Blood and CSF metabolomics identifies tryptophan catabolism and polyamine synthesis as drivers of CAR T-cell-associated neurotoxicity

Abstract Background: Anti-CD19 chimeric antigen receptor (CAR) T-cell therapies have demonstrated durable responses and are approved for the treatment of relapsed/refractory large B-cell lymphoma (LBCL), B-cell acute lymphoblastic leukemia, mantle cell lymphoma, and follicular lymphoma. However, severe neurological events (NEs), whose underlying mechanisms remain elusive, have limited effective interventions. Here we investigated the relationship between metabolic signatures and severity of NE following treatment with axicabtagene ciloleucel (axi-cel) or brexucabtagene autoleucel (brexu-cel). Methods: We performed global metabolomic profiling on 3,788 longitudinal serum and plasma samples from a multi-trial meta-cohort of 686 patients treated with axi-cel (ZUMA-1, -5, -7, -12, -24) or brexu-cel (ZUMA-2). Additionally, we analyzed 60 cerebrospinal fluid (CSF) samples from 26 LBCL patients in ZUMA-1. Associations between metabolite levels and NE severity or clinical outcomes were assessed using mixed-effects models. Correlations were performed between metabolites and protein inflammatory markers in blood and CSF, immune cell types in blood and CSF, CAR T cell pharmacokinetics (PK) in blood, product attributes and blood chemistry laboratory values. Multivariate machine learning (ML) models were employed with the dataset split between discovery and validation cohorts. Results: Patients who developed high-grade (grade ≥ 3) NE exhibited markedly increased tryptophan catabolism in peripheral blood, resulting in elevated levels of excitotoxic metabolites such as quinolinate both pre-CAR T cell infusion (baseline) and post-infusion (P < 0.0001). This accumulation can promote oxidative neuronal damage through generation of reactive oxygen species, mitochondrial dysfunction, and lipid peroxidation. Similarly, pre-treatment and post-infusion plasma or serum from patients with high-grade NE showed increased flux through the arginine–citrulline pathway, favoring elevated production of urea and acetylated polyamines including N1,N12-diacetylspermine (P < 0.0001). This metabolic shift could reflect enhanced arginase-mediated conversion of arginine to ornithine, followed by polyamine biosynthesis via ornithine decarboxylase. We further identified alterations in other pathways: the glutamine–glutamate cycle, which maintains neurotransmitter homeostasis in the CNS; NAD⁺ metabolism, linked to tryptophan catabolism through the kynurenine pathway; glycerophospholipids synthesis, which can be released by phospholipase activity during excitotoxic states; and balance of neurotransmitter levels. Consistently, in CSF, individuals with severe NE exhibited elevated glutamate levels (P = 0.019), along with increased concentrations of kynurenine and quinolinate (P = 0.003) at the time of the neurotoxic event. These alterations suggest disruption of the glutamine-glutamate cycle driven by accumulation of extracellular glutamate and quinolinate which may contribute to excitotoxic activation of N-methyl-D-aspartate (NMDA) receptors. Based on these observations, we developed metabolic pathway scores derived from the above altered metabolites. At the time of NE, these scores demonstrated a stronger association with severe NE than a composite inflammatory score derived from serum proteins (e.g., IL-6, TNFα), which are known correlates of CAR T-cell–mediated NE. Notably, increased tryptophan catabolism leading to higher levels of quinolinate, and elevated levels of polyamines, such as N1, N12-diacetylspermine were also associated with disease progression (P = 0.046, HR = 0.78 [0.61–1.00] and P < 0.0001, HR = 2.1 [1.64–2.68], respectively), suggesting potential therapeutic targets. At the time of NE, quinolinate and other tryptophan pathway intermediates (such as xanthurenate, picolinate) correlated not only with inflammatory markers, CAR T-cell PK, immune cell proliferation and effector markers, but also with CD14+ monocytes in CSF (P<0.0001, R = 0.44). Finally, ML models corroborated the relevance of the tryptophan pathway for association with severe NE. Conclusions: This study represents the largest metabolomic analysis in the context of CAR-T cell therapy. We identified metabolic pathways associated with severe neurotoxicity following anti-CD19 CAR T-cell therapy. The results support strategies for early identification and mitigation of neurotoxicity, including targeting tryptophan catabolism, NMDA receptor-mediated excitotoxicity and acetylated polyamine synthesis.