TCPA v3.0: An Integrative Platform to Explore the Pan-Cancer Analysis of Functional Proteomic Data

Reverse-phase protein arrays represent a powerful functional proteomics approach to characterizing cell signaling pathways and understanding their effects on cancer development. Using this platform, we have characterized ∼8,000 patient samples of 32 cancer types through The Cancer Genome Atlas and built a widely used, open-access bioinformatic resource, The Cancer Proteome Atlas (TCPA). To maximize the utility of TCPA, we have developed a new module called “TCGA Pan-Cancer Analysis,” which provides comprehensive protein-centric analyses that integrate protein expression data and other TCGA data across cancer types. We further demonstrate the value of this module by examining the correlations of RPPA proteins with significantly mutated genes, assessing the predictive power of somatic copy-number alterations, DNA methylation, and mRNA on protein expression, inferring the regulatory effects of miRNAs on protein expression, constructing a co-expression network of proteins and pathways, and identifying clinically relevant protein markers. This upgraded TCPA (v3.0) will provide the cancer research community with a more powerful tool for studying functional proteomics and making translational impacts. Reverse-phase protein arrays represent a powerful functional proteomics approach to characterizing cell signaling pathways and understanding their effects on cancer development. Using this platform, we have characterized ∼8,000 patient samples of 32 cancer types through The Cancer Genome Atlas and built a widely used, open-access bioinformatic resource, The Cancer Proteome Atlas (TCPA). To maximize the utility of TCPA, we have developed a new module called “TCGA Pan-Cancer Analysis,” which provides comprehensive protein-centric analyses that integrate protein expression data and other TCGA data across cancer types. We further demonstrate the value of this module by examining the correlations of RPPA proteins with significantly mutated genes, assessing the predictive power of somatic copy-number alterations, DNA methylation, and mRNA on protein expression, inferring the regulatory effects of miRNAs on protein expression, constructing a co-expression network of proteins and pathways, and identifying clinically relevant protein markers. This upgraded TCPA (v3.0) will provide the cancer research community with a more powerful tool for studying functional proteomics and making translational impacts. Functional proteomics is a powerful approach to characterizing cell signaling pathways and understanding their phenotypic effects on cancer development. Reverse-phase protein arrays (RPPAs) 1The abbreviations used are: RPPA, reverse-phase protein arrays; TCGA, The Cancer Genome Atlas; TCPA, The Cancer Proteome Atlas; FDR, false discovery rate; SCNAs, somatic copy-number alterations; SMGs, significantly mutated genes.1The abbreviations used are: RPPA, reverse-phase protein arrays; TCGA, The Cancer Genome Atlas; TCPA, The Cancer Proteome Atlas; FDR, false discovery rate; SCNAs, somatic copy-number alterations; SMGs, significantly mutated genes. represent a cutting-edge proteomics technology that can quantitatively assess a large number of protein markers in thousands of samples in a cost-effective, sensitive, and high-throughput manner (1.Sheehan K.M. Calvert V.S. Kay E.W. Lu Y. Fishman D. Espina V. Aquino J. Speer R. Araujo R. Mills G.B. Liotta L.A. Petricoin 3rd, E.F. Wulfkuhle J.D. Use of reverse phase protein microarrays and reference standard development for molecular network analysis of metastatic ovarian carcinoma.Mol. Cell. Proteomics. 2005; 4: 346-355Abstract Full Text Full Text PDF PubMed Scopus (269) Google Scholar, 2.Spurrier B. Ramalingam S. Nishizuka S. Reverse-phase protein lysate microarrays for cell signaling analysis.Nat. Protoc. 2008; 3: 1796-1808Crossref PubMed Scopus (154) Google Scholar, 3.Lu Y. Ling S. Hegde A.M. Byers L.A. Coombes K. Mills G.B. Akbani R. Using reverse-phase protein arrays as pharmacodynamic assays for functional proteomics, biomarker discovery, and drug development in cancer.Semin. Oncol. 2016; 43: 476-483Crossref PubMed Scopus (39) Google Scholar). Using the RPPA platform, we have characterized ∼8,000 patient samples across 32 cancer types through The Cancer Genome Atlas (TCGA) project and >650 independent cancer cell lines of 19 cell lineages (4.Li J. Lu Y. Akbani R. Ju Z. Roebuck P.L. Liu W. Yang J.Y. Broom B.M. Verhaak R.G. Kane D.W. Wakefield C. Weinstein J.N. Mills G.B. Liang H. TCPA: A resource for cancer functional proteomics data.Nat. Methods. 2013; 10: 1046-1047Crossref PubMed Scopus (314) Google Scholar, 5.Li J. Zhao W. Akbani R. Liu W. Ju Z. Ling S. Vellano C.P. Roebuck P. Yu Q. Eterovic A.K. Byers L.A. Davies M.A. Deng W. Gopal Y.N. Chen G. von Euw E.M. Slamon D. Conklin D. Heymach J.V. Gazdar A.F. Minna J.D. Myers J.N. Lu Y. Mills G.B. Liang H. Characterization of human cancer cell lines by reverse-phase protein arrays.Cancer Cell. 2017; 31: 225-239Abstract Full Text Full Text PDF PubMed Scopus (141) Google Scholar, 6.Hoadley K.A. Yau C. Hinoue T. Wolf D.M. Lazar A.J. Drill E. Shen R. Taylor A.M. Cherniack A.D. Thorsson V. Akbani R. Bowlby R. Wong C.K. Wiznerowicz M. Sanchez-Vega F. Robertson A.G. Schneider B.G. Lawrence M.S. Noushmehr H. Malta T.M. Cancer Genome Atlas N. Stuart J.M. Benz C.C. Laird P.W. Cell-of-origin patterns dominate the molecular classification of 10,000 tumors from 33 types of cancer.Cell. 2018; 173: 291-304.e296Abstract Full Text Full Text PDF PubMed Scopus (1050) Google Scholar). To better utilize the RPPA data and serve a broad biomedical research community, we developed an open-access bioinformatics resource, The Cancer Proteome Atlas (TCPA). This web-based platform not only releases the most updated data that are generated by our RPPA platform but also provides a user-friendly interface allowing users to analyze and visualize RPPA data in a rich context (7.Li J. Akbani R. Zhao W. Lu Y. Weinstein J.N. Mills G.B. Liang H. Explore, visualize, and analyze functional cancer proteomic data using the Cancer Proteome Atlas.Cancer Res. 2017; 77: e51-e54Crossref PubMed Scopus (60) Google Scholar), which has substantially reduced the computational barriers to analyzing complex RPPA data in large-scale sample sets. However, analytic modules in the previous TCPA versions have focused on the protein-based analyses in individual cancer types only and do not provide integrative analyses of RPPA data with other types of molecular data. Using those modules, it is difficult to explore the similarities and differences among cancer types, which limits the full potential of TCPA. To maximize its utility, we have updated TCPA by adding a newly developed module called “TCGA Pan-Cancer Analysis.” This module provides comprehensive protein-centric analyses that integrate association studies between protein data and other types of molecular and clinical data from TCGA. These data include somatic mutations, somatic copy-number alterations (SCNAs), DNA methylation, mRNA expression and miRNA expression, as well as patient survival, tumor subtype, and disease stage. With the new module, users can easily identify protein markers that show interesting patterns across cancer types. Overall, the current TCPA (v3.0) represents a comprehensive, cutting-edge, protein-centric pan-cancer analytic platform that is freely available at http://tcpaportal.org. The RPPA data were first quantified in the samples collected from TCGA project, which included 7,694 patient samples across 32 cancer types (6.Hoadley K.A. Yau C. Hinoue T. Wolf D.M. Lazar A.J. Drill E. Shen R. Taylor A.M. Cherniack A.D. Thorsson V. Akbani R. Bowlby R. Wong C.K. Wiznerowicz M. Sanchez-Vega F. Robertson A.G. Schneider B.G. Lawrence M.S. Noushmehr H. Malta T.M. Cancer Genome Atlas N. Stuart J.M. Benz C.C. Laird P.W. Cell-of-origin patterns dominate the molecular classification of 10,000 tumors from 33 types of cancer.Cell. 2018; 173: 291-304.e296Abstract Full Text Full Text PDF PubMed Scopus (1050) Google Scholar, 8.Tibes R. Qiu Y. Lu Y. Hennessy B. Andreeff M. Mills G.B. Kornblau S.M. Reverse phase protein array: Validation of a novel proteomic technology and utility for analysis of primary leukemia specimens and hematopoietic stem cells.Mol. Cancer Ther. 2006; 5: 2512-2521Crossref PubMed Scopus (522) Google Scholar, 9.Hennessy B.T. Lu Y. Gonzalez-Angulo A.M. Carey M.S. Myhre S. Ju Z. Davies M.A. Liu W. Coombes K. Meric-Bernstam F. Bedrosian I. McGahren M. Agarwal R. Zhang F. Overgaard J. Alsner J. Neve R.M. Kuo W.L. Gray J.W. Borresen-Dale A.L. Mills G.B. A technical assessment of the utility of reverse phase protein arrays for the study of the functional proteome in non-microdissected human breast cancers.Clin. Proteomics. 2010; 6: 129-151Crossref PubMed Scopus (177) Google Scholar, 10.Akbani R. Ng P.K. Werner H.M. Shahmoradgoli M. Zhang F. Ju Z. Liu W. Yang J.Y. Yoshihara K. Li J. Ling S. Seviour E.G. Ram P.T. Minna J.D. Diao L. Tong P. Heymach J.V. Hill S.M. Dondelinger F. Städler N. Byers L.A. Meric-Bernstam F. Weinstein J.N. Broom B.M. Verhaak R.G. Liang H. Mukherjee S. Lu Y. Mills G.B. A pan-cancer proteomic perspective on The Cancer Genome Atlas.Nat. Commun. 2014; 5: 3887Crossref PubMed Scopus (352) Google Scholar). In total, 258 protein markers (including total and phosphorylated proteins) were assayed. The raw RPPA data were further standardized and normalized by SuperCurve, median polish, and replicate-based normalization (10.Akbani R. Ng P.K. Werner H.M. Shahmoradgoli M. Zhang F. Ju Z. Liu W. Yang J.Y. Yoshihara K. Li J. Ling S. Seviour E.G. Ram P.T. Minna J.D. Diao L. Tong P. Heymach J.V. Hill S.M. Dondelinger F. Städler N. Byers L.A. Meric-Bernstam F. Weinstein J.N. Broom B.M. Verhaak R.G. Liang H. Mukherjee S. Lu Y. Mills G.B. A pan-cancer proteomic perspective on The Cancer Genome Atlas.Nat. Commun. 2014; 5: 3887Crossref PubMed Scopus (352) Google Scholar, 11.Ju Z. Liu W. Roebuck P.L. Siwak D.R. Zhang N. Lu Y. Davies M.A. Akbani R. Weinstein J.N. Mills G.B. Coombes K.R. Development of a robust classifier for quality control of reverse-phase protein arrays.Bioinformatics. 2015; 31: 912-918Crossref PubMed Scopus (33) Google Scholar) for downstream analyses. All the clinical and molecular data were obtained from the TCGA Pan-Cancer Atlas project consisting of (i) DNA data: somatic mutations, DNA methylation, and SCNAs; (ii) RNA data: RNA-seq and miRNA-seq-based gene expression data; and (iii) clinical data: stage, tumor grade, overall survival, and progression-free survival data. Specifically, SCNA data were downloaded from Synapse (https://www.synapse.org/) with accession number syn5049520.1. All the other data were retrieved from the TCGA Pan-Cancer Atlas website (https://gdc.cancer.gov/about-data/publications/pancanatlas). In addition, all the tumor subtype information were collected from the TCGA marker publications (12.Li J. Han L. Roebuck P. Diao L. Liu L. Yuan Y. Weinstein J.N. Liang H. TANRIC: An interactive open platform to explore the function of lncRNAs in cancer.Cancer Res. 2015; 75: 3728-3737Crossref PubMed Scopus (377) Google Scholar). Multiple statistical methods were used to examine pan-cancer associations between RPPA data and other types of molecular and clinical data: (i) for continuous data (including DNA methylation, SCNA, mRNA expression, miRNA expression, RPPA, and pathway scores), Spearman's rank correlation analyses were performed; (ii) for dichotomous variables (e.g. wild type versus mutated), t-tests were performed; and (iii) ANOVA tests were applied to those with more than two levels (e.g. tumor subtype, stage, and grade). For survival analysis, both Cox proportional-hazards model and log-rank test were used to calculate the correlations between the levels of protein markers and patient survival times. All the p values derived from multiple comparisons were adjusted by false discovery rate (FDR). We reported significant results at FDR < 0.1. Eleven core cancer pathway scores were calculated based on (signed) average of the RPPA levels for each cancer type before being applied in the network analysis (13.Wang Y. Xu X. Maglic D. Dill M.T. Mojumdar K. Ng P.K. Jeong K.J. Tsang Y.H. Moreno D. Bhavana V.H. Peng X. Ge Z. Chen H. Li J. Chen Z. Zhang H. Han L. Du D. Creighton C.J. Mills G.B. Cancer Genome Atlas Research Network Camargo F. Liang H. Comprehensive molecular characterization of the Hippo signaling pathway in cancer.Cell Rep. 2018; 25: 1304-1317.e5Abstract Full Text Full Text PDF PubMed Scopus (190) Google Scholar). The protein members and their functional directions (i.e. up- or down-regulation) within each pathway were defined in Akbani et al. (10.Akbani R. Ng P.K. Werner H.M. Shahmoradgoli M. Zhang F. Ju Z. Liu W. Yang J.Y. Yoshihara K. Li J. Ling S. Seviour E.G. Ram P.T. Minna J.D. Diao L. Tong P. Heymach J.V. Hill S.M. Dondelinger F. Städler N. Byers L.A. Meric-Bernstam F. Weinstein J.N. Broom B.M. Verhaak R.G. Liang H. Mukherjee S. Lu Y. Mills G.B. A pan-cancer proteomic perspective on The Cancer Genome Atlas.Nat. Commun. 2014; 5: 3887Crossref PubMed Scopus (352) Google Scholar). The 11 pathways included were apoptosis, core reactive, cell cycle, DNA damage response, epithelial-mesenchymal transition, hormone a, hormone b, PI3K/Akt, Ras/MAPK, RTK, and TSC/mTOR. To show the correlations between protein levels and mutated genes, a bipartite network was generated based on the of significantly mutated and the 11 pathway In total, were obtained from et al. C. E. S. D. J. B. Ng P. Jeong K.J. S. Z. J. Q. F. Liu E.M. L. C. N. I. Liang J.M. D. C. J.Y. E. E.M. Liang H. Cancer Genome Atlas Research Network Lawrence M.S. N. Stuart J. D. G. Chen K. Lazar A.J. Mills G.B. R. L. Comprehensive characterization of cancer and 2018; Full Text Full Text PDF PubMed Scopus Google Scholar), which were used to the mutated and sample in each cancer The correlations between pathways and were by t-tests with FDR < 0.1. In the bipartite the of each represents the number of cancer types with a significant correlation < and the of each represents the of The total proteins were first from the RPPA data for the correlation total in SCNA, DNA methylation, and mRNA data. multiple were to the the the most correlation with the mRNA expression was The were and were For each cancer Spearman's rank correlation analysis was to assess both the and All the p values were adjusted by The correlation of and were in with The between and were by We first obtained miRNA from V. J.W. in 2015; 4: PubMed Scopus Google Scholar). total proteins were used in this the were to the which generated miRNAs and genes. The miRNAs with at were included in the further analysis miRNAs For each Spearman's rank correlations were To the miRNA effects in cancer types, the between the of and correlations were by the The of the was calculated to the or in a cancer The miRNAs were by the number of cancer types and only those with directions in at two cancer types were in the A network was generated to show the co-expression among protein markers and RPPA In the each represents a and each represents the number of cancer types in which the Spearman's rank correlations between two were significant < For the with in cancer types were in the The were further by the in R. All the analytic results were first by statistical analyses. The with both molecular and clinical were further to and The pan-cancer analyses were in R. The interface of the pan-cancer analytic module was in All the results were by and all the were generated by we TCPA a interactive platform with a newly protein-centric analytic module for to visualize and analyze RPPA data from a pan-cancer perspective This new RPPA pan-cancer data with 258 protein markers quantified from 7,694 TCGA patient samples across 32 cancer types primary tumors and 258 metastatic The protein markers both total and phosphorylated proteins) all signaling pathways relevant to human as PI3K/Akt, Ras/MAPK, and Hippo signaling To maximize the utility of the RPPA we collected other types of molecular and clinical data from TCGA Pan-Cancer Atlas to analyses. The pan-cancer analytic module is which users to examine the associations between protein expression and all other types of clinical and molecular (i.e. and protein The pan-cancer analytic module of (i) the protein expression with patient clinical patient survival, tumor subtype, stage, and (ii) the associations between protein expression and gene SCNAs, and DNA (iii) In the the correlations between protein expression and expression are and the protein All the analytic results can by interactive can easily and or in a the results of all the cancer types are a it is for users to test their pan-cancer In addition, each in the has a the data which can easily in or PDF For and are generated to visualize survival analysis, and Specifically, with the users can examine a protein marker with patient overall or progression-free survival by Cox proportional-hazards model and log-rank through a or a protein expression among tumor or by through This can potential markers across cancer types. For with protein expression show better overall survival than those with expression in cell and it better progression-free survival in cell The users to examine the expression of a protein marker is by somatic in a gene by test among the mutated and sample through or its expression with DNA and by Spearman's rank through For mutated to significantly protein expression in breast and The users to examine a protein marker correlations with RNA of its or other genes, or miRNA as by Spearman's rank through For and have a correlation in breast and other cancer types, for and the provides an for users to examine the expression levels of two protein markers with each other across cancer types by Spearman's rank through For each associations are in a The first is the cancer types for the by the protein markers and their relevant clinical or molecular and the and p All significant results can downloaded through the To demonstrate the utility of this newly developed pan-cancer analytic module, we further the significant of the pan-cancer results in TCPA To assess the associations between protein expression and somatic mutations, we focused on in a study C. E. S. D. J. B. Ng P. Jeong K.J. S. Z. J. Q. F. Liu E.M. L. C. N. I. Liang J.M. D. C. J.Y. E. E.M. Liang H. Cancer Genome Atlas Research Network Lawrence M.S. N. Stuart J. D. G. Chen K. Lazar A.J. Mills G.B. R. L. Comprehensive characterization of cancer and 2018; Full Text Full Text PDF PubMed Scopus Google Scholar) are cancer and the sample of their mutated are large for statistical we were with the levels of their in was the most protein between and mutated patient samples across cancer types, a significant in 11 cancer types. in all the 11 cancer types, protein expression was in the mutated a potential in and expression in the mutated which to RNA C. Lu J. Xu Z. in the gene in the of protein and protein PubMed Scopus Google Scholar). we pathways were with the of in the correlations among all the of the pathways were with in cancer types, that as a in signaling of cancer Specifically, DNA damage response, apoptosis, and cell were in multiple cancer types in mutated all the pathways, was by the number of total SMGs, as and across cancer types. all were with of These results provide an overall of potential effects of somatic on their protein and pathway expression is by a of and SCNAs, DNA methylation, and mRNA expression are to with protein we a analysis to which most to protein in all the correlations in significantly from test p values < Specifically, mRNA expression correlations with protein to the average from our previous TCGA pan-cancer by Akbani et al. (10.Akbani R. Ng P.K. Werner H.M. Shahmoradgoli M. Zhang F. Ju Z. Liu W. Yang J.Y. Yoshihara K. Li J. Ling S. Seviour E.G. Ram P.T. Minna J.D. Diao L. Tong P. Heymach J.V. Hill S.M. Dondelinger F. Städler N. Byers L.A. Meric-Bernstam F. Weinstein J.N. Broom B.M. Verhaak R.G. Liang H. Mukherjee S. Lu Y. Mills G.B. A pan-cancer proteomic perspective on The Cancer Genome Atlas.Nat. Commun. 2014; 5: 3887Crossref PubMed Scopus (352) Google also effects but to a than mRNA and DNA more correlations with their We patterns within cancer types, the correlation from protein to protein We to identify the for the protein a for the samples with all the data types we collected all the significant correlations < and and a from SCNA, methylation, or mRNA as the its correlation the other in the expression of most proteins can by mRNA in all cancer types, for and for a number of the protein was better by DNA as cell tumors or as which to potential of gene expression for those with or These results assess the of on protein To better the miRNA on protein expression, we an integrative analysis to their correlations in cancer in by the correlations of versus genes, we 19 miRNAs as potential across cancer types by using the approach The two miRNAs is cancer which has reported to function as an in cell L. Li X. Y. J. Zhao L. Xu J. Xu W. X. Li H. Yang S. Y. Y. of as an and biomarker in J. Res. 2018; 10: Google Scholar) and cancer C. Zhang X. H.M. Liu Zhang B. as an by in cancer.Cell 2017; PubMed Scopus Google Scholar) and cancer a tumor in cancer J. F. Y. J. C. Liang Y. R. T. S. T. C. Yang G. Liu X. M. Y. Liu Y. F. Y. Zhang B. Liu Y. X. Zhang J. Liu L. by and of PubMed Scopus Google Scholar). These results their regulatory in cell cell and To the miRNA we the regulatory across cancer types and using those with a significantly Spearman's rank correlation < The to have a on and significant effects on protein and miRNAs can gene expression through mRNA or translational and regulatory Full Text Full Text PDF PubMed Scopus Google Scholar), we which regulatory miRNAs by correlation in only proteins but not significantly a of translational is the of The results that the RPPA data provide information to identify miRNA and the regulatory To better the functional associations among we first calculated pathway scores and built a co-expression network based on the correlations between pathways and proteins The network of two (i) the network pathways are among cancer types, in which each represents a pathway and each represents the of significant correlations in cancer (ii) the network the between protein members in each in which each is a protein marker with the pathway and each represents the of significant the pathways, we two of PI3K/Akt, RTK, and pathways, all correlations with each other in at cancer types, and the other of apoptosis, epithelial-mesenchymal transition, and core with all correlations in at cancer types. The patterns the of two were by consisting of apoptosis, and was with the other two These results a of among pathway members and potential between pathways in we the clinical of RPPA protein markers by assessing their correlations with patient survival tumor subtype, and disease across cancer types. in we that protein markers with overall patient survival in at cancer types < log-rank test or Cox proportional-hazards the protein the marker a and significant in cancer types expression with patient survival with the reported patterns in and and cancer types B. N. L. S. and of and in breast cancer 2014; PubMed Scopus Google Scholar, S. L. D. C. S. S. L. E. S. S.M. S. M. P. E. of Commun. 2017; PubMed Scopus Google Scholar, W. E. R. P. Yang Y. expression of is with cell and the signaling pathway in Rep. 2016; PubMed Scopus Google Scholar), results that serve as a robust marker across cancer types. for correlations with tumor we that of the 258 protein markers were among the tumor (e.g. for breast the power of RPPA protein markers in tumor within a cancer type we proteins the expression patterns with tumor in a manner expression with tumor their in tumor results demonstrate the translational potential of protein markers in and tumor analyses using TCGA data have to identify and clinically we TCPA as a interactive platform to explore and analyze TCGA pan-cancer protein expression data. The new pan-cancer analytic module provides a for biomedical to test their across a broad of cancer types. on the analytic results obtained from this new module, we have molecular and clinical that show significant associations with protein markers in cancer These further demonstrate the utility of the pan-cancer analytic module by users novel and We that the current pan-cancer RPPA data only protein which limits the of using functional proteomics to For for only a number of miRNA were We are in the of the to proteins all cancer We this new of TCPA to a bioinformatic resource for the cancer research All the raw data are available at TCGA Pan-Cancer Atlas website (https://gdc.cancer.gov/about-data/publications/pancanatlas). The data and results are also available at TCPA website We from the TCGA Research