CAR-T efficacy in large B-cell lymphoma is shaped by neoantigen-driven epitope spreading and limited by HLA loss and genomic instability

Abstract INTRODUCTION CD19-directed chimeric antigen receptor T-cell therapy (CAR-T) has revolutionized large B cell lymphoma (LBCL) treatment. Resistance to CAR-T cannot be fully explained by LBCL loss of CD19 target (Plaks, Blood 2017). Emerging but inconclusive evidence suggests CAR-T promotes epitope spreading (ES), characterized by endogenous T cell (eCD8) activity against tumor antigens (Ag). Using orthogonal approaches including 1) bioinformatics of LBCL whole genome sequencing (WGS), 2) mathematical modeling recapitulating human data, 3) interferon gamma (IFNγ) ELISPOTs against patients (pts) autologous tumor, and 4) immunocompetent murine CAR-T lymphoma models, we reveal the critical role immunogenic tumor antigen recognized by MHC plays in CAR-T response. METHODS 1) We analyzed 61 LBCL WGS samples from 54 CAR-T treated pts (Jain, Blood 2022). Our robust workflow identified HLA class I mono- or biallelic loss. pVACseq predicted immunogenic clonal NeoAg, accounting for HLA alleles. 2) Optimization of an established math model of CAR-T dynamics (Kimmel, Proc B 2021) enumerated in silico enumeration of NeoAg in the context of HLA and/or CD19 loss. 3) Baseline (BL) and post-CAR-T (Post, CAR-depleted) LBCL tumor infiltrating lymphocyte (TIL) or peripheral blood mononuclear cells (PBMC) were cocultured with autologous tumor cell suspensions (T cell depleted) for 48h IFNγ ELISPOT. 4) Two immunocompetent murine models were employed, first Balb/c, with wild type (WT) A20 lymphoma or B2m knockout (B2MKO) A20 that lacked MHC-I expression; and WT or CD8KO C57BL/6 mice with BL3750 tumor. CAR-T were infused after tumor establishment and mice were followed for tumor growth and/or survival. In some experiments, mice were euthanized at two weeks and eCD8 or eCD4 T cells were FACS sorted and cocultured with A20 WT, A20 CD19KO (CD19-loss) or A20 B2MKO (MHC-I-loss), for IFNγ ELISPOT. RESULTS. We compared the number of predicted NeoAg between pts who experienced disease progression (progressors) and those who did not (responders). No significant difference was observed between groups. Interestingly, progressors exhibited a bimodal distribution of NeoAg burden. Within the high NeoAg subgroup (n=10), 6 patients harbored biallelic deletions in HLA class I complex (HLA-A, HLA-B, HLA-C and B2M) and/or CD58, impairing Ag presentation. After adjusting NeoAg counts to account for these cases, progressors showed a significantly lower NeoAg burden compared to responders (p=0.0157). These findings suggest the critical role of intact NeoAg presentation machinery in mediating the efficacy of CAR T-cell therapy in rrLBCL. 2) Our dynamic math model of CAR-T, previously parameterized on different pt data, proved the biological plausibility of cooperation between CAR-T (vs. CD19) and eCD8 (vs. NeoAg) to effect response, recapitulating our pt data: without LBCL CD19 or HLA loss, higher pre-treatment NeoAg associates with response. 3) In a precious set of surgical excisional LBCL biopsies taken shortly before CAR-T, post-infusion PBMC samples produced more IFNγ in response to autologous tumor compared to effectors from TIL or PBMC at BL (n=5 pts, p=0.0006), indicating enhanced tumor-specific reactivity of the eCD8 consistent with CAR-T–mediated epitope spreading. 4) Supporting these findings, in immunocompetent mice harboring eCD8 T cells, decreased response to CAR-T was seen against B2MKO vs WT tumor, and in CD8KO vs WT mice, highlighting a key role of eCD8 in lymphoma-clearance post-CAR-T. ELISPOT revealed that both non-adoptively transferred eCD4 and eCD8 from CAR-T treated mice had increased IFNγ response vs WT (eCD4 p=0.0002, eCD8 p=0.0009) and CD19KO (eCD4 p=0.0002, eCD8 p=0.0475) tumors compared to controls. eCD8 response was impaired against MHC-I deficient tumors (B2MKO) (p0.0001), indicating the antitumoral response is dependent on TCR recognition of non-CAR targets. CONCLUSION CAR T cell therapy facilitated the activation of eCD8 T cells against diverse LBCL Ag beyond the CAR target, highlighting a mechanism that contributes to CAR-T efficacy. This mechanism of response is thwarted in some LBCL patients via genomic instability impairing NeoAg expression and recognition by eCD8 T cells. Understanding and harnessing epitope spreading induced by CAR-T may offer new strategies to enhance therapeutic efficacy and overcome tumor Ag escape in patients with LBCL. CF, BZ, VS, and CG contributed equally AA, JS, FM, and FL are co-senior authors