E402: A Pan-cancer Comparative Analysis of The Cancer Genome Atlas Genetic Immune Signatures

Numerous transcriptomics-based immune signatures were created from The Cancer Genome Atlas (TCGA) and studies of tumor-infiltrating lymphocytes (TIL). Some signatures correlate with prognosis and response to immunotherapy in selected cancers. However, no comprehensive analysis exists to compare the performance of these signatures.

Methods:

Transcriptomics-derived signatures from TCGA analysis and benchtop TIL research were curated. Samples from TCGA RNA-sequencing recount3 project were downloaded for 33 cancer types and 11,348 samples and processed via the Monorail system. The GSVA R/ Bioconductor package was used to summarize gene sets and calculate individual gene scores. We used overall survival (OS) and progression-free interval (PFI) coefficients to determine the applicability of each immune signature. Analysis was conducted across pan-cancer, tissue germ cell origins, and histology.

Results:

Literature review returned 146 molecular signatures comprising 3088 unique genes with nearly half (1432) shared across multiple signatures.  The most overlapped genes across signatures were ENTPD1, PDCD1, and HAVCR2. The median number of genes per signature was 50.

Across pan-cancer, all 146 signatures had comparable OS and PFI coefficients suggesting equivocal performance. The top signatures correlating with OS and PFI coefficients by germ cell origin included Oh.CD8.CM, Zhang.CD8.TCS, and Caushi.Stem-Like memory for ectoderm, endoderm, and mesoderm-derived cancers, respectively. The top signatures correlating with OS and PFI coefficient in neural crest-derived cancers were Shi-TCS and Zhang.CDS.TCS, respectively. Only 8.4% of genes (n=12/143) were shared across these signatures.

Grouping samples by histology showed variability across OS coefficient; however, the two highest correlated signatures were Tang_Ferroptosis and Zhang.CD8.TCS each as top prognosticators of OS in 4 cancer types. The most common overlapped genes across these signatures were IL7R, CD69, ENTPD1, GPR183, IL7R, KLF3, PABPC1, SC5D, SELL and TCF7. For PFI coefficient, the same signatures Zhang.CD8.TCS and Tang_Ferroptosis were the top prognosticators of PFI in 5 and 4 cancer types respectively. The most common overlapped genes across these immune signatures were IL7R, TCF7, GPR183, CCR7, and KLF3.

Conclusions:

Despite a high level of overlap, selected immune signatures from transcriptomic data were prognostic of OS and PFI when examining samples by germ cell origin and histology. The use of these selected signatures by histology may aid prognostication and guide patient selection for immunotherapy.