Integration of Circulating Immune Checkpoint Proteins and Osteopontin Refined Risk Stratification in Osteosarcoma

Osteosarcoma remains the most common primary malignant bone tumor in children and adolescents, with largely unchanged survival outcomes in recent decades. Non-invasive biomarkers for preoperative risk stratification are urgently needed. Circulating immune checkpoint proteins (ICPs) and bone remodeling factors (BRFs) represent promising candidates; however, their combined prognostic value in osteosarcoma remains unclear.

We prospectively analyzed plasma samples from 47 patients with osteosarcoma, selecting the earliest available sample before disease progression or the last follow-up. A panel of ICPs and BRFs was quantified using multiplex immunoassays. sHVEM and sCD27 were used to construct a two-marker ICP signature (ICP2); subtypes were defined by unsupervised consensus clustering (k = 2) applied to log₂-transformed sHVEM and sCD27 concentrations, and sOPN was included for integrative analysis. Survival associations were assessed using Kaplan-Meier analysis and Firth's penalized Cox regression. The model performance was evaluated using Harrell's C-index.

Two ICP2 subtypes were identified (type1, n = 26; type2, n = 21). ICP2-type2 was associated with significantly worse progression free survival (PFS) (p = 0.0045) and overall survival (OS) (p = 0.0024) and remained an independent predictor after adjusting clinicopathological factors. The ICP2 model demonstrated high discriminative performance (C-index 0.924 for PFS and 0.903 for OS). Internal validation by leave-one-out cross-validation (LOOCV), in which consensus clustering was re-derived at each fold, yielded corrected C-indices of 0.612 for PFS and 0.806 for OS (versus apparent estimates of 0.687 and 0.689, respectively), confirming acceptable-to-good discrimination after accounting for overfitting. Elevated sOPN was associated with poorer survival in the exploratory analyses, although its prognostic value was less consistent under a pre-specified cutoff. In the combined models, sOPN provided modest additional value for OS stratification, with the grdgfdgICP2-type2/OPN-high subgroup exhibiting the poorest outcome.

The circulating sHVEM/sCD27 signature defines prognostically distinct osteosarcoma subtypes with a strong discriminative performance. This minimally invasive model may support preoperative risk stratification. Integration with sOPN suggests a potential biological link between immune regulation and bone remodeling, warranting validation in larger cohorts.