Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Proteogenomics connects somatic mutations to signalling in breast cancer


Somatic mutations have been extensively characterized in breast cancer, but the effects of these genetic alterations on the proteomic landscape remain poorly understood. Here we describe quantitative mass-spectrometry-based proteomic and phosphoproteomic analyses of 105 genomically annotated breast cancers, of which 77 provided high-quality data. Integrated analyses provided insights into the somatic cancer genome including the consequences of chromosomal loss, such as the 5q deletion characteristic of basal-like breast cancer. Interrogation of the 5q trans-effects against the Library of Integrated Network-based Cellular Signatures, connected loss of CETN3 and SKP1 to elevated expression of epidermal growth factor receptor (EGFR), and SKP1 loss also to increased SRC tyrosine kinase. Global proteomic data confirmed a stromal-enriched group of proteins in addition to basal and luminal clusters, and pathway analysis of the phosphoproteome identified a G-protein-coupled receptor cluster that was not readily identified at the mRNA level. In addition to ERBB2, other amplicon-associated highly phosphorylated kinases were identified, including CDK12, PAK1, PTK2, RIPK2 and TLK2. We demonstrate that proteogenomic analysis of breast cancer elucidates the functional consequences of somatic mutations, narrows candidate nominations for driver genes within large deletions and amplified regions, and identifies therapeutic targets.

Your institute does not have access to this article

Relevant articles

Open Access articles citing this article.

Access options

Buy article

Get time limited or full article access on ReadCube.


All prices are NET prices.

Figure 1: Direct effects of genomic alterations on protein level.
Figure 2: Effects of CNAs on mRNA, protein, and phosphoprotein abundance.
Figure 3: Proteomic and phosphoproteomic subtypes of breast cancer and subtype-specific pathway enrichment.
Figure 4: Example analyses of aberrantly regulated kinases in human breast cancer.


  1. Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature 490, 61–70 (2012)

  2. Curtis, C. et al. The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature 486, 346–352 (2012)

    CAS  Article  Google Scholar 

  3. van ’t Veer, L. J. et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530–536 (2002)

    Article  Google Scholar 

  4. Chin, K. et al. Genomic and transcriptional aberrations linked to breast cancer pathophysiologies. Cancer Cell 10, 529–541 (2006)

    CAS  MathSciNet  Article  Google Scholar 

  5. Ellis, M. J. et al. Connecting genomic alterations to cancer biology with proteomics: the NCI Clinical Proteomic Tumor Analysis Consortium. Cancer Discov . 3, 1108–1112 (2013)

    CAS  Article  Google Scholar 

  6. Zhang, B. et al. Proteogenomic characterization of human colon and rectal cancer. Nature 513, 382–387 (2014)

    CAS  Article  Google Scholar 

  7. Perou, C. M. et al. Molecular portraits of human breast tumours. Nature 406, 747–752 (2000)

    CAS  ADS  Article  Google Scholar 

  8. Sorlie, T. et al. Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc. Natl Acad. Sci. USA 100, 8418–8423 (2003)

    CAS  ADS  Article  Google Scholar 

  9. Parker, J. S. et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J. Clin. Oncol. 27, 1160–1167 (2009)

    Article  Google Scholar 

  10. Li, S. et al. Endocrine-therapy-resistant ESR1 variants revealed by genomic characterization of breast-cancer-derived xenografts. Cell Reports 4, 1116–1130 (2013)

    CAS  ADS  Article  Google Scholar 

  11. Polyak, K. Heterogeneity in breast cancer. J. Clin. Invest. 121, 3786–3788 (2011)

    CAS  Article  Google Scholar 

  12. Bertos, N. R. & Park, M. Breast cancer — one term, many entities? J. Clin. Invest. 121, 3789–3796 (2011)

    CAS  Article  Google Scholar 

  13. Symmans, W. F., Liu, J., Knowles, D. M. & Inghirami, G. Breast cancer heterogeneity: evaluation of clonality in primary and metastatic lesions. Hum. Pathol. 26, 210–216 (1995)

    CAS  Article  Google Scholar 

  14. Yoshihara, K. et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nat. Commun. 4, 2612 (2013)

    ADS  Article  Google Scholar 

  15. Mertins, P. et al. Ischemia in tumors induces early and sustained phosphorylation changes in stress kinase pathways but does not affect global protein levels. Mol. Cell Proteomics 13, 1690–1704 (2014)

    CAS  Article  Google Scholar 

  16. Ruggles, K. V. et al. An analysis of the sensitivity of proteogenomic mapping of somatic mutations and novel splicing events in cancer. Mol. Cell Proteomics 15, 1060–1071 (2015)

    Article  Google Scholar 

  17. Scheffner, M., Huibregtse, J. M., Vierstra, R. D. & Howley, P. M. The HPV-16 E6 and E6-AP complex functions as a ubiquitin-protein ligase in the ubiquitination of p53. Cell 75, 495–505 (1993)

    CAS  Article  Google Scholar 

  18. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005)

    CAS  ADS  Article  Google Scholar 

  19. Silva, G. O. et al. Cross-species DNA copy number analyses identifies multiple 1q21-q23 subtype-specific driver genes for breast cancer. Breast Cancer Res. Treat. 152, 347–356 (2015)

    CAS  Article  Google Scholar 

  20. Lamb, J. et al. The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science 313, 1929–1935 (2006)

    CAS  ADS  Article  Google Scholar 

  21. Peck, D. et al. A method for high-throughput gene expression signature analysis. Genome Biol. 7, R61 (2006)

    Article  Google Scholar 

  22. Duan, Q. et al. LINCS Canvas Browser: interactive web app to query, browse and interrogate LINCS L1000 gene expression signatures. Nucleic Acids Res. 42, W449–W660 (2014)

    CAS  Article  Google Scholar 

  23. Nakayama, K. I. & Nakayama, K. Ubiquitin ligases: cell-cycle control and cancer. Nat. Rev. Cancer 6, 369–381 (2006)

    CAS  Article  Google Scholar 

  24. Hein, M. Y. et al. A human interactome in three quantitative dimensions organized by stoichiometries and abundances. Cell 163, 712–723 (2015)

    CAS  Article  Google Scholar 

  25. Petralia, F., Song, W. M., Tu, Z. & Wang, P. New method for joint network analysis reveals common and different coexpression patterns among genes and proteins in breast cancer. J. Proteome Res. 15, 743–754 (2016)

    CAS  Article  Google Scholar 

  26. Loi, S. et al. PIK3CA mutations associated with gene signature of low mTORC1 signaling and better outcomes in estrogen receptor-positive breast cancer. Proc. Natl Acad. Sci. USA 107, 10208–10213 (2010)

    CAS  ADS  Article  Google Scholar 

  27. Vasudevan, K. M. et al. AKT-independent signaling downstream of oncogenic PIK3CA mutations in human cancer. Cancer Cell 16, 21–32 (2009)

    CAS  Article  Google Scholar 

  28. Wu, X. et al. Activation of diverse signalling pathways by oncogenic PIK3CA mutations. Nat. Commun. 5, 4961 (2014)

    CAS  ADS  Article  Google Scholar 

  29. Krzywinski, M. et al. Circos: an information aesthetic for comparative genomics. Genome Res. 19, 1639–1645 (2009)

    CAS  Article  Google Scholar 

  30. Blazek, D. et al. The Cyclin K/Cdk12 complex maintains genomic stability via regulation of expression of DNA damage response genes. Genes Dev. 25, 2158–2172 (2011)

    CAS  Article  Google Scholar 

  31. Shrestha, Y. et al. PAK1 is a breast cancer oncogene that coordinately activates MAPK and MET signaling. Oncogene 31, 3397–3408 (2012)

    CAS  Article  Google Scholar 

  32. Chen, Y. et al. Identification of druggable cancer driver genes amplified across TCGA datasets. PLoS One 9, e98293 (2014)

    ADS  Article  Google Scholar 

  33. Prudnikova, T. Y., Rawat, S. J. & Chernoff, J. Molecular pathways: targeting the kinase effectors of RHO-family GTPases. Clin. Cancer Res. 21, 24–29 (2015)

    CAS  Article  Google Scholar 

  34. Jiang, W. et al. Differential phosphorylation of DNA-PKcs regulates the interplay between end-processing and end-ligation during nonhomologous end-joining. Mol. Cell 58, 172–185 (2015)

    CAS  Article  Google Scholar 

  35. Agrawal, P. B. et al. SPEG interacts with myotubularin, and its deficiency causes centronuclear myopathy with dilated cardiomyopathy. Am. J. Hum. Genet. 95, 218–226 (2014)

    CAS  Article  Google Scholar 

  36. Borges, S. et al. Effective Targeting of estrogen receptor-negative breast cancers with the protein kinase D inhibitor CRT0066101. Mol. Cancer Ther. 14, 1306–1316 (2015)

    CAS  Article  Google Scholar 

  37. Walkinshaw, D. R. et al. The tumor suppressor kinase LKB1 activates the downstream kinases SIK2 and SIK3 to stimulate nuclear export of class IIa histone deacetylases. J. Biol. Chem. 288, 9345–9362 (2013)

    CAS  Article  Google Scholar 

  38. Jiang, X. et al. Numb regulates glioma stem cell fate and growth by altering epidermal growth factor receptor and Skp1-Cullin-F-box ubiquitin ligase activity. Stem Cells 30, 1313–1326 (2012)

    CAS  Article  Google Scholar 

  39. Carey, L. A. et al. TBCRC 001: randomized phase II study of cetuximab in combination with carboplatin in stage IV triple-negative breast cancer. J. Clin. Oncol. 30, 2615–2623 (2012)

    CAS  Article  Google Scholar 

  40. Ong, C. C. et al. Small molecule inhibition of group I p21-activated kinases in breast cancer induces apoptosis and potentiates the activity of microtubule stabilizing agents. Breast Cancer Res. 17, 59 (2015)

    Article  Google Scholar 

  41. Carr, S. A. et al. Targeted peptide measurements in biology and medicine: best practices for mass spectrometry-based assay development using a fit-for-purpose approach. Mol. Cell Proteomics 13, 907–917 (2014)

    CAS  Article  Google Scholar 

Download references


This work was supported by National Cancer Institute (NCI) CPTAC awards U24CA160034 (Broad Institute; Fred Hutchinson Cancer Research Center), U24CA160036 (Johns Hopkins University), U24CA160019 (Pacific Northwest National Laboratory), U24CA159988 (Vanderbilt University), U24CA160035 (Washington University, St. Louis; University of North Carolina, Chapel Hill). P.W. and F.P. were also supported by SUB-R01GM108711 and MJE by CPRIT grant RR140033. M.J.E. is also a McNair Foundation Scholar. D.F. was supported by Leidos contract 13XS068. Primary genomics data for this study were generated by The Cancer Genome Atlas pilot project established by the NCI and the National Human Genome Research Institute. Resequencing of select samples conducted in this study was supported by National Cancer Institute (NCI) CPTAC award U24CA160035. Information about TCGA and the investigators and institutions that constitute the TCGA research network can be found at We also acknowledge the expert assistance of J. Snider, P. Erdmann-Gilmore and R. Connors for the preparation of the tumour tissues for solubilization. We thank the Alvin J. Siteman Cancer Center at Washington University School of Medicine and Barnes-Jewish Hospital in St. Louis, for the use of the Tissue Procurement Core, which provided accessioning, histologic processing and review for the TCGA samples included in this study. The Siteman Cancer Center is supported in part by an NCI Cancer Center Support Grant #P30 CA91842 (see more at We also thank the HAMLET Core at The Washington University in St. Louis for providing breast cancer xenograft tumors. The HAMLET Core was supported in part by grants from NIH/NCRR Washington University-ICTS (UL1 RR024992) and Susan G. Komen for the Cure (KG 090422). F.M. was also supported by The Swedish Research Council (Dnr 2014-323). We also thank A. Subramanian, C. Flynn and J. Asiedu at the Broad Institute for their guidance and assistance in accessing LINCS to run a large number of enrichment queries.

Author information

Authors and Affiliations




P.M., D.R.M., M.A.G., K.R.C., and S.A.C. designed the proteomic analysis experiments, data analysis workflow, and proteomic-genomic data comparisons. P.M., M.A.G., J.W.Q., and S.A.C. directed and performed proteomic analysis of breast tumour and quality control samples. P.M., D.R.M., K.V.R., K.R.C., P.W., X.W., S.C., E.K., F.P., Z.T., J.T.L., M.L.G., M.W., V.Y., K.H., C.L., M.D.M., P.Y., J.W., B.Z., and D.F. performed proteomic-genomic data analyses. D.R.M., P.W., and S.J.S. provided statistical support. D.R.M., K.V.R., K.R.C., K.K. and D.F. performed analyses of mass spectrometry data and adapted algorithms and software for data analysis. S.R.D., R.R.T and M.J.E. developed and prepared breast xenografts used as quality control samples. P.M. and F.M. prepared and analyzed cell lines for correlative functional annotation of frequently mutated genes. P.M., D.R.M., M.A.G., and S.A.C designed strategy for quality control analyses. M.A.G., S.R.D., C.R.K., M.M., and H.R. coordinated acquisition, distribution and quality control evaluation of TCGA tumour samples. P.M., M.A.G., C.M.P., L.D., A.G.P., and M.J.E. interpreted data in the context of breast cancer biology. P.M., D.R.M, M.A.G., K.R.C., P.W., A.G.P, M.J.E. and S.A.C. wrote the manuscript.

Corresponding authors

Correspondence to Philipp Mertins, Matthew J. Ellis or Steven A. Carr.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Additional information

All primary mass spectrometry data are deposited at the CPTAC Data Portal as raw and mzML files and complete protein assembly data sets for public access ( In addition, a set of ancillary files such as dataset G1/P1, G3/P3, G4/P4, G5/P5, G7/P7, CNA correlation tables for CNA–mRNA, CNA–proteome and CNA–phosphoproteome, CNA data, and RNA-seq expression data have also been deposited at the CPTAC Data Coordinating Center (DCC). Two browsers for the results: one provides track hubs for viewing the identified peptides in the UCSD genome browser (; the other is an online tool for proteogenomic data exploration, accessed at (see Supplementary Methods for descriptions).

A list of participants and their affiliations appears in the Supplementary Information..

Extended data figures and tables

Extended Data Figure 1 Experimental and data analysis workflows and longitudinal data generation quality control.

a, iTRAQ 4-plex global proteome and phosphoproteome analysis workflow. 105 TCGA breast tumours were analysed in 35 iTRAQ 4-plex experiments (plus one replicate and one normal sample experiment), with three tumours of different subtypes compared to a fourth common internal reference sample in each experiment. The reference sample comprised 10 individual tumours of each of the four major breast cancer intrinsic subtypes and served as an internal standard for all proteins and phosphoproteins quantified in this study. Each iTRAQ MS/MS spectrum measures a peptide from four samples (3 individual patients and the reference sample mix of 40 patients). More than 400,000 distinct peptides were identified and quantified in ~14 million MS/MS spectra. Personalized tumour-specific protein databases were generated in the QUILTS software package using whole-exome-sequencing-derived variant calls and RNA-seq-derived transcript information. All mass spectrometry data was analysed using the Spectrum Mill software package. b, Overview of proteome and phosphoproteome data sets. The table provides a summary of the data sets used in specific analyses, including the filters applied to derive the proteins and phosphosites/phosphoproteins that constitute each data set; the protein, phosphosite or phosphoprotein count; and the methods that employ the respective data sets. c, Distribution of sequence coverage of the identified proteins with tryptic peptides detected by MS/MS, whiskers show the 5–95 percentiles. d, e, Robust and accurate proteome/phosphoproteome platform. Longitudinal performance was tested by repeated proteome and phosphoproteome analysis of patient-derived xenograft tumours. Scatter plots, histograms and Pearson correlations comparing individual replicate measurements are shown.

Extended Data Figure 2 Tumour sample quality control.

a, Remark diagram showing sample processing and partitioning. Initial quality review encompassed histopathological examination of tissue slices stained with haematoxylin and eosin. *For 3 samples, no tumour cells were seen on histopathology (BH-A0E9, BH-A0C1, A2-A0SW). These samples were nevertheless included in the proteome analysis as other quality control standards were met (see below) and samples with 0% tumour cellularity on top or bottom sections were included in TCGA analyses. b, Correlation of TCGA (top or bottom sections) and CPTAC histological assessment of neoplastic cellularity for samples (n = 105). The average and range of neoplastic cellularities were identical for CPTAC and TCGA histological assessments. Averages (s.d.) for neoplastic cellularity were 76% (±17) for CPTAC, 76% (±15) for TCGA_Top, and 75% (±18) for TCGA_Bottom histopathology slides (Supplementary Table 2). Note that in three CPTAC cases where no tumour cells were identified by histopathological assessment, numbers of protein-level somatic variants were similar to all other tumours. The identified mutated proteins were TP53_R273C, NOP58_Q23E, TAGLN2_G154R, TUBA1B_D116H, and MRPL48_I173K (Supplementary Table 5), indicating presence of tumour cells in these samples. c, Proteome iTRAQ tumour to internal reference ratio heat map for all CPTAC samples (8,028 proteins without missing values) including passed and failed proteomic quality control (QC) samples. d, Global tumour to reference proteome ratio distributions for samples that passed and failed proteomic quality control analysis. e, Degradation-related gene sets were enriched in tumours that failed proteomic quality control analysis. f, Variant allele frequency (VAF) analysis of re-sequenced CPTAC tumours and comparison to original TCGA data. Overall VAFs for failed quality control samples were lower compared to passed samples suggesting lower purity.

Extended Data Figure 3 Tumour sample quality control.

a, There was high concordance (94.6%) between DNA variants reported by TCGA and CPTAC re-sequenced tumours. Most point mutations reported by TCGA could be identified across the eight re-sequenced samples used in the study. b, A high overall correlation (mean = 0.77) was observed for the CPTAC VAF (x axis) and TCGA VAF (y axis) across the eight samples used in the study. c, Agglomerative hierarchical clustering (Supplementary Methods section 3.8) used to co-cluster protein and RNA tumour expression data after filtering to retain 4,291 proteins and genes with moderate to high protein–RNA correlation (Pearson correlation > 0.4) with results displayed as a circular dendrogram (fanplot). The proteome (.P) and RNA (.R) components of each sample are labelled using the same colour. The outer ring shows proteome samples in light grey and RNA samples in dark grey. High concordance between RNA and protein expression is evident from the colour adjacency in the inner ring and alternating colour in the outer ring showing that RNA and protein components co-cluster for a large proportion of samples (62 out of 80). d, Co-clustering of MS/MS and RPPA tumour data. 126 RPPA readouts were mapped to gene names. These genes were intersected with the genes observed in the MS/MS proteome, filtered to 48 proteins with moderate or higher RPPA–MS/MS protein correlation, and analysed for co-clustering as in c. 47 of 80 RPPA–MS/MS protein pairs co-cluster. Although this is a smaller proportion than for RNA-protein analysis, the number of genes used in the clustering is significantly smaller for RPPA (48 versus 4,291 for RNA). e, ESTIMATE tumour purity comparison between mRNA, RNA-seq, and proteome data. ANOVA is used to assess the difference in distribution (−log10(P value)) of ESTIMATE, stromal, immune, and tumour purity scores across mRNA (microarray), RNA-seq and proteome data. The only significant P value (0.02) is for the cluster 3 stromal score, and higher stromal scores for the proteome drive that difference. f, Ischaemia score analysis. Comparison of ischaemia scores of 77 CPTAC tumours, 3 normal samples, and patient-derived xenografts. CPTAC tumours had generally lower ischaemia scores than PDX samples subjected to 30 min of cold ischaemia. Median ischaemia scores are less than 30 min for each subtype and no significant differences were observed across subtypes. Effects due to cold ischaemia therefore appear to be negligible in this CPTAC sample collection.

Extended Data Figure 4 Protein–protein, protein–CNA, and protein–mRNA correlation analyses.

a, Identification of UBE3A as an E3 ubiquitin ligase that negatively correlates to p53 on the protein level. Pearson correlation and Benjamini–Hochberg-corrected P value are shown. b, Analysis of counter-regulated genes with negative correlation of CNA–RNA as well as CNA–protein levels. Negative Pearson correlations are shown with Benjamini–Hochberg-corrected P values for CNA–protein correlations. Depicted genes have significant negative correlations at FDR <0.05 in the CNA–RNA and CNA–protein analyses. c, Global mRNA–protein correlation and gene set enrichment analysis.

Extended Data Figure 5 Global CNA effects and comparison of CNA trans-effects to knockdown signatures in the LINCS database.

a, CNA landscape in the CPTAC tumour collection. The segment-based CNAs of 77 samples were downloaded from TCGA Firehose, including 18 Basal, 12 Her2, 23 Luminal A and 24 Luminal B subtypes. Copy number amplifications were marked in red and deletions in blue. The bottom colour key represents the log2-transformed copy number value, with CNA = 2 centred at 0. Specific CNA events are seen for chromosome 5q and 10p regions in basal-like tumours. b, Correlations of copy number alterations (x axis) to phosphoprotein levels (y axis) highlight new CNA cis- and trans-effects. Significant (FDR < 0.05) positive (red) and negative (green) correlations between CNA and phosphoproteins are indicated. Histograms show the fraction (%) of significant CNA trans-effects for each CNA gene. c, LINCS CMap analysis facilitates identification of novel functional candidates for CNA trans-effects. Knockdown profiles were compared with CNA–protein trans-effects for 502 genes. Genes with a connectivity score >|90| were considered connected and significant cis-effects were annotated at an FDR <0.05. d, Basal-like tumour-specific CNAs are candidate regulatory events for EGFR and SRC expression levels. Oncogenic kinases with significant CNA–protein trans-effects (left panel), that were regulated in LINCS short hairpin RNA experiments (right panel; 4 cell lines) and directly measured as LINCS landmark genes, are shown alongside candidate regulatory genes CETN3 and SKP1. Clinical ER, PR, and HER2 annotation and PAM50 classification are shown in the header rows of each column.

Extended Data Figure 6 Proteome cluster heat map and stability analysis.

a, K-means consensus clustering of proteome and phosphoproteome data identifies three subgroups: basal-enriched, luminal-enriched, and stromal-enriched. The heat map represents all 1,521 proteins used for clustering (data set G8). b, Identification of optimal proteome clusters for quality-control-passed CPTAC breast cancer tumours. Proteome clusters were derived using consensus clustering based on 1,000 resampled data sets, exploring the range of 2 to 6 K-means clusters. Visualization of consensus matrices from K-means consensus clustering for K = 3, 4, 5 and 6 target clusters. Consensus clustering was performed on 1,521 proteins with no missing values and s.d. > 1.5. c, Silhouette plots were generated to evaluate the coherence of the clustering. Silhouette plots for K = 3 and K = 4 clusters showing a cleaner separation of clusters for K = 3. d, On the basis of both visual inspection of the consensus matrix and the delta plot assessing change in consensus cumulative distribution function (CDF) area, three robustly segregated groups were observed. Consensus CDF and delta area (change in CDF area) plots for 2–6 clusters.

Extended Data Figure 7 Proteome cluster markers and enriched pathways.

a, Markers (based on SAM analysis; FDR <0.01) discriminate between proteome clusters 1, 2 and 3 (compare to heat map of proteins used to derive clusters depicted in Extended Data Fig. 6a). b, Applying a Fisher-exact-test-based enrichment analysis to the proteome, phosphoproteome and mRNA data, gene sets from MSigDB were identified that were unique for each proteome cluster. Heat map showing specific pathways comprising dominant biological themes that are significantly differential by enrichment analysis between basal-enriched and luminal-enriched tumours (Fisher exact test Benjamini–Hochberg-corrected P values are shown; enrichment test performed on marker sets identified using SAM analysis; see Methods; compare to Fig. 3c). c, Heat map showing a selection of gene sets significant in basal-enriched or luminal-enriched tumours exclusively by mRNA, protein or phosphoprotein expression. Cytokine signatures, for example, were strongly captured at the mRNA level, but were seen to only a limited degree at the global protein level, probably because of their typically low protein abundance. By contrast, the vast majority of significant gene sets annotated as ‘signaling’ were enriched only at the phosphoprotein level. d, Global heat map representing all gene sets significantly enriched in at least one of the proteomic breast cancer subtypes. The stromal-enriched group was characterized by breast cancer normal-like, adipocyte differentiation, smooth muscle, toll-like receptor signalling and endothelin gene sets, supporting the clustering-based annotation of high stromal and/or adipose content in these tumours (see Supplementary Table 13).

Extended Data Figure 8 Phosphoproteome pathway clustering, kinase-phosphosite multivariate regression, and protein co-expression networks.

a, Phosphoproteome pathway clustering. Using phosphorylation state as a proxy for activity, deep phosphoproteome profiling allows development of a breast cancer molecular taxonomy on the basis of signalling pathways. K-means consensus clustering was performed on pathways derived from single sample GSEA analysis of phosphopeptide data (908 pathways shown). Of four robustly segregated groups, subgroups 2 and 3 substantially recapitulated the stromal- and luminal-enriched proteomic subgroups, respectively. Subgroup 4 included a significant majority of tumours from the basal-enriched proteomic subgroup, but was admixed particularly with luminal-enriched samples. This subgroup was defined by high levels of cell cycle and checkpoint activity. All basal and a majority of non-basal samples in this subgroup had TP53 mutations. Subgroup 1 was a novel subgroup defined exclusively in the phosphoproteome pathway activity domain, with no enrichment for either proteomic or PAM50 subtypes. It was defined by G protein, G-protein-coupled receptor, and inositol phosphate metabolism signatures, as well as ionotropic glutamate signalling. b, Analysis of the regulatory relationship between outlier kinases (see Supplementary Table 19) and phosphopeptides by regulatory multivariate regression analysis (see Methods) identified CDK1 as the most highly connected of the outlier cyclin-dependent kinases, with highest centrality (based on node-degree; see Methods) among the outlier CDKs and seventh highest centrality among all the outlier kinases considered in the remMap analysis. Each line represents a phosphosite–kinase relationship. cf, Analysis of differences in the co-expression patterns among genes/proteins across different subgroups. A Joint Random Forest method was applied to simultaneously build gene co-expression and protein co-expression networks (Supplementary Table 17, and Methods). Modules in these networks revealed different interaction patterns between basal-enriched and luminal-enriched subgroups. c, Network module P1 of the protein co-expression network, defined chiefly in the proteome space. This module contained 12 genes connected by 39 edges, among which 34 were protein-specific and 5 were shared by both the protein and mRNA co-expression networks. Many edges were supported by published information and were contained in the STRING database. Edges in red are specific to the protein co-expression network; edges in green are shared by both protein and gene co-expression networks; edges indicated by double lines are contained in the STRING database with confidence score greater than 0.15. MMP9, one of the central proteins in this module, contributes to metastatic progression and is a potential target for anti-metastatic therapies for basal-like/triple-negative breast cancer. d, Heat maps of the absolute correlation across each pair of genes in module P1 (shown in c), based on either protein or gene expression data for samples in the basal-enriched and luminal-enriched subgroups, respectively. The MMP9 protein was strongly co-expressed with the other members of the module only in the basal-enriched subgroup. Notably, this observation is dependent on protein data; the correlation at the mRNA level for this module was consistently low in both the basal-enriched and luminal-enriched subgroups indicating that these events coherently occur at the proteomic level. e, Co-expression network based on proteomics data. The network contains 693 proteomic network-specific edges (grey) and 792 edges shared with the RNA-seq network (green). For each module, the most enriched category and corresponding Benjamini–Hochberg-adjusted P value is reported. Pie charts adjacent to each module show the proportion of proteomics-specific edges (grey area) and edges shared between proteomics and RNA-seq data (green area). f, RNA-seq network.

Extended Data Figure 9 Phosphoproteome signatures of PIK3CA- and TP53-mutated tumours highlight activated key regulators and indicate frequency of activation.

a, c, Phosphosites upregulated in mutated tumours (SAM FDR < 0.05 across all tumours and independently also across luminal tumours; average phosphosite signal for all markers shown as bar graph). To avoid confounding by intrinsic subtype-specific distinctions, only markers that were significantly identified both in analyses covering all tumours and analyses restricted to luminal tumours were selected (FDR < 0.05). Colour bars in the margins indicate FDRs for grouped analysis of different mutation classes and indicate kinase substrates of known kinases in the respective pathways. Significantly regulated kinase phosphosites are annotated. The average phosphorylation signal of the marker phosphosites provides a read-out for PI3K and TP53 pathway activity in mutated tumours (histogram below heat map). A 95% prediction confidence interval (indicated by dashed lines) across the average signal in non-mutated tumours was chosen in order to discriminate active from non-active tumours. The most strongly activated PIK3CA kinase domain mutant tumour differed from the other nine kinase domain mutant tumours, as it contained an amino acid side chain charge neutral H1047L instead of the more common positively charged H1047R mutation. Among the 62 phosphosites identified that were significantly upregulated in PIK3CA-mutated tumours, 13 phosphosites were found on phosphoproteins that are known substrates of well-annotated kinases in the PIK3CA pathway (a, right column). In the mutant TP53 analysis, a total 20 phosphosites were found on phosphoproteins that are known substrates of well annotated kinases in the p53 pathway (c, right column). b, d, Upregulated phosphosite sets were derived from isogenic PIK3CA and TP53 mutant versus wild-type cell-line pairs and tested for enrichment within mutant versus wild-type CPTAC tumours using single sample GSEA. Significantly enriched phosphosite sets are shown (P < 0.05).

Extended Data Figure 10 Pircos plots, kinase outliers and outliers in the ERBB2 pathway.

a, Pircos (proteogenomics circos) plots for 8q and 17q showing median CNA, RNA, protein, and phosphosite expression for 20 tumours with amplification in 8q based on RIPK2 CNA >1; 23 tumours with amplification in 8q based on PTK2 CNA >1; 15 tumours with amplification in 17q based on CDK12 CNA >1; and 10 tumours with amplification in 17q based on TLK2 CNA >1. Red indicates expression >1, blue <−1, and grey between −1 and 1. Genes with both copy number amplification (CNA >1) and increased phosphosite expression (p-site >1) are labelled. b, Phosphosite outliers in known ERBB2 signalling genes. To better understand the downstream effects of ERBB2 amplification, phosphosite outliers in known ERBB2 signalling genes (MSigDB’ pathway set, ‘KEGG_ERBB_SIGNALING PATHWAY’) were identified for the 15 samples that had ERBB2 phosphosite outlier status. Forty-one genes were identified as having a phosphosite outlier in at least one of the ERBB2-amplified samples. PAK4 and ARAF phosphosite outlier status were found in seven of the 15 ERBB2 kinase outlier samples; GSK3B outliers were found in 6 samples; and EIF4EBP1, MAP2K2, ABL1 and AKT1 outlier status was found in 5 of the 15 samples. c, Proteogenomic outlier expression analysis for TLK2 and RIPK2. Samples with outlier phosphosite (red), protein (yellow), RNA (green) and copy number (purple) expression are shown. Phosphosite squares indicate per-sample outlier phosphosites.

Supplementary information

Supplementary Information

This file contains a Supplementary Discussion, Supplementary Methods, the legends for Supplementary Tables 1-19 (see separate zipped file), Supplementary References and Supplementary Notes (see Contents for more details). (PDF 1847 kb)

Supplementary Tables

This zipped file contains Supplementary Tables 1-19. (ZIP 135465 kb)

PowerPoint slides

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Mertins, P., Mani, D., Ruggles, K. et al. Proteogenomics connects somatic mutations to signalling in breast cancer. Nature 534, 55–62 (2016).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing