Multi-modal profiling identifies CD4+CXCR5+PD-1- Tfh cells as prognostic and predictive biomarkers for response to R-CHOP therapy in human DLBCL

Despite the improvements in clinical outcomes for patients with Diffuse Large B-Cell Lymphoma (DLBCL), a significant proportion of those patients still face challenges with refractory/relapsed (R/R) disease after receiving first-line R-CHOP treatment. Characterizing the heterogeneity of the tumor microenvironment (TME) in diffuse large B cell lymphoma (DLBCL) is crucial for understanding relapsed/refractory disease. However, the complex and diverse nature of the TME has impeded progress. Here, using single-cell RNA sequencing(scRNA-seq), we explored the DLBCL landscape at single-cell resolution, profiling 77,344 cells from both primary and relapsed DLBCL patients. We further investigated the shared and distinct molecular and cellular features of tumor microenvironment in both primary and relapsed DLBCL tumors by integrating next-generation DNA sequencing data and multiple scRNA-seq datasets from total 72,351 cells. Our results demonstrated that there was a significant decrease of CD4+CXCR5+PD-1- Tfh cells, along with excessive activation of the TNF-NFκB signaling pathways in malignant B cells in R/R disease. Furthermore, by using multiplex IHC, we confirmed that CD4+CXCR5+PD-1- Tfh cells are prognostic and predictive biomarkers for response to R-CHOP treatment. As a proof of concept, we successfully generated mutilplexed images of CD4+CXCR5+PD-1- Tfh cells from a single DAPI staining in human DLBCL tissues through using generative artificial intelligence. Our study offers critical insights into the heterogeneity and molecular features of DLBCL, shedding light on the crucial role of CD4+CXCR5+PD-1- Tfh cells in the recurrent process.