Patterns, risk factors and management of CD19-directed chimeric antigen receptor T-cell therapy failure in CNS lymphoma

Abstract Background: CD19-directed Chimeric Antigen Receptor T-cell therapy (CD19-CAR) is highly effective in recurrent CNS lymphomas (CNSL), but most patients ultimately experience disease progression (PD) despite initial responses. Clear prognostic markers are currently lacking to identify patients at risk for CD19-CAR failure. Optimal salvage therapies, particularly those that could exhibit synergistic effects with CD19-CAR, remain to be defined. Methods: Clinical and radiological characteristics of CD19-CAR failure were analyzed in CNSL patients treated at Massachusetts General Hospital from 2018-2024. In-depth radiomic analyses of high-resolution contrast-enhanced magnetic resonance imaging (MRI) obtained at time of CD19-CAR infusion and at time of PD were carried out. PD patterns were defined as local (involving/adjacent to initial disease site) or distant (new disease site). CNS-progression-free survival from CD19-CAR infusion (CNS-PFS1) and first subsequent progression (CNS-PFS2) were analyzed using multivariable Cox proportional hazards models. Results: Our cohort included 60 patients with recurrent CNSL (primary 20/60, secondary 40/60) treated with CD19-CAR after a median of 3 (IQR: 2-3) prior lines of CNS-directed therapy including high-dose methotrexate (56/60) and rituximab (50/60) in most. CD19-CAR achieved a 60% overall response rate (ORR: 45% complete (CR), 15% partial response). Median CNS-PFS1 was 4 months with radiographic PD in 36/60 patients (progression identified as local 14/60; local and distant 10/60; distant 12/60). Local PD occurred early at a median of 1 month (IQR: 1-1) and was typically found in CD19-CAR refractory disease. Distant relapses developed significantly (p = 0.02) later at a median of 5 months (IQR: 2-11) and were associated with initial CR. To identify predictive biomarkers for CD19-CAR refractory disease, primary non-responders with progressive CNSL were compared to patients achieving a CR. While demographic, clinical, and histological characteristics were comparable in both groups, CNSL lesions with peripheral contrast enhancement (CE) patterns were significantly enriched in primary refractory disease (47% vs 0%; p < 0.001). Peripherally CE lesions were associated with lower ORR (28% vs 68%; p = 0.03) and shorter CNS-PFS1 (median 1 vs 8 months; p < 0.01) on univariate analysis. Aiming to delineate risk factors for progression after initial CR, patients with sustained CR (>12 months after CD19-CAR) were compared to those with early disease recurrence (≤12 months). Baseline leptomeningeal disease (LMD) was associated with early recurrence after initial CR (78% vs 12%; p = 0.02). While ORR to CD19-CAR was not significantly different, LMD correlated with shorter CNS-PFS1 (median 1.5 vs 8 months; p < 0.01) on univariate analysis. Peripheral CE and LMD at time of CD19-CAR infusion were forwarded into a multivariate Cox proportional hazard regression analysis. The model corrected for sociodemographic factors (age, sex), diagnosis (primary vs secondary CNSL), Eastern Cooperative Oncology Group (ECOG) performance status, and response to bridging therapy. Both peripheral contrast enhancement pattern (HR: 2.84; 95%-CI: 1.11-6.99; p = 0.030) and LMD (HR: 2.54; 95%-CI: 1.11-5.86; p = 0.027) independently correlated with shorter CNS-PFS1 on multivariate analysis. ECOG < 2 at infusion (HR 0.30, 95%-CI: 0.10-0.87; p = 0.027) and occurrence of tumor inflammation associated neurotoxicity (HR 0.29, 95%-CI: 0.07-0.98; p = 0.047) were associated with longer CNS-PFS1. Outcome after CD19-CAR failure was overall unfavorable with a median CNS-PFS2 of one month. Notably, peripheral CD19+-B-cell aplasia suggested persistence of CD19-CAR T-cells in 93% of patients at PD. Salvage immune checkpoint inhibition (pembrolizumab, nivolumab), and lenalidomide with rituximab or tafasitamab yielded prolonged responses in a subset of patients, often exceeding 5 months. Conclusions: Our study identifies novel radiological risk factors for CD19-CAR failure in patients with CNSL, namely peripheral CE and LMD prior to CD19-CAR, which may guide prognostic stratification at baseline. Outcome after CD19-CAR failure remains poor, underlining the need for rational salvage treatments. In patients progressing after CD19-CAR therapy, we noted encouraging responses after salvage ICI and lenalidomide combined with rituximab/tafasitamab, which warrant further investigation in prospective studies.