Novel RB1-Loss Transcriptomic Signature Is Associated with ... · 11Departments of Cancer Biology and Medical Oncology, Thomas Jefferson University, Philadelphia, Pennsylvania. 12Department

Precision Medicine and Imaging

Novel RB1-Loss Transcriptomic Signature IsAssociated with Poor Clinical Outcomes acrossCancer TypesWilliam 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,as it has been shown to predict outcomes in multiple cancers.We sought to develop a transcriptomic signature for detectingbiallelic RB1 loss (RBS) that could be used to assess the clinicalimplications of RB1 loss on a pan-cancer scale.

Experimental Design: We utilized data from the CancerCell Line Encyclopedia (N ¼ 995) to develop the first pan-cancer transcriptomic signature for predicting biallelic RB1loss (RBS). Model accuracy was validated using The CancerGenome Atlas (TCGA) Pan-Cancer dataset (N¼ 11,007). RBSwas then used to assess the clinical relevance of biallelic RB1loss in TCGA Pan-Cancer and in an additional metastaticcastration-resistant prostate cancer (mCRPC) cohort.

Results: RBS outperformed the leading existing signaturefor detecting RB1 biallelic loss across all cancer types in TCGAPan-Cancer (AUC, 0.89 vs. 0.66). High RBS (RB1biallelic loss)

was associated with promoter hypermethylation (P ¼ 0.008)and gene body hypomethylation (P ¼ 0.002), suggesting RBScould detect epigenetic gene silencing. TCGA Pan-Cancerclinical analyses revealed that high RBS was associated withshort progression-free (P<0.00001), overall (P¼0.0004), anddisease-specific (P < 0.00001) survival. Onmultivariable anal-yses, high RBS was predictive of shorter progression-freesurvival in TCGA Pan-Cancer (P¼ 0.03) and of shorter overallsurvival in mCRPC (P¼ 0.004) independently of the numberof DNA alterations in RB1.

Conclusions: Our study provides the first validated tool toassess RB1 biallelic loss across cancer types based on geneexpression. RBS can be useful for analyzing datasets with orwithout DNA-sequencing results to investigate the emergingprognostic and treatment implications of Rb-pathwaydisruption.

IntroductionRB1 is a tumor suppressor that has been implicated in the

pathogenesis of numerous cancer types. In addition to causingpediatric retinoblastoma, RB1 alterations have been shown toplay a major role in the progression of osteosarcoma (1), lym-phoma (2), and breast (3–5), lung (6, 7), and prostate (8, 9)malignancies. Moreover, recent studies have highlighted RB1 lossas an important clinical prognostic factor in specific cancer types.For example, RB1 loss has been shown to be associated with pooroverall survival (OS) in osteosarcoma (1), glioblastoma (10), andlung cancers (11) and has been shown to predict resistance orsensitivity to various small-cell lung cancer (7), pancreatic can-cer (12), and breast cancer therapies (3, 13).

In order to study the clinical implications of Rb-pathwaydisruption, onemust first be able to confidently assess RB1 status.Next-generation DNA-sequencing approaches are well suited foridentifying mutations, copy-number alterations, and structuralvariants. However, there is often uncertainty as to whether a DNAalteration truly inactivates the affected allele. Moreover, othermechanisms of gene inactivation exist that may not be capturedby DNA-sequencing techniques (e.g., epigenetic, posttranscrip-tional, or posttranslational modifications). An alternativeapproach to assessing gene inactivation is to examine the sequelaeof genomic alterations by assessing the resulting expression ofrelated, downstream genes.

1Helen Diller Family Comprehensive Cancer Center, University of California, SanFrancisco, San Francisco, California. 2Yale School of Medicine, New Haven,Connecticut. 3Dana-Farber Cancer Institute andBrighamandWomen's Hospital,Boston, Massachusetts. 4Department of Radiation Oncology, University ofMichigan, Ann Arbor, Michigan. 5GenomeDx Biosciences, Vancouver, BritishColumbia, Canada. 6Department of Epidemiology, Harvard T.H. Chan School ofPublic Health, Boston, Massachusetts. 7Department of Medicine, Memorial SloanKettering Cancer Center, New York, New York. 8McDonnell Genome Institute,Washington University in St. Louis, St. Louis, Missouri. 9Department of InternalMedicine,WashingtonUniversity in St. Louis, St. Louis, Missouri. 10Department ofBiomedical Engineering, Washington University in St. Louis, St. Louis, Missouri.11Departments of Cancer Biology and Medical Oncology, Thomas JeffersonUniversity, Philadelphia, Pennsylvania. 12Department of Medicine, University ofCalifornia, San Francisco, San Francisco, California. 13Departments of RadiationOncology and Urology, University of California, San Francisco, San Francisco,California.

Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).

W.S. Chen and M. Alshalalfa contributed equally to this article.

P.L. Nguyen and F.Y. Feng contributed equally to this article.

CorrespondingAuthor:Felix Y. Feng, University of California, SanFrancisco, SanFrancisco, CA 94143. Phone: 415-885-7627; Fax: 415-353-9883; E-mail:[email protected]

Clin Cancer Res 2019;XX:XX–XX

doi: 10.1158/1078-0432.CCR-19-0404

�2019 American Association for Cancer Research.

ClinicalCancerResearch

www.aacrjournals.org OF1

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

There exist a few RB1 gene sets (genes theorized to be collec-tively indicative of RB1 status) and two gene signatures (combi-natorial expression pattern of the genes in a gene set) for predict-ing RB1 loss (14, 15). However, they all share the key limitationthat they consist largely of cell-cycle genes (whose expression isnot specific to RB1 loss). Moreover, because these gene sets andsignatures were primarily developed using breast cancer data,their generalizability to different cancer types has not been val-idated. Our first aim was to develop a novel pan-cancer RB1biallelic loss gene signature (RBS) that outperformed existingRB1-loss signatures and accurately predicted biallelic RB1 lossacross cancer types.

After generating and validating RBS, we then sought to use it toassess RB1 loss as a prognostic factor across all major cancer typesusing The Cancer Genome Atlas (TCGA) Pan-Cancer database(N ¼ 11,007). Because RB1 loss was known to be clinicallyimportant inmetastatic prostate cancer (not included in the TCGAPan-Cancer dataset), we examined the prognostic significance ofRBS in an independent metastatic castration-resistant prostatecancer (mCRPC) cohort.

Materials and MethodsVariable definitions

Wedefined "RB1 loss" in our article as predicted biallelic loss ofRB1. For the purposes of training and testing our RB1-loss clas-sifier (RBS), ground-truth labels ofRB1 status for each tumorwereassigned based on the number of DNA alterations (i.e., nonsilentexonic mutations, copy-number loss, and inactivating structuralvariants) observed in RB1. For these ground-truth labels, RB1 losswas defined as presence of at least two DNA alterations in RB1.

RB1-loss gene signature (RBS) development and validationusing the CCLE and TCGA Pan-Cancer datasets

Taking an unbiased approach to selecting genes indicative ofRB1 loss, we leveraged microarray log2-normalized RPKM geneexpression data of 951 pan-cancer cell lines from the Cancer CellLine Encyclopedia (CCLE; ref. 16). We extracted GISTC2.0 (17)and whole-exome sequencing (WES)–based mutation calls from

UCSC Xena Browser to annotate RB1 copy number (CN) andmutation calls (18). Cell lines with GISTC score < -0.8 wereannotated as deep (two-copy) deletion (CN-2), and cell lineswithGISTC score between -0.8 and -0.4were annotated as shallow(single-copy) deletion (CN-1). The remaining cell lines wereannotated as two-copy intact (CN-0). To build anmRNA classifierto predictRB1 functional loss, we defined the tumor cell lineswithpredicted biallelic loss (i.e., deep deletion, shallow deletion withadditional DNAmutation, or 2þDNAmutations) as theRB1-lossgroup and remaining cell lines as theRB1-intact group. To identifydifferentially expressed genes between the two groups, we usedthe Wilcoxon Mann–Whitney test with an adjusted P valuethreshold of P < 1 � 10�10.

We then used a nearest shrunken centroid approach (PAM;ref. 19) to generate our gene signature based on the expressionpattern of the genes selected as described above. We trained themodel by applying PAM to CCLE expression data, using posteriorclass probabilities for RB1-loss class predictions. The model wastrained using 10-fold cross validation to optimize the PAMshrinkage parameter.

RBSwas then validatedon the TCGAPan-Cancer RNA-sequenc-ing (RNA-seq) expression dataset of 11,007 tumor samples span-ning 33 cancer types, downloaded from UCSC Xena Browserusing the Synapse platform (syn4976369). RB1 copy-numbercalls and mutation data for these samples were obtained fromUCSC Xena Browser, and the same GISTIC2.0 copy-numberthresholds and mutation criteria as used in the CCLE training setwere applied to the validation set. Final RB1-loss annotationswere defined based on the number of RB1 DNA alterationsobserved: 2 alterations (deep deletion, shallow deletion with onemutation, or 2þ mutations), 1 alteration (shallow deletion withnomutations or onemutation with no deletion), 0 alteration (nodeletions or mutations). Model accuracy was assessed based onarea under the ROC curve (AUC), benchmarked against theleading existing RB1-loss signature (14).

RBS pathway enrichment analysisThe EnrichR web tool was used to identify genomic pathways

enriched in the RBS gene set. Candidate gene sets were defined asall pathways in the KEGG, Reactome, WikiPathways, and Bio-Carta databases. Pathways were considered significantly enrichedif their adjusted P values were less than a predetermined signif-icance level of 0.05.

Differential expression analysis of RB1 loss due to two or moreRB1 mutations

Differential expression analysis between CN-0 tumors with nomutations and CN-0 tumors with two or more mutations wasperformed to identify genes that were differentially expressed intumors with 2þ RB1 mutations. Given that there were far fewertumors with 2þmutations than therewere with nomutations, werandomly subsampled a set of CN-0 tumors with no mutationsequal in size to the subset of tumors with 2þmutations. We thenperformed a differential expression analysis between the tumorswith 2þ mutations and the tumors with no mutations using theWilcoxonMann–Whitney test with an adjusted P-value thresholdof P < 0.001. For statistical robustness, we performed a boot-strapped analysis with 1,000 different subsamples. Genes wereconsidered significantly differentially expressed if, in >95% of allcomparisons, they demonstrated the same directionality of over-

Translational Relevance

RB1 loss is a recurrent genomic alteration that has beenshown to predict response to various treatments includingradiotherapy, platinum-based chemotherapy, and CDK4/6inhibitors in multiple cancer types. Leveraging the transcrip-tomic and DNA-sequencing data of over 11,000 cancer celllines and clinical tumor samples, we identified a novel pan-cancer transcriptomic signature for identifying RB1 loss (RBS).RBS ismore accurate than existing transcriptomic signatures indetecting RB1 loss and can be used alongside DNA sequencingto identify Rb-loss tumors more comprehensively. Using RBS,we found that RB1 loss was associated with impaired survivalacross cancer types, supporting the notion that RB1 lossconstitutes a biologically and clinically distinct subgroup ofcancers. Our novel transcriptomic signature can be used tofurther investigate the clinical implications of RB1 loss andmay be coupled with treatment response data to help developpersonalized cancer treatment regimens.

Chen et al.

Clin Cancer Res; 2019 Clinical Cancer ResearchOF2

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

versus underexpression and had adjusted P values below thepredetermined significance level of 0.001.

Promoter and gene body methylation analysisTo assess the utility of RBS in detecting gene silencing due to

methylation, we downloaded TCGA Pan-Cancer methylationdata for 49RB1methylation probes from theUCSCXena Browser.We first filtered out probes that were previously identified to be oflow quality (20). We then computed Spearman correlation coef-ficients between RBS score and Illumina DNA methylation 450Karray beta values for each RB1methylation probe. To test whetherthe correlations between RBS score andmethylation probe valueswere significant in the RB1 promoter and gene body regions, wegenerated a null model by computing the correlation betweenRBS score andmethylation in thepromoter and genebody regionsof 20 other random tumor suppressors not known to be related toRB1. For this analysis, a large set of tumor suppressors (N¼1,217)was downloaded from the Tumor SuppressorGeneDatabase (21)and those not located on the same chromosome as RB1 (i.e., noton chromosome 13) and not included in RBS were used ascandidate genes for the null model. Spearman correlation coeffi-cients computed between RBS and eachmethylation probe in thepromoter region of a gene (defined as� 1.5kb of the transcriptionstart site; ref. 22) were then modeled as a distribution. Thedistribution of correlations between RBS and RB1 promotermethylation probes was compared with the distribution of cor-relations between RBS and non-RB1 promoter methylationprobes using the Kolmogorov–Smirnov test with a two-sidedsignificance level of 0.05. Analogous analyses were performedfor the gene body region, where gene body was defined as theregion 1.5 kb downstream of the transcription start site to thetranscription terminator. Transcription start sites and terminatorswere defined using the "biomaRt" R package (23).

Characterizing the prognostic value of RB1 loss across cancertypes

Clinical outcomes data [progression-free survival (PFS), OS,anddisease-specific survival (DSS)]were obtained from the TCGAPan-Cancer Clinical Data Resource (24). All patients with avail-able log2-normalized RPKM RNA-seq data and clinical outcomesdata were included in survival analyses. Microarray expressiondata were log2-normalized and scaled prior to RBS analysis. Datafor themCRPC cohort were obtained fromapreviously publishedstudy (25). This cohort consisted of 101patientswithdeepwhole-genome sequencing, whole-transcriptome RNA-seq, and clinicaloutcomes data available. The mCRPC RNA-seq data were log2-normalized FPKM values. The clinical endpoint examined wasOS, with time of study entry defined as date of mCRPC diagnosis.

The threshold of RBS score used to assign binary RB1-impairedversus RB1-intact status in both cohorts was determined by usingthe Youden index (computed using the "OptimalCutpoints" Rpackage; ref. 26) to select a threshold that maximized predictionaccuracy in the CCLE training dataset. Cox proportional hazardmodels were used to model time-to-event data. All survivalanalyses were performed using R version 3.5.0.

ResultsRB1-loss gene signature development and validation usingCCLE and TCGA Pan-Cancer data

To define our RB1-loss gene set, we identified genes that weredifferentially expressed between CCLE cell lines that demonstrat-

ed RB1 loss and cell lines that had intact RB1. Note that 951 of the995 total cell lines had both copy-number andmicroarray expres-sion data available. Of these 951, 126 were identified as havingbiallelic RB1 loss (99 harbored two-copy deletions, 23 harboredsingle-copy deletions with an additional mutation, and 11 har-bored 2þmutations) and 797 were identified as RB1 intact. Ourunbiased approach to defining an RB1-loss gene set using CCLEdata identified a final set of 186 genes that were indicative of RB1loss (Supplementary Table S1A). Of note, only 7 of the 186 genesoverlapped with genes in the existing RB1-loss signature (14).

To assess the potential utility of our 186-gene signature forpredicting RB1 loss, we first performed t-SNE dimensionalityreduction on the CCLE training data (N ¼ 951). Visualizationof the t-SNE embedding revealed that cell lines with 2þ DNAalterations in RB1 tended to map to similar parts of the embed-ding, suggesting that these cell lines had similar 186-gene expres-sion profiles (Fig. 1A). This finding supported the hypothesis thatthe 186 genes were useful in differentiating between RB1-loss andRB1-intact samples.

The expression values of the 186 genes nominated as describedabove were then used in a supervised learning approach (PAM) tocompute an RBS score for predicting RB1 loss. The model wastrained using the gene expression profiles of CCLE cell lines withknown RB1 status (i.e., RB1-loss vs. RB1-intact). The modelidentified 144 genes whose expression values were most predic-tive ofRB1 status—these geneswere used to compute thefinal RBSscore (Fig. 1B; Supplementary Table S1B). RB1 and CCND1 wereamong the genes expressed at relatively low levels in RB1-losssamples, whereas CDKN2A was among the genes expressed atrelatively high levels inRB1-loss samples. Thiswas consistentwithprior studies which found a high ratio of CDKN2A to CCND1expression to be associated with RB1 loss in multiple cancertypes (27, 28). Because we noticed that prior RB1 gene sets andgene signatures largely consisted of cell proliferation genes, weassessed the association between RBS and a previously publishedcell proliferation activity score (29). Although a previously pub-lished RB1-loss signature (14)was found to be strongly correlatedwith the cell proliferation score (r¼ 0.93), we found that RBSwasonly weakly correlated with the cell proliferation score (r¼ 0.03).These findings suggested that RBS was not a surrogate marker forcell proliferation and was potentially more specific to RB1 lossthan existing signatures. Moreover, EnrichR pathway enrichmentanalysis revealed that RBS was enriched for genes not only in thecyclin D–CDK4/6 and cell-cycle–related pathways but also in theDNA damage response and TP53 signaling pathways (Supple-mentary Table S2). Altogether, these results were consistent withrecent literature that suggests RB1 may play a role in processesother than cell-cycle control (30).

To validate RBS as an accurate model for predicting biallelicRB1 loss, we used the TCGA Pan-Cancer atlas expression datasetcontaining RNA expression data for 11,007 tumors spanning 33cancer types with known mutation and copy-number annota-tions. Note that 698 of these samples were annotated as havingtwoormoreRB1DNA alterations [559 had deep deletion (CN-2),89 had shallow deletion with mutation (CN-1/mut), and 50 hadtwo or more mutations with no deletions], 1,514 as having oneRB1 alteration [1,332 with shallow deletion and no mutation(CN-1/no-mut), and 182 with one mutation and no deletions(CN-0/mut)], and 7,727 as having no RB1 DNA alterations. RBSachieved an AUC of 0.89 for predicting RB1 biallelic loss in thisvalidation set—far superior to an AUC of 0.66 achieved by

Pan-Cancer RB1-Loss Signature Associated with Poor Survival

www.aacrjournals.org Clin Cancer Res; 2019 OF3

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

applying the leading existing RB1-loss signature (14) to the samedataset (Fig. 2A andB). RBS also outperformed a predictivemodelbased solely on the ratio of CDKN2A to CCND1 expression (AUC¼ 0.72), which was previously reported to be associated with RB1loss. Genes including CAMK2N2, CDKN2A, and GPR137C werepositively correlated with RBS score (i.e., high expression in RB1loss), whereas genes including MED4 and RB1 were most nega-tively correlated with RBS score (Fig. 2C).

RBS was highly accurate at identifying RB1 loss due to deepdeletion and due to shallow deletion with additional mutation,which comprised the largemajority ofRB1-loss tumors. However,RBS was less effective at detecting the few RB1-loss tumors with2þ RB1 mutations, suggesting that these tumors may have adistinct gene expression profile. To investigate this further, weperformed a bootstrapped differential expression analysis toidentify genes over- and underexpressed in CN-0 tumors with

Figure 1.

A, t-SNE embedding of CCLE cell lines colored by number of DNA alterations in RB1. Embedding was constructed based on expression levels of the 186 genesfound to be differentially expressed between RB1-impaired and RB1-intact cell lines. B, Heatmap visualizing expression values of 186 genes (rows) in 951 CCLE celllines (columns). Cell lines are ordered from left to right in terms of increasing RBS score, where high RBS score denotes impaired RB1. Orange represents highexpression, and blue represents low expression.

Figure 2.

Boxplots showing accuracy of (A) RBS in predicting biallelic RB1 loss compared with (B) the leading existing model. C, Heatmap of TCGA Pan-Cancer datashowing mRNA expression profiles of 186 genes (rows) in 11,007 patients (columns). Patients are ordered from left to right in terms of increasing RBS score,where high RBS score denotes impaired RB1. Orange represents high expression, and blue represents low expression.

Chen et al.

Clin Cancer Res; 2019 Clinical Cancer ResearchOF4

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

two or more RB1 mutations compared to tumors with no RB1mutations (Materials and Methods). We identified 448 genessignificantly overexpressed and 245 genes significantly under-expressed in the tumors with two or more RB1 mutations(Supplementary Table S3). Of these, 16 overexpressed genes(including CCNE2 and CDKN2A) and 3 underexpressed genes(most notably RB1) were also found in RBS. In addition, severalknown regulators or effectors of RB1 such as CCNE1, CDK2,EZH2, HOXB7, and select E2F-family genes were not in RBSbut were differentially expressed in the tumors with two ormore mutations in RB1 (30–34). Altogether, these findingssuggested that there are some transcriptomic similarities butalso notable differences between RB1 loss due to deletion anddue to biallelic RB1 mutations.

RBS can be useful for capturing the effects of gene inactivationdue to epigenetic modification

To assess the utility of RBS in capturing the effects ofepigenetic events on gene expression, we examined the corre-lation between RBS score and the methylation scores of 39methylation probes in the Pan-Cancer cohort (Fig. 3). To testwhether the pattern of correlation between RBS and methyla-tion probe values was significant in the RB1 promoter and genebody regions, we compared our results with the correlationbetween RBS score and methylation in the promoter and genebody regions of 20 other random tumor suppressors unrelatedto RB1 (Supplementary Table S4). We found that the positivecorrelation between RBS score and RB1 promoter methylationand negative correlation between RBS score and RB1 gene body

Figure 3.

High RBS (impaired RB1) is (A)positively correlated withmethylation of CpGs in the RB1promoter region and (B) negativelycorrelated with methylation of CpGsin the RB1 gene body. Given priorreports of promoterhypermethylation and gene bodyhypomethylation being associatedwith gene inactivation, these resultssuggest RBS may detect tumorswith impaired RB1 due tomethylation-based gene silencing.

Pan-Cancer RB1-Loss Signature Associated with Poor Survival

www.aacrjournals.org Clin Cancer Res; 2019 OF5

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

methylation were significant (P ¼ 0.0077 and P ¼ 0.0016,respectively). The directionality of correlation was also consis-tent with existing literature, which suggests that promotermethylation is associated with decreased gene expression andgene body methylation is associated with increased geneexpression in tumor suppressors (22). These findings supportedthe hypothesis that RBS could detect the downstream effects ofRB1 loss due to multiple etiologies, including those (such asmethylation) that may not be captured using DNA-sequencingtechniques.

RBS highlights RB1 loss as a recurrent genomic event andprognostic factor across cancer types

After assessing the accuracy of RBS for predicting RB1 loss, wesought to use RBS to investigate the prognostic significance ofRB1loss across cancer types. For this analysis, we included patients inthe TCGA Pan-Cancer dataset with available clinical follow-up.High RBS was defined as scores above a threshold of 0.6, deter-mined based on the Youden Index approach applied to theCCLE training dataset. Of note, we found that the majority ofcancer types had an RB1 2-hit prevalence of greater than 5%,suggesting that RB1 loss was common and potentially importantinmany cancer types. In our pooled analysis of all patient samplesacross cancer types, we found that RB1 loss defined using RBSwas

predictive of short PFS [median PFS, 36 vs. 56 months; HR, 1.3;95% confidence interval (CI), 1.18–1.44; P < 0.0001; Fig. 4A],short DSS (median DSS, 88 vs. 219 months; HR, 1.34; 95% CI,1.17–1.55; P < 0.0001; Fig. 4B), and short OS (median OS, 70 vs.94months; HR, 1.23; 95%,CI, 1.09–1.38; P¼ 0.0004; Fig. 4C). Ina multivariable survival model including both RBS and cancertype, high RBSwas found to be independently prognostic of shortPFS (HR, 1.12; 95% CI, 1.02–1.26; P ¼ 0.04). These findingssupported the hypothesis that RB1 loss is clinically importantacross cancer types andmay indicatemore advanced or aggressivedisease in general.

We additionally assessed the prognostic significance of a DNA-sequencing–based definition of RB1 loss, namely, having at leasttwo DNA alterations in RB1. We found that similarly to high RBS,presence of 2þ DNA alterations in RB1 was associated with shortOS, PFS, and DFS compared with presence of 0 or 1 DNAalterations in RB1 (Fig. 4D–F). These findings suggested that ourdefinition of "RB1 loss" as predicted biallelic loss of RB1 wasclinically meaningful.

RBS is predictive of poor clinical outcomes independently ofthe number of DNA alterations in RB1

In our methylation analysis, we showed that RBS could poten-tially be used to detect RB1 loss through mechanisms that could

Figure 4.

Differences in (A) PFS, (B) DSS, and (C) OS between high-RBS (RNA-seq profile consistent with impaired RB1) and low-RBS cancers. Differences in (D) PFS, (E)DSS, and (F) OS between patients with 0, 1, and 2 DNA alterations in RB1.

Chen et al.

Clin Cancer Res; 2019 Clinical Cancer ResearchOF6

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

not be detected by DNA sequencing. In addition, it is known thatnot all DNA mutations and copy-number loss events in a genehave the same effect on the affected allele (i.e., resulting proteinmay still be partly or completely functional). Because RBS mea-sures the downstream effects of DNA and non-DNA alterations atthe gene expression level, we hypothesized that RBS may offerinformation on Rb-pathway disruption that is independent ofDNA-sequencing results.

To explore this hypothesis, we assessed the prognostic sig-nificance of high RBS for predicting survival in the TCGA Pan-Cancer cohort independently of number of observed DNAalterations. For these analyses, we focused on PFS as our clinicalendpoint of interest due to a prior study that found that PFSwas generally the most accurate endpoint collected across allcancer types in the TCGA Pan-Cancer dataset (24). On multi-variable analysis adjusting for number of DNA alterations inRB1, high RBS was independently predictive of short PFS (HR,1.14; 95% CI, 1.02–1.29; P ¼ 0.03). This suggested that RBSmay help distinguish patients with a more pronounced RB1-impaired clinical phenotype from those with a less-pronouncedphenotype independently of the number of DNA alterationsobserved in the gene. Moreover, using a criterion of high RBS or2þ DNA alterations in RB1 to select RB1-impaired patientsresulted in a 73% increase in group size as compared with usingthe criteria of just 2þ DNA alterations (Supplementary Fig.S1A). Thus, RBS may be useful for identifying a more compre-hensive group of patients with Rb-pathway disruption than canbe recovered using DNA sequencing alone.

To explore this concept further, we examined a previouslypublished cohort of patients with mCRPC (25)—the lethal sub-type of prostate cancer not represented in the TCGA Pan-Cancercohort. RB1 loss (as defined based on detected DNA alterations inRB1) has been shown to be associated with short survival inmCRPC (35). Interrogating the mCRPC cohort of 101 patientswithbothwhole-genome sequencing andRNA-seqdata available,we aimed to assess whether high RBS might be predictive of short

OS independently of the number of DNA alterations present.First, we examined the degree of concordance between RB1 statusas defined based on number of DNA alterations observed and asdefined based on RBS score. We found that although RBS scorewas strongly related to the number of DNA alterations observed(AUC ¼ 0.90), not all tumors with high RBS score harbored 2þDNA alterations and vice versa (Fig. 5A). By expanding the DNA-sequencing–based definition of RB1-loss (2þ RB1 DNA altera-tions) to include tumorswith fewer than 2DNAalterations inRB1butwith high RBS, one could recover 50%more tumorswithRB1-impaired status (Supplementary Fig. S1B). Next, we examined theprognostic significance of high RBS in the mCRPC cohort. Wefound that RB1 loss as defined by high RBSwas predictive of shortOS in mCRPC (median OS, 15.0 vs. 42.0 months; HR, 2.93; 95%CI, 1.47–5.83; P ¼ 0.001; Fig. 5B). Finally, to assess whether theRNA-seq (high RBS) and DNA-sequencing (number of DNAalterations in RB1) results were independently predictive ofsurvival outcomes, we performed amultivariable analysis includ-ing both the RNA-seq and DNA-sequencing definitions of pre-dicted RB1 loss. We found that both the RNA-seq and DNA-sequencing definitions were independently predictive of short OS(P ¼ 0.0036 and P ¼ 0.046, respectively), suggesting that bothRNA-seq and DNA sequencing offered unique information onRB1 status that could be used to detect a clinical phenotype ofRB1-impaired, clinically aggressive mCRPC.

DiscussionIn order to assess the clinical implications of RB1 loss across

cancer types, we developed a pan-cancer RB1-loss signature (RBS)that predicted biallelic loss of RB1 based on gene expression data.We found that RBS was highly accurate at predicting RB1 lossacross cancer types compared with existing RB1 gene signatures.Moreover, RBS was able to capture RB1 inactivation due to bothDNA and epigenetic changes. Using pan-cancer (N¼ 10,486) andmetastatic prostate cancer (N ¼ 101) cohorts, we demonstrated

Figure 5.

A, Boxplot showing distribution of RBS scores for mCRPC tumors stratified by number of DNA alterations in RB1. AUC represents accuracy of RBS in predictingpresence of 2þDNA alterations in RB1. B, Difference in OS between Rb-loss (high-RBS) and Rb-intact (low-RBS) patients.

Pan-Cancer RB1-Loss Signature Associated with Poor Survival

www.aacrjournals.org Clin Cancer Res; 2019 OF7

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

that high RBS was predictive of poor clinical outcomes indepen-dently of the number of DNA alterations in RB1.

There are several possible explanations as towhyRBSwasmuchmore accurate than the leading existing RB1 signature at predict-ing biallelic loss of RB1 (AUC of 0.89 vs. 0.66). For one, RBS wasthe only RB1-loss signature that was designed to be applied acrosscancer types. Because RBS was trained on CCLE cell-line dataderived from many different primary tissue types, it was well-suited to assess RB1 loss in the TCGA Pan-Cancer validation set,which also included patient samples frommany different diseasesites. Moreover, in contrast to existing RB1-loss signatures, whichincluded genes largely or exclusively based on prior annotations,the RBS gene set was selected in an unbiased, unsupervisedmanner. Our approach nominated genes from the set of allexisting genes that were most differentially expressed in ourpan-cancer, RB1-loss training set samples. A final methodologicalstrength of RBSwas that it was trained on a very large dataset (N¼995) including many samples with known RB1 loss (N ¼ 133)that could be collectively used to represent a distinct RB1-lossexpression pattern.

It is important to note that that the "accuracy" of our model forAUC analyses was defined as concordance between (RBS-based)RB1-loss calls and DNA-sequencing–based variant calls (i.e.,mutation, copy-number, and structural variant data when avail-able). This was because DNA-sequencing results are commonlyused to predict gene functional status and were the only dataavailable for comparison. However, DNA-sequencing calls do notcapture certain forms of gene inactivation such as DNA methyl-ation of the RB1 promoter. Although RBS demonstrated highconcordance with DNA-sequencing calls in our pan-cancer andmCRPC-specific analyses (AUCs of 0.89 and 0.87, respectively),the differences in RB1-loss assignments may not be due to errorbut rather improved identification of RB1 gene inactivation.

This study is not without limitations. We evaluated RBS as apotential tool to identify RB1 loss due to DNA-sequence altera-tions and DNAmethylation at the RB1 locus. However, still othermechanisms of RB1 inactivation exist, such as CDK phosphory-lation of the Rb protein (36, 37). It is unclear whether thesemechanisms of RB1 inactivation result in a similar pattern of geneexpression and whether RBS can be used to identify these Rb-inactivated tumors. Future work may involve collecting andintegrating phosphoproteomic data with DNA-sequencing andRNA-seq data to study these additional cases of tumors with RB1gene inactivation. In addition, because our analysis was con-ducted primarily using theCCLE and TCGAPan-Cancer databases(which focus on primary cancers), an extension to metastaticcancers is needed. In particular, as RB1 loss and RB1 underexpres-sion have been implicated as predictors of more advanced diseasein various cancers (38–40), future disease-specific studies with arange of indolent and aggressive tumors may leverage RBS tostudy RB1 loss in the context of disease progression.

The data presented here offer several novel insights and con-tributions. First, our study is the first to examine the clinicalimplications of RB1 loss on a pan-cancer scale. We foundthat RB1 loss was associated with shorter PFS, OS, and DSS,highlighting the widespread clinical importance of the genomicevent. Second, our novel transcriptomic signature (RBS) ishighly accurate at predicting RB1 loss and can be used as atool in future studies to shed new light on the biological andclinical impact of RB1 loss. This is especially relevant in light ofrecent studies which suggest that RB1 loss may associated with

response to various cancer therapies including radiothera-py (3, 41), platinum-based chemotherapy (3, 7), and CDK4/6 inhibitors (13, 15) in breast, prostate, and small-cell lungcancers. RBS may be useful for detecting differential response tospecific cancer therapies for an even broader range of therapiesand cancer types than has been already studied. Third, RBS isspecific to RB1-loss and not strongly correlated with cell pro-liferation scores (in contrast to existing RB1-loss signatures).Altogether, our study along with others suggest RB1 may haveimportant functions aside from regulating cell proliferation,such as DNA damage repair (41–43). Additional studies areneeded to assess this in greater detail. Fourth, our transcrip-tomic signature may be used to identify RB1-impaired tumorsthat may not be detected using standard DNA-sequencing–based definitions of predicted RB1 loss. The results of ourmultivariable analyses on two independent cohorts suggestthat both RNA-seq and DNA-sequencing results may be usefulto identify a more complete set of RB1-impaired patients.

Our approach to developing an RB1-loss signature is general-izable to studying a wide range of genomic alterations and mayserve as a paradigm for generating expression-based gene signa-tures in an unbiased manner. Because RBS is an expression-basedsignature, it is complementary to and potentially more holisticthan DNA-sequencing–based approaches, which may fail tocapture the full spectrum of genomic events that can result in aspecific gene expression profile or phenotype. Given the plethoraof studies highlighting RB1 loss as a driver event in a number ofcancer types, the potential clinical implications, and the increas-ing availability of gene expression data for both retrospective andprospective cohorts, RBS is an immediately useful tool that can beused to assessRB1 loss in a variety of settings.Our analyses and thefindings of others suggest that RB1 lossmay be predictive not onlyof survival but also of response to cytotoxic and targeted therapies.RBS may be invaluable for investigating these relationships fur-ther with the broader goal of developing personalized cancertreatment regimens.

Disclosure of Potential Conflicts of InterestS.G. Zhao and E. Davicioni hold ownership interest (including patents) in

GenomeDx Biosciences. C.A. Maher holds ownership interest (includingpatents) in Illumina. K.E. Knudsen reports receiving commercial researchgrants from Celgene and CellCentric, speakers bureau honoraria fromCelgene, and is a consultant/advisory board member for CellCentric. E.J.Small holds ownership interest (including patents) in Harpoon Therapeuticsand Fortis Therapeutics, and is a consultant/advisory board member forJanssen, Fortis Therapeutics, and Beigene. P.L. Nguyen reports receivingcommercial research grants from Janssen and Astellas, holds ownershipinterest (including patents) in Augmenix, and is a consultant/advisory boardmember for Augmenix, Boston Scientific, Ferring, Bayer, Dendreon, BlueEarth Diagnostics, Astellas, GenomeDx Biosciences, Nanobiotix, and Cota.F.Y. Feng reports receiving commercial research grants from Zenith, isa consultant/advisory board member for Bayer, Blue Earth Diagnostics,Celgene, Clovis, Janssen, EMD Serono, Sanofi, Dendreon, Ferring, andAstellas, and reports receiving other remuneration from PFS Genomics. Nopotential conflicts of interest were disclosed by the other authors.

Authors' ContributionsConception anddesign:W.S. Chen,M. Alshalalfa, S.G. Zhao, Y. Liu, B.A.Mahal,P.W. Kantoff, E.J. Small, P.L. Nguyen, F.Y. FengDevelopment of methodology: W.S. Chen, M. Alshalalfa, S.G. Zhao, Y. Liu,B.A. Mahal, F.Y. FengAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): W.S. Chen, M. Alshalalfa, Y. Liu, E. Davicioni,E.J. Small

Chen et al.

Clin Cancer Res; 2019 Clinical Cancer ResearchOF8

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

Analysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): W.S. Chen, M. Alshalalfa, S.G. Zhao, Y. Liu,D.A. Quigley, E. Davicioni, T.R. Rebbeck, C.A. Maher, K.E. Knudsen,P.L. Nguyen, F.Y. FengWriting, review, and/or revision of the manuscript:W.S. Chen, M. Alshalalfa,S.G. Zhao, Y. Liu, B.A. Mahal, D.A. Quigley, T. Wei, E. Davicioni, T.R. Rebbeck,P.W. Kantoff, C.A. Maher, K.E. Knudsen, E.J. Small, P.L. Nguyen, F.Y. FengAdministrative, technical, ormaterial support (i.e., reporting or organizingdata, constructing databases): W.S. Chen, E. Davicioni, T.R. Rebbeck,F.Y. FengStudy supervision: T.R. Rebbeck, E.J. Small, P.L. Nguyen, F.Y. Feng

AcknowledgmentsS.G. Zhao, B.A. Mahal, D.A. Quigley, E.J. Small, and F.Y. Feng are supported

by the Prostate Cancer Foundation.

The costs of publication of this articlewere defrayed inpart by the payment ofpage charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received January 31, 2019; revised March 27, 2019; accepted April 17, 2019;published first April 22, 2019.

References1. Ren W, Gu G. Prognostic implications of RB1 tumour suppressor gene

alterations in the clinical outcome of human osteosarcoma: a meta-analysis. Eur J Cancer Care 2017;26:e12401.

2. Pinyol M, Bea S, Pl�a L, Ribrag V, Bosq J, Rosenwald A, et al. Inactivation ofRB1 in mantle-cell lymphoma detected by nonsense-mediated mRNAdecay pathway inhibition and microarray analysis. Blood 2007;109:5422–9.

3. Robinson TJW, Liu JC, Vizeacoumar F, SunT,MacleanN, Egan SE, et al. RB1status in triple negative breast cancer cells dictates response to radiationtreatment and selective therapeutic drugs. PLoS One 2013;8:e78641.

4. Jones RA, Robinson TJ, Liu JC, Shrestha M, Voisin V, Ju Y, et al. RB1deficiency in triple-negative breast cancer induces mitochondrial proteintranslation. J Clin Invest 2016;126:3739–57.

5. Witkiewicz AK, Knudsen ES. Retinoblastoma tumor suppressor pathway inbreast cancer: prognosis, precision medicine, and therapeutic interven-tions. Breast Cancer Res 2014;16:207.

6. Meuwissen R, Linn SC, Linnoila RI, Zevenhoven J, Mooi WJ, Berns A.Induction of small cell lung cancer by somatic inactivation of both Trp53and Rb1 in a conditional mouse model. Cancer Cell 2003;4:181–9.

7. Dowlati A, Lipka MB, McColl K, Dabir S, Behtaj M, Kresak A, et al. Clinicalcorrelation of extensive-stage small-cell lung cancer genomics. Ann Oncol2016;27:642–7.

8. Ku SY, Rosario S, Wang Y, Mu P, Seshadri M, Goodrich ZW, et al. Rb1 andTrp53 cooperate to suppress prostate cancer lineage plasticity, metastasis,and antiandrogen resistance. Science 2017;355:78–83.

9. Mu P, Rosario S, Wang Y, Mu P, Seshadri M, Goodrich ZW, et al. SOX2promotes lineage plasticity and antiandrogen resistance in TP53- and RB1-deficient prostate cancer. Science 2017;355:84–8.

10. B€acklund LM, Nilsson BR, Goike HM, Schmidt EE, Liu L, Ichimura K, et al.Short postoperative survival for glioblastoma patients with a dysfunctionalRb1 pathway in combination with no wild-type PTEN. Clin Cancer Res2003;9:4151–8.

11. Seidel D, Zander T, Heukamp LC, Peifer M, Bos M, Fern�andez-Cuesta L,et al. Clinical Lung Cancer Genome Project (CLCGP) & Network GenomicMedicine (NGM). A genomics-based classification of human lung tumors.Sci Transl Med 2013;5:209ra153.

12. Hijioka S, HosodaW,Matsuo K, UenoM, FurukawaM, Yoshitomi H, et al.Rb loss and KRAS mutation are predictors of the response to platinum-based chemotherapy in pancreatic neuroendocrine neoplasmwith grade 3:a Japanesemulticenter pancreaticNEN-G3 study. ClinCancer Res 2017;23:4625–32.

13. Condorelli R, Spring L,O'Shaughnessy J, Lacroix L, Bailleux C, Scott V, et al.Polyclonal RB1mutations and acquired resistance toCDK4/6 inhibitors inpatients with metastatic breast cancer. Ann Oncol 2018;29:640–5.

14. Ertel A, Dean JL, Rui H, Liu C, Witkiewicz AK, Knudsen KE, et al. RB-pathway disruption in breast cancer: differential association with diseasesubtypes, disease-specific prognosis and therapeutic response. Cell Cycle2010;9:4153–63.

15. Malorni L, Piazza S, Ciani Y, Guarducci C, Bonechi M, Biagioni C, et al. Agene expression signature of retinoblastoma loss-of-function is a predictivebiomarker of resistance to palbociclib in breast cancer cell lines and isprognostic in patients with ER positive early breast cancer. Oncotarget2016;7:68012–22.

16. Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S,et al. The Cancer Cell Line Encyclopedia enables predictive modelling ofanticancer drug sensitivity. Nature 2012;483:603–7.

17. Mermel CH, Schumacher SE, Hill B, Meyerson ML, Beroukhim R, Getz G.GISTIC2.0 facilitates sensitive and confident localization of the targets offocal somatic copy-number alteration in human cancers. Genome Biol2011;12:R41.

18. Goldman M, Craft B, Kamath A, Brooks A, Zhu J, Haussler D, et al.The UCSC Xena Platform for cancer genomics data visualization andinterpretation. BioRxiv 326470 [Preprint]. 2018. Available from:https://www.biorxiv.org/content/10.1101/326470v1.

19. Tibshirani R, Hastie T, Narasimhan B, Chu G. Diagnosis of multiple cancertypes by shrunken centroids of gene expression. Proc Nat Acad Sci U S A2002;99:6567–72.

20. Naeem H, Wong NC, Chatterton Z, Hong MK, Pedersen JS,Corcoran NM, et al. Reducing the risk of false discovery enablingidentification of biologically significant genome-wide methylationstatus using the HumanMethylation450 array. BMC Genomics2014;15:51.

21. ZhaoM, Kim P,Mitra R, Zhao J, Zhao Z. TSGene 2.0: an updated literature-based knowledgebase for tumor suppressor genes. Nucleic Acids Res 2016;44:D1023–31.

22. Yang X, Han H, De Carvalho DD, Lay FD, Jones PA, Liang G. Gene bodymethylation can alter gene expression and is a therapeutic target in cancer.Cancer Cell 2014;26:577–90.

23. Durinck S, Spellman PT, Birney E, Huber W. Mapping identifiers for theintegrationof genomicdatasetswith theR/Bioconductor package biomaRt.Nat Protoc 2009;4:1184–91.

24. Liu J, Lichtenberg T, Hoadley KA, Poisson LM, Lazar AJ, CherniackAD, et al. An integrated TCGA pan-cancer clinical data resourceto drive high-quality survival outcome analytics. Cell 2018;173:400–16.e11.

25. Quigley DA, Dang HX, Zhao SG, Lloyd P, Aggarwal R, Alumkal JJ, et al.Genomic hallmarks and structural variation in metastatic prostate cancer.Cell 2018;174:758–69.e9.

26. L�opez-Rat�on M, Rodríguez-�Alvarez MX, Su�arez CC, Sampedro FG. Opti-malCutpoints: an R package for selecting optimal cutpoints in diagnostictests. J Stat Soft 2014;61. doi: 10.18637/jss.v061.i08.

27. Mizuarai S, Machida T, Kobayashi T, Komatani H, Itadani H, Kotani H.Expression ratio of CCND1 to CDKN2A mRNA predicts RB1 status ofcultured cancer cell lines and clinical tumor samples. Mol Cancer 2011;10:31.

28. Tsai H, Morais CL, Alshalalfa M, Tan H-L, Haddad Z, Hicks J, et al. CyclinD1 loss distinguishes prostatic small-cell carcinoma from most prostaticadenocarcinomas. Clin Cancer Res 2015;21:5619–29.

29. Cuzick J, Swanson GP, Fisher G, Brothman AR, Berney DM, Reid JE, et al.Prognostic value of an RNA expression signature derived from cell cycleproliferation genes in patients with prostate cancer: a retrospective study.Lancet Oncol 2011;12:245–55.

30. McNair C, Xu K, Mandigo AC, Benelli M, Leiby B, Rodrigues D, et al.Differential impact of RB status on E2F1 reprogramming in human cancer.J Clin Invest 2017;128:341–58.

31. Siu KT, RosnerMR,Minella AC. An integrated view of cyclin E function andregulation. Cell Cycle 2012;11:57–64.

32. Ishak CA, Marshall AE, Passos DT, White CR, Kim SJ, Cecchini MJ, et al. AnRB-EZH2 complex mediates silencing of repetitive DNA sequences.Mol Cell 2016;64:1074–87.

33. Chile T, Zanella Fortes MAH, Correa-Giannella MLC, Brentani HP, MariaDA, Puga RD, et al. HOXB7 mRNA is overexpressed in pancreatic ductal

www.aacrjournals.org Clin Cancer Res; 2019 OF9

Pan-Cancer RB1-Loss Signature Associated with Poor Survival

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

adenocarcinomas and its knockdown induces cell cycle arrest and apo-ptosis. BMC Cancer 2013;13:451.

34. Kent LN, Bae S, Tsai SY, Tang X, Srivastava A, Koivisto C, et al. Dosage-dependent copy number gains in E2f1 and E2f3 drive hepatocellularcarcinoma. J Clin Invest 2017;127:830–42.

35. Wyatt AW, Azad AA, Volik SV, Annala M, Beja K, McConeghy B, et al.Genomic alterations in cell-free DNA and enzalutamide resistance incastration-resistant prostate cancer. JAMA Oncol 2016;2:1598.

36. Rubin SM. Deciphering the retinoblastoma protein phosphorylation code.Trends Biochem Sci 2013;38:12–9.

37. Dyson NJ. RB1: a prototype tumor suppressor and an enigma. Genes Dev2016;30:1492–502.

38. Beltran H, Prandi D, Mosquera JM, Benelli M, Puca L, Cyrta J, et al.Divergent clonal evolution of castration-resistant neuroendocrine prostatecancer. Nature Medicine 2016;22:298–305.

39. Macleod KF. The RB tumor suppressor: a gatekeeper to hormone inde-pendence in prostate cancer? J Clin Invest 2010;120:4179–82.

40. Goldhoff P, Clarke J, Smirnov I, Berger MS, Prados MD, James CD, et al.Clinical stratification of glioblastoma based on alterations in retinoblas-toma tumor suppressor protein (RB1) and association with the proneuralsubtype. J Neuropathol Exp Neurol 2012;71:83–9.

41. Thangavel C, Boopathi E, Ciment S, Liu Y, O'Neill R, Sharma A, et al. Theretinoblastoma tumor suppressormodulatesDNA repair and radiorespon-siveness. Clin Cancer Res 2014;20:5468–82.

42. Cook R, Zoumpoulidou G, Luczynski MT, Rieger S, Moquet J, Spans-wick VJ, et al. Direct involvement of retinoblastoma family proteinsin DNA repair by non-homologous end-joining. Cell Rep 2015;10:2006–18.

43. Huang PH, Cook R, Mittnacht S. RB in DNA repair. Oncotarget 2015;6:20746–7.

Clin Cancer Res; 2019 Clinical Cancer ResearchOF10

Chen et al.

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404

Published OnlineFirst April 22, 2019.Clin Cancer Res   William S. Chen, Mohammed Alshalalfa, Shuang G. Zhao, et al.   Poor Clinical Outcomes across Cancer TypesNovel RB1-Loss Transcriptomic Signature Is Associated with

  Updated version

  10.1158/1078-0432.CCR-19-0404doi:

Access the most recent version of this article at:

  Material

Supplementary

http://clincancerres.aacrjournals.org/content/suppl/2019/04/20/1078-0432.CCR-19-0404.DC1Access the most recent supplemental material at:

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

  Subscriptions

Reprints and

  [email protected] at

To order reprints of this article or to subscribe to the journal, contact the AACR Publications

  Permissions

  Rightslink site. (CCC)Click on "Request Permissions" which will take you to the Copyright Clearance Center's

.http://clincancerres.aacrjournals.org/content/early/2019/05/22/1078-0432.CCR-19-0404To request permission to re-use all or part of this article, use this link

Research. on June 26, 2020. © 2019 American Association for Cancerclincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2019; DOI: 10.1158/1078-0432.CCR-19-0404