Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404 Precision Medicine and Imaging Clinical Cancer Research Novel RB1-Loss Transcriptomic Signature Is Associated with Poor Clinical Outcomes across Cancer Types William S. Chen1,2, Mohammed Alshalalfa1,3, Shuang G. Zhao4, Yang Liu5, Brandon A. Mahal3, David A. Quigley1, Ting Wei1, Elai Davicioni5, Timothy R. Rebbeck6, Philip W. Kantoff7, Christopher A. Maher8,9,10, Karen E. Knudsen11, Eric J. Small1,12, Paul L. Nguyen3, and Felix Y. Feng1,13 Abstract Purpose: Rb-pathway disruption is of great clinical interest, was associated with promoter hypermethylation (P ¼ 0.008) as it has been shown to predict outcomes in multiple cancers. and gene body hypomethylation (P ¼ 0.002), suggesting RBS We sought to develop a transcriptomic signature for detecting could detect epigenetic gene silencing. TCGA Pan-Cancer biallelic RB1 loss (RBS) that could be used to assess the clinical clinical analyses revealed that high RBS was associated with implications of RB1 loss on a pan-cancer scale. short progression-free (P < 0.00001), overall (P ¼ 0.0004), and Experimental Design: We utilized data from the Cancer disease-specific(P < 0.00001) survival. On multivariable anal- Cell Line Encyclopedia (N ¼ 995) to develop the first pan- yses, high RBS was predictive of shorter progression-free cancer transcriptomic signature for predicting biallelic RB1 survival in TCGA Pan-Cancer (P ¼ 0.03) and of shorter overall loss (RBS). Model accuracy was validated using The Cancer survival in mCRPC (P ¼ 0.004) independently of the number Genome Atlas (TCGA) Pan-Cancer dataset (N ¼ 11,007). RBS of DNA alterations in RB1. was then used to assess the clinical relevance of biallelic RB1 Conclusions: Our study provides the first validated tool to loss in TCGA Pan-Cancer and in an additional metastatic assess RB1 biallelic loss across cancer types based on gene castration-resistant prostate cancer (mCRPC) cohort. expression. RBS can be useful for analyzing datasets with or Results: RBS outperformed the leading existing signature without DNA-sequencing results to investigate the emerging for detecting RB1 biallelic loss across all cancer types in TCGA prognostic and treatment implications of Rb-pathway Pan-Cancer (AUC, 0.89 vs. 0.66). High RBS (RB1 biallelic loss) disruption. 1Helen Diller Family Comprehensive Cancer Center, University of California, San Introduction 2 Francisco, San Francisco, California. Yale School of Medicine, New Haven, RB1 is a tumor suppressor that has been implicated in the Connecticut. 3Dana-Farber Cancer Institute and Brigham and Women's Hospital, 4 pathogenesis of numerous cancer types. In addition to causing Boston, Massachusetts. Department of Radiation Oncology, University of RB1 Michigan, Ann Arbor, Michigan. 5GenomeDx Biosciences, Vancouver, British pediatric retinoblastoma, alterations have been shown to Columbia, Canada. 6Department of Epidemiology, Harvard T.H. Chan School of play a major role in the progression of osteosarcoma (1), lym- Public Health, Boston, Massachusetts. 7Department of Medicine, Memorial Sloan phoma (2), and breast (3–5), lung (6, 7), and prostate (8, 9) Kettering Cancer Center, New York, New York. 8McDonnell Genome Institute, malignancies. Moreover, recent studies have highlighted RB1 loss 9 Washington University in St. Louis, St. Louis, Missouri. Department of Internal as an important clinical prognostic factor in specific cancer types. Medicine, Washington University in St. Louis, St. Louis, Missouri. 10Department of For example, RB1 loss has been shown to be associated with poor Biomedical Engineering, Washington University in St. Louis, St. Louis, Missouri. 11Departments of Cancer Biology and Medical Oncology, Thomas Jefferson overall survival (OS) in osteosarcoma (1), glioblastoma (10), and University, Philadelphia, Pennsylvania. 12Department of Medicine, University of lung cancers (11) and has been shown to predict resistance or California, San Francisco, San Francisco, California. 13Departments of Radiation sensitivity to various small-cell lung cancer (7), pancreatic can- Oncology and Urology, University of California, San Francisco, San Francisco, cer (12), and breast cancer therapies (3, 13). California. In order to study the clinical implications of Rb-pathway Note: Supplementary data for this article are available at Clinical Cancer disruption, one must first be able to confidently assess RB1 status. Research Online (http://clincancerres.aacrjournals.org/). Next-generation DNA-sequencing approaches are well suited for W.S. Chen and M. Alshalalfa contributed equally to this article. identifying mutations, copy-number alterations, and structural P.L. Nguyen and F.Y. Feng contributed equally to this article. variants. However, there is often uncertainty as to whether a DNA alteration truly inactivates the affected allele. Moreover, other Corresponding Author: Felix Y. Feng, University of California, San Francisco, San mechanisms of gene inactivation exist that may not be captured Francisco, CA 94143. Phone: 415-885-7627; Fax: 415-353-9883; E-mail: [email protected] by DNA-sequencing techniques (e.g., epigenetic, posttranscrip- tional, or posttranslational modifications). An alternative Clin Cancer Res 2019;XX:XX–XX approach to assessing gene inactivation is to examine the sequelae doi: 10.1158/1078-0432.CCR-19-0404 of genomic alterations by assessing the resulting expression of Ó2019 American Association for Cancer Research. related, downstream genes. www.aacrjournals.org OF1 Downloaded from clincancerres.aacrjournals.org on September 23, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404 Chen et al. UCSC Xena Browser to annotate RB1 copy number (CN) and Translational Relevance mutation calls (18). Cell lines with GISTC score < -0.8 were RB1 loss is a recurrent genomic alteration that has been annotated as deep (two-copy) deletion (CN-2), and cell lines shown to predict response to various treatments including with GISTC score between -0.8 and -0.4 were annotated as shallow radiotherapy, platinum-based chemotherapy, and CDK4/6 (single-copy) deletion (CN-1). The remaining cell lines were inhibitors in multiple cancer types. Leveraging the transcrip- annotated as two-copy intact (CN-0). To build an mRNA classifier tomic and DNA-sequencing data of over 11,000 cancer cell to predict RB1 functional loss, we defined the tumor cell lines with lines and clinical tumor samples, we identified a novel pan- predicted biallelic loss (i.e., deep deletion, shallow deletion with cancer transcriptomic signature for identifying RB1 loss (RBS). additional DNA mutation, or 2þ DNA mutations) as the RB1-loss RBS is more accurate than existing transcriptomic signatures in group and remaining cell lines as the RB1-intact group. To identify detecting RB1 loss and can be used alongside DNA sequencing differentially expressed genes between the two groups, we used to identify Rb-loss tumors more comprehensively. Using RBS, the Wilcoxon Mann–Whitney test with an adjusted P value À we found that RB1 loss was associated with impaired survival threshold of P < 1 Â 10 10. across cancer types, supporting the notion that RB1 loss We then used a nearest shrunken centroid approach (PAM; constitutes a biologically and clinically distinct subgroup of ref. 19) to generate our gene signature based on the expression cancers. Our novel transcriptomic signature can be used to pattern of the genes selected as described above. We trained the further investigate the clinical implications of RB1 loss and model by applying PAM to CCLE expression data, using posterior may be coupled with treatment response data to help develop class probabilities for RB1-loss class predictions. The model was personalized cancer treatment regimens. trained using 10-fold cross validation to optimize the PAM shrinkage parameter. RBS was then validated on the TCGA Pan-Cancer RNA-sequenc- ing (RNA-seq) expression dataset of 11,007 tumor samples span- RB1 There exist a few gene sets (genes theorized to be collec- ning 33 cancer types, downloaded from UCSC Xena Browser RB1 tively indicative of status) and two gene signatures (combi- using the Synapse platform (syn4976369). RB1 copy-number natorial expression pattern of the genes in a gene set) for predict- calls and mutation data for these samples were obtained from RB1 ing loss (14, 15). However, they all share the key limitation UCSC Xena Browser, and the same GISTIC2.0 copy-number that they consist largely of cell-cycle genes (whose expression is thresholds and mutation criteria as used in the CCLE training set fi RB1 not speci cto loss). Moreover, because these gene sets and were applied to the validation set. Final RB1-loss annotations signatures were primarily developed using breast cancer data, were defined based on the number of RB1 DNA alterations their generalizability to different cancer types has not been val- observed: 2 alterations (deep deletion, shallow deletion with one fi RB1 idated. Our rst aim was to develop a novel pan-cancer mutation, or 2þ mutations), 1 alteration (shallow deletion with biallelic loss gene signature (RBS) that outperformed existing no mutations or one mutation with no deletion), 0 alteration (no RB1 RB1 -loss signatures and accurately predicted biallelic loss deletions or mutations). Model accuracy was assessed based on across cancer types. area under the ROC curve (AUC), benchmarked against the After generating and validating RBS, we then sought