The Immune Landscape of Cancer

Spatial Organization and Molecular Correlation of Tumor-Infiltrating Lymphocytes Using Deep Learning on Pathology Images

“…A wide range in spatial infiltrate is seen among the TCGA tumor types (Figure 3A and Table S1), with high infiltrates in gastric cancer (STAD) with a mean of 14.6%, rectal cancer (READ) at 13.0%, squamous cell carcinoma in the lung (LUSC) at 11.6%, while uveal melanoma (UVM) has only 1% estimated TIL fraction, consistent with its inclusion as negative control (Figure S3A). Wide differences are also seen grouping tumors by the nature of the immune response, according to a recent immune characterization of all TCGA tumors (Thorsson et al, 2018)(Figure 3B). The most immunologically active immune subtypes (e.g., C1, C2) tend to have the greatest spatial infiltration of lymphocytes.…”

“…For example, EBV-positive GI cancers are enriched in the “ Brisk, diffuse ” phenotype (14; 5; 6 × 10 −3 ). Differences are also seen among immune subtypes (Figure S5B) defined in the TCGA pan-immune analysis (Thorsson et al, 2018), where the C4 subtype is enriched in the “ Non-brisk, focal ” (82; 44; 6 × 10 −8 ). This is noteworthy, as subtype C4 is relatively richer in cells of the monocyte/macrophage lineage, which may play a role in sculpting the TME as evidenced in these patterns.…”

“…See Additional Resources for listing and acronyms. We included uveal melanoma (UVM) as one of the 13 cancer types essentially as a type of negative control (Figure S3A), since it has the fewest immune cells among TCGA tumors (Thorsson et al, 2018). Tumor types were selected to represent a range of known positive involvement of lymphocytes and immunogenicity from literature and from molecular estimates of lymphocyte content.…”

“…The spatial fraction of TILs was compared with molecular estimates of TIL content from molecular genomics assays (Thorsson et al, 2018). The molecular estimate of TIL fraction is obtained by multiplying an estimate of the overall leukocyte fraction, based on DNA methylation arrays, with an estimate of the lymphocyte proportion within the immune compartment obtained by applying CIBERSORT (Newman et al, 2015) to RNA sequencing data.…”

“…Recent publications (Charoentong et al, 2017; Li et al, 2016; Rooney et al, 2015) have presented comprehensive analyses of TCGA data on the basis of immune content response. A recent study (Thorsson et al, 2018) reports on a series of immunogenomic characterizations that include assessments such as total lymphocytic infiltrate, immune cell type fractions, immune gene expression signatures, HLA type and expression, neoantigen prediction, T cell and B cell repertoire, and viral RNA expression. From these base-level results, integrative analyses were performed to derive six immune subtypes, spanning tumor types and subtypes.…”