Open Full Page

Published OnlineFirst March 14, 2013; DOI: 10.1158/1078-0432.CCR-12-1000

Clinical Cancer Predictive Biomarkers and Personalized Medicine Research

Molecular Proﬁling of Aromatase Inhibitor–Treated Postmenopausal Breast Tumors Identiﬁes Immune-Related Correlates of Resistance

Anita K. Dunbier1,2,5, Zara Ghazoui1,2, Helen Anderson1,2, Janine Salter1, Ashutosh Nerurkar1, Peter Osin1, Roger A'hern3, William R. Miller4, Ian E. Smith1, and Mitch Dowsett1,2

Abstract Purpose: Estrogen withdrawal by treatment with aromatase inhibitors is the most effective form of endocrine therapy for postmenopausal estrogen receptor–positive (ERþ) breast cancer. However, response to therapy varies markedly and understanding of the precise molecular effects of aromatase inhibitors and causes of resistance is limited. We aimed to identify in clinical breast cancer those genes and pathways most associated with resistance to aromatase inhibitors by examining the global transcriptional effects of AI treatment. Experimental Design: Baseline and 2-week posttreatment biopsies were obtained from 112 postmenopausal women with ERþ breast cancer receiving neoadjuvant anastrozole. Gene expression data were obtained from 81 baseline and 2-week paired samples. Pathway analysis identified (i) the most prevalent changes in expression and (ii) the pretreatment genes/pathways most related to poor antiproliferative response. Results: A total of 1,327 genes were differentially expressed after 2-week treatment (false discovery rate < 0.01). Proliferation-associated genes and classical estrogen-dependent genes were strongly downregulated whereas collagens and chemokines were upregulated. Pretreatment expression of an inflammatory signature correlated with antiproliferative response to anastrozole and this observation was validated in an independent study. Higher expression of immune-related genes such as SLAMF8 and TNF as well as lymphocytic infiltration were associated with poorer response (P < 0.001) and validated in an independent cohort. Conclusions: The molecular response to aromatase inhibitor treatment varies greatly between patients consistent with the variable clinical benefit from aromatase inhibitor treatment. Higher baseline expression of an inflammatory signature is associated with poor antiproliferative response and should be assessed further as a novel biomarker and potential target for aromatase inhibitor-treated patients. Clin Cancer Res; 19(10); 2775–86. 2013 AACR.

Introduction inhibitors is the most effective form of endocrine therapy, Approximately 80% of human breast carcinomas present but response to them varies markedly (1–3). Mechanisms of as estrogen receptor a–positive (ERþ). In postmenopausal resistance appear to be multiple but are poorly character- In vitro women, estrogen withdrawal by treatment with aromatase ized. studies of acquired resistance to estrogen deprivation have identiﬁed several putative mechanisms which largely involve growth factor–related signal trans- Authors' Afﬁliations: 1Royal Marsden Hospital; 2Breakthrough Breast duction pathways (4–7) but there is limited clinical evi- 3 Cancer Research Centre, Institute of Cancer Research; Cancer Research dence to support these. It is also notable that these models UK Clinical Trials and Statistics Unit, Section of Clinical Trials, Institute of Cancer Research, London; 4Breast Research Group, University of Edin- do not involve an assessment of the contribution of human burgh, Edinburgh, United Kingdom; and 5Department of Biochemistry, stromal elements. University of Otago, Dunedin, New Zealand The presurgical/neoadjuvant setting provides an excep- Note: Supplementary data for this article are available at Clinical Cancer tionally valuable scenario for linking biology to clinical Research Online (http://clincancerres.aacrjournals.org/). response and to study mechanisms of resistance; within this A.K. Dunbier, Z. Ghazoui, and H. Anderson contributed equally to this work. setting, the proliferation marker Ki67 is a validated phar- Corresponding Author: Anita K. Dunbier, Department of Biochemistry, macodynamic indicator of response to endocrine therapy University of Otago, PO Box 56, Dunedin 9016, New Zealand. Phone: (8–10). Treatment-dependent changes in sequential mea- 643479-9258; Fax: 643479-7738; E-mail: [email protected] surements of Ki67 in neoadjuvant trials of endocrine ther- doi: 10.1158/1078-0432.CCR-12-1000 apy in postmenopausal women (IMPACT, p024, and 2013 American Association for Cancer Research. Z1031; refs. 9, 11, 12) have all revealed differences or lack

www.aacrjournals.org 2775

Dunbier et al.

Materials and Methods Translational Relevance Patient samples Most postmenopausal women with estrogen receptor Fourteen-gauge core-cut tumor biopsies were obtained (ER)–positive breast cancer receive an aromatase inhib- from 112 postmenopausal women with stage I to IIIB ERþ itor at some stage during their treatment. However, early breast cancer before and after 2-week anastrozole responsiveness to aromatase inhibitors varies greatly. monotherapy in a neoadjuvant trial (17, 18). Patient demo- Previous studies of resistance mechanisms have been graphics are shown in Supplementary Table S1. Tissue was conducted in cell lines and have highlighted the impor- stored in RNAlater at 20C. Two 4-mm sections from the tance of signal transduction pathways, but these studies core were stained with hematoxylin and eosin (H&E) and omit the stromal influences that are increasingly recog- examined by a pathologist (A. Nerurkar and P. Osin) to nized to have a major influence on tumor biology. In confirm the presence of cancerous tissue, assess histopa- contrast, we report genome-wide expression profiling of thology, and the presence or absence of lymphocytic infil- 81 ER-positive breast carcinomas including stromal tis- tration. Tumors were deemed to be positive for lymphocytic sues before and during treatment with an aromatase infiltration if intraepithelial mononuclear cells could be inhibitor in the neoadjuvant setting. We identify an seen within tumor cell nests or in direct contact with tumor inflammatory signature as the strongest correlate of poor cells. Biopsies in which lymphocytic infiltrate could be seen antiproliferative response and confirm this finding in an without direct contact to tumor cells were not considered to independent set of tumors as well as using gross lym- have lymphocytic infiltration. Samples were assessed blind- phocytic infiltration as an alternative measure of inflam- ly by both pathologists and the concordance rate was 81%. matory activity. This work provides a potential new Discordant samples were then reassessed by A. Nerurkar avenue for drug development and for identifying who was blinded to the result of the initial assessment and patients with a reduced likelihood of response to aro- the more frequent of the 3 assessments was used in the matase inhibition. analysis. Total RNA was extracted using RNeasy (Qiagen). RNA quality was checked using an Agilent Bioanalyser: samples with RNA integrity values of less than 5 were excluded from further analysis. ER (H-score) and Ki67 (% cells positive) values by centralized immunohistochem- of differences that were parallel to recurrence-free survival istry were already available (17). (RFS) differences in the equivalent adjuvant trials (ATAC, An additional 71 stored sections from paraffin-embed- BIG 1-98, and MA27; refs. 1, 13, 14), respectively. In ded, formalin-fixed core-cut biopsies were obtained from a addition, Ki67 of patients after 2 weeks’ treatment predicts subset of patients from the IMPACT trial who received RFS more closely than pretreatment values (8). neoadjuvant anastrozole or tamoxifen (15). Sections were Nearly all patients on an aromatase inhibitor show H&E-stained, and the presence or absence of detectable reduction in Ki67 expression, suggesting that some benefit lymphocytic infiltration was assessed by a pathologist (A. is derived, although this may be modest (9). This contin- Nerurkar). Ki67 data were obtained from previous analysis uous marker of response is well-suited to assessment of of these patients (8, 9). mechanisms of resistance that are also likely to have a nonbinary effect. This may explain why changes in Ki67 have proven to be better predictors of benefit from endo- Gene expression analysis and data preprocessing crine therapy than clinical response during neoadjuvant RNA amplification, labeling, and hybridization on endocrine therapy (15, 16). HumanWG-6 v2 Expression BeadChips (Illumina) were Gene expression profiling of tumor biopsies before and conducted according to the manufacturer’s instructions at during treatment has the potential to enable the identifica- a single Illumina BeadStation facility. Tumor RNA of suf- tion of the most important genes/pathways involved in the ficient quality and quantity was available to generate expres- response to estrogen deprivation therapy and the pretreat- sion data from 104 pretreatment biopsies and 85 two-week ment determinants of response and resistance. Availability biopsies (Supplementary Fig. S1). Data were extracted using of comprehensive expression datasets, as provided here, will BeadStudio software and normalized with variance-stabi- allow the evaluation of candidate genes from experimental lizing transformation (VST) and Robust Spline Normaliza- research for their clinical relevance. tion (RSN) method in the Lumi package (19). Probes that P > We present, to our knowledge, the largest study of the were not detected in any samples (detection 1%) were global transcriptional effects of aromatase inhibitor treat- discarded from further analysis. Gene expression data from ment in the neoadjuvant setting with the aims of identify- this study is deposited (20). ing: (i) transcriptional features of response to aromatase inhibitors and (ii) phenotypic and genotypic determinants Data analysis of benefit from aromatase inhibitors. This approach Genes differentially expressed between baseline and 2- revealed for the first time the importance of immune/ week samples were identified using a multivariate permu- inflammatory influences that could not have been detected tation test (21) implemented in BRB-Array Tools (22). by studies of cell lines. Random variance t-statistics were calculated for each gene

2776 Clin Cancer Res; 19(10) May 15, 2013 Clinical Cancer Research

Genomic Proﬁling Predicts Response to Aromatase Inhibitor Treatment

Table 1. Genes differentially expressed between pretreatment and 2 weeks in 81 pairs of patients treated with anastrozole

Fold change Proportion Gene symbol Description (post/pre) of cases Downregulated genes 1 TFF1 Trefoil factor 1 0.34 84% 2 UBE2C Ubiquitin-conjugating enzyme E2C 0.42 89% 3 TOP2A Topoisomerase (DNA) II alpha 170kDa 0.45 94% 4 IQGAP3 IQ motif containing GTPase-activating protein 3 0.47 88% 5 UHRF1 Ubiquitin-like, containing PHD and RING finger domains, 1 0.50 90% 6 Hs.388347 mRNA; cDNA DKFZp686J0156 0.50 79% 7 SERPINA3 Serpin peptidase inhibitor, clade A, member 3 0.50 92% 8 CDC20 Cell division cycle 20 homolog (S. cerevisiae)0.5188% 9 PDZK1 PDZ domain containing 1 0.51 88% 10 FGFR3 Fibroblast growth factor receptor 3 0.51 84% 11 DTL Denticleless homolog (Drosophila) 0.52 85% 12 NUSAP1 Nucleolar and spindle-associated protein 1 0.52 85% 13 PRC1 Protein regulator of cytokinesis 1 0.53 88% 14 Hs.159264 Human clone 23948 mRNA sequence 0.53 83% 15 KIAA0101 KIAA0101 0.53 90% 16 SUSD3 Sushi domain containing 3 0.54 89% 17 AGR2 Anterior gradient homolog 2 (Xenopus laevis)0.5584% 18 CCNB2 Cyclin B2 0.56 85% 19 ASPM Asp (abnormal spindle) homolog, microcephaly associated 0.56 89% 20 CDCA5 Cell division cycle associated 5 0.57 92% 22 MYB V-myb myeloblastosis viral oncogene homolog (avian) 0.58 85% 23 STC2 Stanniocalcin 2 0.58 76% 36 STC1 Stanniocalcin 1 0.65 77% 43 MCM4 Minichromosome maintenance complex component 4 0.66 83% 44 GREB1 GREB1 protein 0.66 84% 45 IL17RB Interleukin 17 receptor B 0.66 85% 46 CCNB1 Cyclin B1 0.67 85% 47 PBK PDZ-binding kinase 0.67 88% 50 MYC V-myc myelocytomatosis viral oncogene homolog (avian) 0.67 78% Upregulated genes 1 LOC651278 Similar to serine hydrolase-like 2 (LOC651278). 1.61 82% 2 THRA Thyroid hormone receptor, alpha (v-erb-a) 1.56 82% 3 PIB5PA Phosphatidylinositol (4,5) bisphosphate 5-phosphatase, A 1.47 80% 4 FBLN1 Fibulin 1 1.45 80% 5 TGFB3 Transforming growth factor, beta 3 1.32 76% 6 PLEKHF1 Pleckstrin homology domain containing, family F member 1 1.30 80% 7 Hs.202577 cDNA FLJ34585 fis, clone KIDNE2008758 1.30 79% 8 DCN Decorin 1.30 78% 9 IQGAP2 IQ motif containing GTPase-activating protein 2 1.28 80% 10 TINF2 TERF1 (TRF1)-interacting nuclear factor 2 1.27 77% 11 CRY2 Cryptochrome 2 (photolyase-like) 1.27 82% 12 DBP D site of albumin promoter (albumin D-box)-binding protein 1.22 80% 13 DHRS10 Hydroxysteroid (17-beta) dehydrogenase 14 1.18 79% 14 SCARB2 Scavenger receptor class B, member 2 1.14 78% 15 CPXM Carboxypeptidase X (M14 family), member 1 1.59 76% 16 HTRA1 HtrA serine peptidase 1 1.39 73% 17 BCAS1 Breast carcinoma–amplified sequence 1 1.33 77% 18 ACACB Acetyl-Coenzyme A carboxylase beta 1.33 76% 19 OLFML1 Olfactomedin-like 1 1.32 79% 20 CDKN1C Cyclin-dependent kinase inhibitor 1C (p57, Kip2) 1.28 78%

NOTE: Genes were selected by class comparison analysis at a FDR of 1%. All genes shown in this table have univariate P 1 107. The top 20 genes downregulated by anastrozole plus selected additional downregulated genes of interest and the top 20 genes upregulated are shown.

www.aacrjournals.org Clin Cancer Res; 19(10) May 15, 2013 2777

Dunbier et al.

(23). Geneset comparison analysis (24) was conducted prominent. Pathway analysis using Geneset Comparison using Biocarta pathways (25). (24) revealed that 19 of the top 30 statistically significantly Multiple correlation analysis was conducted in BRB-Array altered pathways were proliferation-associated (Table 2). Tools. A statistical significance level for each gene for testing Pathways related to estrogen signaling, apoptosis, and the the hypothesis that the Spearman’s correlation between gene complement pathway also changed. expression and change in Ki67 was calculated to be 0 and P values were then used in a multivariate permutation test (21) Tumors exhibit heterogeneity in transcriptional from which false discovery rates (FDR) were computed. response to estrogen deprivation Proportional 2-week change in Ki67 was defined as (2-week To visualize the degree of variability between the 81 Ki67/baseline Ki67) 100. Pathway analysis using Ingenu- tumors in their transcriptional response to estrogen depri- ity Pathways Analysis (Ingenuity Systems) was conducted on vation, we conducted unsupervised hierarchical clustering P the list of genes correlated with 0.005. Other statistical analysis of the values representing the change (posttreat- analyses were conducted in SPSS for Windows (SPSS Inc.), ment baseline) in expression of each of the 1,327 genes and Graphpad Prism (Graphpad Software Inc.). upon aromatase inhibitor treatment (Fig. 1). No single Multivariable analyses were undertaken with log propor- gene was up- or downregulated in all tumors; the most tional change in Ki67 as the dependent variable and the consistently altered gene TOP2A was downregulated in 94% ESR1 immune metagene, , ER H-score, and grade as inde- of cases. No gene was upregulated in more than 82% of pendent variables. Backward selection was used. Cases with cases. missing values for any of the variables in the model were No strong patterns emerged according to this clustering excluded from analysis. for PgR or HER2 status or for DKi67 (Fig. 1B). Most tumors Normalized data from the validation set (Edinburgh) showed strong downregulation of classical estrogen-regu- were downloaded from Gene Expression Omnibus (ref. 26; lated genes (ERG; Fig. 1C) and proliferation-associated ¼ Accession number GSE20181). Expression of the inflam- genes (Fig. 1D). However, while a small number of tumors matory metagene was derived by extracting data for the showed consistently poor suppression of proliferation- component genes from the normalized series matrix file. associated genes, variability in the suppression of ERGs Ki67 was obtained from a previous study of these samples differed markedly between tumors (Fig. 1C). The majority (27). of tumors showed consistent upregulation of the clusters of Exploratory analysis of the type of immune cell repre- genes coding for collagens and chemokines (Fig. 1E and F). sented by the signature associated with change in Ki67 was Although a small number of tumors with poor Ki67 carried out in prediction analysis of microarrays (PAM; responses also showed lesser increases in the chemokine ref. 28). Relative expression of the correlated genes was and collagen clusters, this inverse association was not seen determined by taking the square of correlation coefficient of in all tumors. Both supervised segregation of the tumors the positively correlated genes and rescaling to the training based on DKi67 and HER2 status and unsupervised clus- population of all immune cell types from the Reference tering of specific clusters showed an association of little Database of Immune Cells (29) from which more than 4 decrease in proliferation genes for those tumors not show- examples were available. Leave-one-out cross-validation ing a decrease in Ki67 but only a subtle difference for HER2- was used to determine the accuracy of the classifier and the positive tumors (Supplementary Fig. S2). normalized expression of the genes correlated with change in Ki67 was entered as an "unknown" to determine the likely identity of the unknown profile. Relationship between pretreatment expression of ESR1 and change in Ki67 This variation in transcriptional response to aromatase Results inhibitor treatment is consistent with the variation Estrogen deprivation induces profound reductions in observed in antiproliferative response as measured by proliferation and estrogen-associated genes immunohistochemical assessment of Ki67 (Fig. 2A). Good quality gene expression data were available at Levels of ESR1 could be expected to contribute to the baseline and 2 weeks postanastrozole treatment from 81 responsiveness of tumors to estrogen deprivation and this matched pairs of samples (Supplementary Fig. S1). Using has previously been shown both by immunohistochem- multiple testing corrected class comparison analysis, 1,327 istry and mRNA analyses in tamoxifen-treated patients genes were identified that were significantly differentially (30–32). In this study, ESR1 expression showed a rela- expressed at an FDR of 1% (Table 1; Supplementary Table tively weak but statistically significant correlation with the S2A). Of these, 926 were downregulated and 401 upregu- proportional 2-week change in Ki67 (Spearman r ¼ lated. Proliferation-associated genes such as TOP2A, 0.29, P ¼ 0.012; Fig. 2B). One patient showing poor CCNB2 and classical estrogen-dependent genes such as change in Ki67, indicated in Fig. 2A with an asterisk, was TFF1 and PDZK1 were highly represented among the down- excluded from further analyses as this patient’s decrease regulated genes. Less consistency in function was observed in plasma estradiol was more than 3 SDs less than the among upregulated genes, however, collagens and stromal mean and hence did not meet criteria for estrogen components including immune-related molecules were deprivation.

2778 Clin Cancer Res; 19(10) May 15, 2013 Clinical Cancer Research

Genomic Proﬁling Predicts Response to Aromatase Inhibitor Treatment

Table 2. Pathways affected by aromatase inhibitor treatment

Pathway description Number of genes LS permutation (P) 1 cdc25 and chk1 regulatory pathway in response to DNA damage 9 0.00001 2 Cyclins and cell-cycle regulation 23 0.00001 3 Estrogen-responsive protein Efp controls cell-cycle 16 0.00001 and breast tumors growth

4 Cell cycle: G1–S checkpoint 27 0.00001

5 Cell cycle: G2–M checkpoint 24 0.00001 6 CDK regulation of DNA replication 18 0.00001 7 How progesterone initiates the oocyte maturation 18 0.00001 8 Sonic hedgehog (SHH) receptor Ptc1 regulates cell cycle 9 0.00001 9 Role of Ran in mitotic spindle regulation 9 0.00001 10 RB tumor suppressor/checkpoint signaling in response to DNA damage 13 0.00001 11 Activation of Src by protein tyrosine phosphatase alpha 10 0.00001 12 Regulation of p27 phosphorylation during cell-cycle progression 12 0.00018 13 p53 signaling pathway 16 0.00021 14 Mechanism of protein import into the nucleus 11 0.0004 15 Downregulated of MTA-3 in ER-negative breast tumors 15 0.00054 16 Role of BRCA1, BRCA2, and ATR in cancer susceptibility 22 0.00084 17 Cycling of Ran in nucleocytoplasmic transport 5 0.00085 18 Tumor suppressor Arf inhibits ribosomal biogenesis 17 0.00112 19 E2F1 destruction pathway 9 0.00122 20 BRCA1-dependent Ub ligase activity 8 0.00152 21 Cyclin E destruction pathway 8 0.00166 22 BTG family proteins and cell-cycle regulation 10 0.0018 23 Cystic ﬁbrosis transmembrane conductance regulator 8 0.00198 (CFTR) and beta 2 adrenergic receptor (b2AR) pathway 24 Classical complement pathway 12 0.00228 25 Stathmin and breast cancer resistance to antimicrotubule agents 17 0.00251 26 p38 MAPK signaling pathway 34 0.00284 27 Regulation of eIF4e and p70 S6 kinase 24 0.00292 28 Internal ribosome entry pathway 6 0.00424 29 Regulation of cell-cycle progression by Plk3 8 0.00438 30 Spliceosomal assembly 14 0.00447 31 Apoptotic DNA fragmentation and tissue homeostasis 10 0.00557

NOTE: Pathways identiﬁed as differentially expressed between pre- and post aromatase inhibitor treatment by gene set comparison analysis are shown. Gray-highlighted rows depict proliferation-associated pathways.

Identification of genes that are correlated with Pathway analysis of the list of 471 correlated probes antiproliferative response to aromatase inhibitors revealed that inflammatory response–related pathways Quantitative trait analysis (QTA) by Spearman correla- were significantly overrepresented (Table 3; Supplementary tion was used to identify 471 genes whose expression in Table S3A). The most significantly overrepresented network baseline tumor biopsies correlated with proportional 2- included modules focused upon Inflammatory Response, week change in Ki67 at P < 0.005 (Table 3; Supplementary Inflammatory Disease, and Immunological Disease (P ¼ Table S2B). The list of genes associated with poor antipro- 6.83 10 23 to 1.83 10 2). liferative response to aromatase inhibitor treatment was dominated by immune-related genes with SLAMF8, a CD2 Validation of correlation of inflammatory response family member, the most highly correlated gene, and TNF, signature with antiproliferative response to aromatase interleukins and receptors and other cytokines all among inhibitors the top 20 genes (Table 3; Fig. 3A). Less consistency of For further analysis, we focused on the 45-gene immune function was observed among the genes predicting for a response signature which was most significantly overrepre- good response to aromatase inhibitors but estrogen signal- sented in the genes correlated with antiproliferative ing–associated genes such as GATA3 and STC2 both fea- response (Supplementary Table S3). The median baseline tured within the top 50 correlated genes (Table 3; Fig. 3B). expression of these genes correlated significantly with

www.aacrjournals.org Clin Cancer Res; 19(10) May 15, 2013 2779

Dunbier et al.

A B IHC Data color key: PgR positive HER2 positive Ki67 increase on treatment

PR Ki67 decrease <50% Ki67 decrease 50–75% Her2 Δ Ki67 decrease>75% Ki67

Sushi domain containing 3 C N-acetyl transferase 1 Leucine-a-2-glycoprotein DA375949 Phorbol-12-myristate induced 1 FLJ14346 Serum/glucocorticoid regulated kinase 3 Cadherin EGF receptor 2 Pl-P3-dependent RAC exchanger 1 Sema doma (Ig) short chain 7-dehydrocholesterol reductase PDZ domain containing 1 PDZ domain containing 1 PDZ domain containing 1 RNA binding motif protein 24 Reticulon 1 Stanniocalcin 1 Retinoic acid induced 2 Acetyl-Coenzyme A oxidase 2 SEC14-like 2 Fructose-2,6-biphosphatase 3 RAB30 member RAS oncogene family RAB30 member RAS oncogene family Interleukin 17 receptor B Transmembrane protease, serine 3 Trefoil factor 1 Trefoil factor 3 Protein kinase Ib Aminocarboxymuconate DC Solute carrier family 27 member 2 CYTP4502B7 pseudogene 1 Testis expressed 14 Heat shock protein 22 protein 8 Tubulin, alpha 3e Tubulin, alpha 2, transcript var 2 Serine peptidase inhibitor, 4 Prostaglandin E synthase

IQ containing GTPase activating 3 D Centromere protein F, 350/400ka Cell division cycle 20 homolog Cell division cycle associated 5 Asp (abnormal spindle) homolog Non-SMC condesin complex G Kinesin family member 20A Protein regulator of cytokinesis 1 Topoisomerase II alpha 170 kDa Ubiquitin-conjugaing enzyme E2C Cyclin B2 Nucleoloar and spindle associated 1 Centrosomal protein 55kDa Aurora kinase A Cell division assoicated 3 TPX2, microtubule-associated, homolog Cyclin B1 Stathmin 1/oncoprotein 18 E2F transcription factor 2 Chromosome 18 ORF 56 Ubiquitin-conjugaing enzyme E2T Exonuclease 1 Fanconi anaemia, group I TTK protein kinase RAD51 associated protein 1 Ubiquitin-like 1 KIAA0101 Kinesin family member 11 Thymidylate synthetase Pituitary tumour-transforming 1 Maternal embryonic leucine zipper kinase Cell division cycle 2, G1 to S and G2 to M Cyclin A2 BUB1 homolog beta Hyaluronan-mediated mobility receptor Discs, large homolog 7 PDZ binding kinase Centromere protein A Cell division cycle 25 homolog C Thyroid hormone receptor interactor 13 Centromere protein E, 313kDa

Fibronectin leucine rich protein 2 E Runt-related transcription factor 1 cDNA clone IMAGE:5922621 Basic helix-loop-helix domain, B Similar to hypothetical protein Transforming growth factor beta 3 Sortillin related receptor 2 Leucine rich repeat containing 17 Nectin homolog Zinc finger protein 521 Coiled-coil domaing containing 80 Chromosome 6 open reading frame 65 Protein associated with Tlr4 Ring finger protein 144A Matrix metallopeptidase 2 Carboxypeptidase Z Procollagen C-endopeptidase enhance Cathepsin K Decorin Secreted frizzled-related protein 2 cDNA clone IMAGE:5261213 CKLF-like 3 Predicted: KIAA1644 protein Anthrax toxin receptor Transforming, coiled-coil containing 1 V-maf musculoaponeurotic fibrosarcoma B Angiopoietin-like 2 Collagen, type II, alpha 1 Collagen, type I, apha 1 Collagen, type VI, alpha 3

Complement factor H F Ra1 GND stimulator-like 1 Transient receptor cation channel, C1 Faftlin family member 2 Chromosome 21 open reading frame 34 Sine oculis binding protein homolog Phosphodiesterase 1A La ribonucleoprotein, member 6 Heat shock 27 kDa protein 2 Sema domain protein Chromosome 6 open reading frame 189 High density lipoprotein-binding protein Claudin 5 LIM domain binding 2 Heat shock 70kD protein 12B MHC, class II, DP beta 1 Phospholipase D family member 3 MHC, class II, DM alpha Complement component 3 GTPase, IMAP family member 4 CXADR adhesion molecule, 1 Chromosome 6 open reading frame 192 Pecanex homolog (Drosophila) Ubiquitin-activating enzyme E1-like Protein phosphatase 1M WAS/WASL interacting protein, 1 Apolipoprotein C-I MHC, class II, DM beta

2780 Clin Cancer Res; 19(10) May 15, 2013 Clinical Cancer Research

Genomic Proﬁling Predicts Response to Aromatase Inhibitor Treatment

Figure 2. Ki67 changes in response to aromatase inhibitor treatment. A, proﬁles of %Ki67 staining before and 2 weeks after aromatase inhibitor treatment as measured by immunohistochemistry. B, relationship between pretreatment expression of ESR1 and change in Ki67.

2-week proportional change in Ki67 (r ¼ 0.44, P 0.0001; n ¼ 52; Fig. 3F). No significant association was observed Fig. 3C). Multivariable analysis including grade, ER H-score, using the anastrozole- or tamoxifen-only subgroups of this and baseline Ki67 found that the inflammatory response cohort (n ¼ 29 and 32 patients, respectively). signature was an independent predictor of change in Ki67 (Supplementary Table S4). In an independent set of 58 Genes associated with response are correlated with the tumors treated with the aromatase inhibitor letrozole (33) transcriptional profile of dendritic cells baseline expression of the immune response metagene was The association between lymphocytic infiltration and significantly correlated with change in Ki67 (r ¼ 0.41, P ¼ change in Ki67 suggests that the gene expression signature 0.0045; Fig. 3D) as in our own data. predicting response to aromatase inhibitors may be derived from infiltrating immune cells, rather the tumor cells them- Exploration of association between lymphocytic selves. To investigate the likely source of the inflammatory infiltration and antiproliferative response to signature predictive of change in Ki67, we defined PAM aromatase inhibitors centroids typifying the main immune cell types using pro- One form of inflammatory activity that can be readily files from publicly available gene expression data (29). assessed in histologic specimens is lymphocytic infiltration. Leave-one-out cross-validation of the centroids identified Two pathologists (A. Nerurkar and P. Osin) identified the hematopoietic stem cells, B cells, T cells, natural killer (NK) presence of visible lymphocytic infiltration in 15 and cells, and dendritic cells with 100% accuracy. Prediction absence in 64 tumors. Consistent with the gene expression analysis using the normalized relative expression of the observations, tumors with lymphocytic infiltration showed genes in our 471-gene response predictor identified the significantly poorer antiproliferative response to aromatase profile as being most closely aligned to that of dendritic inhibitor treatment (Mann–Whitney, P ¼ 0.011, n ¼ 79; Fig. cells (Supplementary Table S5). 3E). The presence of lymphocytic infiltration correlated with expression of the immune response metagene (Sup- Discussion plementary Fig. S3A). In addition, the proportion of sam- In this study, the largest reported of the global transcrip- ples containing a lymphocytic infiltrate in this cohort tional consequences of aromatase inhibitor treatment in increased over the duration of treatment (Supplementary breast tumors, we have identified an inflammatory gene Fig. S3B). We also examined lymphocytic infiltration in a expression signature in baseline samples that is indepen- subset of material from the tamoxifen and anastrozole arms dently associated with poor antiproliferative response to of the IMPACT trial (15). Although the relationship was not neoadjuvant aromatase inhibitor treatment (8, 15). The significant, possibly due to the smaller number assessed, a signature we identified appears to contain the transcrip- similar effect size was observed (Mann—Whitney, P ¼ 0.15, tional fingerprint of infiltrating immune cells, a possibility

Figure 1. Heatmap of changes of expression of genes which alter upon aromatase inhibitor treatment. A, heatmap of changes in 1327 genes differentially expressed at FDR < 1%. B, dendrogram of 81 pairs of tumors color coded with immunohistochemical information corresponding to each sample. Heatmaps of genes comprising the core of the (C) estrogen-associated gene cluster; (D) proliferation-associated gene cluster; (E) ECM cluster; and (F) immune cluster. Red denotes upregulation, green denotes downregulation.

www.aacrjournals.org Clin Cancer Res; 19(10) May 15, 2013 2781

Dunbier et al.

Table 3. Genes and pathways associated with antiproliferative response to anastrozole treatment

Correlation Parametric Rank Gene symbol Description coefﬁcient p-value or score Genes associated with poor response 1 SLAMF8 Signaling lymphocytic activation molecule 8 0.520 3.0E-06 2 P2RY6 Pyrimidinergic receptor P2Y, G-protein coupled, 6 0.507 5.6E-06 3 Hs.370503 mRNA; cDNA DKFZp313O229 0.505 5.9E-06 4 ZBED2 Zinc ﬁnger, BED-type containing 2 0.492 1.11E-05 5 PITPNM1 Phosphatidylinositol transfer protein, 1 0.487 1.40E-05 6 IL21R Interleukin 21 receptor 0.481 1.81E-05 7 LAIR2 Leukocyte-associated immunoglobulin-like R2 0.476 2.29E-05 8 RGS19 Regulator of G-protein signaling 19 0.474 2.54E-05 9 HAVCR2 Hepatitis A virus cellular receptor 2 0.465 3.75E-05 10 IL32 Interleukin 32 0.464 3.92E-05 11 ADAM8 ADAM metallopeptidase domain 8 0.464 3.84E-05 12 PLCL3 Phospholipase C, eta 1 0.464 3.77E-05 13 FPRL2 Formyl peptide receptor-like 2 0.462 4.25E-05 14 LAG3 Lymphocyte activation gene 3 0.461 4.31E-05 15 SGK Serum/glucocorticoid regulated kinase 0.460 4.66E-05 16 TNF Tumor necrosis factor 0.458 4.93E-05 17 Hs.560728 cDNA clone IMAGE:38786 3 0.457 5.12E-05 18 CARD9 Caspase recruitment domain family, member 9 0.457 5.24E-05 19 TRAF3 TNF receptor–associated factor 3 0.452 6.47E-05 20 AKR1B1 Aldo-keto reductase family 1, member B1 0.452 6.38E-05 24 TNFAIP3 Tumor necrosis factor, alpha-induced protein 3 0.438 0.000113 25 CD53 CD53 molecule 0.438 0.000112 33 IRF8 Interferon regulatory factor 8 0.427 0.000173 35 CD86 CD86 molecule 0.426 0.000177 36 IL10RA Interleukin 10 receptor, alpha 0.426 0.000177 37 CD84 CD84 molecule 0.425 0.000184 40 ITGAL Integrin, alpha L 0.423 0.000199 43 IGSF6 Immunoglobulin superfamily, member 6 0.420 0.000226 46 ITGB2 Integrin, beta 2 0.419 0.000228 48 LPXN Leupaxin 0.418 0.000242 49 SLAMF1 Signaling lymphocytic activation molecule 1 0.417 0.000253 50 TNFSF7 CD70 molecule 0.417 0.000251

Genes associated with good response 1 LOC441425 LOC441425 0.522 1.9E-06 2 SLC3A1 Solute carrier family 3, member 1 0.495 7.3E-06 3 LOC645636 LOC645636 (similar to AIP1) 0.470 2.29E-05 4 TMC4 Transmembrane channel-like 4 0.443 7.18E-05 5 Hs.99785 cDNA: FLJ21245 ﬁs, clone COL01184 0.438 8.92E-05 6 EML2 Echinoderm microtubule–associated protein like 2 0.437 9.31E-05 7 C1orf182 Chromosome 1 open reading frame 182 0.433 0.000109 8 GATA3 GATA-binding protein 3 0.431 0.000118 9 LOC388564 Hypothetical gene supported by BC052596 0.428 0.000133 10 C17orf79 Chromosome 17 open reading frame 79 0.425 0.000148 11 GSTO2 Glutathione S-transferase omega 2 0.422 0.000165 12 PEX19 Peroxisomal biogenesis factor 19 0.421 0.000178 13 PAK6 P21(CDKN1A)-activated kinase 6 0.417 0.000206 14 Hs.570838 FLJ36037 0.408 0.000284 15 Hs.58384 BX090362 0.407 0.000292 16 NME5 Non-metastatic cells protein 5 0.406 0.000302 17 HIST1H2AC Histone cluster 1, H2ac 0.406 0.000306 (Continued on the following page)

2782 Clin Cancer Res; 19(10) May 15, 2013 Clinical Cancer Research

Genomic Proﬁling Predicts Response to Aromatase Inhibitor Treatment

Table 3. Genes and pathways associated with antiproliferative response to anastrozole treatment (Cont'd )

Correlation Parametric Rank Gene symbol Description coefﬁcient p-value or score 18 CYB5D2 Cytochrome b5 domain containing 2 0.403 0.000339 19 CELSR2 Cadherin, EGF LAG seven-pass G-type receptor 2 0.403 0.00034 20 POLR2L Polymerase (RNA) II polypeptide L, 7.6kDa 0.400 0.000381 50 STC2 Stanniocalcin 2 0.361 0.001688

Top networks associated with change in Ki67 1Inflammatory response, inflammatory disease, immunological disease 36 2 Hematologic system development and function, tissue morphology, 32 cell-to-cell signaling and interaction 3Inflammatory response, cell-to-cell signaling and interaction, 21 hematologic system development and function 4 Gene expression, cellular growth and proliferation, cell morphology 21 5 Cell-to-cell signaling and interaction, hematologic system development 19 and function, immune cell trafficking

NOTE: The top 20 genes associated with poor response, selected immune-related genes from the top 50, the top 20 genes associated with good response and the top gene networks as defined by Ingenuity Pathway Analysis of the complete list of 471 genes are shown. that is supported by our observation that tumors with However, further work is needed to define the contribution detectable lymphocytic infiltration appear to obtain lesser of these cells to aromatase inhibitor response. benefit from aromatase inhibitors. Importantly, the rela- The prominence of the immune system as a determinant tionship of the signature with change in Ki67 was validated of resistance to endocrine therapy contrasts with evidence in an independent set of tumors. from cell lines of growth factor pathways as the major The immune system has conflicting potential roles in determinants (4–7). This is readily explained by the cell both suppressing tumor growth by destroying cancer cells line work, even if conducted in rodent models, excluding and promoting tumors through the production of cytokines human stromal components. It is also clear, however, that and growth factors (34). Investigation of the association immune influences as characterized by the current work can between tumor-associated lymphocytes and response to explain resistance in only a proportion of patients and less neoadjuvant chemotherapy has revealed that the presence statistically prominent pathways may be important deter- of lymphocytes is an independent predictor of good minants of resistance. Larger patient numbers and a greater response to cytotoxic chemotherapy (35) in patients with focus on the pathways activated in individual patients are breast cancer. Similarly, Mahmoud and colleagues recently required to define the proportional influence of the various showed that tumor-infiltrating CD8þ lymphocytes are indi- putative mechanisms. The public availability of our com- cators of good clinical outcome in patients with ER breast prehensive molecular characterization of this set of tissues cancer (36). Our observation that infiltrating immune cells from a carefully monitored clinical trial provides a reference are associated with poor response to endocrine therapy service for other experimentalists to assess the clinical suggests that their role may be significantly different in relevance for their findings in relation to estrogen depriva- tumors treated with endocrine therapy as opposed to che- tion therapy. motherapy and between the ERþ and ER subgroups. The The changes in gene expression were very substantial but differential cytotoxicities of chemotherapy and endocrine variable between tumors as previously reported (40, 41). therapy on lymphocytes and the effect of estrogen on the Some clusters of proliferation, estrogen-regulated, extracel- immune system may partially explain these differences with lular matrix (ECM), and immune genes were obvious, but chemotherapeutic agents potentially killing infiltrating within these clusters, major variability between the changes immune cells, hence negating their proproliferative effect. in gene expression existed. While proliferation gene changes Exploratory analysis of the inflammatory signature sug- correlated with immunohistochemical Ki67 changes, gested that dendritic cells could be involved in the poor changes in estrogen-regulated genes showed little obvious response of tumors with high expression of the signature to pattern but this may be due to almost all patients showing estrogen deprivation. Dendritic cells have been implicated suppression of nearly all of these genes and thus creating in promoting breast tumorigenesis by polarizing CD4þ T limited opportunity for segregation of patient groups on cells (37) and hence could aid resistance through this this basis. The most consistently downregulated gene was manner. In addition, a dendritic cell metagene (38) has TOP2A, a key target for anthracycline-based therapies. been shown to be associated with endocrine resistance in This observation supports the current practice of sequenc- high proliferation, high estrogen-related score tumors (39). ing aromatase inhibitor treatment after chemotherapy,

www.aacrjournals.org Clin Cancer Res; 19(10) May 15, 2013 2783

Dunbier et al.

Figure 3. Immune involvement and antiproliferative response to Ki67. A, scatter plot of relationship between pretreatment expression of SLAMF8 and proportional change in Ki67. B, relationship between pretreatment expression of GATA3 and proportional change in Ki67. C, relationship between pretreatment expression of the inflammatory response metagenes and proportional change in Ki67 in our discovery cohort. D, relationship between pretreatment expression of the inflammatory response metagenes and proportional change in Ki67 in the Edinburgh patient cohort. E, plot of proportional change in Ki67 in tumors with detectable lymphocytic infiltration versus those without in our discovery cohort. F, plot of proportional change in Ki67 in tumors with lymphocytic infiltration versus those with no detectable lymphocytic infiltration in the IMPACT series. Dotted lines are shown at 100% (no change in Ki67 after treatment) and 50% (data points above this have a less than 50% decrease in Ki67 and as such are defined as nonresponders; ref. 10).

particularly anthracyclins which directly target Topo- with 1 of the 3 third-generation aromatase inhibitors found isomerase-2. It should be noted, however, that proapop- that both luminal A and B subgroups responded, well totic genes are also downregulated, hence the balance of although the luminal B group remained at poorer prognosis molecular effects in relation to their benefit or detriment posttreatment (12). This study did not report on the explo- for combination with specific chemotherapy is difficult to ration of other signatures to segregate responsive and non- predict. responsive subgroups. A small number of other groups have explored genomic The use of Ki67 change as an intermediate endpoint for profiles in relation to clinical response to aromatase inhi- treatment benefit may be considered a limitation of this bitors. Consistent with the current report, Miller and col- study. However, this approach allows the identification of leagues (40) found that the most prominent downregulated genes/signatures that are specific to response. In contrast, genes by aromatase inhibitors, in their case, letrozole, to be clinical response requires objective regression of tumors those related to proliferation and those that are increased to and this has dependence on initial growth rate as well the include many stroma-related genes. However, in that study, anti-growth effects of therapy. In addition, the continuous the genes most associated with resistance to the aromatase nature of Ki67 expression provides greater statistical power inhibitors were related to cellular biosynthetic processes, in than the categorical nature of clinical response analyses. particular those coding for ribosomal proteins (33). This Finally, as pointed out above, Ki67 on treatment is a better difference from the current set of findings may be due to predictor for long-term outcome than clinical response to clinical response rather than Ki67 change being the end- endocrine treatment (8, 15). point for response in the Miller study: the need to get In conclusion, as well as immune-related functions hav- shrinkage of tumor for assignment of response may place ing recently described prognostic value and predictive value an increased emphasis on metabolic processes. A recent for chemotherapy in early breast cancer, they are related to study of 377 patients randomized to neoadjuvant treatment reduced antiproliferative response to aromatase inhibitor

2784 Clin Cancer Res; 19(10) May 15, 2013 Clinical Cancer Research

Genomic Proﬁling Predicts Response to Aromatase Inhibitor Treatment

therapy. Further study is needed to determine whether these Writing, review, and/or revision of the manuscript: A.K. Dunbier, Z. Ghazoui, H. Anderson, A. Nerurkar, R. A’hern, W.R. Miller, I.E. Smith, M. relationships are causative and therefore potentially subject Dowsett to intervention. Administrative, technical, or material support (i.e., reporting or orga- nizing data, constructing databases): Z. Ghazoui, H. Anderson, J. Salter Study supervision: A.K. Dunbier, M. Dowsett Disclosure of Potential Conﬂicts of Interest H. Anderson has ownership interest (including patents) in shares in AstraZeneca. M. Dowsett has a commercial research grant, honoraria from Grant Support speakers’ bureau, and is a consultant/advisory board member of AstraZe- A.K. Dunbier, H. Anderson, and Z. Ghazoui were supported by the Mary- neca. No potential conﬂicts of interest were disclosed by the other authors. Jean Mitchell Green Foundation. This work was also supported by a Break- through Breast Cancer Research Grant (to M. Dowsett), a Health Research Council of New Zealand Sir Charles Hercus Fellowship (to A.K. Dunbier), Authors' Contributions and National Health Service funding to the NIHR Biomedical Research Conception and design: A.K. Dunbier, Z. Ghazoui, P. Osin, M. Dowsett Centre. Development of methodology: A.K. Dunbier, Z. Ghazoui, H. Anderson The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked Acquisition of data (provided animals, acquired and managed patients, advertisement provided facilities, etc.): A.K. Dunbier, Z. Ghazoui, H. Anderson, A. in accordance with 18 U.S.C. Section 1734 solely to indicate Nerurkar, W.R. Miller, I.E. Smith this fact. Analysis and interpretation of data (e.g., statistical analysis, biosta- tistics, computational analysis): A.K. Dunbier, Z. Ghazoui, H. Anderson, Received March 26, 2012; revised February 11, 2013; accepted March 1, R. A’hern, W.R. Miller, M. Dowsett 2013; published OnlineFirst March 14, 2013.

References 1. Mouridsen H, Gershanovich M, Sun Y, Perez-Carrion R, Boni C, 12. Ellis MJ, Suman VJ, Hoog J, Lin L, Snider J, Prat A, et al. Randomized Monnier A, et al. Superior efficacy of letrozole versus tamoxifen phase II neoadjuvant comparison between letrozole, anastrozole, and as first-line therapy for postmenopausal women with advanced exemestane for postmenopausal women with estrogen receptor-rich breast cancer: results of a phase III study of the International stage 2 to 3 breast cancer: clinical and biomarker outcomes and Letrozole Breast Cancer Group. J Clin Oncol 2001;19:2596– predictive value of the baseline PAM50-based intrinsic subtype–ACO- 606. SOG Z1031. J Clin Oncol 2011;29:2342–9. 2. Nabholtz JM, Buzdar A, Pollak M, Harwin W, Burton G, Mangalik A, 13. Baum M, Budzar AU, Cuzick J, Forbes J, Houghton JH, Klijn JG, et al. et al. Anastrozole is superior to tamoxifen as first-line therapy for Anastrozole alone or in combination with tamoxifen versus tamoxifen advanced breast cancer in postmenopausal women: results of a North alone for adjuvant treatment of postmenopausal women with early American multicenter randomized trial. Arimidex Study Group. J Clin breast cancer: first results of the ATAC randomised trial. Lancet Oncol 2000;18:3758–67. 2002;359:2131–9. 3. Bonneterre J, Thurlimann B, Robertson JF, Krzakowski M, Mauriac L, 14. Goss PE, Ingle JN, Pritchard KI, Ellis MJ, Sledge GW, Budd GT, et al. Koralewski P, et al. Anastrozole versus tamoxifen as first-line therapy Exemestane versus anastrozole in postmenopausal women with early for advanced breast cancer in 668 postmenopausal women: results of breast cancer: NCIC CTG MA.27–A randomized controlled phase III the Tamoxifen or Arimidex Randomized Group Efficacy and Tolera- trial. J Clin Oncol 2013;31:1398–404. bility study. J Clin Oncol 2000;18:3748–57. 15. Smith IE, Dowsett M, Ebbs SR, Dixon JM, Skene A, Blohmer JU, et al. 4. Miller TW, Hennessy BT, Gonzalez-Angulo AM, Fox EM, Mills GB, Neoadjuvant treatment of postmenopausal breast cancer with anastro- Chen H, et al. Hyperactivation of phosphatidylinositol-3 kinase pro- zole, tamoxifen, or both in combination: the Immediate Preoperative motes escape from hormone dependence in estrogen receptor-pos- Anastrozole, Tamoxifen, or Combined with Tamoxifen (IMPACT) mul- itive human breast cancer. J Clin Invest 2010;120:2406–13. ticenter double-blind randomized trial. J Clin Oncol 2005;23:5108–16. 5. Chan CM, Martin LA, Johnston SR, Ali S, Dowsett M. Molecular 16. Dowsett M, Smith IE, Ebbs SR, Dixon JM, Skene A, Griffith C, et al. changes associated with the acquisition of oestrogen hypersensitivity Short-term changes in Ki-67 during neoadjuvant treatment of primary in MCF-7 breast cancer cells on long-term oestrogen deprivation. breast cancer with anastrozole or tamoxifen alone or combined cor- J Steroid Biochem Mol Biol 2002;81:333–41. relate with recurrence-free survival. Clin Cancer Res 2005;11:951s–8s. 6. Martin LA, Farmer I, Johnston SR, Ali S, Marshall C, Dowsett M. 17. Smith IE, Walsh G, Skene A, Llombart A, Mayordomo JI, Detre S, et al. A Enhanced estrogen receptor (ER) alpha, ERBB2, and MAPK signal phase II placebo-controlled trial of neoadjuvant anastrozole alone or transduction pathways operate during the adaptation of MCF-7 cells with gefitinib in early breast cancer. J Clin Oncol 2007;25:3816–22. to long term estrogen deprivation. J Biol Chem 2003;278:30458–68. 18. Dunbier AK, Anderson H, Ghazoui Z, Folkerd EJ, A'Hern R, Crowder 7. Lupien M, Meyer CA, Bailey ST, Eeckhoute J, Cook J, Westerling T, RJ, et al. Relationship between plasma estradiol levels and estrogen- et al. Growth factor stimulation induces a distinct ER(alpha) cistrome responsive gene expression in estrogen receptor-positive breast can- underlying breast cancer endocrine resistance. Genes Dev 2010;24: cer in postmenopausal women. J Clin Oncol 2010;28:1161–7. 2219–27. 19. Du P, Kibbe WA, Lin SM. lumi: a pipeline for processing Illumina 8. Dowsett M, Smith IE, Ebbs SR, Dixon JM, Skene A, A'Hern R, et al. microarray. Bioinformatics 2008;24:1547–8. Prognostic value of Ki67 expression after short-term presurgical 20. Synapse [cited 2013 Mar 1]. Available from: http://synapse.sagebase endocrine therapy for primary breast cancer. J Natl Cancer Inst 2007; .org/#Dataset:16243. 99:167–70. 21. Korn EL, Troendle JF, McShane LM, Simon R. Controlling the number 9. Dowsett M, Ebbs SR, Dixon JM, Skene A, Griffith C, Boeddinghaus I, of false discoveries: application to high-dimensional genomic data. et al. Biomarker changes during neoadjuvant anastrozole, tamoxifen, or J Stat Plann Inference 2004;124:379–98. the combination: influence of hormonal status and HER-2 in breast 22. Simon R, Team B-AD. BRB-ArrayTools 2011; [cited 2011 Mar 12]. cancer-a study from the IMPACT trialists. J Clin Oncol 2005;23:2477–92. Available from: http://linus.nci.nih.gov/BRB-ArrayTools.html. 10. Dowsett M, Smith I, Robertson J, Robison L, Pinhel I, Johnson L, et al. 23. Wright GW, Simon RM. A random variance model for detection of Endocrine therapy, new biologicals, and new study designs for pre- differential gene expression in small microarray experiments. Bioin- surgical studies in breast cancer. J Natl Cancer Inst Monogr 2011; formatics 2003;19:2448–55. 2011:120–3. 24. Efron B, Tibshirani R. On testing the significance of sets of genes. Ann 11. Ellis MJ, Coop A, Singh B, Tao Y, Llombart-Cussac A, Janicke F, et al. Appl Stat 2007;1:107–29. Letrozole inhibits tumor proliferation more effectively than tamoxifen 25. Biocarta. [cited 2012 Jul 11]. Available from: http://cgap.nci.nih.gov/ independent of HER1/2 expression status. Cancer Res 2003;63:6523–31. Pathways/BioCarta_Pathways.

www.aacrjournals.org Clin Cancer Res; 19(10) May 15, 2013 2785

Dunbier et al.

26. National Centre for Biotechnology Information. Gene Expression clinically responsive or resistant to letrozole. J Clin Oncol 2009;27: Omnibus. [cited 2010 Nov 20]. Available from: www.ncbi.nlm.nih. 1382–7. gov/geo/. 34. Schreiber RD, Old LJ, Smyth MJ. Cancer immunoediting: integrating 27. Miller TW, Balko JM, Ghazoui Z, Dunbier A, Anderson H, Dowsett immunity's roles in cancer suppression and promotion. Science M, et al. A gene expression signature from human breast cancer 2011;331:1565–70. cells with acquired hormone independence identifies MYC as a 35. Denkert C, Loibl S, Noske A, Roller M, Muller BM, Komor M, et al. mediator of antiestrogen resistance. Clin Cancer Res 2011;17: Tumor-associated lymphocytes as an independent predictor of 2024–34. response to neoadjuvant chemotherapy in breast cancer. J Clin Oncol 28. Tibshirani R, Hastie T, Narasimhan B, Chu G. Diagnosis of multiple 2010;28:105–13. cancer types by shrunken centroids of gene expression. Proc Natl 36. Mahmoud SM, Paish EC, Powe DG, Macmillan RD, Grainge MJ, Lee Acad Sci U S A 2002;99:6567–72. AH, et al. Tumor-infiltrating CD8þ lymphocytes predict clinical out- 29. Hijikata A, Kitamura H, Kimura Y, Yokoyama R, Aiba Y, Bao Y, et al. come in breast cancer. J Clin Oncol 2011;29:1949–55. Construction of an open-access database that integrates cross-ref- 37. Aspord C, Pedroza-Gonzalez A, Gallegos M, Tindle S, Burton EC, Su D, erence information from the transcriptome and proteome of immune et al. Breast cancer instructs dendritic cells to prime interleukin 13- cells. Bioinformatics 2007;23:2934–41. secreting CD4 þT cells that facilitate tumor development. J Exp Med 30. Byar DP, Sears ME, McGuire WL. Relationship between estrogen 2007;204:1037–47. receptor values and clinical data in predicting the response to endo- 38. Bianchini G, Qi Y, Alvarez RH, Iwamoto T, Coutant C, Ibrahim NK, et al. crine therapy for patients with advanced breast cancer. Eur J Cancer Molecular anatomy of breast cancer stroma and its prognostic value in 1979;15:299–310. estrogen receptor-positive and -negative cancers. J Clin Oncol 31. EBCTCG. Tamoxifen for early breast cancer: an overview of the 2010;28:4316–23. randomised trials. Early Breast Cancer Trialists' Collaborative Group. 39. Bianchini G, Pusztai L, Qi Y, Iwamoto T, Kelly CM, Zambetti M, et al. A Lancet 1998;351:1451–67. dendritic metagene that predicts prognosis and endocrine resistance 32. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, et al. Expression of in breast cancer. J Clin Oncol 2012;30:545. the 21 genes in the Recurrence Score assay and tamoxifen clinical 40. Miller WR, Larionov AA, Renshaw L, Anderson TJ, White S, Murray J, benefit in the NSABP study B-14 of node negative, estrogen receptor et al. Changes in breast cancer transcriptional profiles after treatment positive breast cancer [abstract nr 510]. In: Proceedings of the Amer- with the aromatase inhibitor, letrozole. Pharmacogenetics Genomics ican Society for Oncology Annual Meeting; 2005 May 13–17; Orlando, 2007;17:813–26. FL. Alexandria, VA: ASCO; 2005. 41. Mackay A, Urruticoechea A, Dixon JM, Dexter T, Fenwick K, Ashworth 33. Miller WR, Larionov A, Renshaw L, Anderson TJ, Walker JR, Krause A, A, et al. Molecular response to aromatase inhibitor treatment in primary et al. Gene expression profiles differentiating between breast cancers breast cancer. Breast Cancer Res 2007;9:R37.

2786 Clin Cancer Res; 19(10) May 15, 2013 Clinical Cancer Research

Molecular Profiling of Aromatase Inhibitor−Treated Postmenopausal Breast Tumors Identifies Immune-Related Correlates of Resistance

Anita K. Dunbier, Zara Ghazoui, Helen Anderson, et al.

Clin Cancer Res 2013;19:2775-2786. Published OnlineFirst March 14, 2013.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-12-1000

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2013/03/14/1078-0432.CCR-12-1000.DC1

Cited articles This article cites 36 articles, 21 of which you can access for free at: http://clincancerres.aacrjournals.org/content/19/10/2775.full#ref-list-1

Citing articles This article has been cited by 20 HighWire-hosted articles. Access the articles at: http://clincancerres.aacrjournals.org/content/19/10/2775.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at Subscriptions [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/19/10/2775. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.