Somatic Mutation Trajectories Define Prognostically Distinct Subtypes and Shape the Tumor Microenvironment in Gastric Cancer

Objective: Gastric cancer (GC) is characterized by molecular heterogeneity, yet current classifications are largely based on cross-sectional molecular profiles and do not account for the temporal order of mutation accumulation. This study aimed to reconstruct somatic mutation trajectories to identify prognostically distinct subtypes and to examine transcriptomic and microenvironmental features associated with these inferred trajectories. Methods: We applied the Subtype and Stage Inference (SuStaIn) algorithm to TCGA-STAD somatic mutation data to infer the temporal sequence of mutation accumulation. Stage-correlated gene expression analysis was performed to identify genes whose expression levels changed with evolutionary stage. The tumor microenvironment (TME) was characterized using EcoTyper and single-cell RNA sequencing deconvolution, while drug sensitivity was estimated through transcriptome-based IC50 prediction. The clinical relevance of the inferred trajectories was further evaluated in three independent external transcriptomic cohorts. Results: We identified two distinct evolutionary trajectories: the Accelerated Path (AP, 65%) and the Gradual Path (GP, 35%). In the AP, TP53 mutations were positioned at an earlier evolutionary stage (Stage 3) compared to the GP (Stage 8). AP patients had significantly worse overall survival (Hazard Ratio = 1.437, p = 0.044, adjusted for clinical stage and molecular subtypes). The AP was associated with stage-correlated downregulation of the sodium channel gene SCN4A (ρ = −0.36, p < 0.001) and an increase in a squamous-associated gene expression score, while the GP showed stage-correlated expression changes in the mitochondrial gene SDHD (ρ = −0.35, p < 0.001). The AP was further characterized by higher inferred abundance of extracellular matrix CAFs (eCAFs) and lower inferred immune cell scores, whereas the GP was associated with higher inferred signatures of activated B cells and effector memory T cells. Computational drug sensitivity modeling predicted a negative correlation between AP stage and IC50 values for 5-Fluorouracil and Docetaxel. Conclusions: Two distinct mutational ordering patterns identified by SuStaIn are associated with divergent transcriptomic features, TME compositions, and clinical outcomes in gastric cancer. The AP subtype is characterized by early TP53 mutations, SCN4A downregulation, and a stromal-enriched microenvironment, while the GP subtype is associated with later TP53 mutations, SDHD-correlated expression, and higher inferred immune cell scores. The reproducibility of these associations was confirmed in independent cohorts. The computational drug sensitivity predictions and the proposed mechanistic links between gene expression patterns and clinical outcomes should be viewed as hypothesis-generating findings that require prospective and functional validation.