A cross-platform transcriptomic risk score integrating LymphoMAP microenvironment archetypes and immune...

Diffuse large B-cell lymphoma (DLBCL) is clinically heterogeneous and existing prognostic indices incompletely capture its biological diversity. The present study aimed to investigate how transcriptome-inferred LymphoMAP archetypes (lymph node-like, fibroblast-macrophage-rich and T cell-exhausted) relate to genetic subtypes and immune evasion-associated programs, and whether integrating these dimensions improves risk stratification beyond the International Prognostic Index (IPI). Using the DLBCL-2018 cohort (n=562), LymphoMAP archetypes were inferred and immune programs were quantified by single-sample gene set enrichment analysis, summarizing them into an immune evasion-associated index. In the immunochemotherapy-treated subset (n=234), overall survival was assessed and an elastic-net Cox model was developed to derive a transcriptomic risk score (RScore-Expr). The score was externally evaluated in four independent Gene Expression Omnibus cohorts (n=1,173) with a random-effects meta-analysis performed across studies. LymphoMAP archetypes were non-randomly associated with genetic subtypes (χ²=20.85, P=0.0076) and exhibited distinct immune-program patterns. While archetypes alone were not prognostic in the treated subset (log-rank P=0.67), an integrated model identified high-risk patients (log-rank P=0.0026) with modest discrimination (C-index: 0.624; time-dependent area under the curve values: 0.662/0.628/0.642 at 12/36/60 months). When added to the IPI, the score improved corrected discrimination (0.639 to 0.687), improved model fit (likelihood-ratio χ²=12.32, P=4.48×10^-4), and yielded higher net benefit on decision-curve analysis at 24 and 60 months. In external cohorts, RScore-Expr showed a consistent association with survival (pooled hazard ratio per 1 standard deviation, 1.13; 95% confidence interval 1.01-1.26; P=0.033). These results support a cross-platform framework linking microenvironment archetypes, immune evasion-associated programs and tumor genetics. The novelty of the current study lies in integrating these features into an externally evaluated transcriptomic risk score and formally demonstrating added value beyond the IPI in DLBCL.