Humanized PDX models reveal Lymphoma–Host interactions and enhanced T-cell therapies via epigenetic modulation

Abstract

Abstract Background: Diffuse large B-cell lymphoma (DLBCL) is a biologically and molecularly heterogeneous malignancy with variable therapeutic responses. Distinct transcriptional and genetic profiles drive the interactions between malignant cells and the host immune system. To explore this complex ecosystem, we analyzed DLBCL biopsies from patients receiving chemotherapy (R-CHOP) or immunotherapy (chimeric antigen receptor CAR T cells) using integrated omics and functional assays, together with fully humanized mouse models (HuMice). Methods: We analyzed 68 DLBCL cases (48 post R-CHOP and 20 post CAR T-cell–treated) and 48 matched patient-derived xenografts (PDX) by bulk RNA-seq, single-cell RNA/TCR-seq, and WES/CAPP-seq. Custom bioinformatics pipelines were also applied to large DLBCL datasets to assess prediction classifiers (Lacy et al., Blood 2020; Schmitz et al., NEJM 2018; Chapuy et al., Nat Med 2018; Li X, et al, Cancer Cell 2025). Humanized mice (HuMice) were generated by engrafting CD34+ Human Stem Progenitor Cell (HSPC) from heterologous/allogeneic donors into NSG-SGM3xMHCI/II-DKO and NSG-SGM3xNBSGW strains. Multiplex flow cytometry and SignalStar™ multiplex IHC (20 Abs) defined human immune reconstitution and spatial phenotypes. CosMx SMI profiling (6000 RNA, 64 proteins) was performed on an extended cohort of primary tumors (112 R-CHOP, 20 CAR T), and treated PDX (9 models), HuMice, and HuPDX with/without CART19 or BAFFR. Results: Using dimensionality reduction, we identified a 12-gene signature of epigenetic regulators and DNA repair genes (e.g., DNMT1, EZH2, HELLS, etc.), distinguishing validated immuno-deserted (ID) vs. immuno-rich (IR) lymphoma microenvironments based on mIHC and CosMx host immune representations. The ID-DLBCL patients displayed poor outcomes among both R-CHOP-(p < 0.0001) and CART19-treated DLBCL patients (p = 0.002) when compared to IR classified patients. To investigate host–lymphoma interactions, we engrafted CD34+ HSPCs to obtain humanized mice and subsequently implanted ID- and IR-PDX models. Mimicking the clinical outcomes, humanized mice with IR-DLBCL PDX showed robust anti-lymphoma immunity and robust T-cell expansions from both heterologous and allogeneic donors, resulting in tumor control. HuMice implanted with ID-DLBCL failed to mount an immune response and showed unimpeded growth. Flow cytometry, mIHC, and CosMX analyses revealed that ID-DLBCL-HuPDX were immune-depleted, with sparse Tumor-Infiltrating Lymphocytes (TILs) at lesion margins, whereas IR-DLBCL-HuPDX displayed abundant TILs (p < 2e-04) and clonal expansion. TILs were primarily represented by CD8+ central memory and exhausted subsets, but were unable to kill matched DLBCL cells in vitro, on rechallenge. However, splenic T-cells from IR-DLBCL-HuPDX exhibited strong proliferation, clonal expansion, and cytotoxicity against matched tumors. To enhance responses in ID-DLBCL-HuPDX, we performed dose-response screens of single- and dual-drug combinations using clinical phase compounds shown to improve immune responses (e.g., valemetostat, DNMT1i, GSK-3685032, azacytidine, belinostat, romidepsin, and lenalidomide) and identify synergistic associations (Vale+GSK, Vale+ Lena: Synergy Score >20). Drug exposure led to significant epigenetic and chromatin changes, fostering transcriptional shifts by converting ID-DLBCL into IR-DLBCL (p < 0.05–0.0005) and modulating key lymphoma pathways. Treatments were associated with improved CART killing. We then tested two different CART products (CD19 BAFFR) in vivo in nine ID- and IR-PDX/DX-HuMice models. Both CART therapy effectively controlled IR-DLBCL but not ID-DLBCL. Neither valemetostat nor lenalidomide (50 mg/kg, 14–21 days) significantly enhanced CART responses in IR-DLBCL (p < 0.05). Drug-treated ID-DLBCL-HuPDX showed significant transcriptional changes, with enrichment of glycolysis, MYC, E2F, G2M, and inflammatory signaling pathways (TGF, IFNγ/α, TNFα), supporting a TILs/CART expansion and activation and tumor killing. Conclusions: Epigenetic therapies reprogram the transcriptome, converting ID- into IR-DLBCL, and enhancing T-cell infiltration and tumor immunorecognition in HuMice. Humanized PDX accurately reflect the DLBCL physiopathology and immune infiltration profiles, providing a strong platform for validating next-generation immunotherapies and studying host-lymphoma interactions. A novel classifier underscores the value of companion diagnostic classifiers in improving patient care.