Abstract 6655: Tumor microenvironment evaluation and tumor intrinsic genomic features predict anti-PD-1 response of metastatic gastric cancer: Results from phase II clinical trial and multi-omics data

Background: Clinical studies support the efficacy of immune checkpoint blockades (ICBs) in a subset of patients with metastatic gastric cancer (mGC). With the aim of identifying determinants of response to ICBs, we performed molecular characterization of tissues from 61 patients with mGC who were treated with pembrolizumab as salvage treatment in a prospective phase II clinical trial (NCT#02589496). Methods: Of 61 patients, 60 patients underwent pretreatment biopsy and 45 specimens were of sufficiently high quality for RNA sequencing. TMEscore, which was previously established to quantify the tumor microenvironment (TME), was used to estimated TME of pretreatment specimens. Additional ACRG and TCGA multi-omics data were applied to validate the aforementioned results. Intrinsic mutations and virus infections correlated to the TMEscore were further identified via analyzation of TCGA and ACRG data. Results: We established a methodology (TMEscore) to evaluate the TME of GC patients, which was previously found to be a robust prognostic and predictive biomarker for patients treated with ICBs. By applying ROC curve analysis, the TMEscore was found to be the best predictive biomarker (TMEscore: AUC = 0.891; CPS: AUC = 0.830; TMB: AUC = 0.672; MSI status: AUC = 0.708; EBV status: AUC = 0.727; respectively). Moreover, TMEscore was the most significant gene signature that correlated with tumor response (TMEscore: P = 1.7e-5; GEPs: P = 0.00035; ImmunoScore: P = 0.29106; CD8+ T cell fraction: P = 0.00011; respectively). A higher TMEscore was significantly associated with EBV+ and MSI-High TCGA molecular subtypes (Kruskal-Wallis test, P = 0.002) which were reported to benefit from ICBs of GC. Moreover, analysis of the TCGA and ACRG data reproductively supported the predictive value of TMEscore and its latent power in identifying GC patients with MSI-High characteristics and EBV+ from other subtypes. Additionally, the study of TCGA dataset revealed that TMEscore was significantly associated with tumor neoantigen load (Spearman test, P = 2.8e-10, r = 0.441). Notably, it highlighted that GC patients with ARID1A and PIK3CA mutations exhibited higher TMEscore (Wilcoxon test, ARID1A: P = 8.4e-10, PIK3CA: P = 4e-10) in both TCGA and ACRG cohorts. Furthermore, TCGA genomic data indicated that patients with ARID1A p.D1850Tfs*33 and p.F2141Sfs*59 mutations exhibited the highest TMEscore than patients with other mutational sites and wild-typed patients (Kruskal-Wallis test, P = 9e-11). Conclusions: These findings indicate that the assessment of TMEscore via high throughput-sequencing and PCA algorithm provides a robust biomarker for the selection of GC patients who may derive greater benefit from pembrolizumab. Our data also suggest that TMEscore may be a more accurate predictive biomarker than TMB, MSI and EBV status, and this resource may help facilitate the development of precision immunotherapy. Citation Format: Dongqiang Zeng, Wangjun Liao, Kyoung Mee Kim, Min Shi, Rui Zhou, Yunfang Yu, Zilan Ye, Jiani Wu. Tumor microenvironment evaluation and tumor intrinsic genomic features predict anti-PD-1 response of metastatic gastric cancer: Results from phase II clinical trial and multi-omics data [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 6655.