Paper Details

ABSTRACT

Background: Immune effector cell-associated neurotoxicity remains a major complication of CD19 CAR T-cell therapy, with no effective preventive treatments available to date. Lymphodepleting chemotherapy is administered prior to CAR T-cell infusion to promote expansion and antitumor activity by eliminating endogenous lymphocytes and the homeostatic cytokine sink. Importantly, approved CD19 CAR T-cell products differ not only in CAR construct and manufacturing but also in lymphodepletion regimens, which may contribute to differences in toxicity profiles. We sought to determine whether lymphodepletion-induced cytokine perturbations identify a biologically and clinically relevant pathway linking lymphodepletion depth to neurotoxicity. Methods: We performed an integrated analysis of serum, CSF, and clinical data from a discovery meta-cohort of 585 patients with B-cell non-Hodgkin lymphoma treated with axicabtagene ciloleucel (axi-cel) across ZUMA-1, ZUMA-5, ZUMA-7, and ZUMA-12. More than 3,000 longitudinal serum samples were analyzed. In addition, immune cell populations were characterized in 307 CSF samples from 139 patients and protein profiling was performed in 289 CSF samples from 142 patients. Twenty-six serum analytes were quantified at baseline (pre-lymphodepletion) and day 0 (post-lymphodepletion, pre-infusion). Key findings were validated in an independent axi-cel cohort (n=54; ZUMA-6/ZUMA-14) and an external cohort of patients (n=71) treated with 4-1BB-based CAR T-cell therapies, including lisocabtagene maraleucel, tisagenlecleucel, and CTL019 at the University of Pennsylvania. High-dimensional Olink® proteomics was performed in a subset of patients (n=241). Associations with toxicity were assessed using covariate-adjusted regression models. Survival outcomes were evaluated using log-rank testing, multivariable Cox proportional hazards modeling, and competing-risk analyses. Results: Among all analytes evaluated before CAR T-cell infusion, day 0 IL-15 emerged as the biomarker most strongly associated with the severity of neurotoxicity (FDR-adjusted P=9.31×10?7). This association remained significant after adjustment for disease histology, age, sex, prior therapies, ECOG performance status, disease stage, tumor burden, ferritin, LDH, and peak CAR T-cell expansion (standardized OR =1.69, P=1.4×10?5). In contrast, baseline analytes showed absent or modest associations with toxicity. IL-15 was largely undetectable before lymphodepletion but increased nearly ten-fold by day 0, representing the largest change among all analytes evaluated. Findings were independently reproduced in both axi-cel and 4-1BB validation cohorts. Olink profiling demonstrated that IL-15 exhibited the greatest increase following lymphodepletion among approximately 1,500 proteins assessed in axi-cel cohorts and approximately 5,000 proteins assessed in the 4-1BB validation cohort. Importantly, IL-15 induction scaled with lymphodepletion intensity. IL-15 fold changes after lymphodepletion were highest following cyclophosphamide/fludarabine 500/30 mg/m² administered for three days (used in axi-cel trials), intermediate following lower-intensity cyclophosphamide/fludarabine regimens, and lowest following bendamustine-based lymphodepletion (used in the 4-1BB cohort). Elevated day 0 IL-15 showed no association with CAR T-cell expansion. Instead, day 0 IL-15 and peak CAR T-cell expansion independently and additively contributed to severe neurotoxicity risk. CSF analyses demonstrated coordinated inflammatory networks in patients with severe neurotoxicity, peaking at day 7, characterized by elevations in IL-15, CAR T cells, monocytes, IFNγ, and IL-6. Systemic day 0 IL-15 levels correlated with the magnitude of subsequent CNS inflammation, linking lymphodepletion-induced perturbations to neurotoxicity pathogenesis. While day 0 IL-15 was not associated with treatment response, elevated levels were associated with inferior overall survival driven by increased non-relapse mortality. The 5-year cumulative incidence of non-relapse mortality was 26% in patients with high day 0 IL-15 compared with 4% in patients with low levels. Conclusions: These findings identify lymphodepletion intensity as a potentially modifiable determinant of severe neurotoxicity, support prospective evaluation of individualized or reduced-intensity conditioning strategies, and establish pre-infusion IL-15 as a practical biomarker for risk stratification and patients’ monitoring.