Atezolizumab

After receiving institutional review board approval from the Memorial Sloan Kettering Cancer Center, institutional pharmacy records were used to identify patients who received at least one dose of immunotherapy (atezolizumab, avelumab, durvalumab, ipilimumab, nivolumab, pembrolizumab, or tremelimumab) and then cross-referenced with patients who had MSK-IMPACT testing done in the context of routine clinical care. Cancer types with greater than 35 patients on initial collection were selected for further analysis in the cohort. The majority of patients who received MSK-IMPACT testing on tumor tissue are enrolled on an institutional IRB-approved research protocol (NCT01775072) with the remaining patients receiving testing as part of routine clinical care; all patients provided informed consent permitting return of results from sequencing analyses and broader characterization of banked specimens for research.Details of tissue processing and next generation sequencing and analysis have been previously described. ^{11 (link)} Importantly, concurrent sequencing of germline DNA from peripheral blood is performed for all samples to identify somatic tumor mutations. Patients enrolled on ongoing clinical trials for which publication of outcomes data was prohibited were removed as well as a small proportion of patients with localized disease treated in the neoadjuvant setting(n=9) or who had localized disease. Other preceding or concurrent non-ICI treatments were not recorded or accounted for in the analysis. The timing of tissue pathology on which MSK-IMPACT was performed relative to ICI administration is also heterogenous with a small portion of patients with testing after ICI administration.

Whole-transcriptome profiles were generated for 263 patients using TruSeq RNA Access technology (Illumina). RNA-seq reads were first aligned to ribosomal RNA sequences to remove ribosomal reads. The remaining reads were aligned to the human reference genome (NCBI Build 38) using GSNAP^{53 (link),54 (link)} version 2013-10-10, allowing a maximum of two mismatches per 75 base sequence (parameters: ‘-M 2 -n 10 -B 2 -i 1 -N 1 -w 200000 -E 1 –pairmax-rna = 200000 -clip-overlap). To quantify gene expression levels, the number of reads mapped to the exons of each RefSeq gene was calculated using the functionality provided by the R/Bioconductor package GenomicAlignments^{55 (link)}.
Gene signatures were defined as follows: Angio^{23 (link)}: VEGFA, KDR, ESM1, PECAM1, ANGPTL4, and CD34; T_eff^{24 (link)}: CD8A, EOMES, PRF1, IFNG, and CD274; myeloid inflammation^{29 (link)–33} : IL-6, CXCL1, CXCL2, CXCL3, CXCL8, and PTGS2. These three gene expression signatures were defined based on previously published associations with their respective biology.
To calculate scores for each of these signatures, counts were first normalized using edgeR’s normalization factors^{56 (link)}, followed by filtering out genes with low coverage (i.e., not reaching 0.25 CPM (counts per million) in at least one-tenth of available samples) and log₂-transformation using limma’s voom^{57 (link)}. Then for each sample, the average expression of all genes in a given signature was computed, and is reported as the sample’s signature score. For each gene signature, patients were divided into two groups based on the median gene signature score of all tumors: high gene signature expression was defined as expression at or above median levels, and low gene signature expression was defined as expression below the median.
For the heatmap (Fig. 2a), each patient was placed into high or low groups for all three gene expression signatures: Angio, T_eff, and myeloid inflammation (based on median expression, as described above). Subsequently, patients were sorted by the combination of these three groups: first T_eff^HighAngio^Low patients are shown, sorted by myeloid inflammation low/high; then T_eff^HighAngio^High patients are shown, sorted by myeloid inflammation high/low; then T_eff^LowAngio^High patients are shown, sorted by myeloid inflammation low/high; finally, T_eff^LowAngio^Low patients are shown, sorted by myeloid inflammation high/low. Also, the ordering of the genes was predetermined, based on biological function. Z-score-transformed normalized counts are shown.