Open Full Page

Published OnlineFirst September 29, 2015; DOI: 10.1158/1078-0432.CCR-15-0876

Cancer Therapy: Preclinical Clinical Cancer Research A Patient-Derived, Pan-Cancer EMT Signature Identiﬁes Global Molecular Alterations and Immune Target Enrichment Following Epithelial-to-Mesenchymal Transition Milena P. Mak1,2, Pan Tong3, Lixia Diao3, Robert J. Cardnell1, Don L. Gibbons1,4, William N. William1, Ferdinandos Skoulidis1, Edwin R. Parra5, Jaime Rodriguez-Canales5, Ignacio I.Wistuba5, John V.Heymach1, John N.Weinstein3, Kevin R.Coombes6, Jing Wang3, and Lauren Averett Byers1

Abstract

Purpose: We previously demonstrated the association between EMT markers across diverse tumor types and identifies differences epithelial-to-mesenchymal transition (EMT) and drug response in in drug sensitivity and global molecular alterations at the DNA, lung cancer using an EMT signature derived in cancer cell lines. RNA, and protein levels. Among those changes associated with Given the contribution of tumor microenvironments to EMT, we EMT, pathway analysis revealed a strong correlation between EMT extended our investigation of EMT to patient tumors from 11 and immune activation. Further supervised analysis demonstrat- cancer types to develop a pan-cancer EMT signature. ed high expression of immune checkpoints and other druggable Experimental Design: Using the pan-cancer EMT signature, we immune targets, such as PD1, PD-L1, CTLA4, OX40L, and PD-L2, conducted an integrated, global analysis of genomic and protein tumors with the most mesenchymal EMT scores. Elevated omic profiles associated with EMT across 1,934 tumors including PD-L1 protein expression in mesenchymal tumors was confirmed breast, lung, colon, ovarian, and bladder cancers. Differences in by IHC in an independent lung cancer cohort. outcome and in vitro drug response corresponding to expression Conclusions: This new signature provides a novel, patient- of the pan-cancer EMT signature were also investigated. based, histology-independent tool for the investigation of EMT Results: Compared with the lung cancer EMT signature, the and offers insights into potential novel therapeutic targets for patient-derived, pan-cancer EMT signature encompasses a set of mesenchymal tumors, independent of cancer type, including core EMT genes that correlate even more strongly with known immune checkpoints. Clin Cancer Res; 22(3); 609–20. 2015 AACR.

Introduction phenotype, a process known as "epithelial-to-mesenchymal transition" (EMT). Studies have shown that EMT plays an important Over the past decade, multiple lines of evidence have suggested biologic role in cancer progression, metastasis, and drug resistance that epithelial cancers can transform into a more mesenchymal (1–3). Tools that facilitate the study of EMT, therefore, will provide new insights into molecular regulation and evolution 1Department of Thoracic/Head and Neck Medical Oncology, The Uni- during oncogenesis and may help to improve treatments for versity of Texas MD Anderson Cancer Center, Houston,Texas. 2Medical mesenchymal cancers. The development of EMT signatures or Oncology, Instituto do Cancer do Estado de Sao Paulo, Faculdade de other molecular markers to identify whether a cancer has under- Medicina da Universidade de Sao Paulo, Sao Paulo, Brazil. 3Depart- ment of Bioinformatics and Computational Biology, The University of gone EMT is an area of active research. Most studies in this area, Texas MD Anderson Cancer Center, Houston, Texas. 4Department of however, have focused on a single tumor type and/or on preclin- Molecular and Cellular Oncology,The University of Texas MDAnderson ical models (3–7). Cancer Center, Houston, Texas. 5Department of Translational and Molecular Pathology, The University of Texas MD Anderson Cancer We previously developed a robust, platform-independent EMT Center, Houston, Texas. 6Department of Biomedical Informatics, Ohio signature based on a set of 54 lung cancer cell lines (hereafter State University, Columbus, Ohio. called the lung cancer EMT signature). In vitro, this lung cancer Note: Supplementary data for this article are available at Clinical Cancer EMT signature predicted resistance to EGFR and PI3K/Akt inhi- Research Online (http://clincancerres.aacrjournals.org/). bitors and identified AXL as a potential therapeutic target for M.P. Mak and P. Tong share first authorship. overcoming resistance to EGFR inhibitors [a common treatment – Corresponding Authors: Lauren Averett Byers, The University of Texas MD for non small cell lung cancer (NSCLC)]. We and others observed fi Anderson Cancer Center, 1515 Holcombe Blvd, Unit 0432, Houston, TX 77030. signi cantly greater resistance to EGFR inhibitors in lung cancers Phone: 713-792-6363; Fax: 713-792-1220; E-mail: [email protected]; and that had undergone EMT, as determined by either their baseline Jing Wang, Department of Bioinformatics and Computational Biology, The EMT signature (3) or the expression of specific EMT markers after University of Texas MD Anderson Cancer Center, Houston, TX. Phone: 713- the development of acquired EGFR inhibitor resistance (8). 794-4190; Fax: 713-563-4242; E-mail: [email protected] Although it is appealing to apply the lung cancer EMT signature doi: 10.1158/1078-0432.CCR-15-0876 to diverse tumor types, this approach may be limited by tumor 2015 American Association for Cancer Research. type-specific differences in EMT or discrepancies between cell line

www.aacrjournals.org 609

Mak et al.

CCLE by SNP fingerprinting at multiple steps, profiles were Translational Relevance compared with existing profiles (10, 11). We used level-3 TCGA Epithelial-to-mesenchymal transition (EMT) is associated pan-cancer data (12, 13), including RNAseqV2, reverse phase with resistance to many approved drugs and with tumor protein array (RPPA), miR, copy number, mutation, and clinical progression. Here, we sought to characterize the common data. A summary of the TCGA data can be found in Supplemen- biology of EMT across multiple tumor types and identify tary Table S1. The Profiling of Resistance Patterns and Oncogenic potential therapeutic vulnerabilities in mesenchymal tumors. Signaling Pathways in Evaluation of Cancers of the Thorax and A patient-derived, pan-cancer EMT signature was developed Therapeutic Target Identification (PROSPECT) dataset of surgi- using 11 distinct tumor types from The Cancer Genome Atlas. cally resected NSCLC has been previously described (9). Array- Mesenchymal tumors had similar patterns of gene, protein, based expression profiling of PROSPECT tumors was performed and miRNA expression independent of cancer type. Tumors using the Illumina Human WG-6 v3 BeadChip according to the with mesenchymal EMT scores not only had a higher expres- manufacturer's protocol and gene expression data have been sion of the receptor tyrosine kinase Axl (previously implicated previously deposited in the GEO repository (GSE42127). with EMT and associated with response to Axl inhibitors), but also expressed high levels of multiple immune checkpoints Developing the pan-cancer EMT signature including PD-L1, PD1, CTLA4, OX40L, and PD-L2. This novel We adopted an approach similar to that used by Byers and finding, which was validated in an independent patient colleagues (3) to derive the pan-cancer EMT signature. We used cohort, highlights the possibility of utilizing EMT status— four established EMT markers, namely, CDH1 (epithelial marker, independent of cancer type—as an additional selection tool to E type), CDH2 (mesenchymal marker, M type), VIM (M type), and select patients who may benefit from immune checkpoint FN1 (M type) as seeds to derive the pan-cancer EMT signature on blockade. the basis of TCGA pan-cancer RNAseq data. In particular, we computed correlations (Pearson correlation, r) between all mRNAs in the RNAseq data and each of the established EMT markers for each individual tumor type. models [the starting point of the lung cancer EMT signature (3)] fi and clinical patient samples—especially in regard to the interplay Identi cation of EMT-associated genomic features between EMT, tumor microenvironments, and immune response We computed Pearson correlation between the EMT score and at different disease sites (9). Therefore, we built upon our previous individual genomic features because these features are normally approach to derive a pan-cancer EMT signature by leveraging distributed. The genomic features per gene included the mRNA molecular datasets from 11 tumor types (n ¼ 1,934 tumors expression, RPPA expression (either total protein or phosphory- overall; see Supplementary Table S1) using clinical patient sam- lated protein), and miRNA expression levels (5p and 3p mature ples from The Cancer Genome Atlas (TCGA). The datasets were strands). obtained from primarily epithelial malignancies, such as breast, colorectal, and endometrial cancer, and two cohorts of lung Association between EMT score and other covariates cancer (adenocarcinoma and squamous cell carcinoma). We applied the log-rank test to assess the association between We tested the performance of the pan-cancer EMT signature by the dichotomized EMT score (with a cutoff of 0) and overall comparing its association with established, but independent, survival. We used the ANOVA test to assess the association EMT markers at the proteomic, microRNA (miRNA), and mRNA between EMT score and tumor grade. level and then applied the signature as a tool for exploring cross- platform molecular and clinical features associated with EMT Pathway analysis across multiple cancer types. Given the previously described We performed a functional annotation of the pan-cancer EMT association of EMT with drug response (both sensitivity and signature as well as pathway enrichment analysis using QIAGEN's resistance), we then further evaluated the therapeutic relevance Ingenuity Pathway Analysis (14). of the pan-cancer EMT signature in preclinical models using two publicly available drug sensitivity databases: the Cancer Cell Line Quantitative IHC Encyclopedia (CCLE; ref. 10) and the Genomics of Drug Sensi- Tissue sections (4 mM thick) were cut from formalin-fixed, tivity in Cancer (GDSC; ref. 11). paraffin-embedded blocks containing representative tumor and processed for IHC, using an automated staining system (Leica Materials and Methods Bond Max, Leica Microsystems). For assessment of PD-L1 expres- Datasets sion, the PD-L1 (E1L3N) XP rabbit monoclonal antibody (Cell We downloaded cell line drug sensitivity databases from the Signaling Technology) was applied at 1:100 dilution followed by CCLE (10) and GDSC (11). This included gene expression data detection using the Leica Bond Polymer Refine detection kit (Leica (Affymetrix U133 Plus 2 array from CCLE and Affymetrix HT HG Microsystems), DAB staining, and hematoxylin counterstaining. U133A array from GDSC) and drug sensitivity data (IC50 values). For quantification, stained slides were digitally scanned using the The CCLE data included 1,035 cell lines and the GDSC data Aperio ScanScope Turbo slide scanner (Leica Microsystems). included 654 cell lines. In total, 425 cell lines were present in both Captured images (200 magnification) were visualized using drug sensitivity databases. Twenty-four targeted drugs were pro- ImageScope software (Leica Microsystems,) and analyzed using filed in CCLE and 138 drugs (both targeted and cytotoxic) were Aperio Image Toolbox (Aperio, Leica Microsystems). For PD-L1 profiled in GDSC with a cell viability assay. The GDSC confirmed analysis in tumor and nontumor cells, five randomly selected identity of cell lines by short tandem repeat (STR) analysis and the square regions (1 mm2) in the core of each tumor were evaluated.

610 Clin Cancer Res; 22(3) February 1, 2016 Clinical Cancer Research

Pan-Cancer EMT Molecular and Immune Alterations

B Pan-cancer EMT signature genes Type Origin A Lung EMT signature E CDH1 Known EMT (cell line derived) Mesenchymal Epithelial M CDH2 markers VIM FN1 CDH2 CDH1 FN1 Type VIM FN1

CDH1 origin

VIM Density 0.00 0.05 0.10 0.15 0.20 0.25 468101214 Expression

Pan-cancer EMT signature (patient tumor derived)

Pan-cancer TCGA analysis of EMT-associated markers Gene expression, miRNAs, proteins, CNAs, mutations

Potential therapeutic targets in EMT In vitro drug sensitivity

1.0 0.8 0.6 0.4 0.2 0.2 0.4 0.6 0.8 1.0

C 1.0 KIRC 0.5 (N = 354) D 0.0 1.0 1.25 1.00 Basal (r = 0.811) 0.75 0.50 BRCA 0.25 (N =503) BLCA (r = 0.929) 0.00 2.0 BRCA (r = 0.883) 1.5 1.0 LUSC 4 COAD (r = 0.811) 0.5 (N =104) 0.0 HNSC (r = 0.946) 2.0 1.5 KIRC (r = 0.759) P < 0.01 1.0 Basal LUAD (r = 0.847) 0.5 0.5 (N =103) 0.0 1.00 LUSC (r = 0.915) 0.75 OVCA (r = 0.893) 0.50 HNSC 2 0.25 (N =203) READ (r = 0.826) 0.00 1.5 UCEC (r = 0.792) 1.0 LUAD 0.5 (N =131) 0.0 2.5 2.0 0.0 1.5

Density OVCA 1.0 0 0.5 (N =123) 0.0 1.5 1.0 BLCA P > 0.05 ECADHERIN (protein) 0.5 (N =88) 0.0 FIBRONECTIN (protein) 2.0 miR-200a 1.5 UCEC - 1.0 2 miR-200b 0.5 (N =189) -0.5 0.0 TWIST1 (mRNA) 1.5 1.0 COAD TWIST2 (mRNA) 0.5

(N =89) EMT score (Pan-cancer signature) 0.0 3 2

READ Correlation using Pan-cancer EMT signature 1 (N =47) -2 0 2 4 -0.5 0.0 0.5 1.0 0 −2 0 2 4 EMT score (Lung EMT signature) Correlation using Lung EMT signature EMT score (Pan-cancer EMT signature)

Figure 1. Development and application of the pan-cancer EMT signature. A, schematic flow of how the pan-cancer EMT score was derived from known EMT markers (CDH1, VIM, FN1, and CDH2) using patient tumor-derived samples from TCGA and subsequently utilized to identify molecular markers and potential therapeutic targets. B, genes in the pan-cancer EMT signature preserved strong correlation with known EMT markers across different tumors. "Type:" indicates whether the signature gene was epithelial-like (E) or mesenchymal-like (M) and "Origin:" indicates which seed gene was used to identify the signature gene. E markers (red) had positive correlation with CDH1 (Pearson correlation mean ¼ 0.44, median ¼ 0.43); M markers (blue) had positive correlation with VIM, FN1,orCDH2 (Pearson correlation mean ¼ 0.67, median ¼ 0.71). OVCA, ovarian cancer (n ¼ 123); BLCA, bladder urothelial carcinoma (n ¼ 88); HNSC, head and neck squamous cell carcinoma (n ¼ 203); COAD, colon adenocarcinoma (n ¼ 89); UCEC, uterine corpus endometrial carcinoma (n ¼ 189); BRCA, breast invasive carcinoma (n ¼ 503); LUSC, lung squamous cell carcinoma (n ¼ 104); Basal, basal-like breast cancer (n ¼ 103); LUAD, lung adenocarcinoma (n ¼ 131). C, TCGA pan-cancer tumor types exhibit a range of epithelial (score < 0) and mesenchymal (score > 0) samples. D, comparison of lung cancer and pan-cancer EMT signatures. The EMT scores from each signature were highly concordant, except for KIRC. Using TWIST1/TWIST2 mRNA, miR-200a/b, and E-cadherin/fibronectin RPPA data as external validation, the pan-cancer EMT signature performed significantly better (P < 0.01) for fibronectin and TWIST1/2 and similarly well for miR-200a/b and E-cadherin.

Analysis of PD-L1 expression speciﬁcally in tumor cells was based ysis matched to existing STR proﬁles, as reported previously on assessment of the whole section. The cell membrane staining (10, 11). algorithm was used to obtain the PD-L1 Histo-score (H-score; Additional more detailed methods are included in Supplemen- 0–300), which is computed on the basis of both extent and tary Materials and Methods. intensity of PD-L1 staining.

Association with drug sensitivity data Results For each drug sensitivity database, we calculated the EMT scores Development and evaluation of the pan-cancer EMT signature for the cell lines and associated them with IC50 values using the We derived the pan-cancer EMT signature using an approach Spearman rank correlation. Cell lines were obtained from com- similar to that for the lung cancer EMT signature, as shown mercial vendors and authentication was conﬁrmed by STR anal- in Fig. 1A (3). We selected candidate genes from the mRNAs that

www.aacrjournals.org Clin Cancer Res; 22(3) February 1, 2016 611

Mak et al.

best correlated with four established EMT markers "the seed," tion, cell-to-cell signaling, and cellular assembly and organiza- E-cadherin (CDH1), vimentin (VIM), fibronectin (FN1), and tion. Another enriched canonical pathway was leukocyte extrav- N-cadherin (CDH2), across nine distinct tumors types (see Fig. asation signaling, suggesting immune activation. 1; Supplementary Table S1 for TCGA tumor types and sample sizes). For this initial process, we excluded kidney renal clear cell An integrated molecular analysis of EMT across tumor types: carcinoma (KIRC) and rectal adenocarcinoma (READ) tumors gene expression and potential therapeutic targets because they contained predominantly mesenchymal (KIRC) or Using the pan-cancer EMT signature, we evaluated the impact of epithelial (READ) tumors (i.e., low dynamic range of EMT; see EMT on the mutational landscape, copy number alterations Materials and Methods). Using this approach, we identified a set (CNA), mRNA, miRNA, and protein expression to identify events of 77 unique genes as the pan-cancer EMT signature (see Materials associated with EMT in patient tumors. To assess how EMT and Methods and Supplementary Table S2). Figure 1B shows the impacts the overall transcriptome, we correlated EMT scores with correlation between these 77 signature genes and the seed com- mRNA expression data for each tumor type (see Supplementary ponent (one of the four established markers) that identified them. File S1). Genes expressed at higher levels in mesenchymal tumors We observed strong and consistent correlations with the seed were highly conserved across multiple tumor types, indicating across different tumor types. This suggests that the pan-cancer greater post-EMT biologic homogeneity; whereas those expressed EMT signature encompasses core EMT markers that function at higher levels in epithelial tumors tended to be more specificto across different tumors. individual tumor types (Supplementary Fig. S2A–S2C). We observed a wide dynamic range of EMT scores calculated Given our previously reported association of EMT with drug from the pan-cancer EMT signature across the 1,934 tumors that resistance (3, 4), individual genes or pathways overexpressed in represent 11 distinct tumor types (Fig. 1C). As expected, EMT mesenchymal tumors may have clinical utility as therapeutic scores (calculated from the resulting pan-cancer EMT signature) targets. For example, we previously showed that the receptor successfully identified KIRC as mostly mesenchymal and READ as tyrosine kinase AXL is highly expressed in lung cancers that have mostly epithelial. With the exception of KIRC (90.1% mesenchy- undergone EMT and, therefore, is a top candidate drug target in mal), READ (87.2% epithelial), and colon adenocarcinoma mesenchymal lung cancers (3). In the pan-cancer EMT signature (COAD; 94.4% epithelial), each tumor type included a significant derived here, AXL was again identified as a signature gene based number of both epithelial (EMT score <0) and mesenchymal on its strong correlation with established EMT markers. AXL, (EMT score >0) samples. therefore, represent a hallmark gene that is positively correlated Fourteen genes overlapped between the original lung cancer with EMT scores across all tumor types (r ¼ 0.53–0.95, median ¼ EMT signature and the new pan-cancer EMT signature (Supple- 0.674; Supplementary Fig. S2D). Other potential therapeutic mentary Fig. S1A). EMT scores computed from the pan-cancer targets that were highly expressed in mesenchymal tumors include EMT signature were highly correlated with scores calculated from the a-type platelet-derived growth factor receptor (PDGFRa; the lung cancer EMT signature for individual tumor types (r ¼ r ¼ 0.22–0.78, median ¼ 0.65), PDGFRb (r ¼ 0.59–0.92, median 0.76–0.95; overall r ¼ 0.82; Fig. 1D). The most notable outlier ¼ 0.81), and discoidin domain-containing receptor-2 (DDR2; r ¼ when comparing the two signatures was KIRC, which had the 0.29–0.84, median ¼ 0.63), inhibitors of which are in clinical lowest correlation (r ¼ 0.759), possibly due to a relatively narrow studies or have been approved for cancer treatment (17–20). range of EMT scores due to its mesenchymal tissue of origin. To investigate whether the pan-cancer EMT signature per- Immune checkpoint expression in mesenchymal tumors formed better than the original lung cancer EMT signature, we To better understand pathways globally dysregulated in the correlated the EMT scores computed from each signature with six setting of EMT, we next performed a pathway analysis of genes putative EMT markers. The markers included two proteins that correlated with the pan-cancer EMT signature in all 11 (E-cadherin and fibronectin) quantified by RPPA, two miRNAs tumor datasets, using a cutoff of an absolute r > 0.3. In addition [miR-200a, miR-200b; well established as regulators of EMT (15)] to functions related to EMT regulation (e.g., cellular movement, as measured by RNAseq, and two EMT-associated transcription cell–cell signaling), the top activated pathways were related to factors (TWIST1 and TWIST2) represented by mRNA expression immune cell signaling (Fig. 2B). Given recently published data levels (Fig. 1D). These six putative markers were selected on from our group that suggests a role of EMT in regulating the basis of their established roles in EMT (1, 2, 16). They served immune escape in lung cancer (9) and the strong therapeutic as an independent validation set because they (1) include data potential of emerging immunotherapies, we performed a more from different (nonsignature) platforms (e.g., protein, miRNA) focused analysis to investigatetheassociationbetweenEMT and/or (2) were not components of either mRNA-based EMT and key genes involved in the immune response across the 11 signature. Figure 1D shows that correlations with the three puta- cancer types. tive mesenchymal markers, fibronectin, TWIST1, and TWIST2 The expression of 20 potentially targetable immune checkpoint (positive correlation with EMT scores), were significantly stronger genes [based on current drug inhibitors that are in preclinical for the pan-cancer EMT signature (one-sided t test, P < 0.01); development, clinical trials, or which have been approved by the whereas correlations with the three epithelial markers, E-cadherin, FDA for specific cancer types (Table 1)] were correlated with the miR-200a, and miR-200b, were similar for both signatures (one- EMT scores for each cancer type. We observed a consistent and sided t test, P > 0.05; see also Supplementary Fig. S1B). strong positive correlation between EMT score and the expression We then performed a pathway analysis to investigate the levels of the immune checkpoint genes across all cancer types, biologic function of the genes identified in the new pan-cancer with more mesenchymal tumors expressing higher levels of these EMT signature (Fig. 2A). Genes in the pan-cancer EMT signature immune targets (Fig. 2C). Notably, high expression of pro- were significantly associated with several molecular functions, grammed cell death 1 (PD1), cytotoxic T-lymphocyte-associated such as cellular movement, morphology, growth, and prolifera- protein 4 (CTLA4), and TNF receptor superfamily, member 4

612 Clin Cancer Res; 22(3) February 1, 2016 Clinical Cancer Research

Pan-Cancer EMT Molecular and Immune Alterations

AB5 4 3 2 1

–log (B-H p-value) 0

signaling α ILK signaling

HIF1

ComplementEicosanoid system signaling Regulation of the EMT

Bladder cancer signaling Dendritic cellPhospholipase maturation signaling

Glioma invasiveness signaling Phenylethylamine degradation

Leukocyte extravasation signaling Inhibition of matrix metalloproteases Granulocyte adhesion and diapedesis Colorectal cancer metastasis signaling Agranulocyte adhesion and diapedesis Intrinsic prothrombin activation pathway Caveolar-mediated endocytosis signaling C 0.20 0.43 0.74 0.64 0.47 0.32 0.87 0.66 0.42 0.67 0.58 TGFB1 0.49 0.55 0.53 0.41 0.44 0.63 0.55 0.03 0.40 0.28 0.12 CD137 0.48 0.51 0.54 0.42 0.49 0.63 0.72 -0.03 0.47 0.24 0.31 CCL2 0.52 0.49 0.65 0.50 0.52 0.72 0.65 -0.13 0.41 0.37 0.20 TIM3 0.65 0.72 0.68 0.58 0.71 0.65 0.67 0.35 0.62 0.59 0.46 OX40L 0.36 0.34 0.45 0.27 0.35 0.54 0.68 0.16 0.34 0.27 0.42 ADORA2A 0.53 0.48 0.51 0.34 0.39 0.71 0.52 0.27 0.58 0.32 0.32 IL10 Pan-cancer EMT signature pathways 0.45 0.44 0.49 0.44 0.54 0.60 0.79 0.41 0.42 0.33 0.47 OX40 0.48 0.54 0.56 0.45 0.47 0.57 0.74 0.29 0.39 0.34 0.20 CD4 0.47 0.49 0.64 0.51 0.56 0.66 0.65 0.26 0.47 0.30 0.35 PDL2 -0.13 0.23 -0.03 0.28 0.37 0.13 0.03 0.18 0.08 -0.02 -0.01 IL1A 0.24 0.23 0.45 0.28 0.41 0.38 0.11 -0.16 0.17 0.12 0.02 PDL1 0.31 0.40 0.18 0.12 0.14 0.28 0.44 -0.01 -0.12 0.05 -0.06 ICOSLG 0.23 0.29 0.40 0.14 0.35 0.59 0.21 0.16 0.23 0.09 0.37 LAG3 0.22 0.36 0.34 0.23 0.33 0.59 0.32 0.14 0.33 0.12 0.43 PD1 Correlation of immune checkpoint expression 0.35 0.49 0.41 0.26 0.36 0.56 0.20 0.15 0.38 0.25 0.39 CTLA4 with EMT score by tumor type 0.39 0.47 0.31 0.24 0.37 0.57 0.27 0.11 0.35 0.20 0.24 ICOS Pearson correlation 0.25 0.24 0.43 0.30 0.47 0.67 0.39 0.56 0.42 0.16 0.09 IL6 −0.16 0.0 0.2 0.4 0.6 0.8 0.55 0.36 0.27 0.41 0.43 0.41 0.49 0.66 0.12 0.23 0.07 B7-H3 0.50 0.47 0.43 0.34 0.45 0.57 0.41 0.53 0.20 0.17 0.42 CXCR4 KIRC Basal BLCA LUSC LUAD BRCA UCEC READ HNSC OVCA COAD

Figure 2. The pan-cancer EMT signature is enriched with different molecular functions and pathways. A, functional annotation of the pan-cancer EMT signature using QIAGEN's Ingenuity Pathway Analysis (14) found ﬁve molecular functions (cyan) and six pathways (purple) that were signiﬁcantly enriched in the pan-cancer EMT signature. Signature gene names are shown in red (genes associated with mesenchymal phenotype) and blue (genes associated with epithelial phenotype) circles, where text size represents average correlation to EMT signature score across tumors. B, enriched pathway for genes (using RNAseq data) strongly correlating with EMT score (Ingenuity Pathway Analysis). Immune pathways ranked top among the enriched pathways derived from commonly expressed genes across different mesenchymal tumor types (red arrows). C, correlation between immune checkpoint genes and EMT score across different tumor types.

(OX40L, also known as CD134) were associated with high EMT Finally, we validated the association between the mesenchy- score regardless of tumor type (PD1 median r ¼ 0.33, median P ¼ mal phenotype and elevated expression of PD-L1 immunohis- 2.110 4; CTLA4 median r ¼ 0.36, median P ¼ 1.310 5; OX40L tochemically in chemotherapy-na€ve lung adenocarcinomas median r ¼ 0.65, median P ¼ 2.410 13). It has been reported that and squamous lung cancers that were surgically resected at the OX40L regulates T-cell response, which has led to the study of University of Texas MD Anderson Cancer Center (PROSPECT OX40L inhibition in conjunction with other checkpoint block- dataset). For this analysis, we compared tumors in the top and ades (21, 22). bottom tertiles for EMT score, using a PD-L1 H-score ranging Looking at individual tumor types, samples with higher EMT from 0 to 300. In agreement with their gene expression profiles scores expressed the immune checkpoint genes in a coordinated that confirmed enrichment for immune-related genes (Fig. 3B way, even in predominantly epithelial tumors, such as READ and C, top panels), mesenchymal adenocarcinomas from (Fig. 3A). This enrichment of immune target expression in tumors PROSPECT exhibited higher PD-L1 H-score than epithelial with more mesenchymal characteristics corroborated our recent tumors (P ¼ 0.003, Mann–Whitney U test;Fig.3B,bottom findings in lung cancer, namely, that lung adenocarcinomas with left). Significantly, this finding was upheld when the analysis higher lung cancer EMT signature scores expressed higher levels of was limited to tumor cells (P ¼ 0.002, Mann–Whitney PD-L1, which is a target of miR-200 (a suppressor of EMT and U test; Fig. 3B, bottom right). We further confirmed the signif- metastasis; ref. 9). Clinically, our findings are consistent with icant, positive correlation between the EMT score and PD-L1 results of immunotherapy trials that have shown the greatest H-score in the entire cohort of adenocarcinomas from PROS- activity of immune checkpoint inhibitors in cancer types with PECT, using Spearman rank correlation coefficient (P ¼ 0.002 the strongest mesenchymal phenotypes, such as melanoma for tumor and nontumor cells and P ¼ 0.011 when the analysis (23–25). was limited to tumor cells).

www.aacrjournals.org Clin Cancer Res; 22(3) February 1, 2016 613

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2016 American Association for Cancer Research. 614 lnCne e;2()Fbur ,2016 1, February 22(3) Res; Cancer Clin a tal. et Mak Downloaded from

Table 1. Clinical development of enriched immune targets in mesenchymal tumors Basal BLCA BRCA COAD HNSC KIRC LUAD LUSC OVCA READ UCEC Clinical Target rPrPrPrPrPrPrPrPrPrPrPDrugs development ADORA2A 0.27 0.006 0.54 <0.001 0.27 <0.001 0.45 <0.001 0.36 <0.001 0.16 0.002 0.35 <0.001 0.34 0.001 0.34 <0.001 0.68 <0.001 0.42 <0.001 ZM 241385 Preclinical Published OnlineFirstSeptember29,2015;DOI:10.1158/1078-0432.CCR-15-0876

clincancerres.aacrjournals.org CCL2 0.24 0.016 0.63 <0.001 0.42 <0.001 0.54 <0.001 0.48 <0.001 0.03 0.625 0.49 <0.001 0.51 <0.001 0.47 <0.001 0.72 <0.001 0.31 <0.001 Carlumab Phase II (prostate) CD274 (PD-L1) 0.12 0.228 0.38 <0.001 0.28 <0.001 0.45 <0.001 0.24 0.001 0.16 0.002 0.41 <0.001 0.23 0.017 0.17 0.066 0.11 0.480 0.02 0.762 Pembrolizumaba Phase III (MK3475)a, MPDL3280A, MDX-1105, BMS-936559, MSB0010718C CD276 (B7-H3) 0.23 0.021 0.41 <0.001 0.41 <0.001 0.27 0.010 0.55 <0.001 0.66 <0.001 0.43 <0.001 0.36 <0.001 0.12 0.173 0.49 <0.001 0.07 0.340 MGA-271, 8H9 Phase I CD4 0.34 <0.001 0.57 <0.001 0.45 <0.001 0.56 <0.001 0.48 <0.001 0.29 <0.001 0.47 <0.001 0.54 <0.001 0.39 <0.001 0.74 <0.001 0.20 0.005 Zanolimumab Phase II CXCR4 0.17 0.083 0.57 <0.001 0.34 <0.001 0.43 <0.001 0.50 <0.001 0.53 <0.001 0.45 <0.001 0.47 <0.001 0.20 0.028 0.41 0.005 0.42 <0.001 BMS-936564, Phase I MSX-122 CTLA4 0.25 0.010 0.56 <0.001 0.26 <0.001 0.41 <0.001 0.35 <0.001 0.15 0.005 0.36 <0.001 0.49 <0.001 0.38 <0.001 0.20 0.175 0.39 <0.001 Ipilimumaba, Phase III Tremelimumab HAVCR2 (TIM3) 0.37 <0.001 0.72 <0.001 0.50 <0.001 0.65 <0.001 0.52 <0.001 0.13 0.014 0.52 <0.001 0.49 <0.001 0.41 <0.001 0.65 <0.001 0.20 0.005 MoAb Preclinical ICOS 0.20 0.048 0.57 <0.001 0.24 <0.001 0.31 0.003 0.39 <0.001 0.11 0.038 0.37 <0.001 0.47 <0.001 0.35 <0.001 0.27 0.065 0.24 0.001 MoAbb Preclinical on September 28, 2021. © 2016American Association for Cancer ICOSLG 0.05 0.609 0.28 0.008 0.12 0.006 0.18 0.094 0.31 <0.001 0.01 0.912 0.14 0.103 0.40 <0.001 0.12 0.172 0.44 0.002 0.06 0.402 MoAb Preclinical

Research. IL1A 0.02 0.834 0.13 0.236 0.28 <0.001 0.03 0.756 0.13 0.063 0.18 0.001 0.37 <0.001 0.23 0.021 0.08 0.374 0.03 0.844 0.01 0.888 Anakinra Phase I IL6 0.16 0.097 0.67 <0.001 0.30 <0.001 0.43 <0.001 0.00 <0.001 0.56 <0.001 0.47 <0.001 0.24 0.016 0.42 <0.001 0.39 0.007 0.09 0.204 Tocilizumaba Phase I/II (ovarian cancer) IL10 0.32 0.001 0.71 <0.001 0.34 <0.001 0.51 <0.001 0.53 <0.001 0.27 <0.001 0.39 <0.001 0.48 <0.001 0.58 <0.001 0.52 <0.001 0.32 <0.001 MoAb Preclinical LAG3 0.09 0.380 0.59 <0.001 0.14 0.001 0.40 <0.001 0.23 0.001 0.16 0.003 0.35 <0.001 0.29 0.003 0.23 0.011 0.21 0.152 0.37 <0.001 IMP321, BMS- Phase III 986016, MoAb (breast), Phase I PDCD1 (PD1) 0.12 0.226 0.59 <0.001 0.23 <0.001 0.34 0.001 0.22 0.002 0.14 0.008 0.33 <0.001 0.36 <0.001 0.33 <0.001 0.32 0.028 0.43 <0.001 Nivolumaba, CT011, Phase III AMP224 PDCD1LG2 (PD-L2) 0.30 0.002 0.66 <0.001 0.51 <0.001 0.64 <0.001 0.47 <0.001 0.26 <0.001 0.56 <0.001 0.49 <0.001 0.47 <0.001 0.65 <0.001 0.35 <0.001 Nivolumaba,b , Phase III Pembrolizumaba (MK3475), CT011, AMP224 TGFB1 0.67 <0.001 0.32 0.002 0.64 <0.001 0.74 <0.001 0.20 0.005 0.66 <0.001 0.47 <0.001 0.43 <0.001 0.42 <0.001 0.87 <0.001 0.58 <0.001 SB-431542, GC 1008 Phase I TNFRSF4 (OX40) 0.33 0.001 0.60 <0.001 0.44 <0.001 0.49 <0.001 0.45 <0.001 0.41 <0.001 0.54 <0.001 0.44 <0.001 0.42 <0.001 0.79 <0.001 0.47 <0.001 Anti-OX40 MoAb Phase I TNFSF4 (OX40L) 0.59 <0.001 0.65 <0.001 0.58 <0.001 0.68 <0.001 0.65 <0.001 0.35 <0.001 0.71 <0.001 0.72 <0.001 0.62 <0.001 0.67 <0.001 0.46 <0.001 Anti-OX40 MoAbb Phase I TNFRSF9 (CD137) 0.28 0.004 0.63 <0.001 0.41 <0.001 0.53 <0.001 0.49 <0.001 0.03 0.564 0.44 <0.001 0.55 <0.001 0.40 <0.001 0.55 <0.001 0.12 0.102 Anti-CD137 MoAb, Phase I, II BMS-663513 (NSCLC CRT) (Urelumab) Abbreviations: MoAb, monoclonal antibody; r, Pearson correlation. aApproved by the FDA. lnclCne Research Cancer Clinical bIndirect targeting. Published OnlineFirst September 29, 2015; DOI: 10.1158/1078-0432.CCR-15-0876

Pan-Cancer EMT Molecular and Immune Alterations

A TGFB1 TGFB1 TGFB1 TGFB1 CD137 CD137 CD137 CD137 CCL2 CCL2 CCL2 CCL2 TIM3 TIM3 TIM3 TIM3 OX40L OX40L OX40L OX40L ADORA2A ADORA2A ADORA2A ADORA2A IL10 IL10 IL10 IL10 OX40 OX40 OX40 OX40 CD4 CD4 CD4 CD4 PDL-2 PDL-2 PDL-2 PDL-2 IL1A IL1A IL1A IL1A PDL-1 PDL-1 PDL-1 PDL-1 ICOSLG ICOSLG ICOSLG ICOSLG LAG3 LAG3 LAG3 LAG3 PD1 PD1 PD1 PD1 CTL4A CTL4A CTL4A CTL4A ICOS ICOS ICOS ICOS IL6 IL6 IL6 IL6 B7-H3 B7-H3 B7-H3 B7-H3 CXCR4 CXCR4 CXCR4 CXCR4 Basal BLCA BRCA COAD

TGFB1 TGFB1 CD137 TGFB1 TGFB1 CD137 CCL2 CD137 CD137 CCL2 TIM3 CCL2 CCL2 TIM3 OX40L TIM3 TIM3 OX40L ADORA2A OX40L OX40L ADORA2A IL10 ADORA2A ADORA2A IL10 OX40 IL10 IL10 OX40 CD4 OX40 OX40 CD4 PDL-2 CD4 CD4 PDL-2 IL1A PDL-2 PDL-2 IL1A PDL-1 IL1A IL1A PDL-1 ICOSLG PDL-1 PDL-1 ICOSLG LAG3 ICOSLG ICOSLG LAG3 PD1 LAG3 LAG3 PD1 CTL4A PD1 PD1 CTL4A ICOS CTL4A CTL4A ICOS IL6 ICOS ICOS IL6 B7-H3 IL6 IL6 B7-H3 CXCR4 B7-H3 B7-H3 CXCR4 CXCR4 CXCR4 HNSC KIRC LUAD LUSC

TGFB1 TGFB1 TGFB1 CD137 CD137 CD137 CCL2 CCL2 CCL2 TIM3 TIM3 TIM3 OX40L OX40L OX40L EMT score ADORA2A ADORA2A ADORA2A IL10 IL10 IL10 E M OX40 OX40 OX40 CD4 CD4 CD4 PDL-2 PDL-2 PDL-2 IL1A IL1A IL1A PDL-1 PDL-1 PDL-1 ICOSLG ICOSLG ICOSLG LAG3 LAG3 LAG3 PD1 PD1 PD1 Expression (mRNA) CTL4A CTL4A CTL4A ICOS ICOS ICOS IL6 IL6 IL6 B7-H3 B7-H3 B7-H3 CXCR4 CXCR4 CXCR4 OVCA READ UCEC

BCTGFB1 TGFB1 CD137 CD137 CCL2 CCL2 TIM3 TIM3 OX40L OX40L ADORA2A ADORA2A IL10 IL10 OX40 OX40 CD4 CD4 PDL-2 PDL-2 IL1A IL1A PDL-1 PDL-1 ICOSLG ICOSLG LAG3 LAG3 PD1 PD1 CTL4A CTL4A ICOS ICOS IL6 IL6 B7-H3 B7-H3 CXCR4 CXCR4 Lung adenocarcinoma Lung squamous

200 P = 0.003 150 P = 0.002 250 P = 0.019 150 P = 0.110 200 150 100 100 150 100 50 100 50 50 50 (tumor cells) (tumor cells) 0 0 PD-L1 H-Score PD-L1 H-Score PD-L1 H-Score 0 PD-L1 H-Score 0 (tumor + non-tumor) (tumor + EM EM non-tumor) (tumor + EM EM

Figure 3. Immune target enrichment in mesenchymal tumors. A, expression of immune checkpoint genes was found in the pan-caner tumor types. Tumor samples within each cancer type are ordered by EMT score. Mesenchymal tumors have higher expression than epithelial tumors of immune checkpoint genes. B and C, heatmap representation of mRNA expression levels of selected immune checkpoint genes in 151 lung adenocarcinomas (B) and 57 lung squamous carcinomas (C) from the PROSPECT study, rank ordered on the basis of their EMT score (top). Comparison of PD-L1 H-score in epithelial (E) and mesenchymal (M) tumors that were included in the PROSPECT tumor microarray. Automated quantiﬁcation of extent and intensity of PD-L1 staining was performed in tumor and nontumor cells or limited to tumor cells (B and C, bottom). Lung adenocarcinomas and squamous carcinomas with EMT score in the top/bottom tertile were included in this analysis. Statistical comparisons are based on the Mann–Whitney U test. , P 0.05; , P 0.01.

www.aacrjournals.org Clin Cancer Res; 22(3) February 1, 2016 615

Mak et al.

A C

10 10 CDH1

miR-200a

ERBB3 8 miR-141 8 miR-200b miR-200c F11R

MYO6 6 miR-378 6 TOB1 ZFPM2 ZEB2 miR-429 FOXC2 TAGLN miR-375 WNT2 THBS1 CD46 COL1A1 TWIST2 miR-148b miR-199a ERBB2 SAMD4A AKT3 FLI1 MEF2C FBN1 FN1 miR-1266 THY1 CSF1R GFPT2 GPR160 MRC2 PXDN SREBF1 GFPT1 LRPPRC FERMT2 miR-577 miR-199b PPFIBP2 CLINT1 FOXA1 STMN2 TGFB3 MYB GPD2 TMEM41B TSHZ3 LHFP PKN2 MDM2 STAT6 GSTP1 RECK SERPINE1 VIM miR-96 miR-30d miR-127 FGFR3 KLF5 HDGF CLCN3 AQP1 TIMP2 4 CD2AP MCAM 4 HADH RAB40B IRF6 ERG PRPH2 DPT miR-192 miR-143 miR-134 SLC25A10 NUDT19 PA2G4 TIMP3 DCN COMP MMP14 FOXP3 miR-194 miR-145 miR-382 CLDN5 RUNX1T1 COL5A3 MYL9 COL1A2 SOX18 PPARG TLR4 SAMSN1 COL4A2 NTRK1 ACTA2 miR-335 miR-152 SPINK1 ID1 ITGB3 TNFSF4 SFRP2 ELK3 NRG1 PTPRD TWIST1 COL4A1 miR-379 miR-654 LCP1 ATP6V1B2 PDLIM7 NPR3 miR-224 miR-100 miR-381 TCF4 LATS2 PDGFRA PRDM1 IGF1 PDGFRB miR-337 LSAMP MMP13 MMP3 NR5A2 IL11 CTGF PROCR miR-708 miR-376c NTRK3 MAFB IGFBP3 TGFB1 DHH ASPN FBLN5 PEA15 CFH FAM105A MAF MSC FGF7 2 miR-182 miR-889 ETS1 LCK BACE1 CXCL12 SLIT2 miR-99a miR-136 miR-410 2 HSD11B1 CLEC5A PLXND1 CD28 MMP11 GALNT1 CDH5 SYNE1 TBX5 TPM1 GLI3 HGF miR-369 CACNA2D1 FMOD DLC1 FGF1 MAP3K12 FOXF1 miR-195 miR-539 miR-154 PPARGC1A GYPC TPM2 PDPN GNAI2 ILK miR-132 HAND2 TNFSF11 LRRK2 ESR1 TMEM2 miR-409 MLLT11 BIN1 CDK6 CYP27A1 BASP1 SRGN VEGFC miR-493 ACAN AGTR1 AP2M1 GHR AXIN2 GATA3 PECAM1 let-7c miR-323 PITX2 SH3D19 FOXP1 FUT4SOAT1 CXCR4 C5AR1 miR-133a HTR1B CHRD MT2A ANPEP CCR1 LOX RUNX2 miR-487b SNAI2 DYRK2 CACNA1C CCL2 TGFB2 RCAN2 ITGB1 miR-21 UNC5C DLG4 NXN IBSP MMP9 CD68 VCAN miR-34c miR-496 miR-299 miR-758 NGF EGR2 ITGAV SNCA TLR1 NCAM1 QKI SOX17 CD4 GLI2 LRP1 TICAM2 RGS2 LRRC8C HSPB6 CCR4 BOC EPHA3 miR-411 TBX2 PGR COPZ2 SPP1NOG LEF1 STK10 LILRB4 GALNTL2 0 miR-125b miR-431 miR-655 KDR HMGA2 NOS3 CDH11 miR-494 miR-432 miR-370 0 SEC23A SLC38A5 CCL13 CTLA4 MYH11 CSF1 CCL7 GYPE LMO2 DKK2 SPARC miR-22 TGFBI FOXL1 CCL4 PDGFB RASGRP3 SLC12A4 TRPC6 NRK miR-495 miR-214 MITF CD70 MMP16 ITK miR-218 miR-485 PSTPIP1 SPI1 LTBP1 GJA1 NRP1 CD80 ID3 MSR1 SNAI1 OGN RHOG IL6 DPYD ITGAM ZEB1 IL10 Frequency associated with epithelial phenotype epithelial with associated Frequency Frequency associated with epithelial phenotype 86420 1210 86420 1210 Frequency associated with mesenchymal phenotype Frequency associated with mesenchymal phenotype

B D 10

0.4 ECADHERIN 8 HER3 CLAUDIN7 0.0 6 RAB25 P90RSK BRAF −0.4 ACCPS79 4 GATA3 BETACATENIN

Correlation with EMT score HER2 FASN −0.8 2 Basal BLCA BRCA COAD HNSC KIRC LUAD LUSC OVCA READ UCEC BCL2 hsa-mir-199a hsa-mir-381 hsa-mir-337 hsa-mir-145 hsa-mir-141 YB1 hsa-mir-199b hsa-mir-411 hsa-mir-299 hsa-mir-495 hsa-mir-200c LCK TAZ FIBRONECTIN 0 CAVEOLIN1 PAI1 hsa-mir-379 hsa-mir-758 hsa-mir-889 hsa-mir-143 hsa-mir-200a PEA15 P21 hsa-mir-127 hsa-mir-485 hsa-mir-493 hsa-mir-152 hsa-mir-200b PKCALPHAPS657 hsa-mir-382 hsa-mir-654 hsa-mir-410 hsa-mir-655 Frequency associated with epithelial phenotype hsa-mir-134 hsa-mir-369 hsa-mir-154 hsa-mir-376c hsa-mir-214 hsa-mir-487b hsa-mir-100 hsa-mir-378 86420 10 12 hsa-mir-409 hsa-mir-370 hsa-mir-432 hsa-mir-429 Frequency associated with mesenchymal phenotype

Figure 4. miRNA expression is highly correlated with EMT status. A, axes show the number of tumor types for which a miRNA was correlated with EMT score (x-axis: miRNA expressed at higher levels in mesenchymal tumors (red); y-axis: miRNA higher in epithelial tumors (blue) at a cutoff of P 0.3). miR-200 family (200c/b/c and 141) is anticorrelated with EMT (blue), whereas numerous miRNAs are positively correlated with EMT (red). B, correlation between miRNA and EMT score for each cancer type. C, targets of EMT-associated miRNAs (A and B) were identified by miRWalk. Among these, we observed a statistically significant enrichment of targets whose expression correlated with EMT score (Kolmogorov–Smirnov test, P ¼ 1.8 1028). Top miRNA targets associated with EMT score are shown based on an absolute correlation >0.3 in at least four tumors (full list of targets and their correlation with EMT score for each cancer type is provided in Supplementary File S2). D, proteins correlated with EMT score. Similar to the finding at gene expression level, epithelial status correlates with expression of HER3 protein, among others. Proteins with an absolute correlation coefficient higher than 0.25 in more than four tumor types are shown (blue, epithelial; red, mesenchymal).

Similar results were noted in squamous lung cancers, with miRNA expression patterns in mesenchymal tumors higher PD-L1 H-score recorded in mesenchymal tumors (P ¼ To better understand potential factors contributing to the EMT 0.019, Mann–Whitney U test;Fig.3C,bottomleft)anda phenotype, we then investigated changes in other data types significant, positive correlation noted between the EMT and (miRNA, protein, mutation, and copy number) that corre- PD-L1 H-score (P ¼ 0.011). Comparison of tumor cell PD-L1 H- sponded with differences in EMT score across the pan-cancer set. score in tumors with EMT scores in the top versus bottom tertiles Because several miRNAs are known to regulate EMT, we identified did not reach statistical significance in squamous tumors (P ¼ miRNAs that were significantly correlated with EMT scores, either 0.110, Mann–Whitney U test, Fig. 3C, bottom right), likely due positively or negatively, across the pan-cancer sets. Figure 4A and to a limited number of tumors included in this analysis. How- B show the miRNAs that strongly correlate with the EMT score ever, the correlation between the EMT score and PD-L1 H-score (absolute r > 0.3 in four or more tumor types). As expected, high (using all squamous tumors) remained significant (P ¼ 0.028). expression of miR-200 family members (miR-200a/b/c, miR-141,

616 Clin Cancer Res; 22(3) February 1, 2016 Clinical Cancer Research

Pan-Cancer EMT Molecular and Immune Alterations

and miR-429), which are known to suppress ZEB1 and thereby association between mutation and EMT score (Supplementary maintain E-cadherin expression and other hallmarks of an epi- Fig. S4A). We also explored whether mesenchymal status would thelial phenotype, was observed among tumors with the lowest influence tumor mutation and CNA burden by quantifying the (most epithelial) EMT scores (Fig. 4A and B). We observed a mutation rate and percentage of CNAs in each sample per tumor positive correlation between the EMT score and numerous other type, as described by Ciriello and colleagues (29). Using this miRNAs, with the miR-199 family (miR-199a/b) having the approach, we found no correlation between CNAs versus muta- greatest expression in mesenchymal tumors (Supplementary tion distribution and EMT score (Supplementary Fig. S4C), indi- Fig. S3A). The miR-199 family has been shown to regulate skeletal cating that enrichment of CNAs or mutations is also primarily development and to play a role in melanoma invasion and driven by the tumor tissue of origin. metastasis, but has been rarely studied in EMT (26, 27). Despite the tumor-specific association observed between the expression EMT score and clinical outcomes of some miRNAs and the EMT score, positively correlated miRNAs In some cancers, EMT may be associated with worse clinical (i.e., those expressed at higher levels in mesenchymal tumors) outcomes. Therefore, we assessed whether pan-cancer EMT scores were more frequent and homogenous in this analysis than neg- were associated with clinical covariates such as overall survival atively correlated miRNAs (those higher in epithelial tumors; Fig. and tumor grade. We did not observe a significant association 4A and B; Supplementary Fig. S3A), suggesting that miRNA between EMT score and overall survival, although we found a expression may play an important role in regulating EMT. trend toward shorter survival times in patients with mesenchymal To investigate the functional roles of miRNAs identified by ovarian carcinomas [HR, 1.32; 95% confidence interval (CI), this analysis, we next investigated whether expression of EMT- 0.96–1.81; P ¼ 0.08] and uterine corpus endometrial carcinomas associated miRNA target genes also correlated with EMT scores. (HR, 2.07; 95% CI, 0.93–4.63; P ¼ 0.09). We also found higher For the EMT-associated miRNAs in Fig. 4A and B, we down- EMT scores in KIRC and head and neck squamous cell carcinomas loaded their experimentally validated targets using miRWalk with higher tumor grade (P < 0.0001 and P ¼ 0.01, respectively; (28)andobservedasignificant overall association between Supplementary Fig. S5). target expression levels and EMT scores (P < 0.001; Fig. 4C). Taking the miR-200 family as an example, Supplementary Fig. Drug sensitivity in mesenchymal tumors S3B shows the strong associations between the miRNA targets Finally, we tested the ability of the pan-cancer EMT signature to and the EMT score. Collectively, these findings support an identify associations between EMT and patterns of drug response intricate regulation of EMT that is conserved across multiple across 1,689 preclinical models, representing 18 tumor types tumor types by miRNAs beyond the miR-200 family and which (Supplementary Table S3). We curated gene expression data from merit further investigation. two publicly available databases [CCLE (10) and GDSC (11)] to compute an EMT score for each cell line, which we subsequently Protein expression and EMT correlated with drug sensitivity (based on IC50 values). Among Next, we evaluated the correlation between the EMT score and 1,035 cell lines in CCLE and 654 cell lines in GDSC with gene protein expression using RPPA data. RPPA is a highly quantitative expression data, 425 cell lines were common to both sets, allow- platform for the measurement of key total and phosphorylated ing us to compare the concordance of EMT scores derived from proteins. In this analysis, we included 181 proteins for which data independently generated expression datasets. Despite the gene were available across all 11 tumor types. Notably, some proteins expression data having been generated from different versions of relevant to EMT, such as AXL, are excluded from the analysis the Affymetrix array platforms, EMT scores calculated for indi- because they were evaluated in only a subset of the TCGA pan- vidual cell lines were highly concordant, suggesting a robust cancer tumor types included here. performance of our approach across platforms (Fig. 5A). Further- We found higher levels of several protein markers such as E- more, the distribution of EMT scores for the cell line models of cadherin, claudin-7, and Her-3 in epithelial tumors across most each disease type closely mirrored those observed across patient tumor types (8); with higher levels of PAI1 and p21 in mesen- tumors (Figs. 1B and 5B). For example (and as expected), bone chymal tumors (Fig. 4D). Fibronectin-1 was the only marker with sarcoma, glioblastoma multiforme, and melanoma were predom- consistently higher expression in all 11 cancer types, possibly due inantly mesenchymal (Fig. 5B). To determine patterns of drug to the use of fibronectin (FN1) mRNA levels in deriving the pan- sensitivity in mesenchymal tumors, we correlated the EMT score cancer signature (one of the seed genes). In contrast, other from each cell line and IC50 values for targeted drugs tested in the proteins such as protein kinase C-a (PKCa) were associated with CCLE (24 drugs) and GDSC (138 drugs) databases. As previously EMT in a subset of tumor types (4 of 11), reflecting tumor-specific demonstrated with the lung cancer EMT signature (3), we found EMT differences. increased resistance to EGFR inhibitors (e.g., erlotinib). Mesen- chymal models also demonstrated greater resistance to other Mutational landscape and CNAs drugs that target the ErbB family, including the dual EGFR/VEGF To determine the mutational landscape of mesenchymal inhibitor vandetanib (ZD-6474) and the EGFR/Her2 inhibitor tumors, we investigated whether a mutation was associated with lapatinib (Fig. 5C and D). In contrast to ErbB inhibitors, but the EMT score independent of tumor origin. Despite a significant consistent with higher expression of FGFR1 and PDGFR in mes- association, the results can still be primarily attributed to muta- enchymal cancers, we observed increased sensitivity to the FGFR1 tions specific to certain tumor types (Supplementary Fig. S4A). and PDGFR multikinase inhibitor dovitinib (TKI258) in cell lines For example, a VHL mutation primarily observed in KIRC, was with higher EMT scores (Fig. 5C). To evaluate potential therapeu- strongly associated with higher EMT score. Therefore, after adjust- tic targets in the GDSC dataset, we performed an exploratory ing for tumor type and multiple testing, we performed another analysis of the in vitro effect of drugs against the same targets. From analysis (see Materials and Methods), and found no significant this target analysis (Fig. 5D; Supplementary Fig. S6), we observed

www.aacrjournals.org Clin Cancer Res; 22(3) February 1, 2016 617

Mak et al.

A B

3 3 2 Bone sarcoma 2 1 1 0 (BSARC) 0 2.0 2.0 1.5 1.5 1.0 Central nervous system 1.0 0.5 0.5 0.0 (GBM) 0.0 8 5 6 4 4 3 Thyroid carcinoma 2 2 1 0 (THC) 0 3 2 Skin cutaneous melanoma 4 1 2 0 (SKCM) 0 3 3 2 Hematological malignancies 2 1 1 0 (AML/DLBC) 0 3 2 6 Ovarian serous cystadenocarcinoma 4 1 2 0 (OV) 0 5 6 4 3 4 2 Not specified non-small cell lung cancer 1 2 0 (NSCLC NOS) 0 3 1.5 1.0 Small cell lung cancer 2 0.5 1 0.0 (SCLC) 0 1.5 5 4 1.0 3 0.5 Lung squamous cell carcinoma 2 1 0.0 (LUSC) 0

Density 1.5 1.0 3 2 0.5 Lung adenocarcinoma 1 0.0 (LUAD) 0 3 2 Head & neck squamous cell carcinoma 4 1 2 0 (HNSC) 0 4 6 3 4 2 Pancreatic adenocarcinoma 1 2 0 (PAAD) 0 4 3 4 2 Esophageal carcinoma 3 2 1 1 0 (ESCA) 0 4 2 3 1 Breast invasive carcinoma 2 1 0 (BRCA) 0 4 3 4 2 1 Stomach adenocarcinoma 2 0 (STAD) 0 3 4 2 3 Colorectal adenocarcinoma 2 1 1 0 (COAD) 0 -2 -1 0 1 2 3 -2 -1 0 1 2 3 EMT score (CCLE) EMT score (GDSC)

C D BRAF, KDR RAF265 MDM2 Nutlin-3 PDGFR N.A. Topotecan ERBB2 N.A. Irinotecan RAF Vemurafenib EGFR HSP90 17-AAG EGFR Erlonib PARP Src, Bcr-ABL, EGFR AZD0530 GSK3 EGFR, ERBB2 Lapanib EGFR, FGFR1, PDGFR, KDR TKI258 JAK2 KDR, EGFR, RET ZD-6474 CDK4/6 PD-0332991 IKK Gamma-secretase L-685458 TBK1 c-MET PHA-665752 N.A. Paclitaxel PDK1 Bcr-ABL, c-KIT Nilonib IAP LBW242 SRC BRAF, KDR, Flt3 Sorafenib ABL MEK PD-0325901 MEK Selumenib ITK ALK TAE684 IGFR1 AEW541 KIF11 ALK, c-MET PF2341066 BCR-ABL HDAC Pabinostat

Figure 5. Potential therapeutic targets in mesenchymal cell lines. A, EMT scores computed from different datasets are highly concordant. Although gene expression data were generated from different laboratories utilizing distinct platforms, high concordance (Pearson correlation ¼ 0.97, concordance correlation coefﬁcient ¼ 0.96) of the EMT score is found in shared cell lines from both datasets (GDSC, y-axis; CCLE, x-axis), reinforcing the robustness of the signature. B, spectra of EMT scores across different tumor types in CCLE and GDSC. Similar to results from patient-derived tumor samples, cell lines from different tumor types exhibited different spectra of EMT scores. C, correlation between IC50 and EMT score in CCLE. Dot size is proportional to the P value of correlation; color indicates magnitude of correlation. D, correlation between IC50 and EMT score in GDSC. Drugs associated with resistance (red) include EGFR and ERBB2 inhibitors; whereas PDGFR inhibitors have lower IC50's in mesenchymal samples (green).

greater sensitivity to PDGFR inhibitors in mesenchymal tumors. better correlation of the pan-cancer EMT signature with known Another interesting finding was increased sensitivity in the mes- markers of EMT across all tumor types evaluated, verifying the enchymal cell lines to the inhibition of the serine/threonine strength of this approach. Interestingly, commonly expressed protein kinases GSK3 and TBK1 (30). genes correlated with the pan-cancer EMT signature in all 11 tumor types included a number of potential therapeutic targets, Discussion such as the receptor tyrosine kinase AXL, which was also an important target in mesenchymal lung cancers identified in our EMT has long been recognized as playing a major role in lung cancer EMT signature (3). With AXL inhibitors entering cancer progression and metastasis (1, 2). Despite extensive clinical trials, the finding of high AXL expression across mul- research in the field, the best way to characterize this phenom- tiple cancer types in the setting of EMT has important clinical enon, especially in the clinical setting, is still under debate (2). implications (31). Another major finding of this analysis Here, through the use of a gene expression signature developed was the significant enrichment of multiple immune targets in from multiple tumor types, we were able to evaluate global cancers that have undergone EMT, which has potentially impor- molecular alterations associated with EMT, as well as potential tant implications for identifying patients with cancer who therapeutic targets. Comparing our original lung cancer EMT may benefit from immunotherapies. Specifically, we found a signature with the pan-cancer EMT signature, we observed a high correlation of expression of 20 druggable immune targets

618 Clin Cancer Res; 22(3) February 1, 2016 Clinical Cancer Research

Pan-Cancer EMT Molecular and Immune Alterations

and mesenchymal status, as defined by the pan-cancer EMT therapeutic response, especially with the emergence of the field of signature and were able to validate these findings by mRNA immunotherapy. By deriving the pan-cancer EMT signature from expression analysis and IHC in an independent lung cancer patients' tumors, we overcame some of these limitations. cohort. Several reports have shown the importance of miRNA regula- Disclosure of Potential Conflicts of Interest tion in EMT (2, 15, 32). As expected, the miR-200 family was J.V. Heymach reports receiving commercial research support from AstraZe- highly expressed in epithelial tumors. However, a number of neca, Bayer, and GlaxoSmithKline; and is a consultant/advisory board member novel miRNAs were found to be highly expressed in mesenchymal for AstraZeneca, Boerhinger Ingelheim, Exelixis, Genentech, GlaxoSmithKline, tumors, regardless of cancer type, including the miR-199 family. Lilly, Novartis, and Synta. No potential conflicts of interest were disclosed by the Previously implicated in skeletal development and fibrosis other authors. (27, 33), miR-199a seems to have a dual role in cancer, both as Authors' Contributions a tumor suppressor and a promoter, depending on the tissue of Conception and design: M.P. Mak, J.V. Heymach, J. Wang, L.A. Byers origin (27). Although its role in EMT is still unknown, the Development of methodology: P. Tong, J. Rodriguez-Canales, I.I. Wistuba, correlation we found between miR-199 and EMT warrants further J.V. Heymach, K.R. Coombes, J. Wang, L.A. Byers investigation. Acquisition of data (provided animals, acquired and managed patients, Although mesenchymal tumors had similar patterns of gene provided facilities, etc.): P. Tong, E.R. Parra, J. Rodriguez-Canales, I.I. Wistuba, and miRNA expression, the overall mutational landscape and L.A. Byers alterations in copy number seem to be mostly determined by Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Tong, L. Diao, D.L. Gibbons, W.N. William, tumor type. In contrast, protein expression and drug sensitivity F. Skoulidis, J.V. Heymach, K.R. Coombes, J. Wang, L.A. Byers patterns appear to be driven by both tumor type and EMT status. Writing, review, and/or revision of the manuscript: M.P. Mak, P. Tong, From this analysis, we identified several drugs as potentially more R.J. Cardnell, D.L. Gibbons, W.N. William, F. Skoulidis, E.R. Parra, I.I. Wistuba, active in the mesenchymal subgroup of each tumor type. How- J.N. Weinstein, K.R. Coombes, J. Wang, L.A. Byers ever, some limitations of this study include different numbers of Study supervision: W.N. William, J. Wang, L.A. Byers cell lines for each drug tested within each tumor type and multiple targets (both inhibitory and promoter) for many of the drugs Grant Support fi included in the analysis. Nevertheless, we identi ed drugs that This project was partially supported by the NIH through the TCGA (5-U24- may be more active in mesenchymal cancers, which are often CA143883-05), the Lung SPORE (P50 CA070907), DoD PROSPECT grant highly resistant to established targeted therapies (e.g., ErbB2 W81XWH-07-1-0306, and the Cancer Center Support Grant (CCSG family inhibitors used in breast, head and neck, and lung cancers). -CA016672), by the Mary K. Chapman Foundation, and through generous Further validation is necessary to determine the clinical relevance philanthropic contributions to The University of Texas MD Anderson Lung Cancer Moon Shot Program. D.L. Gibbons was supported by Rexanna's Foun- of these drugs in the context of EMT. dation for Fighting Lung Cancer, The Stading Lung Cancer Research Fund, the In summary, the new pan-cancer EMT signature introduced National Institutes of Health (K08-CA151651), and the Cancer Prevention & here identifies global molecular alterations across multiple Research Institute of Texas (RP150405); W.N. William was supported by a cancer types that are associated with EMT. Furthermore, the Conquer Cancer Foundation Career Development Award (4546); I.I. Wistuba identification of the association between pan-cancer EMT sig- was supported by MD Anderson's Institutional Tissue Bank (ITB), Award nature scores and expression of candidate drug targets suggests Number 2P30CA016672 from the NIH National Cancer Institute. J.V. Heymach was supported by NIH/NCI (1 R01 CA168484-04) the LUNGevity foundation, thatEMTmaybeaclinicallyusefulmarkerforselectingpatients The V Foundation for Cancer Research, the David Bruton, Jr. Endowed Chair and fi most likely to respond to speci c cancer-targeting approaches. the Rexana Foundation for Fighting Lung Cancer; J.N. Weinstein was supported The finding that mesenchymal tumors express higher levels of by NIH/NCI (5, 2P50 CA100632-11, 2 UL1RR00037106-A1, and 2 P30 immune checkpoint targets is of particular clinical relevance, CA016672 38), the Cancer Prevention & Research Institute of Texas given the significant clinical activity of immunotherapies, such (RP130397), and The Michael and Susan Dell Foundation; L.A. Byers was as PD1 and PD-L1 inhibitors in lung cancer (34, 35) and other supported by an NCI Cancer Clinical Investigator Team leadership Award (P30 CA016672), The Sidney Kimmel Foundation for Cancer Research, and the malignancies, which currently lack validated biomarkers to Sheikh Khalifa Bin Zayed Al Nahyan Institute for the Personalized Cancer fi select patients most likely to bene t from these targeted immu- Therapy's (IPCT's) Center for Professional Education and Training; D.L. Gib- notherapies. The results described here suggest a possible role bons and L.A. Byers are supported by MD Anderson Cancer Center Physician for using a tumor's EMT phenotype to identify cancers that may Scientist Awards, R. Lee Clark Fellowships of The University of Texas MD be more sensitive to immune targeting strategies. Anderson Cancer Center, supported by the Jeane F. Shelby Scholarship Fund, Except for one previous report (4), other existing EMT signa- and the LUNGvevity Foundation. – The costs of publication of this article were defrayed in part by the payment of tures were developed from a single tumor type (3, 5 7), which is page charges. This article must therefore be hereby marked advertisement in likely to limit their application across diverse cancer types. Fur- accordance with 18 U.S.C. Section 1734 solely to indicate this fact. thermore, many of these were developed in cell line models, which do not reflect the contribution of the tumor microenvi- Received April 13, 2015; revised August 11, 2015; accepted August 28, 2015; ronment and immune response, which play important roles in published OnlineFirst September 29, 2015.

References 1. Singh A, Settleman J. EMT, cancer stem cells and drug resistance: 3. Byers LA, Diao L, Wang J, Saintigny P, Girard L, Peyton M, et al. An an emerging axis of evil in the war on cancer. Oncogene 2010; epithelial-mesenchymal transition gene signature predicts resistance to 29:4741–51. EGFRandPI3Kinhibitorsandidentiﬁes Axl as a therapeutic target for 2. De Craene B, Berx G. Regulatory networks deﬁning EMT during cancer overcoming EGFR inhibitor resistance. Clin Cancer Res 2013;19: initiation and progression. Nat Rev Cancer 2013;13:97–110. 279–90.

www.aacrjournals.org Clin Cancer Res; 22(3) February 1, 2016 619

Mak et al.

4. Tan TZ, Miow QH, Miki Y, Noda T, Mori S, Huang RY, et al. Epithelial- 19. Sternberg CN, Davis ID, Mardiak J, Szczylik C, Lee E, Wagstaff J, et al. mesenchymal transition spectrum quantification and its efficacy in deci- Pazopanib in locally advanced or metastatic renal cell carcinoma: results of phering survival and drug responses of cancer patients. EMBO Mol Med a randomized phase III trial. J Clin Oncol 2010;28:1061–8. 2014;6:1279–93. 20. van der Graaf WT, Blay JY, Chawla SP, Kim DW, Bui-Nguyen B, Casali PG, 5. Loboda A, Nebozhyn MV, Watters JW, Buser CA, Shaw PM, Huang PS, et al. et al. Pazopanib for metastatic soft-tissue sarcoma (PALETTE): a rando- EMT is the dominant program in human colon cancer. BMC Med Geno- mised, double-blind, placebo-controlled phase 3 trial. Lancet 2012;379: mics 2011;4:9. 1879–86. 6. Chung CH, Parker JS, Ely K, Carter J, Yi Y, Murphy BA, et al. Gene 21. Redmond WL, Linch SN, Kasiewicz MJ. Combined targeting of costimu- expression profiles identify epithelial-to-mesenchymal transition and latory (OX40) and coinhibitory (CTLA-4) pathways elicits potent effector T activation of nuclear factor-{kappa}B signaling as characteristics of a cells capable of driving robust antitumor immunity. Cancer Immunol Res high-risk head and neck squamous cell carcinoma. Cancer Res 2006; 2014;2:142–53. 66:8210–8. 22. Guo Z, Wang X, Cheng D, Xia Z, Luan M, Zhang S. PD-1 blockade and OX40 7. AlonsoSR,TraceyL,OrtizP,Perez-GomezB,PalaciosJ,PollanM,etal. triggering synergistically protects against tumor growth in a murine model A high-throughput study in melanoma identifies epithelial-mesenchy- of ovarian cancer. PLoS One 2014;9:e89350. mal transition as a major determinant of metastasis. Cancer Res 23. Robert C, Thomas L, Bondarenko I, O'Day S, Weber J, Garbe C, et al. 2007;67:3450–60. Ipilimumab plus dacarbazine for previously untreated metastatic mela- 8. Zhang Z, Lee JC, Lin L, Olivas V, Au V, Laframboise T, et al. Activation of the noma. N Engl J Med 2011;364:2517–26. AXL kinase causes resistance to EGFR-targeted therapy in lung cancer. Nat 24. Hodi FS, O'Day SJ, McDermott DF, Weber RW, Sosman JA, Haanen JB, et al. Genet 2012;44:852–60. Improved survival with ipilimumab in patients with metastatic melanoma. 9. Chen L, Gibbons DL, Goswami S, Cortez MA, Ahn YH, Byers LA, et al. N Engl J Med 2010;363:711–23. Metastasis is regulated via microRNA-200/ZEB1 axis control of tumour cell 25. Hamid O, Robert C, Daud A, Hodi FS, Hwu WJ, Kefford R, et al. Safety and PD-L1 expression and intratumoral immunosuppression. Nat Commun tumor responses with lambrolizumab (anti-PD-1) in melanoma. N Engl 2014;5:5241. J Med 2013;369:134–44. 10. Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, 26. Pencheva N, Tran H, Buss C, Huh D, Drobnjak M, Busam K, et al. et al. The cancer cell line encyclopedia enables predictive modelling of Convergent multi-miRNA targeting of ApoE drives LRP1/LRP8-dependent anticancer drug sensitivity. Nature 2012;483:603–7. melanoma metastasis and angiogenesis. Cell 2012;151:1068–82. 11. Yang W, Soares J, Greninger P, Edelman EJ, Lightfoot H, Forbes S, et al. 27. Gu S, Chan WY. Flexible and versatile as a chameleon-sophisticated Genomics of drug sensitivity in cancer (GDSC): a resource for thera- functions of microRNA-199a. Int J Mol Sci 2012;13:8449–66. peutic biomarker discovery in cancer cells. Nucleic Acids Res 2013;41: 28. Dweep H, Sticht C, Pandey P, Gretz N. miRWalk–database: prediction of D955–61. possible miRNA binding sites by "walking" the genes of three genomes. 12. Akbani R, Ng PK, Werner HM, Shahmoradgoli M, Zhang F, Ju Z, et al. A pan- J Biomed Inform 2011;44:839–47. cancer proteomic perspective on The Cancer Genome Atlas. Nat Commun 29. Ciriello G, Miller ML, Aksoy BA, Senbabaoglu Y, Schultz N, Sander C. 2014;5:3887. Emerging landscape of oncogenic signatures across human cancers. 13. Hoadley KA, Yau C, Wolf DM, Cherniack AD, Tamborero D, Ng S, et al. Nat Genet 2013;45:1127–33. Multiplatform analysis of 12 cancer types reveals molecular classification 30. Wei C, Cao Y, Yang X, Zheng Z, Guan K, Wang Q, et al. Elevated expression within and across tissues of origin. Cell 2014;158:929–44. of TANK-binding kinase 1 enhances tamoxifen resistance in breast cancer. 14. Ingenuity Pathway Analysis (IPA, QIAGEN Redwood City, www.qiagen. Proc Natl Acad Sci U S A 2014;111:E601–10. com/ingenuity). 31. Sheridan C. First Axl inhibitor enters clinical trials. Nat Biotechnol 15. Gregory PA, Bert AG, Paterson EL, Barry SC, Tsykin A, Farshid G, et al. 2013;31:775–6. The miR-200 family and miR-205 regulate epithelial to mesenchymal 32. Schliekelman MJ, Gibbons DL, Faca VM, Creighton CJ, Rizvi ZH, Zhang transition by targeting ZEB1 and SIP1. Nat Cell Biol 2008;10: Q, et al. Targets of the tumor suppressor miR-200 in regulation of 593–601. the epithelial-mesenchymal transition in cancer. Cancer Res 2011;71: 16. Yang J, Mani SA, Donaher JL, Ramaswamy S, Itzykson RA, Come C, et al. 7670–82. Twist, a master regulator of morphogenesis, plays an essential role in tumor 33. Lino Cardenas CL, Henaoui IS, Courcot E, Roderburg C, Cauffiez C, Aubert metastasis. Cell 2004;117:927–39. S, et al. miR-199a-5p Is upregulated during fibrogenic response to tissue 17. Brunner AM, Costa DB, Heist RS, Garcia E, Lindeman NI, Sholl LM, et al. injury and mediates TGFbeta-induced lung fibroblast activation by target- Treatment-related toxicities in a phase II trial of dasatinib in patients ing caveolin-1. PLoS Genet 2013;9:e1003291. with squamous cell carcinoma of the lung. J Thorac Oncol 2013; 34. Brahmer J, Reckamp KL, Baas P, Crino L, Eberhardt WE, Poddubskaya E, 8:1434–7. et al. Nivolumab versus docetaxel in advanced squamous-cell non-small- 18. Lewis NL, Lewis LD, Eder JP, Reddy NJ, Guo F, Pierce KJ, et al. Phase I study cell lung cancer. N Engl J Med 2015;373:123–35. of the safety, tolerability, and pharmacokinetics of oral CP-868,596, a 35. Gettinger SN, Horn L, Gandhi L, Spigel DR, Antonia SJ, Rizvi NA, et al. highly specific platelet-derived growth factor receptor tyrosine kinase Overall survival and long-term safety of nivolumab (anti-programmed death inhibitor in patients with advanced cancers. J Clin Oncol 2009;27: 1 antibody, BMS-936558, ONO-4538) in patients with previously treated 5262–9. advanced non-small-cell lung cancer. J Clin Oncol 2015;33:2004–12.

620 Clin Cancer Res; 22(3) February 1, 2016 Clinical Cancer Research

A Patient-Derived, Pan-Cancer EMT Signature Identifies Global Molecular Alterations and Immune Target Enrichment Following Epithelial-to-Mesenchymal Transition

Milena P. Mak, Pan Tong, Lixia Diao, et al.

Clin Cancer Res 2016;22:609-620. Published OnlineFirst September 29, 2015.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-15-0876

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2015/10/02/1078-0432.CCR-15-0876.DC1

Cited articles This article cites 34 articles, 10 of which you can access for free at: http://clincancerres.aacrjournals.org/content/22/3/609.full#ref-list-1

Citing articles This article has been cited by 30 HighWire-hosted articles. Access the articles at: http://clincancerres.aacrjournals.org/content/22/3/609.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/22/3/609. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.