Diffuse large B-cell lymphoma (DLBCL) features an immunosuppressive tumor microenvironment (TME), yet the molecular drivers connecting metabolic reprogramming to immune evasion remain poorly defined. Here, we deployed an integrative single-cell transcriptomic analysis combined with a machine learning (ML) framework to systematically identify key immune-suppressive hubs in DLBCL. Through ML-driven prioritization of a 33-gene panel, PAICS emerged as a central node within an immunosuppressive B-cell subgroup. Functional assays confirmed that PAICS promotes lymphoma proliferation, survival, and tumor growth while establishing an immunosuppressive TME-marked by reduced IFN‑γ, elevated TGF‑β and IL‑10, and enhanced CD8⁺ T cell exhaustion. Mechanistically, we uncovered the IRF4-PAICS-LDHA axis: IRF4 transcriptionally activates PAICS, which physically interacts with LDHA to augment its activity, thereby skewing the NAD⁺/NADH balance toward metabolic immunosuppression. Importantly, our AI-aided approach not only identified this axis but also predicted its vulnerability to metabolic intervention: both methotrexate treatment and LDHA knockdown restored metabolic balance, reversed T‑cell exhaustion, and suppressed tumor growth. These findings highlight the power of ML in uncovering multi-targetable metabolic-immune networks and in guiding therapeutic strategies to overcome immune evasion in DLBCL.
Wang et al. (Thu,) studied this question.