Prediction of Doxorubicin Sensitivity in Breast Tumors Based on Gene Expression Proﬁles of Drug-Resistant Cell Lines Correlates with Patient Survival

Home , IRX1

Prediction of doxorubicin sensitivity in breast tumors based on gene expression proﬁles of drug-resistant cell lines correlates with patient survival

Bala´ zs Gyo¨ rffy*,1,2,6, Violeta Serra1,3,6, Karsten Ju¨ rchott1, Rula Abdul-Ghani1, Mitch Garber4, Ulrike Stein5, Iver Petersen1, Hermann Lage1, Manfred Dietel1 and Reinhold Scha¨ fer1

1Charite´, Institute of Pathology, Humboldt University, Schumannstr. 20/21, Berlin D-10117, Germany; 22nd Department of Internal Medicine, Semmelweis University Budapest, Szentkira´lyi u. 46, Budapest H-1088, Hungary; 3Centro Nacional de Investigaciones Oncologicas (CNIO), Melchor Ferna´ndez Almagro 3, Madrid E-28029, Spain; 4Departments of Genetics, Stanford University School of Medicine, Stanford, California, CA 94305, USA; 5Max-Delbru¨ck-Center for Molecular Medicine, Robert-Ro¨ssle-Strasse 10, 13092 Berlin, Germany

Up to date clinical tests for predicting cancer chemotherapy Keywords: cancer chemoresistance; gene expression; response are not available and individual markers have microarray; breast cancer shown little predictive value. We hypothesized that gene expression patterns attributable to chemotherapy-resistant cells can predict response and cancer prognosis. We contrasted the expression profiles of 13different human Introduction tumor cell lines of gastric (EPG85–257), pancreatic (EPP85–181), colon (HT29) and breast (MCF7 and The ability of cancer cells to acquire simultaneous MDA-MB-231) origin and their counterparts resistant to resistance to different drugs is a significant obstacle to the topoisomerase inhibitors daunorubicin, doxorubicin or successful chemotherapy and is commonly designated as mitoxantrone. We interrogated cDNA arrays with 43000 multidrug resistance (MDR). MDR may be achieved by cDNA clones (B30 000 unique genes) to study the a variety of mechanisms including the overexpression of expression pattern of these cell lines. We divided gene ATP-dependent membrane transporters like P-glyco- expression profiles into two sets: we compared the expression protein (P-gp, MDR1, ABCB1), the multidrug resistance patterns of the daunorubicin/doxorubicin-resistant cell lines proteins MRP1 and MRP2 (ABCC1, ABCC2) or the and the mitoxantrone-resistant cell lines independently to the mitoxantrone resistance protein/breast cancer resistance parental cell lines. For identifying predictive genes, the protein (MXR, BCRP, ABCG2) (Gottesman et al., Prediction Analysis for Mircorarrays algorithm was used. 2002). P-gp is involved particularly in drug resistance of The analysis revealed 79 genes best correlated with colon, kidney, adrenocortical and hepatocellular cancers doxorubicin resistance and 70 genes with mitoxantrone as well as of acute myelogenic leukemia (AML) (Gold- resistance. In an independent classification experiment, we stein et al., 1989). In vivo imaging studies of breast applied our model of resistance for predicting the sensitivity carcinomas using the P-gp substrate 99mTc-sestamibi, of 44 previously characterized breast cancer samples. The developed for monitoring cardiac function, showed that patient group characterized by the gene expression profile transporter activity is increased in breast tumors and similar to those of doxorubicin-sensitive cell lines exhibited demonstrated its potential as predictive marker for longer survival (49.7726.1 months, n 21, P 0.034) than ¼ ¼ tumor response towards treatment with anthracyclines 7 n the resistant group (32.9 18.7 months, ¼ 23). The and vinca alkaloids (Alonso et al., 2002; Sciuto et al., application of gene expression signatures derived from 2002). The MRP1 gene is overexpressed in leukemias, doxorubicin-resistant and -sensitive cell lines allowed to esophageal carcinoma and non-small-cell lung cancers predict effectively clinical survival after doxorubicin mono- among others (Nooter et al., 1995). Additionally, drug therapy. Our approach demonstrates the significance of resistance is mediated by defects in intrinsic safeguard in vitro experiments in the development of new strategies mechanisms capable of eliminating cells by apoptosis. for cancer response prediction. The loss of programmed cell death induced by Oncogene (2005) 7542–7551. doi:10.1038/sj.onc.1208908; 24, antitumor drugs can be achieved via disruption of published online 25 July 2005 regulators of DNA damage signaling, such as p53 and bcl-2 (Fisher et al., 1993; Lowe et al., 1994). *Correspondence: B Gyo¨ rffy, Charite´ , Institute of Pathology, Hum- Systemic chemotherapies are widely used for the boldt University, Schumannstr. 20/21, Berlin 10117, Germany; treatment of breast cancer. Efficient response to E-mail: [email protected] 6These authors contributed equally to this work preoperative treatment has prognostic value because it Received 11 November 2004; revised 18 April 2005; accepted 7 June is correlated with increased survival. The estrogen 2005; published online 25 July 2005 receptor status is predictive for response to hormonal Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7543 treatment (Kuerer et al., 1999; Chollet et al., 2002). eight sensitive and drug-resistant cell lines by XTT assay Despite its central role in mammary biology and as described in Materials and methods. The prolifera- carcinogenesis, the estrogen and progesterone receptor tion of HT29 colon cancer cells at increasing drug status does not effectively predict the sensitivity against concentrations is depicted in Figure 1 as a representative cytotoxic drugs. Doxorubicin is widely used in the example. We confirmed previously published data (Lage treatment of breast cancer, it is involved in four of the et al., 2000; Lage and Dietel, 2002) indicating that the six protocols defined by the American Cancer Society resistance pattern is a robust feature upon prolonged (protocols named CAF, AC, AC þ and A-CMF). cultivation. Table 1 summarizes the characteristics of Mitoxantron is used in monotherapy for treating the entire set of multidrug-resistant human cancer cell metastatic breast cancer. The clinical decision for the lines used in this study. For gene expression profiling, application of anthracyclines, taxane-derivatives or RNA was prepared from nontreated parental cells and other combinations of drugs is made regardless whether from resistant derivatives cultured in the presence of the patient’s tumor is likely to respond to it or not. daunorubicin and mitoxantrone at a drug concentration Predictions of therapy sensitivity based on single marker which eliminated the majority of sensitive cells (Figure 1, genes such as P-gp or MRP1 have often been incon- arrows). clusive. The closest correlation between P-gp expression We divided gene expression profiles obtained for all and drug resistance has been reported for AML. cell lines into two sets: we compared the expression Nevertheless, P-gp overexpression does not correctly patterns of the daunorubicin/doxorubicin-resistant cell predict relapse cases (Gottesman et al., 2002). lines and the mitoxantrone-resistant cell lines indepen- High-density microarrays can help to identify relevant dently to the parental cell lines. We selected genes that genes involved in tumor pathogenesis and to classify were present in at least three RNA samples within a different tumor entities and even tumor subtypes. For sample set of four. In all, 37 771 sequences remained in example, Sorlie and others have comprehensively the doxorubicin resistance set and 32 943 in the examined expression profiling data to confirm the mitoxantrone resistance set. The complete filtered and universality of distinguishing basal- and luminal-like normalized data set is shown as supplementary informa- breast cancer subtypes and established a correlation to tion (see Supplementary Table 1). clinical outcome (Sorlie et al., 2003; Sotiriou et al., To select discriminatory genes, we compared the 2003). Moreover, van’t Veer et al. (2003) demonstrated expression profiles of the resistant and parental cell that breast cancer prognosis can be deduced from the gene expression profile of primary tumors. Chang et al. (2003) demonstrated that gene expression profiles of primary breast cancer could predict the response to a 1.2 docetaxel. Thus, the molecular profiles may allow the HT29PAR 1 development of microarray-based assays for drug HT29RDB sensitivity and help to reduce unnecessary treatment of 0.8 women with breast cancer. The results of microarray analysis aiming at elucidat- 0.6 ing gene patterns associated with cytotoxic drug 0.4 resistance may be obscured by the presence of various cell types in heterogeneous cancer specimens. Therefore, relative cell viability 0.2 we firstly generated gene expression patterns of 13 well- 0 defined chemotherapy-resistant and -sensitive cancer 0.001 0.01 0.1 1 10 100 cell lines in order to identify discriminatory genes concentration DB (µg/ml) associated with drug resistance in four different types of solid tumors. Secondly we contrasted the cell line- b 1.2 derived gene expression profiles with those of a set of 1 precharacterized mammary cancer patients (Sorlie et al., 2003) to use molecular signatures for predicting the 0.8 patient response to chemotherapy in conjunction with 0.6 prognosis. 0.4 HT29PAR

relative cell viability 0.2 HT29RNOV Results 0 0.0001 0.001 0.01 0.1 1 10 Identification of discriminatory genes concentration NOV (µg/ml) To identify discriminatory genes for predicting sensitiv- Figure 1 Sensitivity of the resistant HT29RDB, HT29RNOV and ity to doxorubicin and mitoxantrone, we contrasted the sensitive HT29 colon cell lines against topoisomerase inhibitors daunoblastin (a) and mitoxantrone (b). Arrows indicate the gene expression profiles of closely related drug-resistant concentration at which the resistant cell lines were cultured for and -sensitive cancer cell lines. Prior to microarray array analysis. Proliferation was assessed using an XTT Cell analysis, we verified drug-related reactive pattern of Proliferation kit

Oncogene Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7544 Table 1 Summary of the investigated cell lines Cell line ArrayID Tumor origin Relative fold Selecting agent Drug concentration Reference fold resistance to agent of selecting agent (mg/ml)

HT29 shcg143 Colon 1 — HT29RDB shcg146 18 Daunorubicin 0.125 Sinha et al. (1999) HT29RNOV shcg144 100 Mitoxantrone 0.2 Sinha et al. (1999) EPP85–181P shbv233 Pancreatic 1 — EPP-181RDB shbv235 1800 Daunorubicin 2.5 Lage and Dietel (2002) EPP-181RNOV shbv234 8 Mitoxantrone 0.02 Lage and Dietel (2002) EPG-257P shbv236 Gastric 1 — EPG-257RDB shbv225 1857 Daunorubicin 2.5 Lage et al. (2000) EPG-257RNOV shbv237 457 Mitoxantrone 0.2 Dietel et al. (1990) MCF7 shcg156 Breast 1 — MCF7/Adr shcg157 192 Doxorubicin 0.05 Fairchild et al. (1987) MDA-MB-231 shcg149 Breast 1 — MDA-MB-231RNOV shcg150 93 Mitoxantrone 0.02 Not published

lines. During the Prediction Analysis for Microarrays obtained from the Stanford Microarray Database (PAM) training analysis, we applied batch adjustment to (http://genome-www5.stanford.edu/). The tumor speci- filter out the differences between the four different cell mens were selected, because the patients had received types. The prediction analysis of microarrays was doxorubicin monotherapy and the survival parameters performed with the genes differentially expressed were reported. Moreover, gene expression profiling of independent of tissue origin at thresholds 3. Centroid breast cancers was performed on the same technical plots for the top 30 genes associated with doxorubicin platform. Recently, a considerable divergence across and mitoxantrone resistance are shown in Figure 2. We different platforms has been reported (Tan et al., 2003). have listed the resistance-associated genes in Supple- We dismissed one patient, who died within 3 months mentary Tables 2 and 3. The mRNA levels of these after surgery probably due to postoperative complica- genes showed a 15-fold decrease to 20-fold increase in tions. Although Sorlie et al. used the same cDNA the resistant cells compared to the sensitive parental cell microarray facility, the proportion of common cDNA lines. Furthermore, we were able to confirm the clones represented on their arrays was only 16% previously reported altered expression of P-gp, MRP compared to our study, and most of the selected top and BCRP in the investigated cell lines (Ross et al., sequences were not present. For this reason, we 1999; Sinha et al., 1999). performed our prediction independently of the pre- To classify discriminatory genes, we have performed viously selected gene list: first, we constructed a matched hierarchical clustering on the genes associated with drug gene list between our arrays and the arrays used resistance (Figure 3). The clustering dendrograms show by Sorlie et al. This matched gene list contained 5923 the 79 genes associated with doxorubicin resistance and sequences (Supplementary Table 4). Gene sets obtained the 70 genes associated with mitoxantrone resistance. by comparing the parental and resistant cell lines Three different cDNA clones representing the gene were used as a training set for the prediction of the encoding the ABCB1 transporter consistently cluster in clinical samples (test set). For the prediction, we applied the doxorubicin-resistant group. PAM to classify the patients rather than to build a TaqMan analysis of the ABCB1 gene using two new predictor based on this limited gene set. Therefore, independent probes verified these results: the ABCB1 we set the threshold to 1, having about 1500 genes is upregulated in the EPP181RDB, EPG257RDB and analysed for the classification. Based on the correspond- MCF7RADR, but not in the HT29RDB cell line ing expression profiles, we then grouped the patients’ compared to the expression in the parental cell lines tumors as predicted to be doxorubicin resistant or (Figure 4). The upregulation of caspase I in the sensitive. Afterwards, we performed a survival analysis doxorubicin-resistant group (fold change: 5.2, (Figure 5 and Supplementary Table 5). The patients Po0.001), and the downregulation of TOP2A in the whose tumors were classified as being resistant had novantron-resistant group (fold change: À4.3, a mean survival time of 32.9718.7 months (n ¼ 23), Po0.001) was verified by quantitative RT–PCR. while in the group of patients whose tumors were classified as being sensitive, the mean survival was 49.7726.1 months (n 21, resistant vs sensitive: Prediction testing on breast-cancer specimens ¼ P ¼ 0.034). Only three patient tumors (7% of total) We used the gene expression profiles derived from were classified as sensitive, while the patients died earlier doxorubicin-resistant cell lines for an independent than 2 years after tumor resection. We failed to use the classification of clinical specimens. As a test set for this training set of genes related to mitoxantrone resistance classification, we used 44 breast cancer expression for a similar prediction, this was due to the lack of profiles described by Sorlie et al. (2003). The data were reported clinical data.

Oncogene Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7545 ab SLC38A2 W49672 AI160645 SYTL2 PTPRH C12orf2 AA629344 T81317 HS3ST1 AI023525 AA706085 IRX1 CALD1 AI206976 AA610036 EPB41L2 AP4E1 LEMD2 AA857380 N51291 CISH MAFF ABCB1 AI821364 FOXJ1 SFRP4 AI668613 MGC29898 AA459410 RRAGD EPLIN AI073991 AA702568 AA776813 AA622998 RASA1 C5orf13 LAMA2 AA007305 N50659 P2RY6 AA282985 AI217376 AA677700 R35649 MGC21518 FLJ35630 MLN51 AI220690 AA428166 AA701154 ELOVL5 C21orf87 ELAVL4 HMP19 N51117 ABCB1 AMOTL1 ABCB1 AI820815

shrunken average expression shrunken average expression Figure 2 Centroid plot showing the top 30 clones with highest discriminatory potential on doxorubicin (a) and mitoxantrone (b) resistance groups after shrinkage with threshold 3. Gene names or Genbank Accession numbers are shown

Discussion Gottesman et al., 2002) and supports the notion that MDR is highly complex. The generally accepted major We have compared gene expression profiles of drug- candidates for drug resistance are the MDR proteins, sensitive and closely related drug-resistant cell lines of which actively expel cytotoxic drugs from the cell, different origin. We have identified the top 79 genes maintaining the drug level below a cell killing threshold associated with doxorubicin resistance and the top 70 (Bodo et al., 2003). Inhibition of ATP-dependent genes associated with mitoxantrone resistance. Of these membrane transporters can reverse multidrug resistance genes, 80 were upregulated and 69 were downregulated. in preclinical models as well as in cancer patients This confirms previous data (Turton et al., 2001; (Robert and Jarry; 2003; Thomas and Coley, 2003). We

Oncogene Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7546

Figure 3 Hierarchical clustering of sensitive and resistant cell lines showing the expression patterns of discriminating genes. Bright red represents the highest levels and bright green the lowest level of expression. Each column in the ﬁgure represents an individual sample, while each row represents an individual gene. Note the upregulation of multiple clones of the ABCB1 gene in the doxorubicin-resistant group. (a) Doxorubicin resistance (b) mitoxantrone resistance

Oncogene Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7547 15 1.2 Censored 10 1 Resistant 5 Sensitive 0.8 0 0.6 -5 Hs00184491 Probability 0.4 relative expression -10 Hs00184500 0.2 -15 0 0 20406080100 Overall survival (months)

HT29RDB Figure 5 Survival analysis of 44 breast cancer patients. The tumor MCF7RADR EPP181RDB EPG257RDB samples are divided into resistant and sensitive groups based on Figure 4 TaqMan analysis of the ABCB1 gene using two similarity to the expression proﬁle of cell lines resistant to independent probes correlates to the results obtained on the cDNA doxorubicin treatment compared to their parental cells. Log-rank arrays. Relative expression (delta delta Ct) is the log difference to test (Cox–Mantel), P ¼ 0.0334 the parental cell line after normalization to the GAPDH control

have grouped the already identified genes involved in resistant and sensitive predictions shows high represen- resistance against doxorubicin and mitoxantrone into tation of sensitive samples in the luminal A cluster and functional classes comprising elements involved in low representation in the luminal B cluster (Figure 6). signaling, cell migration, transport, apoptosis and Our result suggests that higher drug sensitivity may be detoxification (Table 2). We retrieved additionally a responsible for better patient survival in the luminal A high proportion of differentially expressed sequences cluster. This is also supported by the altered expression without known function. In all, 49% percent of in the genes that have previously been associated with sequences associated with doxorubicin resistance and multidrug resistance (for example MDR1, CYP27A1, 44% of sequences related to mitoxantrone resistance EGFR, CISH, TOP2A). Furthermore, our prediction were ESTs, which might represent novel genes asso- model seems to be applicable for the unclassified ciated with cytostatic multidrug resistance. samples from Sorlie et al., with respect to survival, but Our experiment utilized expression patterns of pre- not histomorphology. characterized and well-defined resistant and sensitive While doxorubicin was used to induce the resistance cell line pairs for therapy response prediction. Up to in our investigated breast cell line, the selective agent for date clinical and pathological markers poorly predict the other cell lines (EPG85–257, EPP85–181 and HT29) response to chemotherapy probably due to the multi- was daunorubicin. Although these are similar anthracy- factorial nature of cancer multidrug resistance. The use cline drugs acting via topoisomerase inhibition, this has of several different cell lines for the construction of a to be taken in consideration when future prediction prediction model enables the application on different analyses are performed. tissues. So far, gene expression profiling has already A recent microarray study correlated in vitro response been used for identifying factors related to the resistance to chemotherapy in four cell lines to response in breast toward cytostatic drugs in cell lines (Kudoh et al., 2000; tumors (Troester et al., 2004). The common discrimina- Staunton et al., 2001) and human cancer xenografts tory genes in their study do not overlap with our results. (Zembutsu et al., 2002). Studies based on cancer The reason for this is probably due to the study design: specimens have concentrated on the classification of while Troester et al. have investigated treatment tumor subtypes and patient prognosis rather than on response after 12-, 24- and 36-h drug administration, drug response (Sorlie et al., 2003; van’t Veer et al., we have focused on the pre-existing characteristics of the 2003). Few studies have correlated expression profiles of resistant cell lines. Although short- and long-term drug breast tumors to their sensitivity to anticancer drugs administration may have different effect on gene (Chang et al., 2003). A novel study identified a expression, both sets of responsive genes could be predictive gene list for the sequential T/FAC (paclitaxel used in predictive tests. A combined study with all and fluorouracil þ doxorubicin þ cyclophosphamide) selected genes could give more insight into the dynamics protocol in breast cancer (Ayers et al., 2004). and course of gene expression changes during treatment. By hierarchical clustering, Sorlie et al. identified Doxorubicin resistance was investigated using micro- different subgroups of breast carcinomas. They found arrays in two recent studies (Suganuma et al., 2003; that patients in the newly identified luminal A cluster Kang et al., 2004). Suganuma et al. investigated genes had better prognosis than patients in the luminal B and related to chemoresistance towards cisplatin, doxorubi- other clusters. We found that our distribution of cin, mitomycin C and 5-fluorouracil in gastric cancer.

Oncogene Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7548 Table 2 Functional description of a selected set of the top genes best associated with resistance Gene symbol Genebank Functional Fold Protein name Notes Accession group change

Doxorubicin resistance ABCB1 AA455911, Transport 5.2 P-gp ATP-dependent drug efflux, leads to decreased drug accumulation AA887211, 4.7 AA135957 9.7 GJA5 AA669536 Transport À4.8 Gap junction protein alpha 5 Integral membrane protein that forms transmembrane channels, small molecule transport, cell–cell communication CASP1 T95052 Apoptosis 20.2 Caspase I Induces cell apoptosis CST1 AI963401 Apoptosis 2.3 Cystatin SN Cysteine proteinase inhibitor CYP27A1 N66957 Detoxification À2.0 Cytochrome P450, family 27, Monooxygenases which catalyse many reactions subfamily A, polypeptide 1 involved in drug metabolism FMO2 AI820530 Detoxification 1.5 Flavin containing NADPH-dependent flavoenzymes that catalyse the monooxygenase 2 oxidation of soft nucleophilic heteroatom centers in drugs, pesticides and xenobiotics PTPRH AI924306 Signaling 8.8 Protein tyrosine phosphatase, Regulates cell growth, differentiation, mitotic cycle and receptor type H oncogenic transformation, possible post-translational phosphorylation of P-gp AP4E1 AI934996 Signaling À2.3 Epsilon 1 subunit of the adaptor- Sorts integral membrane proteins at various stages of related protein complex 4 the endocytic and secretory pathways CISH AA427521 Signaling À1.5 Cytokine-inducible SH2- Cytokine-inducible negative regulators of cytokine containing protein signaling P2RY6 AA778919 Signaling À3.3 Pyrimidinergic receptor P2Y2 G-protein-coupled pyrimidinergic receptor P2Y2 ANGPT1 W84856 Signaling 4.8 Angiopoietin 1 Vascular development and angiogenesis, the secreted glycoprotein activates the endothelial cell-specific tyrosine-protein kinase receptor by inducing its tyrosine phosphorylation CALD1 AA402898 Cell migration 6.8 Caldesmon 1 A calmodulin- and actin-binding protein, and plays a vital role in the regulation of smooth muscle and nonmuscle contraction ADD2 AA019320 Cell migration À1.4 Adducin 2 (beta) Assembly of cell–cell contact in epithelial tissues TIMP2 AA486280 Cell migration 7.6 Regulated tissue inhibitor Expression is largely constitutive, administration of of metalloproteinase 2 TIMP2 showed to modify the invasive behavior of pancreatic cancer and have significant antitumor effects in vivo

Mitoxantrone resistance MAFF T47417 Detoxification 2.1 v-maf musculoaponeurotic Basic leucine zipper (bZIP) transcription factor that fibrosarcoma oncogene lacks a transactivation domain, may also be involved in homolog F the cellular stress response CYP4F11 AA991369 Detoxification 1.5 — A member of the cytochrome P450 superfamily, the specific function of this protein has not been determined SFRP4 AA486838 Apoptosis À15.5 Secreted frizzled-related Soluble modulators of Wnt signaling, the expression in related protein 4 ventricular myocardium correlates with apoptosis-related gene expression TOP2A AA504348 DNA repair À2.3 Topoisomerase II alpha Controls and alters the topologic states of DNA during transcription, the decreased topoisomerase II activity leads to lower inducibility of the apoptotic pathway SMARCE1 AA599120 DNA repair À1.5 SWI/SNF-related, matrix- part of the large ATP-dependent chromatin remodeling associated, actin-dependent complex SWI/SNF, which is required for transcriptional regulator of chromatin, activation of genes normally repressed by chromatin subfamily e, member 1 RRAGD AI095082 Signaling À5.6 Ras-related GTP binding D — RASA1 N22145 Signaling À3.7 RAS p21 protein activator 1 Part of the GAP1 family of GTPase-activating proteins, stimulates the GTPase activity of normal RAS p21 but not its oncogenic counterpart, the protein enhances the weak intrinsic GTPase activity of RAS proteins resulting in the inactive GDP-bound form of RAS, thereby allowing control of cellular proliferation and differentiation DPM1 AA918102 Signaling À1.4 Dolichyl-phosphate mannosyl- A donor of mannosyl residues on the lumenal side of the transferase polypeptide 1, endoplasmic reticulum (ER), lack of Dol-P-Man results catalytic subunit in defective surface expression of GPI-anchored proteins PTPN5 AI183359 Signaling 7.8 The protein tyrosine phospha- — tase, nonreceptor type 5 IFNGR1 AA281497 Signaling 1.4 Interferon gamma receptor 1 — TNFAIP3 R70479 Signaling 2.0 Tumor necrosis factor Zinc-finger protein, and has been shown to inhibit NF-k alpha-induced protein 3 B activation as well as TNF-mediated apoptosis, knockout studies of a similar gene in mice suggested that this gene is critical for limiting inflammation by terminating TNF-induced NF-kB responses

ESTs are excluded.

Oncogene Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7549 12 et al., 2000). Likewise, the human pancreatic carcinoma cell Resistant line EPP85–181P and its drug-resistant sublines were estab- 10 lished in the same way (Lage and Dietel, 2002). In accordance Sensitive drug-resistant cell variants of the human colon carcinoma cell 8 line HT29 with (Chen et al., 1987), and the breast cancer cell lines MCF7 (Soule et al., 1973) and MDA-MB-231 (Cailleau 6 et al., 1974) were also established in our laboratory. Cell culture was performed as described (Lage et al., 2000;

Frequency 4 Lage and Dietel, 2002). The selection agents and conditions are summarized in Table 1. Daunorubicin and doxorubicin 2 were obtained from Farmitalia Carlo Erba (Freiburg, Germany); mitoxantrone was purchased from Lederle (Wol- 0 fratshausen, Germany).

Basal Cell proliferation assay ERBB2+

luminal A luminal B We re-examined the drug resistance of each resistant derivative

Clusters by Sorlie et al Unclassified relative to the sensitive parental line by culturing the cells in the presence of drugs ranging from 0.0001 to 100 mg/ml for a Figure 6 Distribution of resistant and sensitive samples among period of 5 days. Proliferation was assessed using an XTT Cell the groups created by hierarchical clustering (described by Sorlie Proliferation kit following the manufacturer’s instructions et al.) to identify different subgroups of the patients shows high (Roche, Germany) (Roehm et al., 1991). To determine the IC50 representation of sensitive samples in the luminal A cluster and low values, the formazan release of control cells without drug was representation in the luminal B cluster set to 100%. Linear regression was plotted for the linear region of the growth curves obtained in at least three independent experiments for each cell line. However, the 20 specimens analysed in this study mRNA preparation exhibited drug resistance, and none of them was sensitive to drug treatment. Therefore, a list of RNA isolation was performed 24 h after cell seeding from all candidate genes specifically associated with doxorubicin cell lines. The log-phase growing cells were lysed with TRIzols resistance could not be established. In the second study, and RNA was isolated following the manufacturer’s protocol. performed on a different microarray platform, Kang Poly(A) RNA was prepared using the Fast Track 2.0 mRNA isolation kit (Invitrogen Life Technologies). RNA quality from et al. identified gene expression patterns related to each sample was assessed by visualization of the 28S/18S resistance against 5-fluorouracil, cisplatin and doxo- ribosomal RNA ratio using an Agilent 2100 Bioanalyser. rubicin in 14 human gastric cancer cells. We have compared the published set of 250 differentially Array hybridization regulated genes with our prediction profiles, but we have not detected an overlap. These results are in line We interrogated cDNA arrays with 43 009 cDNA clones, with the findings of a recent study demonstrating that representing approximately 30 000 unique genes for studying different gene signatures can achieve similar prediction the expression patterns of the 13 cell lines. Microarrays were hybridized as described previously (Alizadeh et al., 2000; Sorlie success for the same classification problem (Ein-Dor et al., 2003). Briefly, 3 mg of mRNA was reverse transcribed in et al., 2005). Moreover, alternative microarray plat- the presence of Cyt5-dUTP. In all, 50 mg of Universal Human forms may yield different results (Tan et al., 2003). RNA (Stratagene) was Cyt3 labeled as a reference. Hybridi- We designed our study to identify patterns of gene zation was carried out at 651C for 12–16 h. Slides were washed expression that could be used in a predictive test for the for 3 min in 2 Â SSC, 0.075% SDS at 501C, then for 3 min in investigated chemotherapy agents. Our study shows that 1 Â SSC and finally for 2 min in 0.1 Â SSC. Glass slides were DNA microarray technology can effectively be used for promptly centrifugated at 600 r.p.m. for 10 min and the predicting the response to chemotherapy and indirectly fluorescent images of hybridized microarrays were obtained the patient prognosis. The results of the resistance using a GenePix 4000 microarray scanner (Axon Instruments). pattern derived from our cell lines can help in the The images of all scanned slides were inspected for artifacts, and aberrant spots and slide regions were flagged for exclusion development of customized DNA arrays for doxorubi- from analyses. cin and mitoxantrone resistance. To define ultimately the complete molecular portrait of drug sensitivity and Statistical analysis resistance, our results should be validated in a study with a large independent cohort of patients. Log 2 (Cy5/Cy3) was retrieved for each spot prior to further analysis. Cy5/Cy3 is the normalized ratio of the background- corrected intensities. To reduce the effect of nonspecific fluorescence, we filtered spots as follows: the mean background Materials and methods for the red and green signals of each array was determined by calculating the average of the mean background of the Cell lines corresponding color intensities of all spots in the array. The The human gastric carcinoma cell line EPG85–257P and its net signal was determined by subtraction of this local drug-resistant derivatives were established in our laboratory background from the average intensity of each spot. Then, and described in detail previously (Dietel et al., 1990; Lage the signals were normalized by applying a single multiplicative

Oncogene Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7550 factor to all intensities measured for the red dye (Cy5). was performed with total RNA. Each quantitative real-time The normalization factor was chosen in order to achieve the PCR (951C 30 s, 45 cycles of 951C10s,601C10s,721C10s) mean log(Cy5/Cy3) for the spots of 0.0, this effectively defining was carried out using the LightCycler (LightCycler DNA the signal-intensity-weighted ‘average’ spot on each array to Master Hybridization Probes kit, Roche Diagnostics, Man- have a Cy5/Cy3 ratio of 1.0. For identifying predictive genes, nheim, Germany). Expression of caspase I, topoisomerase II the PAM (v.1.12) package was used as described by Tusher alpha and of the housekeeping gene glucose-6-phosphate et al. (2001). PAM uses soft thresholding to produce a dehydrogenase (G6PDH) was determined in parallel, each shrunken centroid, which allows the selection of genes with carried out in duplicate per sample. For caspase I, a 132 bp high predictive potential. Hierarchical clustering was per- amplicon (forward primer: 50-cagcacgttcctggtgttc-30; FITC- formed using the Genesis software (Sturn et al., 2002). labeled probe: 50-ggaagaaacactctgagcaagtccc-30-FITC; Univariate Kaplan–Meier analysis was performed by using LCRed640-labeled probe: LCRed640-50-gatatactacaactcaatg- Winstat for Excel (R Fitch Software, Staufen, Germany). caatctttaacatgt-30; reverse primer: 50-cttgggcagttcttggtattc-30), All hybridization data are stored in the Stanford Micro- for topoisomerase II alpha, 112 bp amplicon (forward primer: array Database and can be accessed at http://genome-www5. 50-atcttggatcaaatgttgtccc-30; FITC-labeled probe: 50-aatc- stanford.edu/. The combined data sets for prediction analysis cagtcctcttagaaattggtttctct-30-FITC; LCRed640-labeled probe: are made available as supplementary information on the LCRed640-50-tttgggagattcagactcagaggcag-30; reverse primer: journal website. 50-ctcttgacctgtcccctctg-30) and for G6PDH, a 113 bp amplicon was produced, detected by gene-specific fluorescein and Validation LCRed640-labeled hybridization probes (caspase I and topoisomerase II alpha primer syntheses: BioTeZ, Berlin, Ger- We have performed TaqMan verification for the ABCB1 gene many; probe syntheses: TIB MOLBIOL, Berlin, Germany; to validate the results obtained on cDNA arrays using an G6PDH primer and probe syntheses: Roche Diagnostics). The Applied Biosystems 7900HT Sequence Detection System. calibrator cDNA derived from EPP181 cells and was employed The results of two independent ABCB1 TaqMan probes in serial dilutions simultaneously in each run. (Hs00184491 and Hs00184500) were normalized to GAPDH (probe Hs99999905) expression. The measurements were performed as described in the 7900HT products user guide Acknowledgements (http://www.appliedbiosystems.com, CA, USA). For TaqMan Violeta Serra and Balazs Gyo¨ rffy are supported by Marie data analysis the SDS 2.2 software was used. Curie fellowships (HPMD-CT-2000-00001). Violeta Serra’s To verify the microarray results for caspase I and for work at Stanford University was supported by the Berliner topoisomerase II alpha at the mRNA level, quantitative real- Krebsgesellschaft. We would like to thank Mrs Schu¨ tze, Mrs time RT–PCR was performed. Reverse transcriptase reaction Schaefer and Mrs Pacyna-Gengelbach for technical help.

References

Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Fairchild CR, Ivy SP, Kao-Shan CS, Whang-Peng J, Rosen N, Rosenwald A, Boldrick JC, Sabet H, Tran T, Yu X, Powell Israel MA, Melera PW, Cowan KH and Goldsmith ME. JI, Yang L, Marti GE, Moore T, Hudson Jr J, Lu L, Lewis (1987). Cancer Res., 47, 5141–5148. DB, Tibshirani R, Sherlock G, Chan WC, Greiner TC, Fisher TC, Milner AE, Gregory CD, Jackman AL, Aherne Weisenburger DD, Armitage JO, Warnke R, Levy R, GW, Hartley JA, Dive C and Hickman JA. (1993). Cancer Wilson W, Grever MR, Byrd JC, Botstein D, Brown PO Res., 53, 3321–3326. and Staudt LM. (2000). Nature, 403, 503–511. Goldstein LJ, Galski H, Fojo A, Willingham M, Lai SL, Alonso O, Delgado L, Nunez M, Vargas C, Lopera J, Gazdar A, Pirker R, Green A, Crist W and Brodeur GM. Andruskevicius P, Sabini G, Gaudiano J, Muse IM and (1989). J. Natl. Cancer Inst., 81, 116–124. Roca R. (2002). Nucl. Med. Commun., 23, 765–771. Gottesman MM, Fojo T and Bates SE. (2002). Nat. Rev. Ayers M, Symmans WF, Stec J, Damokosh AI, Clark E, Hess Cancer, 2, 48–58. K, Lecocke M, Metivier J, Booser D, Ibrahim N, Valero V, Kang HC, Kim IJ, Park JH, Shin Y, Ku JL, Jung MS, Yoo Royce M, Arun B, Whitman G, Ross J, Sneige N, BC, Kim HK and Park JG. (2004). Clin. Cancer Res., 10, Hortobagyi GN and Pusztai L. (2004). J. Clin. Oncol., 22, 272–284. 2284–2293. Kudoh K, Ramanna M, Ravatn R, Elkahloun AG, Bittner Bodo A, Bakos E, Szeri F, Varadi A and Sarkadi B. (2003). ML, Meltzer PS, Trent JM, Dalton WS and Chin KV. Toxicol. Lett., 140–141, 133–143. (2000). Cancer Res., 60, 4161–4166. Cailleau R, Young R, Olive M and Reeves Jr WJ. (1974). J. Kuerer HM, Newman LA, Smith TL, Ames FC, Hunt KK, Natl. Cancer Inst., 53, 661–674. Dhingra K, Theriault RL, Singh G, Binkley SM, Sneige N, Chang JC, Wooten EC, Tsimelzon A, Hilsenbeck SG, Buchholz TA, Ross MI, McNeese MD, Buzdar AU, Gutierrez MC, Elledge R, Mohsin S, Osborne CK, Cham- Hortobagyi GN and Singletary SE. (1999). J. Clin. Oncol., ness GC, Allred DC and O’Connell P. (2003). Lancet, 362, 17, 460–469. 362–369. Lage H and Dietel M. (2002). J. Cancer Res. Clin. Oncol., 128, Chen TR, Drabkowski D, Hay RJ, Macy M and Peterson Jr 349–357. W. (1987). Cancer Genet. Cytogenet., 27, 125–134. Lage H, Jordan A, Scholz R and Dietel M. (2000). Int. J. Chollet P, Amat S, Cure H, de Latour M, Le Bouedec G, Hyperthermia, 16, 291–303. Mouret-Reynier MA, Ferriere JP, Achard JL, Dauplat J and Lowe SW, Bodis S, McClatchey A, Remington L, Ruley HE, Penault-Llorca F. (2002). Br. J. Cancer, 86, 1041–1046. Fisher DE, Housman DE and Jacks T. (1994). Science, 266, Dietel M, Arps H, Lage H and Niendorf A. (1990). Cancer 807–810. Res., 50, 6100–6106. Nooter K, Westerman AM, Flens MJ, Zaman GJ, Scheper RJ, Ein-Dor L, Kela I, Getz G, Givol D and Domany E. (2005). van Wingerden KE, Burger H, Oostrum R, Boersma T and Bioinformatics, 21, 171–178. Sonneveld P. (1995). Clin. Cancer Res., 1, 1301–1310.

Oncogene Prediction of drug sensitivity using cDNA array signatures B Gyo¨rffy et al 7551 Robert J and Jarry C. (2003). J. Med. Chem., 46, 4805–4817. Sturn A, Quackenbush J and Trajanoski Z. (2002). Bioinfor- Roehm NW, Rodgers GH, Hatﬁeld SM and Glasebrook AL. matics, 18, 207–208. (1991). J. Immunol. Methods, 142, 257–265. Suganuma K, Kubota T, Saikawa Y, Abe S, Otani Y, Ross DD, Yang W, Abruzzo LV, Dalton WS, Schneider E and Furukawa T, Kumai K, Hasegawa H, Watanabe M, Lage H. (1999). J. Natl. Cancer Inst., 91, 429–433. Kitajima M, Nakayama H and Okabe H. (2003). Cancer Sinha P, Hutter G, Kottgen E, Dietel M, Schadendorf D and Sci., 94, 355–359. Lage H. (1999). Electrophoresis, 20, 2961–2969. Tan PK, Downey TJ, Spitznagel Jr EL, Xu P, Fu D, Dimitrov Sciuto R, Pasqualoni R, Bergomi S, Petrilli G, Vici P, Belli F, DS, Lempicki RA, Raaka BM and Cam MC. (2003). Nucleic Botti C, Mottolese M and Maini CL. (2002). J. Nucl. Med., Acids Res., 31, 5676–5684. 43, 745–751. Thomas H and Coley HM. (2003). Cancer Control, 10, Sorlie T, Tibshirani R, Parker J, Hastie T, Marron JS, Nobel 159–165. A, Deng S, Johnsen H, Pesich R, Geisler S, Demeter Troester MA, Hoadley KA, Sorlie T, Herbert BS, Borresen- J, Perou CM, Lonning PE, Brown PO, Borresen-Dale AL Dale AL, Lonning PE, Shay JW, Kaufmann WK and Perou and Botstein D. (2003). Proc. Natl. Acad. Sci. USA, 100, CM. (2004). Cancer Res., 64, 4218–4226. 8418–8423. Turton NJ, Judah DJ, Riley J, Davies R, Lipson D, Sotiriou C, Neo SY, McShane LM, Korn EL, Long PM, Styles JA, Smith AG and Gant TW. (2001). Oncogene, 20, Jazaeri A, Martiat P, Fox SB, Harris AL and Liu ET. 1300–1306. (2003). Proc. Natl. Acad. Sci. USA, 100, 10393–10398. Tusher VG, Tibshirani R and Chu G. (2001). Proc. Natl. Acad. Soule HD, Vazguez J, Long A, Albert S and Brennan M. Sci. USA, 98, 5116–5121. (1973). J. Natl. Cancer Inst., 51, 1409–1416. van’t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Staunton JE, Slonim DK, Coller HA, Tamayo P, Angelo MJ, Bernards R and Friend SH. (2003). Breast Cancer Res., 5, Park J, Scherf U, Lee JK, Reinhold WO, Weinstein JN, 57–58. Mesirov JP, Lander ES and GolubTR. (2001). Proc. Natl. Zembutsu H, Ohnishi Y, Tsunoda T, Furukawa Y, Katagiri T Acad. Sci. USA, 98, 10787–10792. and Ueyama Y. (2002). Cancer Res., 62, 518–527.

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

Oncogene