Cancer genomics takes center stage for bioinformatic classification of cancer subtypes providing oncologists with valued information to help tailor personalized patient drug therapy protocol. High throughput transcriptomics provides an alternative/additional mechanism in which to probe cancer subtypes. Here we used statistical analyses of transcriptome-wide gene expression levels, to probe for differential patient survival genes per cancer subtype using large sample data (>11,000 patients ; the Cancer Genome Atlas- TCAG). Transcriptomic survival genes (TSGs) were identified in 27 out of 36 different cancer subtypes (p<0.003). Over 10,000 non-overlapping TSGs were identified and represented at least once across all cancers. Of these, only 797 were found to be shared across four or more cancer subtypes. Probing these shared genes for functional annotation, expectantly we found that many of these overlap with cell cycle (168 genes; P= 3.4e-34). More specific annotations were also enriched, correlating with different signaling pathways (e.g. ERBB3, ATR, WEE1, BRD4 or SRC knockdown). Significant correlations were also found for TSGs and gene signatures related to diverse non-cancer phenotypic/disease indications, including e.g., autoimmunity (diabetes & scleroderma), hypoxia, bacterial & retroviral infection as well as angiogenesis. These annotations can be mapped back to determine in which cancer subtypes they are enriched as differentially expressed survival genes.
TSG analysis was also used to probe the architecture of patient survival for a single cancer subtype. Here we focused upon melanoma (SKCM; 469 patients; 1500 TSGs (p< 0.03)), whose survival genes could be subdivided into five gene expression clusters. Cluster#2 (499 genes) was heavily enriched immune-cell lineage genes, and included PDL1 (CD274) and CD3D. PDL1 is a diagnostic marker to predict better response to anti-PD1 immunotherapy. Thus Cluster#2 genes might provide a robust transcriptomic means to predict favorable patient response to PD-1 blockade.
We propose that each survival gene expression cluster represents alternative manifestations of disease phenotype and thus provides us with the means to use these annotations to help subdivide cancer types into separate categories. This data can complement tumor genotyping analysis to give a better comprehensive understanding of tumor phenotype and potential best tailored therapeutic intervention.