Intron retention marks a myeloid‑skewed immune desert and nominates an ICI nonresponse marker in H. pylori -positive gastric cancer

Background: The benefits of immune checkpoint inhibitors (ICIs) in gastric cancer (GC) vary according to the Helicobacter pylori (Hp) status. While profiling transcriptomes, intron retention (IR) emerged as a major distinguishing feature in Hp-positive disease, but its immunologic significance remains unclear. Methods: We analyzed 66 RNA-seq tumor samples (39 Hp-positive and 27 Hp-negative) from 24 patients. Alternative splicing (AS) were quantified USING PSIsigma, rMATS. AS-derived neoantigens were predicted (SNAF; strict filters), and a quality-weighted neoantigen burden was calculated. Antigen-presentation machinery (APM) was summarized using a purity-adjusted APM score. Tumor microenvironment (TME) features were profiled using ESTIMATE, Hallmark GSEA, ssGSEA, CIBERSORTx, and a four-class subtype (desert/immune-enriched [IE]/IE fibrotic/fibrotic). High IR was defined as three or more IR events. Statistical analysis was performed using the Mann-Whitney test and Spearman correlation, with BH-FDR correction. Results: Across Hp infection status, Hp-positive tumors harbored more IR/AS events and a higher total neoantigen burden under strict filters; however, IR-specific and quality-weighted burdens were not significantly higher. In Hpnegative tumors, higherquality neoantigen burden correlated with intrinsic APM capacity (ρ≈0.41, p=0.034), indicating preserved antigen-APM coupling. Within Hp-positive tumors, IRhigh (n=12) versus IRlow (n=27) tumors showed higher neoantigen metrics (IRderived 4 vs 1, p=4.6×10⁻⁷; weighted 24.1 vs 12.2, p=3.9×10⁻⁴), enrichment of G2M/E2F/MYC/mitotic spindle programs (FDR<0.05), higher tumor purity (p=0.022), lower ImmuneScore (p=0.0156) and StromalScore (p=0.017), and uniform classification as the Desert subtype (12/12). NMD and splicing signatures were significantly elevated (p=0.014/0.034), CD8/IFN signatures were reduced (p=0.034), and the APM score favored IRlow tumors ( p=0.050). A Myeloid/mastcell-skewed pattern was also evident (MDSC signature, p = 0.037; activated mast cells, p=0.0077; resting mast cells, p=0.0027). In Hp-positive tumors overall, quality-weighted neoantigen burden correlated positively with NMD/splicing (ρ≈0.63, p=1.9×10⁻⁵; ρ≈0.67, p=3.7×10⁻⁶), while IR burden correlated negatively with CD8/IFN activity (ρ≈−0.44, p=0.0056). IR events were not enriched in core APM genes. Conclusions: In Hp-positive GC, high IR defines a proliferative, NMD/splicing-stressed, myeloid/mastcell-skewed immunedesert microenvironment characterized by neoantigen-APM uncoupling. This phenotype aligns with limited ICI benefit, identifying high IR as a potential predictive marker of ICI nonresponse, supporting prospective validation and hypothesis-driven evaluation of ICI-based regimens that reprogram myeloid-rich TMEs or modulate splicing/NMD pathways.