Tumor microenvironment-based classification for predicting gastric cancer prognosis

The tumor microenvironment (TME), consisting of tumor-associated stroma and tumor-infiltrating lymphocytes (TIL), is crucial for prognostic information in gastric cancer (GC). Despite its potential, routine clinical adoption remains limited. In a study of 320 GC patients, virtual staining and image processing were applied to hematoxylin & eosin-stained slides. This method quantified the tumor-stroma ratio (TSR) and TIL, leading to a TME-based prediction model (TMEPATH) using a scoring system derived from univariate Cox regression. Subgroups were categorized to predict GC patient survival, with genomic analysis linking TME-based prognostic models to specific genetic alterations. TSR was categorized into TSR_low (n = 113) and TSR_high (n = 207) using a 0.76 cut-off, selected to maximize the concordance index for overall survival prediction. Two TIL subtypes were defined based on a 0.03 cut-off. TMEPATH, a composite biomarker integrating the TSR- and TIL-based subtypes, stratified patients into low-risk (91 patients, 28.4 %), medium-risk (167 patients, 52.2 %), and high-risk (62 patients, 19.4 %) groups, correlating with survival outcomes (hazard ratio [HR] 1.281; 95 % CI 0.957-1.714 for medium vs. low-risk, and HR 1.768; 95 % CI 1.242-2.517 for high vs. low-risk; log-rank P = 0.0061). These findings were validated in a separate cohort (n = 186) with significant clinical relevance (HR 1.389; 95 % CI 0.855-2.257 for medium vs. low-risk, and HR 2.435; 95 % CI 1.380-4.298 for high vs. low-risk; log-rank P = 0.0064). TSR, TIL, and TMEPATH were associated with microsatellite instability, tumor mutation burden, and CDH1 mutations. The classification of GC into three TME subtypes using TSR and TIL provides a reliable prognostic tool for survival prediction.