Oncogene (2007) 26, 6269–6279 & 2007 Nature Publishing Group All rights reserved 0950-9232/07 $30.00 www.nature.com/onc ORIGINAL ARTICLE Oxidative stress pathways highlighted in tumor cell immortalization: association with breast cancer outcome

SH Dairkee1, M Nicolau2,3, A Sayeed1, S Champion1,YJi2,4, DH Moore1, B Yong1,5, Z Meng1 and SS Jeffrey2

1California Pacific Medical Center Research Institute, San Francisco, CA, USA and 2Department of Surgery, Stanford University School of Medicine, Stanford, CA, USA

An improved understanding of cell immortalization and its Keywords: primary tumor culture; telomerase (hTERT); manifestation in clinical tumors could facilitate novel expression therapeutic approaches. However, only rare tumor cells, which maintain telomerase expression in vitro, immorta- lize spontaneously. By expression-profiling analyses of limited-life primary breast tumor cultures pre- and post- hTERT transduction, and spontaneously immortalized Introduction breast cancer cell lines, we identified a common signature characteristic of tumor cell immortalization. A predomi- nant feature of this immortalization signature (ImmSig) Cell immortalization is a hallmark of cancer, induced by was the significant overexpression of physical, chemical and biological agents. An in-depth . In contrast to epithelial cells derived from low understanding of the molecular pathways leading to this histologic grade primary tumors, which required hTERT phenotype, and the variability inherent in its acquisition, transduction for the acquisition of ImmSig, spontaneously could improve clinical cancer management. Reactiva- immortalizing high-grade tumor cultures displayed similar tion of telomerase for maintaining the telomeric ends of molecular changes independent of exogenous hTERT. is an essential prerequisite towards the Silencing the hTERT gene reversed ImmSig expression, immortalization of proliferating tumor cells (Counter increased cellular reactive oxygen species levels, altered et al., 1992). Conversely, the absence of telomerase mitochondrial membrane potential and induced apoptotic activity in model systems results in replicative senescence and proliferation changes in immortalized cells. In clinical (reviewed in Ben-Porath and Weinberg, 2005), which is breast cancer samples, cell-proliferation-pathway genes reversed by ectopic expression of the catalytic subunit of were significantly associated with ImmSig. In these cases, telomerase, hTERT (Bodnar et al., 1998). In addition to ImmSig expression itself was inversely correlated with the classical role of telomerase in telomere elongation, patient survival (P ¼ 0), and was particularly relevant to evidence suggests that it enhances cell survival under the outcome of estrogen receptor-positive tumors. Our conditions of cellular stress by regulating the expression data support the notion that ImmSig assists in surmount- of proliferation and apoptosis-inducing genes (Xiang ing normal barriers related to oxidative and replicative et al., 2002; Zhang et al., 2003; Dudognon et al., 2004), stress response. Targeting a subset of aggressive breast suppresses cell differentiation (Wang et al., 2005), and cancers by reversing ImmSig components could be a contributes to the overall maintenance of chromatin practical therapeutic strategy. and DNA-damage response (Masutomi et al., 2005). Oncogene (2007) 26, 6269–6279; doi:10.1038/sj.onc.1210452; hTERT-induced immortalization of normal human published online 30 April 2007 mammary epithelial cells (HMEC) confers resistance to TGF-b growth inhibition (Stampfer et al., 2001), and reduction in growth factor requirements (Smith Correspondence: Dr SH Dairkee, California Pacific Medical Center, et al., 2003). Research Institute, 475Brannan Street, San Francisco, CA 94107, Our previous work with primary breast tumor cell USA or Dr SS Jeffrey, Department of Surgery, Stanford University culture demonstrated that immortality was not a School of Medicine, Medical School Lab Surge Building, Room P214, constitutive trait of all malignant cells in vitro (Dairkee 1201 Welch Road, M/C 5494, Stanford, CA 94305-5494, USA. et al., 2004). Molecular changes underlying finite mitotic E-mail: [email protected] or [email protected] 3Current address: M Nicolau, Department of Mathematics, Stanford life included high transcript levels of TGFb, and other University, Stanford, CA 94305, USA. negative growth regulating genes accompanied by the 4Current address: Y Ji, Genentech, Inc., 1 DNA Way, South San extinction of telomerase expression and catalytic acti- Francisco, CA 94080, USA. vity. Our principal component analyses based on gene 5Current address: B Yong, Pfizer Pharmaceuticals Limited, 1168 Nan Jing Road, Shanghai 200041, PR China. expression patterns demonstrated that primary tumor Received 5December 2006; revised 23 February 2007; accepted 1 March cultures lacking hTERT clustered apart from short-term 2007; published online 30 April 2007 HMEC, as well as from those rare primary tumor Immortalization signature in breast cancer SH Dairkee et al 6270 cultures that immortalized spontaneously and displayed To induce cell immortality in NSI tumor cultures, we hTERT expression comparable to that of established considered that since they already harbor multiple cancer cell lines. Consistent with the reported literature changes associated with malignant transformation (Kurz et al., 2004), it appears that high oxygen levels, (references cited above), the abrogation of senescence characteristic of routine culture conditions, generate barriers, a characteristic of normal HMEC in vitro, oxidative stress resulting in telomerase extinction and might not be required. Indeed, we observed that the replicative senescence in most primary tumor cultures. introduction of hTERT alone was sufficient to immor- Thus, we reasoned that hTERT transduction might talize cells isolated from NSI tumors (three grade I, five rescue such tumor cell cultures from a suboptimal grade II). hTERT-transduced tumor cells were com- microenvironment, and confer immortality in a single pared with vector only controls for the ability to step. We have identified an immortalization signature proliferate continuously. While NSI control cells dis- (ImmSig) in tumor cell cultures, and demonstrated its played proliferation to p7 passages (B15population manifestation in tumor tissue. Instead of routine doublings), matched hTERT-transduced cells currently proliferation changes associated with immortalization, average 60 passages (B125population doublings, we found that ImmSig was largely composed of genes Supplementary Table 1). Whereas hTERC transcript promoting cellular oxidoreductase activity. These find- levels were relatively unchanged in transduced cultures, ings suggest that although continuous cell proliferation hTERT expression was up to 4000-fold greater than may portray the final effect of immortalization, ImmSig those of control cultures (Figure 1b) resulting in robust identifies the symptomatic cellular changes leading to cell growth (Figure 1c). This indicated that for the it. The application of the unique ImmSig profile to a emergence of cell immortality in NSI cultures, (a) clinical breast cancer database revealed patient subsets endogenous hTERC expression adequately supported that might benefit from anti-telomerase therapies and/or the experimentally increased telomerase activity (data redox modulators in conjunction with other cancer not shown), and/or (b) hTERT overexpression was treatments. largely independent of its DNA synthesis function wherein hTERC served as its template for the addition of new telomeric repeats.

Results Global gene expression changes associated with hTERT induced the immortalization of primary breast immortalization of primary breast tumor cultures tumor cultures Nine isogenic sets of primary tumor cultures pre- and Breast tumor tissue propagated by conventional post-hTERT-transduction were assayed using 42 000 approaches generally results in cultures biologically feature cDNA microarrays. Primary tumor cell cultures resembling normal rather than malignant epithelial without exogenous hTERT were designated as tumor cells. We previously implemented novel methods for control (TC), and those transduced with hTERT,as tissue dissociation and processing to enrich and selec- tumor transduced (TT). Significance analysis of micro- tively expand tumor cells. A major advantage of our array (SAM)-based comparisons were made between TC modified protocols is that they allow invasive tumor and TT cultures of NSI cases. At a median false cells in the tissue to be fractionated, and non-malignant discovery rate (FDR) of 6.70%, 5164 Unigene IDs epithelium confined by an intact basement membrane deemed significant by SAM were selected (3027 up, and to be excluded. Tumor cultures resulting from such 2137 down genes in TT cultures) and applied towards methodological improvements closely resemble the (GO) analysis (Boyle et al., 2004) to original tumor tissue in characteristics, such as the define the underlying biology. Additionally, correspond- presence of aneuploidy, clonal loss of heterozygosity, ing to the minimum FDR (median ¼ 0.08%), 597 genes clonal p53 mutations, ERBB2 amplification and over- (427 up, and 170 down genes in TT cultures) were expression, and in global gene expression patterns as retrieved and used to compute a centroid, designated as measured by cDNA microarrays (Dairkee et al., 1995, the ImmSig. Genes representing this signature were 1997, 2004; Li et al., 1998). In a comprehensive concordantly induced or repressed in hTERT-immorta- quantitative real-time reverse transcription–polymerase lized primary cultures of low- and intermediate-grade chain reaction (QPCR) analysis (Figure 1a) of tumor tumors (Figure 2a). hTERT showed the highest level of cells cultured from primary breast cancers of varying expression (>100-fold) as expected for an experimen- degrees of clinical aggressiveness, we demonstrate here tally transduced gene. Other ImmSig genes were that high transcript levels of hTERT, and hTERC (the upregulated up to sixfold, or downregulated down to RNA component of telomerase), were maintained only 10-fold (Supplementary Dataset 1). GO annotation by high histologic grade-derived tumor cells, which gave analysis of upregulated ImmSig genes confirmed that rise to spontaneously immortalizing (SI) cultures (3/3). in response to immortalization cues, induction of genes Epithelial cultures of low- and intermediate-grade occurred in pathways encompassing multiple mitochon- tumors, where expression of both genes was below the drial functions, in particular, oxidoreductase activity baseline level of non-malignant epithelium, displayed (COX5A, CBR3, OGDH, multiple isoforms of NADH limited proliferation (21/21). These were designated as dehydrogenase (ubiquinone) 1 alpha and beta subcom- non-spontaneously immortalizing (NSI) tumors. plexes, NQO2), catalytic activity (PCYT2, ATP5G2,

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6271 Expression of telomerase in primary tumor cells and immortal cel llines

Cell lines Primary tumor cultures Cell lines Primary tumor cultures Relative expression (fold)

Primary Immortal

b Expression pre-and post-hTERT c Microscopic appearance transduction TC TT Tumor Control hTERT Transf. grade I Relative expression (fold)

Pre-hTERT Post-hTERT grade III grade II

Figure 1 hTERT-induced immortalization of primary breast tumor cells. (a) hTERT and hTERC transcript levels in primary tumor cultures compared with immortal breast cancer cell lines. (b) Increased hTERT but not hTERC expression in hTERT-transduced primary tumor cells. Note higher endogenous transcript levels in the high-grade cases. In (a) and (b) above, gene expression values were normalized to those of epithelial organoids isolated from non-malignant reduction mammoplasty tissue, designated as 1 (average of five samples, SDo10%). (c) Characteristic epithelial phenotype of TC cultures and TT cultures.

UQCRC, FDFT1), RNA binding (DCP2, RPS4X, (20–60, designated as TT1), and highest in late-passage KHSRP), and as structural constituents of the ribosome populations (150–350, designated as TT2) (Figure 2b, top (RPL36, RPL30, MRPL48, RPS5, MRPS5) (Table 1 panel). We then determined whether ImmSig expression and Supplementary Table 2). was displayed by (a) SI cases among our patient samples, and (b) breast cancer cell lines routinely used in the field, which are by definition, spontaneously Synergistic expression of ImmSig in hTERT- immortalized. Twenty-nine TC and TT samples from immortalized and spontaneously immortalized primary cultures representing NSI and SI cases, including early tumor cultures and late passages of a sample for which a matched TC Compared with TC cultures of NSI tumors, correla- was not available (CCdl067TT1 and CCdl067TT2), were tion to ImmSig was higher for early passage TT cultures ranked by their correlation to the ImmSig. Notably, TC

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6272

b

Figure 2 ImmSig expression in primary tumor cultures. (a) Gene expression profile of finite mitotic life, TC cultures compared with matched immortalized TT cultures. SAM-derived ImmSig represents the centroid (mean expression for each gene) for the TT class. (b) Cell cultures ordered by correlation to ImmSig; blue-to-yellow gradient represents increasing correlation to ImmSig. (Top panel) Analysis of matched TC and TT cultures of NSI tumors. Note higher ImmSig correlation displayed by TT cultures (gold font). (Bottom panel) Combined analysis of ImmSig correlations of NSI, SI primary tumor cultures and established breast epithelial cell lines (green font). Note SI-derived TC cultures (marked by asterisks) admixed among immortalized cell lines with high-ImmSig correlations.

cultures (without exogenous hTERT) from SI tumors representative of immortal cell populations that emerge displayed a relatively high correlation to ImmSig by a process of spontaneous selection. (Figure 2b, bottom panel). Similarly, ImmSig correla- The unexpected induction of telomere-independent tions placed established breast cancer cell lines in close pathways in immortalized tumor cultures prompted proximity to hTERT-immortalized cultures among a us to spot-test the microarray data using RNA inter- group of 33 assorted finite life and immortal cultures. ference. Transfection with hTERT siRNA resulted This strongly suggests that the underlying biology in >60% inhibition of hTERT mRNA expression identified by ImmSig, although derived by the analysis (Figure 3a) leading to the increased generation of of experimentally immortalized tumor cells, is also reactive oxygen species (ROS) in the absence of

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6273 Table 1 Gene Ontology overrepresentation analysis Gene category Cluster frequency Gene frequency in Corrected (% of total) background (% of total) P-value

ImmSig genes n ¼ 3023 n ¼ 12775 GO molecular function Structural constituent of ribosome 2.20 0.80 1.52E-14 NADH dehydrogenase activity 0.70 0.20 0.00027 Oxidoreductase activity, acting on NADH or NADPH, 0.70 0.20 0.00027 quinone or similar compound as acceptor NADH dehydrogenase (ubiquinone) activity 0.60 0.20 0.0008 NADH dehydrogenase (quinone) activity 0.60 0.20 0.0008 Oxidoreductase activity 4.00 2.80 0.00229 RNA binding 3.50 2.30 0.00321 Oxidoreductase activity, acting on CH–OH group of donors 0.90 0.40 0.00372 Catalytic activity 23.90 21.00 0.00455 Electron-carrier activity 0.80 0.30 0.00881 Oxidoreductase activity, acting on NADH or NADPH 0.70 0.30 0.01005 Oxidoreductase activity, acting on the CH–OH group of 0.90 0.40 0.01195 donors, NAD or NADP as acceptor

ImmSig genes GO biological process Protein biosynthesis 4.60 2.60 1.47E-09 Cellular biosynthesis 6.70 4.50 7.70E-08 Biosynthesis 7.50 5.20 1.16E-07 Macromolecule biosynthesis 4.80 3.00 2.96E-07 Macromolecule metabolism 21.40 17.60 3.00E-07 Cellular metabolism 33.00 28.50 6.91E-07 Metabolism 35.00 30.60 1.72E-06 Cellular physiological process 43.60 39.60 0.00017 Primary metabolism 31.40 27.70 0.00024 Physiological process 46.60 42.80 0.00128 Cellular protein metabolism 13.80 11.70 0.03067 Protein metabolism 14.80 12.70 0.04562

ImmSig associated genes in tumor tissue GO molecular function n ¼ 87 n ¼ 1538 ATP binding 15.0 4.2 0.00742 Adenyl nucleotide binding 15.0 4.4 0.01388 activity 6.2 0.8 0.03707

ImmSig associated genes in tumor tissue GO biological process M phase 18.8 1.51.20E-12 Mitosis 15.0 1.0 1.60E-11 M phase of mitotic cell cycle 15.0 1.0 6.16E-11 Mitotic cell cycle 16.2 1.4 2.60E-10 Cell cycle 26.2 4.4 3.64E-10 DNA metabolism 18.8 2.7 1.10E-07 Cell division 11.2 0.8 2.51E-07 Biopolymer metabolism 32.59.4 5.00E-07 DNA replication 10.0 0.8 9.20E-06 Organelle organization and biogenesis 15.0 3.3 0.00122 Macromolecule metabolism 36.2 16.4 0.00173 Microtubule-based process 7.50.8 0.00275 Regulation of mitosis 6.2 0.50.00388 Regulation of progression through cell cycle 13.8 3.1 0.00415 Regulation of cell cycle 13.8 3.1 0.00415 Microtubule cytoskeleton organization and biogenesis 5.0 0.3 0.00761 segregation 5.0 0.3 0.00761 Phosphoinositide-mediated signaling 6.2 0.6 0.00838 Cytoskeleton organization and biogenesis 8.8 1.6 0.03118 Mitotic sister chromatid segregation 3.8 0.2 0.03163 Cell cycle checkpoint 3.8 0.2 0.03163 Sister chromatid segregation 3.8 0.2 0.03163 Spindle organization and biogenesis 3.8 0.2 0.03163 DNA repair 6.2 0.8 0.04655 DNA-dependent DNA replication 5.0 0.5 0.0492

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6274

Figure 3 Effect of hTERT silencing on ImmSig gene expression. (a) QRT measurement of hTERT siRNA-induced reduction of hTERT expression in immortalized primary tumor-derived cultures relative to control siRNA-transfected cells. (b) Induction of ROS by hTERT siRNA in NSI tumor-derived culture, CCd1022TT2. (Left panel) ROS measured by mean fluorescent intensity (MFI) of the C400 dye, in three independent experiments, *Po0.001. (Right panel) hTERT siRNA-induced shift in MFI in a single FACS experiment. (c) hTERT siRNA-induced mitochondrial membrane depolarization in CCd1022TT2 cells measured by flow cytometry as a ratio of red-to-green MFI of the JC-1 dye. (Left panel) Bar graphs of mitochondrial membrane potential (DCm) measurements in three independent experiments, *Po0.001. (Right panel) hTERT siRNA-induced depolarization shown in a representative run. Two- color ratios indicated in the upper right hand quadrant. Unstained samples (not shown) were used to set the quadrants. CCCP-treated samples were used to set the compensation settings prior to data acquisition. (d) Induction of apoptosis by hTERT siRNA in an immortalized SI tumor-derived culture, CCdl054TT2. (Left panel) Bar graphs show percentage of Annexin positive and propidium iodide negative cells (average of three independent FACS experiments), *Po0.001. (Right panel) Representative dot plots indicating the percentage populations of live, necrotic and apoptotic cells. Note significant increase in early apoptotic cell population (Annexin V positive, propidium iodide negative) induced by hTERT siRNA. (e) hTERT siRNA-induced reduction in cell density in an hTERT- transduced NSI tumor-derived culture, CCdl022TT2. Data points represent average of triplicate counts. (f) hTERT siRNA-induced reduction of overexpressed ImmSig gene transcripts measured by QPCR. Gene expression is shown relative to control siRNA.

exogenous induction (Figure 3b) and a detectable ImmSig expression and clinical outcome in breast cancer reduction in mitochondrial membrane potential To reveal additional in vivo biology associated with (Figure 3c). hTERT knockdown caused a significant the ImmSig phenotype, we used microarray data on induction of apoptosis (Figure 3d) and >2-fold reduc- tumors derived from primary breast cancer patients tion in cell density relative to non-targeting siRNA in treated at the Nederlands Kanker Instituut (NKI). We immortalized primary tumor cultures (Figure 3e). These first sought to identify genes, not part of ImmSig cellular responses were accompanied by a marked but whose expression levels were highly associated with reduction (up to 80%) in transcript levels of ImmSig the ImmSig phenotype (Figure 4). Enriched GO terms genes in various functional categories as measured by for ImmSig-associated genes in the 95th percentile QPCR (Figure 3f). Together, these findings confirm that (highly correlated) predominantly displayed pathways hTERT expression was critical for the survival and of: macromolecule metabolism (GZMB, HIST1H1B, growth of both SI and NSI tumor cultures, and validate CENPA, RRM2, ORC6L, UBE2C), and cell cycle its association with a spectrum of ImmSig genes related (BIRC5, KIFC1, ESPL1, CCNB1, EXO1, PTTG1, to the dissipation of oxidative stress. AURKB, BUB1, CKS2). Pathways associated with the

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6275 ERBB2 BASAL LUMINAL A LUMINAL B NORMAL-LIKE

cor toTumor ImmSig

MITOTIC CELLCYCLE INTRACELLULAR TRANSPORT ANTI - APOPTOSIS

PHOSPHATE TRANSPORT EXTRACELLULAR MATRIX -2.00 -1.33 -0.67 0.00 0.67 1.33 2.00

Figure 4 ImmSig-associated genes in vivo. Heat map of ImmSig-associated genes in NKI primary breast tumors. Each column represents an independent case. Tumors are ordered in columns by increasing correlation to ImmSig (yellow, higher correlation; blue, lower correlation). Vertical bars above the sample ID indicate molecular subtype of the tumor. bottom 5th percentile (highly anticorrelated) included generally known to represent highly proliferative tumors phosphate transport (COL10A1, COL6A3, COL12A1), (Figure 4). and extra cellular matrix (MATN3, LUM, COL14A1). Finally, we determined whether correlation to ImmSig Intriguingly, in contrast to the distinctive oxidative for tumors in the NKI dataset was indicative of changes identified by ImmSig itself, ImmSig-associated overall survival and time to metastasis by the Cox genes mostly reflected the proliferation-related aberra- proportional hazards model. For both clinical endpoints, tions previously reported in immortal cell lines and patients in the top tertile of ImmSig correlations were aggressive tumors (Table 1). found to be associated with worse outcome for disease- Next, we ranked NKI tumors by their correlation to free survival and death (Po0.001, Figure 5, top panels). ImmSig to determine a pattern of association with When patients were subdivided by correlation to the known breast cancer molecular subtypes (Perou et al., ImmSig in conjunction with the ER status, higher 2000; Sorlie et al., 2001, 2003). Among estrogen receptor ImmSig correlation was found to be significantly (ER)-negative tumors, the ERBB2 overexpressing sub- associated with poor outcome specifically in ER-positive type was relatively uncorrelated, while basal tumors patients (Po0.05, Figure 5, bottom panels). ImmSig showed higher ImmSig correlation. Among ER-positive expression was not predictive of prognosis in ER- tumors, the luminal A subtype displayed lower ImmSig negative tumors (Supplementary Figure 1). Importantly, correlation in comparison to the luminal B subtype, the hTERT gene alone was not prognosticative in this

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6276 Low ImmSig Metastasis-free survival, all cases Overall survival, all cases High ImmSig 1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

Probability of survival p = 0.0001 p = 0 0.0 0.0 0 5 10 15 0 5 10 15

Metastasis -free survival, ER positive cases Overall survival, ER positive cases 1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2 Probability of survival p = 0.0036 p = 0.03 0.0 0.0 0510 15 0 5 10 15 Years Years Figure 5 ImmSig predicts clinical outcome in breast cancer. Kaplan–Meier survival curves for subgroups with high vs low-ImmSig correlations. Patients in the top ImmSig tertile had worse prognosis than those in the bottom tertile.

Table 2 Multivariate analysis of prognostic factors in ER positive tumors Comparison Univariate analysis HR (95% CI) P Multivariate analysis HR (95% CI) P

ImmSig tertile Top vs bottom 3.2 (1.44–7.15) 0.004 3.13 (1.23–7.98) 0.01

Histologic grade II vs I 3.3 (1.1–9.89) 0.03 3.54 (0.74–16.8) 0.11 III vs I 7.72 (2.68–22.21) o0.001 4.49 (0.95–21.07) 0.05

Lymph node status Negative vs positive 1.01 (0.56–1.82) 0.05 0.84 (0.26–2.69) 0.77

Tumor size >2 vs p2 cm 1.66 (0.92–3.0) 0.08 0.72 (0.31–1.6) 0.44

Age p50 vs >50 years 1.05 (0.44–2.49) 0.9 1.26 (0.37–4.29) 0.71

Abbreviations: CI, confidence interval; HR, hazard ratio.

dataset (Supplementary Figure 2) in contrast to a Discussion reported association (Elkak et al., 2005). In the light of a recent study that has questioned the specificity We have shown that comparison of multiple isogenic of hTERT antibodies (Wu et al., 2006), data obtained sets of finite mitotic life and hTERT-immortalized by immunostaining of tumor tissue requires further primary breast tumor cells has provided further verification. molecular understanding of cancer cell immortalization, In ER-positive tumors, only top ImmSig tertile and a key hallmark of malignancy. While cell cultures high histologic grade maintained significance as mea- derived from high histologic grade tumors adapted sured by multivariate analysis of known prognostic readily to in vitro environmental conditions and factors (P ¼ 0.01, and P ¼ 0.05, respectively, Table 2). In continued to proliferate, resulting in spontaneously conclusion, aside from its role in the immortalization of immortalized lines, low- and intermediate-grade tumor primary breast tumor cultures, ImmSig-based separa- cells exhibited telomerase repression leading to a finite tion of patient outcome provides strong evidence that mitotic life. These events have a direct bearing on the this in vitro derived signature captures an in vivo fact that the latter class of tumors remains unrepre- phenotype portraying relevance to clinical disease. sented among currently available cancer cell lines, and

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6277 therefore, understudied functionally and translationally. impairing ER functions related to DNA binding and However, even clinically less aggressive tumor cells can transactivation (Liang et al., 1998). Thus, it seems be immortalized by exogenous hTERT in a single step as plausible that high-ImmSig, ER-positive tumors are shown here. The ImmSig displayed by tumor cells functionally analogous to estrogen-independent, ER- immortalized either experimentally or spontaneously negative tumors and are likely to show less sensitivity to showed significant expression changes of genes involved conventional hormonal therapies. Indeed, conversion to in oxidoreductase activity. These observations fit well a tamoxifen-resistant phenotype is associated with with the notion that high-grade tumor cells, which oxidative stress in the ER-positive MCF-7 cell line maintain hTERT expression, are protected from oxida- (Schiff et al., 2000). Future work with ImmSig genes tive damage generally induced by ambient oxygen levels could provide additional approaches for the manage- (21%) prevalent in the tissue culture environment. ment of tumors resistant to anti-estrogen therapy. Additionally, competence in chromosome end replica- At this time it has not been determined whether tion allows such cells to immortalize spontaneously. cultures derived from all tumors with a higher ImmSig Converging data from low- and intermediate-histologic correlation immortalize spontaneously. Considering the grade tumor cells immortalized by hTERT transduction rarity of spontaneously immortalized cell lines, it is support a novel role for telomerase reactivation in conceivable that clinical samples only at the uppermost conferring the ability to switch to a state that resists dual end of the ImmSig-correlated tertile may be indicative of damage, both oxidative and replicative. this phenotype. ImmSig pathways induced upon the ImmSig portrays a consistent and common biology acquisition of immortalization in vitro support the view between breast cancer cells in laboratory dishes, and in that at a biological level, this phenotype represents the the patient host. Just as immortalized cell cultures proclivity to (a) overcome growth arresting oxidative displayed aggressive dysregulated growth, high ImmSig damage, and (b) sustain cell survival and growth. The correlated tumors identified aggressive clinical disease. subset of high ImmSig breast tumors may well be one Although applied to the same patient cohort as other that harbors a greater aptitude to prevail under adverse prognostic signatures (van de Vijver et al., 2002; van’t microenvironmental conditions including resistance to Veer et al., 2002; Chang et al., 2005; Sotiriou et al., clinical interventions such as treatment with radiation 2006), ImmSig is distinct. Like ImmSig, the fibroblast- and DNA-damaging drugs. Foreseeably, the addition based common serum response signature (Chang et al., of redox modulators and telomerase inhibitors to 2005) is biologically derived from cell culture studies, yet the therapeutic regimen of high-ImmSig cases could elements common to the two signatures, related to cell improve disease outcome. cycle and proliferation, are represented not in ImmSig itself but as ImmSig-associated genes. This is likely due to the fact that fibroblasts are strikingly more resistant Materials and methods to stress than other cell types (Michiels et al., 1990; Serrano and Blasco, 2001). While ImmSig-based out- An expanded version of this section is provided as a come is more relevant to ER-positive breast tumors, Supplementary methods file. other gene lists designed specifically for predicting the outcome of ER-positive tumors (Ma et al., 2004; Paik Tumor cell culture and hTERT transduction et al., 2004), could serve as better prognostic indicators. Fresh primary breast tumor samples and reduction mammo- In our view, ImmSig might be more effective in plasty tissues, collected under IRB-approved guidelines, were improving patient outcome by evoking functional propagated as described previously (Dairkee et al., 1997; Li response to therapeutics highlighted by this signature, et al., 1998). For retroviral infection of primary tumor cultures such as anti-telomerase and antioxidant approaches. at passage 3–4, supernatant from PA317 packaging cells infected with pBabePuro containing hTERT (kindly provided As yet, the molecular and clinical consequences of by Dr Elizabeth Blackburn, UCSF) or control vector was oxidative stress in the tumor microenvironment are not used. Infection efficiency was approximately 10%. Control well understood. Response to oxidative stress is known uninfected cultures were completely non-viable after 7 days of to be variable since, in some systems, it results in treatment with 1.5 mg/ml puromycin (Sigma, St Louis, MO, apoptosis resistant cell populations (Shami et al., 1998), USA). After an additional 7 days of drug selection, hTERT while in others it promotes a senescence phenotype overexpressing stably infected tumor cell populations were accompanied by upregulation of the p53 dependent, p21 continuously expanded and passaged as mass cultures. gene (Chen et al., 1998). Intuitively, it may be surmised that in vivo, oxidative stress plays a key role in the Gene expression and functional assays selection of p53-deficient clones (Storz, 2005), which are Total RNA was extracted from subconfluent primary tumor generally ER-negative (Sorlie et al., 2003). Our study cultures with the RNAeasy Mini kit (Qiagen, Chatsworth, CA, sheds new light on the link between oxidative stress and USA). cDNA was synthesized and analyzed as reported aggressiveness in ER-positive tumors. Overexpression of previously (Dairkee et al., 2004) using QPCR in conjunction with an Applied Biosystems 5700 Sequence Detection System a significant number of oxidoreductase activity genes (Foster City, CA, USA). observed here in ER-positive, highly proliferative For hTERT gene silencing, pre-annealed siRNA duplex was luminal tumors implicates that oxidative stress may be purchased from Dharmacon Research (Lafayette, CO, USA). intrinsic to this subtype. Our observations are synergis- siCONTROL non-targeting siRNA#1 served to evaluate off- tic with the known effects of oxidative stress in target effects of siRNA transfection. PCR primers for ImmSig

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6278 genes are listed in Supplementary Table 3. Additional effects of into tertiles based on whether their correlation to the ImmSig hTERT siRNA were evaluated using standard methods in the was greatest or smallest in the cohort. Kaplan–Meier survival following functional assays: growth rate, apoptotic fraction, curves were computed for overall survival and time to generation of ROS, and mitochondrial depolarization. metastasis. To identify additional genes associated with ImmSig, an Constructing the ImmSig and testing clinical tumor samples ImmSig-correlation vector (ICV) was computed. Genes whose Array data from B42 000 feature cDNA microarrays of NSI correlation to the ICV was in the 95th percentile (most tumor cultures was used to compute ImmSig. Two classes of correlated) or in the 5th percentile (most anti-correlated) were samples, non-transfected TC cultures and TT were compared. identified and used for analysis of enriched GO terms (Boyle Analysis of variance (ANOVA) correction was performed such et al., 2004). To generate a heatmap, tumors were sorted by the that the weight of the TT and TC cultures was the same for ICV, and genes were sorted by their correlation to this vector. each patient. Data were then analysed using significance Associations between ImmSig correlation in top or bottom analysis of microarrays (SAM, Tusher et al., 2000). Corres- tertiles, and clinical parameters were analyzed by the w2-test. ponding to the minimum FDR of 0.08%, 597 Unigene Clone P-values from two-sided tests were considered to be significant IDs were used to compute a centroid (mean expression for each when o0.05. For multivariate analysis, a Cox proportional gene within a class) for each class, TT and TC. The centroid for hazards model was fit to data from NKI patients characterized the TT class (the cultures immortalized by hTERT transfection) by age, tumor size, number of positive nodes and histologic was denoted as the ImmSig. For measuring relative correlation grade. to ImmSig among primary tumor cultures and cell lines, an ANOVA correction was performed on each dataset, and samples were sorted by their Pearson correlation to ImmSig. Acknowledgements Global expression data in a previously published NKI dataset of 295Stage I and II breast cancers from women less We thank Dr Janos Demeter and Dr Wenzhong Xiao (Stanford than 53 years of age (van de Vijver et al., 2002; Chang et al., University) for helpful suggestions on the Bioinformatic aspects 2005) were collapsed by UniGene Cluster ID (Build #183), and of this study. This research was supported by NIH transformed from log10 to log2. R01CA109325(SHD), NIH K01 HG00030 (MN) and NIH After gene mean centering, correlation to the ImmSig was U01 CA85129 (SSJ) and by funds from the California Breast computed for each tumor and rather than clustering, tumors Cancer Research Program of the University of California, were sorted by this correlation. Tumors were then separated grant numbers 8WB-0032 (SHD) and 10 EB-1086 (SSJ).

References

Ben-Porath I, Weinberg RA. (2005). The signals and pathways Dairkee SH, Paulo EC, Traquina P, Moore DH, Ljung BM, activating cellular senescence. Int J Biochem Cell Biol 37: Smith HS. (1997). Partial enzymatic degradation of stroma 961–976. allows enrichment and expansion of primary breast tumor Bodnar AG, Ouellette M, Frolkis M, Holt SE, Chiu CP, cells. Cancer Res 57: 1590–1596. Morin GB et al. (1998). Extension of life-span by introduc- Dudognon C, Pendino F, Hillion J, Saumet A, Lanotte M, tion of telomerase into normal human cells. Science 279: Segal-Bendirdjian E. (2004). Death receptor signaling 349–352. regulatory function for telomerase: hTERT abolishes Boyle EI, Weng S, Gollub J, Jin H, Botstein D, Cherry JM TRAIL-induced apoptosis, independently of telomere main- et al. (2004). GO:TermFinder – open source software for tenance. Oncogene 23: 7469–7474. accessing Gene Ontology information and finding signifi- Elkak AE, Meligonis G, Salhab M, Mitchell B, Blake JR, cantly enriched Gene Ontology terms associated with a list Newbold RF et al. (2005). hTERT protein expression is of genes. Bioinformatics 20: 3710–3715. independent of clinicopathological parameters and c-Myc Chang HY, Nuyten DS, Sneddon JB, Hastie T, Tibshirani R, protein expression in human breast cancer. J Carcinog 4: 17. Sorlie T et al. (2005). Robustness, scalability, and integra- Kurz DJ, Decary S, Hong Y, Trivier E, Akhmedov A, tion of a wound-response gene expression signature in Erusalimsky JD. (2004). Chronic oxidative stress compro- predicting breast cancer survival. Proc Natl Acad Sci USA mises telomere integrity and accelerates the onset of 102: 3738–3743. senescence in human endothelial cells. J Cell Sci 117: Chen QM, Bartholomew JC, Campisi J, Acosta M, Reagan JD, 2417–2426. Ames BN. (1998). Molecular analysis of H2O2-induced Li Z, Bustos V, Miner J, Meng ZH, Zlotnikov G, Ljung BM senescent-like growth arrest in normal human fibroblasts: et al. (1998). Propagation of genetically altered tumor cells p53 and Rb control G1 arrest but not cell replication. derived from fine-needle aspirates of primary breast carci- Biochem J 332: 43–50. noma. Cancer Res 58: 5271–5274. Counter CM, Avilion AA, LeFeuvre CE, Stewart NG, Liang X, Lu B, Scott GK, Chang CH, Baldwin MA, Benz CC. Greider CW, Harley CB et al. (1992). Telomere shortening (1998). Oxidant stress impaired DNA-binding of estrogen associated with chromosome instability is arrested in receptor from human breast cancer. Mol Cell Endocrinol immortal cells, which express telomerase activity. EMBO J 146: 151–161. 11: 1921–1929. Ma XJ, Wang Z, Ryan PD, Isakoff SJ, Barmettler A, Fuller A Dairkee SH, Deng G, Stampfer MR, Waldman FM, Smith HS. et al. (2004). A two-gene expression ratio predicts clinical (1995). Selective cell culture of primary breast carcinoma. outcome in breast cancer patients treated with tamoxifen. Cancer Res 55: 2516–2519. Cancer Cell 5: 607–616. Dairkee SH, Ji Y, Ben Y, Moore DH, Meng Z, Jeffrey SS. Masutomi K, Possemato R, Wong JM, Currier JL, Tothova Z, (2004). A molecular ‘signature’ of primary breast cancer Manola JB et al. (2005). The telomerase reverse transcriptase cultures; patterns resembling tumor tissue. BMC Genomics regulates chromatin state and DNA damage responses. Proc 5: 47. Natl Acad Sci USA 102: 8222–8227.

Oncogene Immortalization signature in breast cancer SH Dairkee et al 6279 Michiels C, Toussaint O, Remacle J. (1990). Comparative Stampfer MR, Garbe J, Levine G, Lichtsteiner S, Vasserot AP, study of oxygen toxicity in human fibroblasts and endothe- Yaswen P. (2001). Expression of the telomerase catalytic lial cells. J Cell Physiol 144: 295–302. subunit, hTERT, induces resistance to transforming growth Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M et al. factor beta growth inhibition in p16INK4A(-) human mam- (2004). A multigene assay to predict recurrence of tamoxifen- mary epithelial cells. Proc Natl Acad Sci USA 98: 4498–4503. treated, node-negative breast cancer. N Engl J Med 351: Storz P. (2005). Reactive oxygen species in tumor progression. 2817–2826. Front Biosci 10: 1881–1896. Perou CM, Sorlie T, Eisen MB, Aas T, Geisler S, Johnsen H Tusher VG, Tibshirani R, Chu G. (2000). Significance analysis et al. (2000). Molecular portraits of human breast tumours. of microarrays applied to the ionizing radiation response. Nature 406: 747–752. Proc Natl Acad Sci USA 98: 5116–5121. Schiff R, Reddy P, Ahotupa M, Coronado-Heinsohn E, Grim M, van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Hilsenbeck SG et al. (2000). Oxidative stress and AP-1 activity in Voskuil DW et al. (2002). A gene-expression signature as a tamoxifen-resistant breast tumors in vivo. J Natl Cancer Inst 92: predictor of survival in breast cancer. N Engl J Med 347: 1926–1934. 1999–2009. Serrano M, Blasco MA. (2001). Putting the stress on van’t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, senescence. Curr Opin Cell Biol 13: 748–753. Mao M et al. (2002). Gene expression profiling predicts Shami PJ, Sauls DL, Weinberg JB. (1998). Schedule and clinical outcome of breast cancer. Nature 415: 530–536. concentration-dependent induction of apoptosis in leukemia Wang J, Feng H, Huang XQ, Xiang H, Mao YW, Liu JP et al. cells by nitric oxide. Leukemia 12: 1461–1466. (2005). Human telomerase reverse transcriptase immorta- Smith LL, Coller HA, Roberts JM. (2003). Telomerase lizes bovine lens epithelial cells and suppresses differentia- modulates expression of growth-controlling genes and tion through regulation of the ERK signaling pathway. enhances cell proliferation. Nat Cell Biol 5: 474–479. J Biol Chem 280: 22776–22787. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H Wu YL, Dudognon C, Nguyen E, Hillion J, Pendino F, et al. (2001). Gene expression patterns of breast carcinomas Tarkanyi I et al. (2006). Immunodetection of human telome- distinguish tumor subclasses with clinical implications. Proc rase reverse-transcriptase (hTERT) re-appraised: nucleolin Natl Acad Sci USA 98: 10869–10874. and telomerase cross paths. J Cell Sci 119: 2797–2806. Sorlie T, Tibshirani R, Parker J, Hastie T, Marron JS, Nobel A Xiang H, Wang J, Mao Y, Liu M, Reddy VN, Li DW. (2002). et al. (2003). Repeated observation of breast tumor subtypes Human telomerase accelerates growth of lens epithelial cells in independent gene expression data sets. Proc Natl Acad Sci through regulation of the genes mediating RB/E2F pathway. USA 100: 8418–8423. Oncogene 21: 3784–3791. Sotiriou C, Wirapati P, Loi S, Harris A, Fox S, Smeds J et al. Zhang P, Chan SL, Fu W, Mendoza M, Mattson MP. (2003). (2006). Gene expression profiling in breast cancer: under- TERT suppresses apoptosis at a premitochondrial step by a standing the molecular basis of histologic grade to improve mechanism requiring reverse transcriptase activity and 14-3-3 prognosis. J Natl Cancer Inst 98: 262–272. protein-binding ability. FASEB J 17: 767–769.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).

Oncogene