Integrin-based prognostic model predicts survival, immunotherapy response, and drug sensitivity in gastric cancer

Abstract

Background

Extracellular matrix (ECM) acts as a physical barrier to tumors, resulting in the lysis or delay of drug delivery. Integrins (ITGs) are essential for tumor cell-ECM interactions. Thus, we established a novel prognostic model to predict overall survival, immunotherapy benefits, and therapeutic agents in gastric cancer (GC) based on ITGs-related ECM landscape.

Methods

Using the TCGA-STAD dataset, we studied the genetic and transcriptional changes of ITGs. We used a merged cohort for ITGs survival analysis and determined molecular pattern clusters using consensus unsupervised clustering methodology. We confirmed the distinct ECM landscape between constructed clusters by performing gene set variation and Kaplan-Meier analysis. We utilized prognostic differentially expressed genes between clusters to develop a prognostic model utilizing logistic least absolute shrinkage and selection operator cox regression analysis, followed by stepwise multivariate Cox analysis in the training dataset. The model was validated by receiver operating characteristic curves and Kaplan-Meier analysis in the testing dataset and seven validation datasets. We compared our model to 35 previously published models. To analyze immune infiltration, we used multiple algorithms, which were further confirmed by single-cell RNA-sequencing and fluorescent multiplex immunohistochemistry. We explored tumor mutation burden (TMB), microsatellite instability-high (MSI-H) grade, immunotherapy response, chemotherapy sensitivity, and clinical significance between the low-risk and high-risk groups. Finally, we assessed the risk score in five reported molecular subtypes.

Results

The two ITGs-related clusters were identified, and their ECM landscapes were distinct. The prognostic model was constructed and had shown stable performance in internal and external validation. In addition, our model outperformed 35 previously published models. High-risk patients had a bad prognostic ECM landscape, high stromal cell inflammation, a lower TMB, a lower MSI-H grade, a worse tumor stage, a worse response to immunotherapy, and less sensitivity to chemotherapy. In five reported molecular subtypes, the worse subtypes showed a higher risk score.

Conclusions

The prognostic model could be an effective and promising tool for predicting prognosis and therapy response in GC patients.