapy is likely to be appropriate for any par- sion should perhaps aim to provide 20 mm of clearance ticular case. beyond mammographically detectable microcalcification. This can be problematic for those cases under con- RISK FACTORS FOR LOCAL RECURRENCE sideration for BCT, because the volume of tissue removed is inversely proportional to the final cosmetic Many studies have assessed the relative importance of result, a consideration of particular importance for risk factors for local failure in BCT for DCIS. Positive younger women. This work has led some authors to rec- margins have been shown to be important by some stud- ommend 10 mm as the optimum target for clearance ies, but they are only one part of an equation that deter- (10,29,30) (20 mm for EORTC) (31) when BCT is at- mines the likelihood of success or failure. Clear margins tempted. The recurrence rates quoted between studies are do not imply that there is no disease left behind, nor do difficult to reconcile, given the differing definitions of a positive margins imply extensive residual disease (9), clear margin and with or without additional local radio- especially when narrow margins have been obtained. therapy. Other key factors are the size, grade, architecture, pres- It was the identification of the prognostic importance ence of necrosis, treatment, and the age of the patient of margin status along with size and grade that prompted (Table 2). Other minor factors assessed include biologic the development of the Van Nuys Prognostic Index by markers expressed, lymphocyte infiltrate, and periductal Silverstein et al. (29). This index gives a differential fibrosis. score of the closeness that DCIS approximates to the Boyages et al. (24) performed a detailed meta-analysis margin as “involved,” “close,” or “clear” (1 mm, 1–9 of the risk factors for local recurrence and found differ- mm, and >10 mm, respectively). The final score places ing levels of risk. Most recurrences reported appear to the patient in low-risk, intermediate-risk, or high-risk occur within 3 to 5 years (36,37), although lower-grade

Advances in Anatomic Pathology, Vol. 8, No. 6, November, 2001 342 E. K. A. MILLAR AND A. S.-Y. LEONG

TABLE 2. Factors associated with increased risk of than five foci of atypical ductal hyperplasia or cancer- recurrence of DCIS ization of lobules in association with DCIS within 5 mm Histology: High nuclear grade (24), necrosis (41), micropapillary of a margin were significantly associated with recurrence pattern (23), volume of COL adjacent to margins (27,28), of DCIS. This would appear to suggest then that the periductal fibrosis, and lymphocytic infiltration (42,23) cancerization of lobules seen close to the margin might Positive margins (1–4) Size: >40 mm (29) represent the outermost area of an adjacent zone of large- Age: <40–45 years (37,28) duct DCIS. Multifocality (42) Absence of microcalcification (28,44) Size Family history of breast cancer (39,40) Biologic markers: HER2/neu (42) Silverstein’s data showed that disease volume of less than 15 mm, 16–40 mm or greater than 40 mm had COL, cancerization of lobules. Numbers in parentheses are references. 8-year disease-free survival rates of 90%, 71%, and 32%, respectively (29). However, larger lesions are often of a higher nuclear grade with necrosis and thus these fea- lesions have been found to recur in the long term, some- tures are likely to be of importance in explaining why times up to 25 years after diagnosis (38). extensive disease carries a higher risk.

Family History of Breast Cancer Periductal Fibrosis and Lymphocytic Infiltration The recurrence rate has been found to be 37–40% for Silverstein’s analysis (42) found both periductal fibro- cases with a positive history, 9% for cases with no family sis and lymphocytic infiltration to be associated with a history (39,40). higher recurrence rate, although Bellamy et al. (23) did not find fibrosis to be of value in prediction of local Age failure. Some studies have found an association with younger age: less than 40 years recurrence rate, 13%; more than Biologic Markers 40 years recurrence rate, 2% (37) (both groups treated Local recurrence rate is higher with HER2/neu over- with conservative surgery and radiotherapy). When un- expression, but there was no significant increase for p53 der 45 years the 10 year relapse rate was 23.4% versus expression (42). Both markers are associated with high- 4.3% for over 45 years of age (28). Age was not a risk grade DCIS (43). factor for recurrence in the report by Hiramatsu et al. (39). Multifocality Multifocality (as defined by 10-mm separation be- Histologic Features tween foci of DCIS) is associated with a higher rate of The NSABP study (41) found that the presence of local recurrence but not at a statistically significant level necrosis was associated with recurrence rate (20%) and (42). absent or minimal necrosis recurrence rate (9%, for conservative surgery alone). Micropapillary DCIS is more Absence of Calcification often multicentric and therefore a higher risk for relapse Absence of calcification in the area of DCIS is asso- (23). Nuclear grade has a recurrence rate of 23%, 22.7%, ciated with five times increased risk of recurrence and 8.1% for high, intermediate, and low, respectively, (28,44). for conservative surgery alone and 9.5%, 7.5%, 3.7%, respectively, for conservative surgery with radiotherapy TREATMENT AND RECURRENCE RISK (24). Recent reports have implicated not only the distance from the margin but more importantly the volume Close or positive margins can still be considered for of DCIS, cancerization of lobules, and atypical ductal BCT when managed with adjuvant radiotherapy. In hyperplasia as relevant predictors of recurrence (27,28). terms of recurrence rates, the risk depends on the clear- Cancerization of lobules within 5 mm has been found to ance obtained. Arnesson quoted a relative risk of 38% at be associated with local failure (28). This group of in- less than 5-mm clearance and 6% at less than 5 mm (45). vestigators has drawn attention to the importance of the Boyages meta-analysis data showed overall that relative volume of disease present at a critical distance from any risk is 22.5% with conservative surgery alone, 8.9% for margin as another important indicator of residual disease conservative surgery and radiotherapy, and 1.8% for and likely recurrence. Their data showed margin positiv- mastectomy alone (24). When grade is analyzed in con- ity on its own did not predict recurrence, but that more junction with conservative surgery and radiotherapy,

Advances in Anatomic Pathology, Vol. 8, No. 6, November, 2001 MARGIN STATUS IN DUCTAL CARCINOMA IN SITU OF BREAST 343 relative risk for grades 1, 2, and 3, respectively, were 0, 5. Solin LJ, Recht A, Fourquet A, et al. Ten-year results of breast 10%, and 34% (8). Silverstein’s most recent figures (46) conserving surgery and definitive irradiation for intraductal carcinoma (duct carcinoma in-situ) of the breast. Cancer 1991;68: has shown that radiotherapy does not influence recur- 2337–44. rence when the margins are greater than 10 mm com- 6. Silverstein MJ, Wiseman JR, Gamagami P, et al. Intraductal car- pared with less than 1-mm clearance, where there was a cinoma of the breast (208 cases): clinical factors influencing treatment choice. Cancer 1990;66:102–8. clear benefit in reduction of relative risk. 7. Fisher ER, Leeming R, Anderson S, et al. Conservative management of intraductal carcinoma (DCIS) of the breast. J Surg Oncol 1991;47:139–47. CONCLUSION 8. Silverstein MJ, Barth A, Poller DN, et al. Ten-year results comparing mastectomy to excision and radiation therapy for ductal In current practice the aim of achieving clear margins carcinoma in-situ of the breast. Eur J Cancer 1995:31A:1425–7. is balanced against good cosmesis. Excision of impal- 9. Vicini FA, Goldstein NS, Kestin LL. Pathologic and technical considerations in the treatment of ductal carcinoma in-situ of the pable lesions makes adequate clearance harder to breast with lumpectomy and radiation therapy. Ann Oncol, 1999; achieve. The use of improved imaging in the future as a 10:883–90. guide to disease extent, such as intraoperative ultrasound 10. Holland PA, Gandhi A, Knox WF, et al. The importance of complete excision in the prevention of local recurrence in ductal car- and/or RODEO-MRI (20), may help to guide the sur- cinoma of the breast. Br J Cancer 1998;77:110–4. geon. The pathologist plays a key role in the assessment 11. Lagios MD. Pathologic procedures for mammographically de- and decision-making process in the management of this tected ductal carcinoma in-situ. In: Silverstein MJ, ed. Ductal car- microscopic disease. Detailed margin assessment is a la- cinoma in-situ of the breast. Baltimore: Williams and Wilkins, 1997:189–193. bor-intensive process that is vital to high-quality patho- 12. Graham RA, Homer MJ, Sigler CJ, et al. The efficacy of specimen logic reporting. The additional costs associated with op- radiography in evaluating surgical margins of in palpable breast timal examination of margins nonetheless would be con- carcinoma. AJR Am J Roentgenol 1994;162:33–6. 13. Lee CH, Carter D. Detecting residual tumor after excisional biopsy siderably cheaper than the cost of treating local of impalpable breast carcinoma: efficacy of comparing preopera- recurrence. Contemporary issues concerning the assess- tive mammograms with radiographs of the biopsy specimen. AJR ment of surgical margins may be summarized as follows: Am J Roentgenol 1995;164:81–6. 14. Guidi AJ, Connolly JL, Harris JR, et al. The relationship between ● Distribution of DCIS: high grade tends to be continu- shaved margin and inked margins status in breast excision specimens. Cancer 1997;79:1568–73. ous and unicentric, whereas low grade is discontinu- 15. The Consensus Conference Committee. Consensus Conference on ous and multicentric (micropapillary). the Classification of Ductal Carcinoma In Situ. Cancer 1997;80: ● Methods of assessment of margins: inked margin is 1798–802. 16. Saarela AO, Paloneva TK, Rissanen TJ, et al. Contaminants of better than shave margin. positive histologic margins and residual tumor after lumpectomy ● Clearance: 5-mm clearance may be adequate; 10-mm for early stage breast cancer; a prospective study with special clearance is optimal. reference to touch preparation cytology. J Surg Oncol 1997;66: ● Volume of DCIS and cancerization of lobules adjacent 248–53. 17. Cox CE, Hyacinthe N, Gonzalez RJ et al. Cytologic evaluation of to closest margin should be incorporated into patho- lumpectomy margins in patients with ductal carcinoma in-situ. Ann logic reports (especially if less than 5 mm). Surg Oncol 1997;4: 644–9. ● Margin status is one index of a number of variables 18. England DW, Chan SY, Stonelake PS, et al. Assessment of excision margins following wide local excision for breast carcinoma used to predict risk of local recurrence. Size, grade, using specimen scrape cytology and tumor bed biopsy. Eur J Surg necrosis, pattern, young age (less than 45 years), and Oncol 1994;20:425–9. family history are also relevant. 19. Klimberg VS, Westbrook KC, Korourian S. Ease of touch preps for diagnosis and evaluation of surgical margins in breast cancer. Ann Surg Oncol 1998;5:220–6. REFERENCES 20. Klimberg VS, Harms S, Korourian S. Assessing margin status. Surg Oncol 1999;8:77–84 1. Fisher ER, Constantino J, Fisher B, et al. Pathologic findings from 21. The Steering Committee on Clinical Practice Guidelines for the the National Surgical Adjuvant Breast Project (NSABP) Protocol Care and Treatment of Breast Cancer. Canadian Association of B-17: Intraductal carcinoma (ductal carcinoma in situ). Cancer Radiation Oncologists. Thee management of ductal carcinoma in- 1995;75:1310–9. situ (DCIS). Can Med Assoc J 1998;158(suppl 3):S27–34. 2. Sneige N, McNeese MD, Atkinson EN, et al. Ductal carcinoma 22. Holland R, Faverly DRG. Whole organ studies. In: Silverstein MJ, in-situ treated with lumpectomy and irradiation: histopathologic ed. Ductal carcinoma in-situ of the breast. Baltimore: Williams analysis of 49 specimens with emphasis on risk factors and long and Wilkins, 1997:233–40. term results. Hum Pathol 1995;26:642–9. 23. Bellamy COC, McDonald C, Salter DM, et al. Noninvasive ductal 3. Cutuli B, Teissier E, Piat J-M, et al. Radical surgery and conser- carcinoma of the breast: the relevance of histologic categorization. vative treatment of ductal carcinoma in-situ of the breast. Eur J Hum Pathol 1993;24:16–23. Cancer 1992;28:649–54. 24. Boyages J, Delaney G, Taylor R. Predictors of local recurrence 4. Cheng L, Al-Kaisi K, Gordon NH, et al. Relationship between the after treatment of ductal carcinoma in-situ. Cancer 1999;85: size and margin status of ductal carcinoma in-situ of the breast and 616–28. residual disease. J Natl Cancer Inst 1997;89:1356. 25. Silverstein MJ, Gierson ED, Colburn WJ. Can intraduct breast

Advances in Anatomic Pathology, Vol. 8, No. 6, November, 2001 344 E. K. A. MILLAR AND A. S.-Y. LEONG

carcinoma be excised completely by local excision? Clinical and ment of ductal carcinoma in-situ of the breast. J Clin Oncol 1996; pathologic predictors. Cancer 1994;73:2985–9. 14:754–63. 26. Fechner RE. Ductal carcinoma involving the lobule of the breast: 37. Fowble B, Hanlon AL, Fein DA, et al. Results of conservative a source of confusion with lobular carcinoma in-situ. Cancer 1971; surgery and radiation for mammographically detected ductal car- 28:274–81. cinoma in-situ (DCIS). Int J Radiat Oncol Biol Phys 1997;38: 27. Goldstein NS, Lacerna M, Vicini F. Cancerization of lobules and 949–57. atypical ductal hyperplasia adjacent to ductal carcinoma in-situ of 38. Page D, Dupont WD, Rodgers LW, et al. Continued local recur- the breast. Am J Clin Pathol 1998;110:357–67. rence of carcinoma 15–25 years after a diagnosis of low grade 28. Kestin LL, Goldstein NS, Lacerna MD, et al. Factors associated ductal carcinoma in situ of the breast treated only by biopsy. Can- with local recurrence of mammographically detected ductal carci- cer 1995;76:1197–200. noma in-situ in patients given breast conserving therapy. Cancer 39. Hiramatsu H, Bornstein BA, Recht A, et al. Local recurrence after 2000;88:596–607. conservative surgery and radiation therapy for ductal carcinoma 29. Silverstein MJ, Lagios, Craig PH, et al. A prognostic index for in-situ. Cancer J Sci Am 1995;1:55–61. ductal carcinoma in-situ of the breast. Cancer 1996;77:2267–74. 40. McCormick B, Rosen PP, Kinne D, et al. Duct carcinoma in-situ of 30. Faverly D, Burgers L, Bult, et al. Three dimensional imaging of the breast: an analysis of local control after conservative surgery mammary ductal carcinoma in-situ: clinical implications. Semin and radiotherapy. Int J Radiat Oncol Biol Phys 1994;28:105–11. Diagn Pathol 1994;11:193–8 41. Page DL, Lagios MD. Pathological analysis of the National Sur- 31. Recht A, Van DJ, Fentiman IS, et al. Third meeting of the DCIS gical Adjuvant Breast Project (NSABP) B-17 Trial. Cancer 1995; working party of the EORTC: conference report. Eur J Cancer 75:1219–22. 1994;28:626–9. 32. Silverstein MJ. Predicting residual disease and local recurrence in 42. Silverstein MJ. Prognostic factors and local recurrence in patients Breast J patients with ductal carcinoma in situ. J Natl Cancer Inst 1997; with ductal carcinoma in-situ of the breast. 1998;4: 89:1330–1. 349–62. 33. Fisher ER, Dignam J, Tan-Chiu E, et al. Pathologic findings from 43. Ellis IO, Poller DN. Ductal carcinoma in-situ (DCIS) of the breast. the National Surgical Advent Breast Project (NSABP) eight year In: Kirkham N, Lemoine NR, eds. Progress in pathology.vol2. update of protocol B-17: intraduct carcinoma. Cancer 1999;86: Edinburgh: Churchill Livingstone, 1995:47–87. 429–38. 44. Fourquet A, Zafrani B, Campana F, et al. Institut Curie experience. 34. Marks LB, Prosnitz LR. Lumpectomy with and without radiation In: Silverstein MJ, ed. Ductal carcinoma in-situ. Baltimore: Wil- for early-stage breast cancer and DCIS. Oncology 1997,11:1361–8. liams and Wilkins, 1997:391–7. 35. Kurtz JM, Jacquemier J, Amalric R, et al. Risk factors for breast 45. Arnesson LG, Smeds S, Fagerburg G, et al. Follow-up of two recurrence in premenopausal and post menopausal patients with treatment modalities for ductal cancer in situ of the breast. Br J ductal cancers treated by conservation therapy. Cancer 1990;65: Surg 1989;76:672–5. 1867–78. 46. Silverstein MJ, Lagios MD, Groshen S, et al. The influence of 36. Solin LJ, Kurtz J, Fourquet A, et al. Fifteen year results of breast margin width on local control of ductal carcinoma in-situ of the conserving surgery and definitive breast irradiation for the treat- breast. NEnglJMed1999;340:1455–61.

Advances in Anatomic Pathology, Vol. 8, No. 6, November, 2001 Histopathology 1999, 34, 491–496

Malignant phyllodes tumours of the breast display increased stromal p53 protein expression

E K A Millar, J Beretov,* P Marr,* M Sarris,† R A Clarke,‡ J H Kearsley‡ & C S Lee§ §Department of Anatomical Pathology, Royal Prince Alfred Hospital and †Department of Pathology, University of Sydney, Camperdown; Discipline of Anatomical Pathology, Hunter Area Pathology Service, John Hunter Hospital and University of Newcastle; and *Department of Anatomical Pathology and ‡Cancer Services St. George Hospital, Kogarah, New South Wales, Australia

Date of submission 7 July 1998 Accepted for publication 1 December 1998

Millar E K A, Beretov J, Marr P, Sarris M, Clarke R A, Kearsley J H & Lee C S (1999) Histopathology 34, 491–496 Malignant phyllodes tumours of the breast display increased stromal p53 protein expression

Aims: To determine the variation in p53 protein or focal weak nuclear positivity of scattered stromal expression in phyllodes tumours and fibroadenomas of cells. the breast. Conclusions: Increased p53 immunoreactivity is present Methods and results: Fifteen phyllodes tumours (six in malignant phyllodes tumours in contrast to benign malignant, nine benign) and 20 fibroadenomas were phyllodes tumours and fibroadenomas. Malignant examined for p53 expression by immunohistochemistry. phyllodes tumours display a distinctive pattern of p53 Five of the six malignant phyllodes tumours showed immunostaining which may be of diagnostic value. moderate or strong p53 positivity at sites of peri- These findings suggest that p53 protein may be epithelial stromal condensation and atypia. All 20 important in the progression of benign to malignant fibroadenomas, nine benign phyllodes tumours and one phyllodes tumours. malignant phyllodes tumour showed either negativity Keywords: fibroadenoma, p53, phyllodes tumour

Introduction present focally or the stroma is not particularly cellular the precise classification of the lesion Phyllodes tumours of the breast are a rare group of becomes problematic. neoplasms showing a spectrum of histological appear- When the term phyllodes tumour is applied there is ances and biological behaviour. There is a tendency for then the problem of determining biological behaviour. local recurrence and very rarely for metastatic poten- 1 Numerous large studies have examined this area and tial. Making the diagnosis of phyllodes tumour is not some criteria have been established relating to the clear-cut and relies on the presence of a combination of stromal cellularity, atypia and mitotic activity.2–4 features. The classical appearance of a cellular stroma However, these criteria are not rigidly defined and with leaf-like papillary projections bulging into cystic vary between studies. Thus the adequacy of local spaces presents no difficulty when present in its pure excision is important in management and the recom- from. However, when considering the histological mendation of a margin of normal tissue taken to prevent features of this group of tumours there is considerable local recurrence even for histologically benign overlap at one end of the spectrum with fibroadenoma. tumours.5 We have investigated differences in the When the characteristic phyllodes pattern is only expression of p53 protein between histologically benign and malignant phyllodes tumours. Mutations Address for correspondence: Professor C. Soon Lee, Department of Anatomical Pathology, Royal Prince Alfred Hospital, Missenden Road, of this tumour suppressor gene are amongst the 6 Camperdown N. S. W. 2050, Australia. commonest detected in human malignancies and

᭧ 1999 Blackwell Science Limited. 492 E K A Millar et al.

much has been published in this regard. Accumulation clinical information regarding tumour size and age of of the protein product, as detected by immuno- patient was available. Recurrence rates and overall histochemistry, has been described in other tumour survival were not. A total of 20 fibroadenomas from types as a marker of neoplastic progression and of years corresponding to those of the phyllodes tumours aggressiveness.7,8 There is, however, a paucity of were also studied. Haematoxylin and eosin (H & E) published work in phyllodes tumours in this regard. A stained sections of each case were independently group of fibroadenomas was also examined to compare reviewed by two pathologists and the histological patterns of expression in these two overlapping entities. features of both phyllodes tumours and fibroadenomas were recorded to include assessment of the following Materials and methods features: nature of margin (pushing or infiltrative), architectural pattern (þ, focal leaf-like projections; þþ, The study group consisted of 15 cases of diagnosed well formed projections into cystic spaces; þþþ, phyllodes tumours. The histological criteria used to uninterrupted stroma with little epithelial component), diagnose phyllodes tumours are well described. We used stromal cellularity (þ, mild; þþ, moderate; þþþ, a combination of architectural pattern, stromal crowded, packed cells), nuclear pleomorphism (þ, cellularity, atypia and mitotic activity.1–5 Only basic mild; þþ, moderate; þþþ, severe), mitotic activity

Table 1. Phyllodes tumours (PT): clinical and histological features

Size Stromal Architectural Mitoses per Stromal Histological Case Margin Age (mm) cellularity pattern 10 hpf (× 400) nuclear pleo. diagnosis Stromal p53

1 Pushing 57 100 þþ/ þþþ þþ 6 þþ/ þþþ Malignant PT þþþ

2 Inﬁltrative 56 10 þþþ þþþ 10 þþ Malignant PT þþ

3 Inﬁltrative 49 30 þþþ þþþ 13 þþ/ þþþ Malignant PT þþþ*

4 Pushing – – þþ þ 7 þþ/ þþþ Malignant PT þþ

5 Pushing 30 50 þþþ þþ 18 þþþ Malignant PT þþ

6 NA 26 Core þþþ þþþ 20 þþþ Malignant PT þþþ

7 Pushing 41 38 þþ þ 0 þþ Benign PT –

8 Pushing 27 40 þþ þþ 1 þþ Benign PT –

9 Pushing 38 15 þþ þþ 0 þ Benign PT –

10 Pushing 48 25 þþ þþ 0 þþ Benign PT þ

11 Pushing 44 30 þþ þþ 0 þþ Benign PT þ

12 Pushing 42 53 þþ þþ 1 þ Benign PT –

13 Pushing 25 33 þþ þþ 1 þþ Benign PT –

14 Pushing – – þþ 0 þþ Benign PT þ

15 Pushing 66 21 þþ þþ 1 þ Benign PT þ

*Lymph node met also p53 þþ. NA, not applicable; –, data were not available (see Materials and methods section).

᭧ 1999 Blackwell Science Ltd, Histopathology, 34, 491–496. p53 in phyllodes tumours of the breast 493

Table 2. Fibroadenomas: histological and immunohisto- nuclear pleomorphism and have five or more mitoses/ chemical profile (see Materials and methods section) 10 high-power fields (hpf). Tumours that showed a pushing margin with stromal cellularity, growth architecture and nuclear pleomorphism of þ to þþ Mitoses per with a mitotic rate of less than 1/10 hpf were classified Stromal 10 hpf Stromal nuclear Stromal p53 1–5 Case cellularity (x 400) pleomorphism expression as benign. Tumours were regarded as ‘borderline’ if they demonstrated some but not all of the features 16 þ 0 þ – of malignant phyllodes tumours and had no more than 2–5 mitoses/10 hpf.1–5 Using these criteria, of the 15 17 þþ 0 þþcases of phyllodes tumours, there were six cases classified as malignant and nine as benign. No cases 18 þþ 1 þþwere classified as ‘borderline’. All cases were immunostained for p53 (DO-7 mono- þþ þþ þ 19 0 clonal, Novocastra, Newcastle-upon-Tyne, UK) and the staining of cells was assessed according to both the 20 þþ 1 þþ – intensity and proportion of cells staining. Tumours were 21 þþ 0 þþ – considered as having no expression (–) of the p53 protein when no stromal cells showed any nuclear 22 þþ 0 þ – immunostaining; þ when less than 33% cell nuclei stained with weak staining intensity; þþ when 34– 23 þþ 1 þþ – 67% cell nuclei stained with weak to moderate staining intensity; and þþþ when more than 67% cells 24 þþ 0 þþ – displayed moderate to strong nuclear immunostaining.

25 þþ 0 þþ þ Results 26 þþ 0 þþ – HISTOLOGY 27 þþ 0 þþ – The results are presented in tabulated form in Tables 1 28 þ 0 þþand 2. As a result of our observations we were able to split the group of phyllodes tumours into histologically 29 þ 0 þþ þ benign and those of an apparently malignant nature. This division was made on the basis of a combination of 30 þ 0 þ – infiltrative margin, increasing stromal cellularity, mitotic activity (greater than five mitoses per 10 high þ þþ 31 0 power fields, × 400) and nuclear pleomorphism. The size of the lesion did not correlate with these features. 32 þ 0 þ – Heterologous elements were not present within any of 33 þ/þþ 0 þþ – the phyllodes tumours. Necrosis was present in one case which showed striking stromal overgrowth (case 6). 34 þ 0 þ – The most notable difference between the benign and malignant groups was the presence of focal areas of 35 þ 0 þþincreased cellularity, often condensed around the epithelial component. These areas often showed increasing nuclear atypia and mitotic activity (Figure 1). Extensive stromal overgrowth was present (mean number of mitoses within the most mitotically in three of the six malignant tumours. active area of the tumour; × 400, per 10 high-power fields). p53 IMMUNOSTAINING We classified phyllodes tumours as malignant when they displayed most of the following features: infiltrative Expression of p53 reflected the histological differences margin, þþto þþþ stromal cellularity, þþto þþþ observed and was notably different between the benign stromal overgrowth architecture, moderate to severe (i.e. benign phyllodes tumours and fibroadenomas) and

᭧ 1999 Blackwell Science Ltd, Histopathology, 34, 491–496. 494 E K A Millar et al.

Figure 1. Malignant phyllodes tumour. Increased nuclear atypia and Figure 4. p53 immunostaining in a malignant phyllodes tumour mitotic activity is notable in areas of subepithelial stromal (case 6). This case showed notable stromal overgrowth with strong condensation (H & E, × 400, case 3). patchy p53 expression (× 400).

Figure 2. p53 immunostaining in a malignant phyllodes tumour Figure 5. p53 immunostaining in a lymph node metastasis (case 3). (case 1). There is characterstic p53 expression in areas of peri- There is continued p53 expression (× 400). epithelial stromal condensation (× 200).

malignant groups. The benign phyllodes tumours displayed minimal p53 expression with only an occasional scattered stromal cell staining weakly. This pattern of staining was also found in all of the fibroadenomas. Malignant tumours in contrast showed a distinctive pattern of staining with the areas of stromal condensation showing p53 expression which was, in five out of six cases, of moderate or strong intensity (Figures 2 and 3). Away from these areas there was patchy focal positivity but often with large areas showing negative staining. Case 3 also showed p53 positivity at the infiltrating edge of the lesion. Two of the three tumours showing notable stromal overgrowth (þþþ for both cellularity and architectural pattern) Figure 3. p53 immunostaining in a malignant phyllodes tumour displayed a patchy strong positivity (Figure 4), the other (case 5). There is strong nuclear expression of p53 in the areas of one displaying only focal weak positivity akin to that of stromal condensation (× 400). the benign cases and fibroadenomas. The single case

᭧ 1999 Blackwell Science Ltd, Histopathology, 34, 491–496. p53 in phyllodes tumours of the breast 495

having a lymph node metastasis (case 3) showed of information pertaining to phyllodes tumours and p53 continued focal moderate positivity of p53 expression expression. In our study, we investigated p53 expression within the node (Figure 5). in fibroadenomas, and benign and malignant phyllodes tumours to identify any difference in expression. The Discussion difference in immunoreactivity observed in malignant phyllodes tumours was clearly confined to areas of Phyllodes tumours of the breast are rare with an stromal condensation surrounding the glandular com- incidence described in a recent series of cases as ponent where there was strong nuclear staining. This representing 1.5% of all breast carcinomas.5 They strong stromal positivity was in comparison to the occur most often in 30–70-year-old females with very benign pattern of negative or only focally weak positive few occurring in adolescents.1 The histological overlap staining. In terms of biology this pattern of accumula- described between benign phyllodes tumours and tion of p53 would correlate with the concept of stromal fibroadenomas requires a thorough examination of induction by the epithelial component. The pattern of the stromal component to identify any evidence of p53 expression is as one would expect and mirrors the increased mitotic activity, with a figure of less than three histological changes of increasing nuclear atypia. It per 10 hpf allowable for fibroadenoma by some shows an association of increased p53 expression in authors.9 However, this figure must take into considera- progression of this neoplasm from benign to malignant. tion the architectural pattern, cellularity and nuclear Only one of the six cases did not show this pattern, atypia which if notable would make the diagnosis of instead displaying only focal positivity akin to that of the phyllodes tumour more likely. There is then a con- benign tumours. In the single case we studied with tinuum of diagnostic features, the weight of each feature the rare occurrence of lymph node metastasis the being variable. metastatic deposit of stroma continued to show strong The clinical distinction between fibroadenoma and p53 positivity. phyllodes tumour is, however, more clear in that Examination of p53 in phyllodes tumours would phyllodes tumours may recur after incomplete excision appear therefore to be potentially useful in providing whereas fibroadenomas will not. Most recurrences further diagnostic weight and confidence to the assign- apparently follow ‘shelling out’ of suspected fibro- ment of a malignant tumour. Conversely absence of adenomas, usually within a 2-year period,10 but strong p53 expression would appear to correlate occasionally longer. Recurrence brings with it a small with the presence of a benign phenotype potentially but ever present risk of malignant transformation useful in cases of marginally increased stromal assessed in one study to be 7%.10 However, there is cellularity, atypia and mitoses. Unfortunately clinical confusion in the literature regarding the use of criteria information regarding the recurrence rates and to differentiate between benign and malignant tumours. overall behaviour of the tumours was not available. A mitotic count of over five per 10 hpf is regarded as We would therefore suggest that further study of a being malignant by some authors but borderline by larger group of cases with detailed clinical follow-up others.3,4 Malignant phyllodes tumours are a rarity, would be of benefit to further evaluate the usefulness comprising 0.18% of all breast malignancies.11 of our observations. Although often reported on histological grounds to be malignant or of borderline malignant (low-grade) References potential one study found that only 23% of malignant 3 1. Page DL, Anderson TJ, Johnson RL. Sarcomas of the breast. In: phyllodes tumours resulted in the death of the patient. Page DL, Anderson TJ, eds. Diagnostic Histopathology of the Breast. Metastatic spread is usually haematogenous in nature Edinburgh: Churchill Livingstone, 1987: 341–350. with lymph node metastasis the exception. As a result of 2. Grigioni WF, Santini D, Grassigli A et al. A clinicopathological the uncertain behaviour and recurrence of benign study of cystosarcoma phyllodes. Twenty case reports. Arch. Anat. phyllodes tumours, the surgical management currently Cytol. Pathol. 1982; 30; 303–306. 3. Hart WL, Bauer RC, Oberman HA. Cystosarcoma phyllodes. consists of wide local excision or mastectomy for large A clinicopathological study of 26 hypercellular periductal 5 lesions without axillary dissection. stromal tumours of the breast. Am. J. Clin. Pathol. 1978; 70; Is there any further information that may help in 211–216. predicting the behaviour of this group of neoplasms? 4. Pietruszka M, Barnes L. Cystosarcoma phyllodes. A clinico- p53 expression has been described in many publications pathological analysis of 42 cases. Cancer 1978; 40; 1974– 1983. relating to prognosis and tumour behaviour in most if 5. Moffat CJC, Pinder SE, Dixon AR, Elston CW, Blamey RW, Ellis IO. not all malignancies in man. It seemed then surprising Phyllodes tumours of the breast: a clinicopathological review of to us that review of the literature has revealed a dearth thirty two cases. Histopathology 1995; 27; 205–218.

᭧ 1999 Blackwell Science Ltd, Histopathology, 34, 491–496. 496 E K A Millar et al.

6. Lane DP. p53 and human cancers. Br. Med. Bulletin 1994; 50; Anderson TJ, eds. Diagnostic Histopathology of the Breast. 582–599. Edinburgh: Churchill Livingstone, 1987: 72–88. 7. Carder P, Wyllie AH, Purdie CA et al. Stabilised p53 facilitates 10. Hadju SI, Espinosa MH, Robbins GF. Recurrent cystosarcoma aneuploid clonal divergence in colorectal carcinoma. Oncogene phyllodes. A clinicopathological study of 32 cases. Cancer 1976; 1993; 8; 1397–1401. 38; 1402–1406. 8. Neshat K, Sanchez CA, Galipeau PC et al. p53 mutations in 11. Vorherr H, Vorherr UF, Kutvirt DM, Key CR. Cystosarcoma Barrett’s adenocarcinoma and high-grade dysplasia. Gastro- phyllodes: epidemiology, pathohistology, pathobiology, diagno- enterology 1994; 106; 1589–1595. sis, therapy and survival. Arch. Gynaecol. 1985; 236; 173– 9. Fechner RE. Fibroadenoma and related lesions. In: Page DL, 181.

᭧ 1999 Blackwell Science Ltd, Histopathology, 34, 491–496. Asia–Paciﬁc Journal of Clinical Oncology 2009; 5: 154–158 doi:10.1111/j.1743-7563.2009.01230.x

CASE REPORT

Metachronous bilateral primary low-grade mucosa-associated lymphoid tissue non-Hodgkins

lymphoma of the breastajco_1230 154..158

M Helene YILMAZ,1 Ewan KA MILLAR,1,2,3,4 Con THEOCHAROUS1 and Peter H GRAHAM3,5 1Department of Anatomical Pathology, South Eastern Area Laboratory Service and 5Department of Radiation Oncology, Cancer Care Centre, St George Hospital, Kogarah, 2Cancer Research Program, Garvan Institute of Medical Research, Darlinghurst, 3Faculty of Medicine, University of New South Wales and 4Faculty of Medicine, University of Western Sydney, Sydney, New South Wales, Australia

Abstract We report the rare occurrence of bilateral low-grade MALT lymphoma of the breast in a 73 year old female. The clinical, pathological and treatment details are discussed, with a review of the literature. Key words: breast, non-Hodgkin’s lymphoma, radiotherapy.

INTRODUCTION localized lumpectomy was performed. Histology revealed a low-grade MALT lymphoma, comprising a Primary non-Hodgkin’s lymphoma (NHL) of breast diffuse infiltrate of small centrocyte-like atypical lym- comprise 0.04–0.53% of primary malignant breast phoid cells, mature plasma cells and numerous reactive 1,2 tumors. Variable frequencies of primary mucosa- lymphoid follicles (Fig. 2). Lymphoepithelial lesions were associated lymphoid tissue (MALT) lymphoma of the present, which were highlighted with immunostains for breast are reported and comprise 4% of all extranodal cytokeratin CAM 5.2 and immunohistochemistry 3 MALT lymphomas in one series. Other more common showed positive staining for CD20, CD79a (Fig. 3) non-gastrointestinal sites of extranodal MALT lympho- CD45, and negativity for CD3, CD5, CD10 and cyclin mas include the head and neck, lung, ocular adnexa, and D1. Molecular genetic analysis was not performed. Local 3 salivary glands. Bilateral, primary MALT lymphoma of excision was complete. The patient’s staging investiga- the breast is exceptionally rare, with only eight cases tions revealed no evidence of extramammary disease. She reported in the published work. The optimal treatment received postoperative localized radiotherapy (30 Gy to is not well established. the anterior chest wall in 15 fractions). The patient was followed up routinely. Four years following her initial CASE REPORT treatment, a 17-mm left breast lesion was detected. A hookwire-localized lumpectomy was performed, reveal- A 73-year-old woman had a suspicious 16-mm lesion ing another low-grade MALT lymphoma with a similar detected in her right breast on a routine mammogram histology and immunophenotype as described in the first (Fig. 1). Fine needle aspiration (FNA) and hookwire- lesion. The patient was re-staged with no evidence of extramammary disease. Local radiotherapy was again utilized. The patient was recently seen at follow-up at Correspondence: Dr Ewan Millar, Department of Anatomical Pathology, South Eastern Area Laboratory Service, St George almost 7 years since the primary diagnosis and had Hospital, Gray Street, Kogarah, NSW 2217, Australia. recently developed a firm swelling within her left eyelid. A Email: [email protected] small incisional biopsy of the lesion showed features Accepted for publication 19 April 2009. suspicious of low-grade MALT lymphoma.

Figure 1 Mammogram of a 16-mm lesion detected the right Figure 3 Immunostain for B-cell marker CD20 highlighting a breast of a 73-year-old woman. lympho-epithelial lesion.

and CD10 and BCL6 negativity exclude follicular lymphoma.4 Various cytogenetic abnormalities have been described in MALT lymphoma, some correlating with the site of disease. Two translocations of the MALT1 gene on 18q21 are known: t(11; 18)(q21; q21) with API2 found mainly in gastric and pulmonary tumors and t(14; 18)(q32; q21) with IGH, which is found predominantly in ocular/adnexal and salivary lesions.5 The relatively rare but well known translocation involving the BCL10 gene t(1; 14)(p22; q32) with IGH is associated with lung and gastric disease and appears to be specific for MALT lymphoma.6 Both MALT1 and BCL10 translocations are associated with activation of the NFkB pathway and cell survival.7 t(3; 14)(p13; q32) Figure 2 Low power view of low-grade mucosa-associated is a newly identified chromosomal translocation in lymphoid tissue non-Hodgkin’s lymphoma infiltrating fat within the breast. MALT lymphoma, which deregulates the expression of the FOXP1 gene and is described in thyroid, ocular, skin and lung disease.8 Trisomy 3 is a relatively non-specific DISCUSSION finding. The genetic features of MALT breast lymphoma are not yet well established. Talwalkar et al. showed no Despite there being no specific immunophenotypic evidence of t(11; 18) and t(14; 18) translocations in marker of MALT lymphoma, the typical immunohis- MALT lymphoma of the breast.9 Additionally, BCL10 tochemical analysis shows atypical centrocyte- staining, which has previously been utilized as an likelymphoid cells that are CD20+, CD79a+, CD4+/-, indirect marker of possible translocations in MALT lym- CD11c+/-, CD5-, CD10- and CD23-.4 Light chain restric- phoma, was found to be discordant with the fluores- tion helps to confirm monoclonality. The differential cence in situ hybridization results.9 This suggests that diagnosis typically includes other small B-cell lympho- a positive BCL10 result should not be interpreted mas, which can usually be excluded on the basis of alone, and cytogenetic presence or absence of the trans- morphology and immunophenotype: CD5 and CD23 location should be confirmed. MALT lymphoma usually negativity exclude small lymphocytic lymphoma, CD5 has an indolent course with recurrences in other and cyclin D1 negativity exclude mantle cell lymphoma extranodal sites after many years as illustrated in our

Table 1 Clinical features of published cases of bilateral mucosa-associated lymphoid tissue (MALT) lymphoma of breast

Time (months) Age Background of for recurrence Reference (years) autoimmune disease Lesion detection Treatment of lesion Staging Outcome

Kraemer DM et al.11 48 Hashimoto’s thyroiditis, Mammography Chemotherapy Simultaneous Cervical, axillary and Complete myalgia, cryoglobulin, GN lesions retroperitoneal remission lymphadenopathy; endoscopy normal Welsh JS et al.12 66 Nil Screening 1. Lumpectomy 12 BM biopsy, LDH and Disease free at mammography 2. Lumpectomy + radiotherapy CT C/A/P normal 36 months Gopal S et al.13 79 Nil Mammography 1. Lumpectomy 3 Incomplete (spinal Disease free at 2. Lumpectomy + radiotherapy lesions present) 12 months Gupta D et al.2 64 SLE Screening 1. Lumpectomy 12 CT A/P negative Disease free at mammography and 2. Lumpectomy + radiotherapy 64 months palpation Farinha P et al.14a,14b 33 DM – Lumpectomy + chemotherapy – – Disease free 76 – – Lumpectomy + chemotherapy – – Dead of disease at 26 months Zobalas B et al.15 68 – Palpable mass Lumpectomy, bilateral axillary Simultaneous Bone scintigraphy, liver Disease free at LN dissection, chemotherapy + lesions U/S, CT C/A/P, 10 months radiotherapy bilateral BM biopsy negative Espinosa LA et al.16 56 – Screening MRI of Bilateral breast radiotherapy 5 CT, BM biopsy, Disease free saPcJCi Oncol Clin J Asia–Pac breasts as part of gallium scan negative clinical trial Present case 73 Uncertain etiology of Mammography 1. Lumpectomy and 48 CT C/A/P, gallium Disease free at hyperthyroidism radiotherapy scan, endoscopy + 24 months 2. Lumpectomy and gastric biopsy, negative radiotherapy

A/P, abdomen pelvis; BM, bone marrow; C/A/P, chest abdomen pelvis; CT, computed tomography; DM, diabetes mellitus; GN, glomerulonephritis; LDH, lactate dehydrogenase; LN, lymph node; MRI, magnetic resonance imaging; SLE, systemic lupus erythematosus; U/S, ultrasound. 2009; HYilmaz MH 5: 154–158 tal. et MALT non-Hodgkin’s lymphoma of the breast 157

case. There are no established markers of outcome in 2 Gupta D, Shidham V, Zemba-Palko V, Keshegian A. this disease, although a recent study indicates that Primary bilateral mucosa-associated lymphoid tissue lym- gastric MALT lymphoma with t(1; 14) or strong BCL10 phoma of the breast with atypical ductal hyperplasia and nuclear expression are unlikely respond to H. pylori localised amyloidosis. Arch Pathol Lab Med 2000; 124: eradication.10 1233–6. 3 Thieblemont C, Bastion Y, Berger F et al. Mucosa- The rarity of bilateral, primary MALT lymphomas of associated lymphoid tissue gastrointestinal and nongas- the breast can make patient management uncertain. trointestinal lymphoma behaviour analysis of 108 patients. There have been eight reported cases of bilateral, primary J Clin Oncol 1997; 15: 1624–30. low-grade MALT lymphoma of the breast, which are 4 Swerdlow SH, Campo E, Harris NL et al., eds. WHO summarized in Table 1. Some key points can be made Classification of Tumours of Haematopoietic and Lym- from the information available in the published work. phoid Tissues, 4th edn. IARC, Lyon 2008. Most of the patients had contralateral recurrence of 5 Du MQ. MALT lymphoma: recent advances in aetiology disease within a few months of initial presentation. and molecular genetics. J Clin Exp Haematol 2007; 47: Although this observation may represent a publication 31–42. bias for bilateral involvement, it highlights the impor- 6 Ye H, Liu H, Attygalle A et al. Variable frequencies of t(11; tance of routine follow-up and regular imaging. There 18)(q21; q21) in MALT lymphomas of different sites : significant association with CagA strains of H. pylori in appear to be no established guidelines regarding treat- gastric MALT lymphoma. Blood 2003; 102: 1012–18. ment. A retrospective study of 70 patients with stage IE or 7 Lucas PC, Yonezumi M, Inohara N et al. Bcl10 and IIE MALT lymphoma generally treated patients with MALT1, independent targets of chromosomal transloca- involved-field radiation therapy alone at doses of tion in MALT lymphoma, cooperate in a novel NF-kappa 25–35 Gy.17 For all cases, the 5-year disease free survival B signaling pathway. J Biol Chem 2001; 276: 19012– and overall survival rates for the entire group were 76 and 19. 96%, respectively.9 Of the five patients treated with 8 Streubel B, Vinatzer U, Lamprecht A, Raderer M, Chott A. surgery alone, two relapsed locally.9 However, in a study t(3;14) (p14.1;q32) involving IGH and FOXP1 is a novel of 86 patients who had achieved complete remission after recurrent chromosomal aberration in MALT lymphoma. initial therapy, 37% relapsed at a median period of Leukemia 2005; 19: 652–8. 47 months (range 14–307 months), suggesting the need 9 Talwalkar SS, Valbuena JR, Abruzzo LV et al. MALT1 gene rearrangements and NF-KB activation involving p65 and for long-term observation of these patients.18 In our case p50 are absent or rare in primary MALT lymphomas of the the patient was treated with a combination of lumpec- breast. Mod Pathol 2006; 19: 1402–8. tomy and radiotherapy on both occurrences of disease. 10 Ye H, Gong L, Liu H et al. Strong BCL10 nuclear Surgery alone is probably inadequate. While radio- expression identifies gastric MALT lymphomas that do therapy alone is probably adequate, surgical excision not respond to H. pylori eradication. Gut 2006; 55: 137– provides definitive histology. Arguably, a core biopsy 8. may provide sufficient histology. Due to current varia- 11 Kraemer DM, Weissinger F, Reimer P, Kunzmann V, tions in possible treatments, it is important that further Rüdiger T, Wilhelm M. Female patient with a history of cases of breast MALT lymphoma are identified so that Hashimoto’s thyroiditis, diagnosed with MALT lymphoma guidelines may be developed to aid clinical management. of both breasts. Onkologie 2003; 26: 277–80. In addition, the prognostic assessment of non- 12 Welsh JS, Howard A, Hong HY, Lucas D, Ho T, Reding DJ. Synchronous bilateral breast mucosa-associated lym- gastrointestinal extranodal MALT lymphoma is yet to be phoid tissue lymphomas addressed with primary radiation clearly established. Arcaini et al. demonstrated that therapy. Am J Clin Oncol 2006; 29: 634–5. factors such as localized disease, lack of nodal involve- 13 Gopal S, Awasthi S, Elghetany MT. Bilateral breast MALT ment, orbital and cutaneous lymphoma, and stage IV lymphoma: a case report and review of the literature. Ann disease with no bone marrow involvement, were signifi- Hematol 2000; 79: 86–9. cantly associated with improved survival.19 However, 14 Farinha P, Andre S, Cabecadas J, Soares J. High frequency variation is seen in other studies.4 of MALT lymphoma in a series of 14 cases of primary breast lymphoma. Appl Immunohistochem Mol Morphol REFERENCES 2002; 10: 115–20. 15 Zobalas B, Sakorafas GH, Kourakli I, Tsiotou AG. Bilat- 1 Cohen PL, Brooks JJ. Lymphoma of the breast: a eral, primary, low-grade, diffuse B-cell lymphoma of clinicopathologic and immunohistochemical study of mucosa-associated lymphoid tissue (MALT) of the breast. primary and secondary cases. Cancer 1991; 67: 1359–69. Breast J 2002; 8: 382.

16 Espinosa LA, Daniel BL, Jeffrey SS, Nowels KW, Ikeda 18 Raderer M, Streubel B, Woehrer S et al. High relapse DM. MRI features of mucosa-associated lymphoid tissue rate in patients with MALT lymphoma warrants lifelong lymphoma in the breast. AJR 2005; 185: 199–202. follow-up. Clin Cancer Res 2005; 11 (9): 3349–52. 17 Tsang RW, Gospodarowicz MK, Pintilie M et al. Stage I 19 Arcaini L, Burcheri S, Rossi A et al. Nongastric and II MALT lymphoma: results of treatment with radio- marginal-zone B cell MALT lymphoma: prognostic therapy. Int J Radiat Oncol Biol Phys 2001; 50 (5): 1258– value of disease dissemination. Oncologist 2006; 11: 64. 285–91.