Abstract
Introduction Primary central nervous system lymphoma (PCNSL) is a rare and aggressive subtype of diffuse large B-cell lymphoma (DLBCL) confined to the CNS. Immune escape in CNS lymphomas (CNSL) limits the efficacy of immune therapies, yet high-resolution immune maps of PCNSL, secondary CNS lymphoma (SCNSL), and DLBCL remain unavailable. Due to the rarity of PCNSL, the largest scRNAseq cohort contains 8 PCNSL patients with only 1 study comparing PCNSL, SCNSL, and DLBCL with 6, 2 and 5 samples, respectively (Liu et al., 2023; Wang et al., 2025). Here, we present the largest and most comprehensive single-nucleus atlas of CNSL to date, doubling the total number of PCNSL and SCNSL cases previously reported across all published studies.
Methods We employed the 10x Genomics Fixed RNA Profiling (FLEX) platform to perform single-nucleus RNA sequencing (snRNA-seq) on fresh frozen (FF) and FFPE biopsies from 20 PCNSL, 4 SCNSL and 5 DLBCL patients. Samples were run using the 16plex barcoding system inbuilt into the FLEX assay. Harmony batch correction was utilized to integrate FF and FFPE samples following quality control in Seurat and doublet removal using the Demuxafy pipeline.
Results We identified 127,824 high-quality nuclei revealing 16 major cell types, including CD19+ B cells, CD4+ and CD8+ T cells, CD14+ myeloid cells, glial, neuronal and endothelial cells. Differential composition analysis using sccomp revealed, no major TME composition differences between PCNSL and SCNSL, except for decreased fibroblasts and increased NK cells in PCNSL. However, PCNSL samples exhibited significantly lower frequencies of M1-like macrophages, C1Q+ macrophages, and dendritic cells. While known drivers of tumor immune evasion, SPP1+ macrophages and SPP1+ microglia were elevated (adj-p<0.05) compared with DLBCL. PCNSL also exhibited significantly reduced naïve CD4+ and CD8+, TH1 and exhausted CD8+ T cells compared with DLBCL, while displaying increased Tfh cell numbers (adj-p<0.05).
Ligand–receptor analysis, revealed distinct interaction networks across disease types, with 218 interactions unique to SCNSL, 32 to PCNSL, and only 9 to DLBCL. While 91 interactions were shared across all three, CNSL shared 192 unique interactions absent from DLBCL, defining a CNS-specific immune architecture. In contrast, PCNSL and DLBCL shared only 11 interactions, and SCNSL and DLBCL shared 16, underscoring divergent immune communication in CNSL vs. DLBCL.
PCNSL and SCNSL exhibited significantly more immune checkpoint interactions than DLBCL (72 and 58 vs 23; PCNSL vs DLBCL p=0.0051, SCNSL vs DLBCL p=0.0128) These interactions primarily involved macrophages signaling with CD4+ or CD8+ T cells through, PD-L1-PD-1 and PD-L2–PD-1 in PCNSL and DLBCL but not in SCNSL. Gal-9–TIM-3 interactions were dominant in SCNSL but also present in PCNSL, and absent in DLBCL. These data suggest CNSL subtypes deploy distinct immune checkpoint pathways, PD-L1-PD-1 and PD-L2–PD-1 dominating in PCNSL and Gal-9–TIM-3 in SCNSL, supporting tailored immunotherapy approaches.
Cyclophilin-A-CD147 interactions between B cells and immune and non-immune cell types were identified in PCNSL, mirroring glioblastoma immune escape pathways (Xu et al., 2020). SCNSL B cells predominantly interacted with endothelial cells via NOTCH1/2–DLL4/CNTN1/JAG2 and VEGFB–NRP1, suggesting roles in angiogenesis and CNS infiltration. Across all subtypes, B cell–myeloid interactions via Gal-9–P4HB and PTPRC–CD22 suggest suppression of cytotoxic T cell activity.
Conclusion This study presents the largest snRNAseq atlas of CNS lymphomas to date, profiling PCNSL, SCNSL, and DLBCL. PCNSL displayed increased SPP1+ myeloid cells compared with DLBCL, which are known drivers of tumor immune evasion. Whilst showing little difference in TME composition with SCNSL, ligand–receptor analysis uncovered CNS-specific immune checkpoint signaling, with PD-L1-PD1, PD-L2–PD-1 and Gal-9–TIM-3 interactions dominating in PCNSL and SCNSL, respectively. This study provides the largest snRNA-seq atlas of CNS lymphomas to date, revealing CNS-specific immune checkpoint pathways (PD-L1–PD-1, PD-L2–PD-1, Gal-9–TIM-3) and SPP1+ myeloid enrichment in PCNSL. These insights nominate distinct immune suppressive programs as tractable therapeutic targets for rational immunotherapy in CNSL.
