Microsatellite-Stable Gastric Cancer Can be Classified into 2 Molecular Subtypes with Different Immunotherapy Response and Prognosis Based on Gene Sequencing and Computational Pathology

Abstract

Most patients with gastric cancer (GC) exhibit microsatellite stability, yet comprehensive subtyping for prognostic prediction and clinical treatment decisions for microsatellite-stable GC is lacking. In this work, RNA-sequencing gene expression data and clinical information of patients with microsatellite-stable GC were obtained from The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO) databases. We employed several machine learning methods to develop and validate a signature based on immune-related genes (IRGs) for subtyping patients with microsatellite-stable GC. Moreover, 2 deep learning models based on the Vision Transformer (ViT) architecture were developed to predict GC tumor tiles and identify microsatellite-stable GC subtypes from digital pathology slides. Microsatellite status was evaluated by immunohistochemistry, and prognostic data as well as hematoxylin and eosin whole-slide images were collected from 105 patients with microsatellite-stable GC to serve as an independent validation cohort. A signature comprising 5 IRGs was established and validated, stratifying patients with microsatellite-stable GC into high-risk (microsatellite-stable-HR) and low-risk (microsatellite-stable-LR) groups. This signature demonstrated consistent performance, with areas under the receiver operating characteristic curve (AUC) of 0.65, 0.70, and 0.70 at 1, 3, and 5 years in the TCGA cohort, and 0.70, 0.60, and 0.62 in the GEO cohort, respectively. The microsatellite-stable-HR subtype exhibited higher levels of tumor immune dysfunction and exclusion, suggesting a greater potential for immune escape compared with the microsatellite-stable-LR subtype. Moreover, the microsatellite-stable-HR/LR subtypes showed differential sensitivities to various therapeutic drugs. Leveraging morphologic differences, the tumor recognition segmentation model achieved an impressive AUC of 0.97, whereas the microsatellite-stable-HR/LR identification model effectively classified microsatellite-stable-HR/LR subtypes with an AUC of 0.94. Both models demonstrated promising results in classifying patients with microsatellite-stable GC in the external validation cohort, highlighting the strong ability to accurately differentiate between microsatellite-stable GC subtypes. The IRG-related microsatellite-stable-HR/LR subtypes had the potential to enhance outcome prediction accuracy and guide treatment strategies. This research may optimize precision treatment and improve the prognosis for patients with microsatellite-stable GC.