Tumor microenvironment-oriented HNGCIscore identifies immunotherapy-sensitive subgroups in ICI retreatment of...

PD-1/PD-L1 inhibitor-based chemoimmunotherapy has become a standard first-line treatment for HER2-negative advanced gastric or gastro-esophageal junction cancer. However, most patients eventually experience disease progression, and optimal post-progression strategies, particularly immune checkpoint inhibitor (ICI)-based retreatment, remain unclear. This study evaluated the real-world feasibility of ICI-based retreatment and developed an exploratory tumor microenvironment (TME)-oriented biomarker framework to stratify patients who may benefit from immunotherapy.

This multicenter retrospective study included 144 patients with HER2-negative advanced gastric cancer who received ICI-based retreatment after progression on first-line ICI-containing therapy between January 2020 and February 2025. Retreatment was classified as cross-line ICI continuation or ICI rechallenge after an ICI-free interval. Clinical outcomes included objective response rate (ORR), disease control rate (DCR), progression-free survival during retreatment (PFS2), overall survival, and safety. TCGA-STAD multi-omics data were used to infer HER2-negative status and identify immune/TME subtypes. Differential expression analysis and weighted gene co-expression network analysis in the PRJEB25780 gastric cancer immunotherapy cohort were used to identify candidate response-associated genes. Machine-learning models were then compared to construct an exploratory immunotherapy-response classifier, termed HNGCIscore. Protein-level validation of selected markers was performed by immunohistochemistry.

Among 144 patients, 73 received cross-line ICI continuation and 71 received ICI rechallenge. In the cross-line group, ORR was 6.85%, DCR was 50.68%, and median PFS2 was 3.0 months. In the rechallenge group, ORR was 5.63%, DCR was 52.11%, and median PFS2 was 5.5 months. These findings were interpreted descriptively because the two groups represented distinct real-world retreatment scenarios rather than randomized comparative cohorts. Any-grade treatment-related adverse events occurred in 58.9% and 77.5% of patients in the cross-line and rechallenge groups, respectively, while grade 3-4 events occurred in 32.0% and 34.0%. No treatment-related deaths were observed. Transcriptomic analyses identified two immune/TME subtypes. Integration of subtype-associated differentially expressed genes with immunotherapy-response-related WGCNA modules yielded 35 candidate genes. Among 14 machine-learning algorithms, XGBoost showed the best overall discrimination, and a seven-gene signature was used to derive HNGCIscore. A simplified three-gene model based on GBP1, IDO1, and CD72 showed moderate internal performance in repeated stratified five-fold cross-validation. Immunohistochemistry further showed higher GBP1, IDO1, and CD72 protein expression in responders than in non-responders, supporting their association with immunotherapy sensitivity.

ICI-based retreatment after first-line immunotherapy failure appears feasible in selected patients with HER2-negative advanced gastric cancer, with measurable disease control and manageable toxicity. The TME-oriented HNGCIscore framework, supported by preliminary protein-level validation of GBP1, IDO1, and CD72, may help identify patients more likely to benefit from immunotherapy. These findings remain exploratory and require validation in larger prospective cohorts with matched transcriptomic, biomarker, and clinical response data.