Humanized DLBCL-PDX models reveal drug-reversible epigenetic immune exclusion driving therapeutic outcomes and patient stratification

Diffuse Large B-cell Lymphoma (DLBCL) is biologically and molecularly heterogeneous. Clinical outcomes remain suboptimal, with limited durable remission in refractory or relapsed patients, even following Chimeric Antigen Receptor T-cell (CART) therapy. No validated biomarker currently predicts treatment response or elucidates the role of the tumor immune microenvironment. We developed an epigenetic core (REPIX) 12-gene classifier that predicts responses to immunochemotherapy or CART therapy and identifies immuno-rich (IR/REPIX-low) and immuno-depleted (ID/ REPIX-high) DLBCL. Validation in humanized DLBCL-PDX models confirmed superior outcomes in IR-DLBCL concordant with REPIX predictions. CART19/BAFFR CART therapy was effective in REPIX-low PDXs; conversely, REPIX-high PDXs were refractory to CART therapy, which was overcome by EZH1/2 inhibition and lenalidomide treatment. Mechanistically, drug exposure induced chromatin remodeling and a switch from REPIX-high to REPIX-low, promoting T-cell infiltration and immunorecognition.The REPIX classifier enables upfront patient-specific regimen selection, and HuMice-PDX-DLBCL dissect the tumor microenvironment, allowing the testing of innovative immune-based strategies.