GSTP1 is associated with NSCLC immunotherapy resistance: multi-omic discovery via transformer-based perturbation across 109 cell types

Immunotherapy has changed the therapeutic landscape of non-small cell lung cancer (NSCLC), but drug resistance is inevitable. Elucidating the functional roles of distinct genes in heterogeneous cell types remains experimentally challenging. Single-cell RNA sequencing (scRNA-seq) data (GSE207422, GSE179994, and GSE176021) of patients with NSCLC who underwent immunotherapy were collected and annotated. Using Geneformer to perform in-silico single-gene knockouts across all cell types to map their impact on immunotherapy efficacy. Key genes were subsequently prioritized by integrating differential-expression analysis with clinical trial sequencing data. Hierarchical clustering delineated functional gene modules, and variance-partitioning quantified each gene’s contribution to inter-patient immunotherapy response. Finally, findings were validated mechanistically in vivo and in vitro. A total of 109 cell types were identified after annotating the scRNA-seq data according to a large-scale pancancer atlas and classic hallmarks. We identified 110 genes associated with immunotherapy efficacy (25 genes associated with better immunotherapy efficacy and 85 associated with worse immunotherapy efficacy) and gained insights into the specific cell types in which these genes play crucial roles. Additionally, we created a website to showcase these 110 genes and developed an immune scoring system based on these genes, as well as a more concise CancerCellScore model based on 7 genes. GSTP1, the top contributor among these 7 genes, was highly effective in predicting the response to NSCLC immunotherapy. The combined use of anti-GSTP1 and anti-PD-1 could significantly inhibit the growth of tumors. An immune scoring system based on these 110 genes was established and verified to be able to well distinguish immunotherapy response population. A website (https://immunotherapy.live/ResisGenes) was deployed to visualize the scoring system and provide real-time sample scores. Mechanistically, GSTP1 expression was negatively associated with the immunotherapy response, targeting GSTP1 may enhance the efficacy of anti-PD-1, potentially through modulation of the ferroptosis pathway.