Pan-cancer immunogenomic analyses reveal genotype-immunophenotype relationships and predictors of response to checkpoint blockade

Abstract

Current major challenges in cancer immunotherapy include identification of patients likely to respond to therapy and development of strategies to treat non-responders. To address these problems and facilitate understanding of the tumor-immune cell interactions we inferred the cellular composition and functional orientation of immune infiltrates, and characterized tumor antigens in 19 solid cancers from The Cancer Genome Atlas (TCGA). Decomposition of immune infiltrates revealed prognostic cellular profiles for distinct cancers, and showed that the tumor genotypes determine immunophenotypes and tumor escape mechanisms. The genotype-immunophenotype relationships were evident at the high-level view (mutational load, tumor heterogenity) and at the low-level view (mutational origin) of the genomic landscapes. Using random forest approach we identified determinants of immunogenicity and developed an immunophenoscore based on the infiltration of immune subsets and expression of immunomodulatory molecules. The immunophenoscore predicted response to immunotherapy with anti-CTLA-4 and anti-PD-1 antibodies in two validation cohorts. Our findings and the database we developed (TCIA-The Cancer Immunome Atlas, http://tcia.at) may help informing cancer immunotherapy and facilitate the development of precision immuno-oncology. ACKNOWLEDGMENTSThe results shown here are in part based upon data generated by the TCGA Research Network: http://cancergenome.nih.gov.

Preclinical efficacy of immune-checkpoint monotherapy does not recapitulate corresponding biomarkers-based clinical predictions in glioblastoma

“…This is because the tendency of these genes (i.e., CD247, CD2, CD3D, CD3G, GPR171, CD27, LCK, LTB, IL2RB and ICOS) to show co-expression has statistically higher (but not absolute) chance of indicating the presence of Tregs. 31,33,34 The principle behind such immune cell-specific metagenes is discussed in details, elsewhere. 31,33,34 In this study, the rational was, if an immune-checkpoint coding gene co-clusters with the Treg-metagene within human GBM tissue than the probability of that immune-checkpoint being expressed on Tregs is higher (but not absolute).…”

“…2,6 We gave due consideration to two currently described scenarios in this regard, i.e., (1) the "canonical" association between immunosuppressive Treg cells and immune checkpoints, considering that the function of Tregs and CTLA4, PD1 and IDO1 in enforcing immunosuppression is overlapping 2,6,17,34,35 ; and (2) the "paradoxical" association between immune checkpoints and T cell-effector markers (like IFNg, Granzyme B-perforin), since activation of T cells eventually causes upregulation of immune checkpoints as an auto-regulatory loop in later stages of effector function (to avoid autoimmunity and resolve inflammation). To this end, we first analyzed the correlation between CTLA4, PDCD1 or IDO1 expression levels and a previously established Treg-specific genetic signature or metagene.…”

“…17,34,35 To this end, we decided to ascertain the prognostic impact of immune-checkpoint gene expression in TCGA and REMBRANDT GBM cohorts. The differential expression of CTLA4, PDCD1 and IDO1 did not show strong association with poor or prolonged overall survival (OS) in GBM patients, neither in TCGA (Figs.…”

“…Pan-cancer T cell subtype-specific gene signatures available from The Cancer Immunome Atlas 34 were used to estimate the total TIL-fractions within 19 different TCGA tumor cohorts (Fig. 4A).…”

“…These coefficient values were used for hierarchical clustering 33 through the Cluster 3.0 software 63 and the gene co-expression matrices were visualized as heatmaps through TreeView. 64 Last but not least, to estimate the basal or pre-existing (total) T cell infiltrates-associated genetic signatures in different cancer types, we used the Cancer Immunome Atlas (https://tcia.at/home), 34 to analyze the absolute amounts of tumor-associated T cell genetic signatures in 19 different TCGA tumor types (mentioned in the figure or figure legends). The "cell type fractions plot" was used to generate absolute fraction values with Cibersoft_LM22 deconvolution methodology.…”