With nearly 70% of esophageal squamous cell carcinoma patients failing to benefit from neoadjuvant immunotherapy, the urgent need for reliable predictive biomarkers has shadowed this field for years. A new framework reported in Cancer Discovery may fundamentally reshape how oncologists select patients for PD-1 blockade — moving beyond tissue biopsy alone toward a blood-based, multimodal prediction system.

Plasma proteomic profiling across patient cohorts identified soluble glycoprotein nonmetastatic melanoma protein B (sGPNMB) as the most consistently elevated circulating protein among immunotherapy non-responders. The mechanistic picture is notable for its specificity: tumor-secreted sGPNMB suppresses CD8+ T-cell receptor signaling through the SDC4-CD148 axis, driving functional exhaustion of anti-tumor immune cells. Upstream, cancer-associated fibroblast–epithelial (CAF-Epi) niche interactions promote SOX2 upregulation in tumor cells, which transcriptionally activates GPNMB expression. Critically, secretion — not just expression — appears necessary for the immunosuppressive effect. In humanized patient-derived xenograft models, GPNMB inhibition showed synergistic benefit when combined with PD-1 blockade. A multimodal predictive model combining circulating GPNMB levels, spatial CAF-Epi niche detection, and clinicopathologic variables demonstrated robust predictive accuracy across both retrospective cohorts and a prospective clinical trial.

This work is significant for several reasons beyond the single finding. GPNMB has previously been explored in melanoma and glioblastoma contexts, but its role as a measurable, functionally active immunosuppressive mediator in a gastrointestinal malignancy — particularly one with limited treatment options — opens a credible new biomarker axis. The integration of spatial tumor microenvironment features with liquid biopsy data into one predictive framework represents an emerging paradigm in precision oncology. Limitations worth noting include the retrospective design of most validation cohorts and the reliance on xenograft rather than fully immunocompetent animal models for mechanistic confirmation. Prospective validation in larger, multi-institutional trials will be essential before clinical adoption. Nonetheless, this is a potentially paradigm-shifting contribution to immunotherapy stratification in esophageal cancer.