A single-cell atlas of large B cell lymphomas reveals distinct malignant archetypes predictive of clinical outcomes.

Large B cell lymphomas (LBCLs) exhibit significant heterogeneity which leads to disparate treatment response, with up to 40% of patients developing refractory disease or relapsing within the first two years. To understand the cellular and molecular basis of this heterogeneity, we performed single-cell RNA-seq, BCR-seq, and TCR-seq on LBCL tumor biopsies, generating an atlas of 63 LBCL cases, mostly classified as diffuse LBCL, not otherwise specified (DLBCL NOS). We identified five conserved transcriptional archetypes of malignant B cells that co-occur within individual tumors, with varying proportions across patients. Notably, high abundance of Archetype 4, characterized by memory B cell features and quiescence markers, correlated with poor event-free survival following standard immunochemotherapy, a finding which was validated in independent cohorts through deconvolution of bulk RNA-seq data. Our study provides a novel framework for patient stratification based on the quantification of malignant cellular states, with implications for precision therapy of LBCLs.