Plasma proteomics for prognostic stratification in newly diagnosed primary central nervous system lymphoma.

BACKGROUND

Baseline biomarkers anticipating outcome in primary central nervous system lymphoma (PCNSL) are limited, and classical scores rely on clinical variables alone. We present a plasma proteomic framework for prognostic stratification of newly diagnosed PCNSL.

METHODS

Baseline plasma from 162 newly diagnosed PCNSL patients (French LOC network; training n=84; test n=78) was profiled using Olink Reveal (1,007 proteins). Consensus k-means clustering applied to the training proteome was projected to the test cohort, evaluated for immune-signature enrichment, and assessed by multivariable Cox regression. An elastic-net classifier predicted early treatment failure, evaluated by held-out and nested cross-validated AUC.

RESULTS

Clustering identified 2 proteomic states, Inflamed and Non-inflamed plasma, that replicated in the test cohort. The Inflamed state was enriched for pre-specified immune signatures, most strongly an Innate/Inflammatory signature. High Innate/Inflammatory scores marked shorter overall (OS) and progression-free survival (PFS) in training (log-rank P=.0014 OS, P=.0037 PFS) and test (P=5.4×10⁻⁴ OS, P=.0014 PFS), remaining significant after adjustment for MSKCC class and sex (test OS HR 2.51, P=.0022). Combining signatures with clinical variables improved test discrimination, most clearly for PFS (C-index 0.68 vs. 0.60). A supervised classifier predicted early treatment failure (PFS < 9 months) at test AUC 0.71, driven by the immune signatures rather than a whole-proteome search.

CONCLUSIONS

Plasma proteomics identifies a reproducible Inflamed state in PCNSL associated with shorter OS and PFS that improves discrimination beyond clinical scores and stratifies the risk of early treatment failure. These findings support prospective evaluation of a focused innate-inflammatory plasma panel for risk stratification.