Abstract
Anti-PD-(L)1 therapies are widely used for the treatment of distinct cancers, but the proportion of patients who benefit from the treatment remains low, and tools for patient selection are suboptimal. In this study, we investigated whether plasma proteomics could explore the benefit of anti-PD-(L)1 antibodies. This prospective study enrolled 56 patients with multiple solid tumor types treated with anti-PD-(L)1 therapies. Pre-treatment plasma samples were profiled with a 92-plex immuno-oncology proximity extension assay panel. Associations with radiographic outcomes and survival were examined. Prediction models for ORR, DCR and OS were evaluated. Analyses were performed in a pooled hypothesis-generating framework. Several proteins, including ADGRG1 and ARG1, were linked to radiographic response. ARG1 achieved ROC-AUCs 0.68 (ORR) and 0.53 (DCR). We generated machine learning models for predicting both radiographic responses and long-term survival (≥ 720 days). For overall survival, the six-protein Cox proportional model separated risk groups (hazard ratio (HR) 4.8; CI 95% confidence interval (CI) 2.1–10.9), while larger panels increased separation, they are prone to over-fit and require validation. Unsupervised clustering did not show dominant segregation by tumor type. Our findings from this small prospective heterogeneous cohort support that plasma proteomics show promising but preliminary results in predicting the benefit from anti-PD-(L)1 therapies in multiple advanced solid tumors. Findings regarding DCR and ORR modeling should be interpreted as exploratory and require external validation, both tumor-specific and pan-cancer, before clinical application. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1038/s41598-025-31229-1.