输注前血浆蛋白质组学识别可预测复发/难治性淋巴瘤 CAR-T 细胞治疗后重度细胞因子释放综合征与神经毒性的内皮-免疫启动特征

Severe cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS) remain frequent, life-threatening complications of CD19 chimeric antigen receptor (CAR) T-cell therapy and constrain its safety, scalability, and outpatient adoption. Existing predictive models lack sufficient external validation for routine clinical use, and pre-infusion biomarkers that capture host susceptibility before infusion are urgently needed.

We applied unbiased mass-spectrometry-based proteomics to pre-infusion biofluids from 98 prospectively-followed adults with relapsed/refractory (r/r) lymphoma at two academic centers (MD Anderson Cancer Center, n = 39, plasma; Moffitt Cancer Center, n = 59, serum). Logistic regression with backward feature selection on the MD Anderson cohort yielded panels for severe (Grade 2) CRS and ICANS that were locked and tested without refitting on the Moffitt cohort. Patients were stratified into low-, intermediate-, and high-risk tertiles. Ingenuity Pathway Analysis defined upstream regulators and canonical pathways. The 17 CRS-associated and 21 ICANS-associated consensus proteins were classified into mechanistic themes for biological interpretation.

A 5-marker CRS panel (SCRIB, MYL6, MTHFD1L, HSP90B1, MMP2) achieved AUCs of 0.85 (95% CI 0.72-0.98) and 0.76 (0.63-0.89) in the discovery and validation cohorts, respectively. An expanded 8-marker ICANS panel (the CRS panel plus SPOCK2, SLC3A2, CD84) achieved AUCs of 0.91 (0.81-1.00) and 0.67 (0.51-0.84). In the combined dataset, high-risk-tertile patients were 13.84-fold (95% CI 4.21-56.26) and 8.59-fold (2.87-29.09) more likely to develop Grade 2 CRS and ICANS, respectively. Pathway analysis converged on AKT-driven inflammation and endothelial activation. Functional clustering of the consensus proteins partitioned into mechanistically coherent themes consistent with a dual-anatomy model: severe CRS reflected peripheral macrophage priming and endothelial activation with surplus complement amplification (HSP90B1 , CSF1 , MMP2 , HEG1 , C3 ) and endotheliopathic coagulation (PROC , F7 ), whereas severe ICANS reflected cerebrovascular junction and basement-membrane stripping (CDH5 , ITGB1 , FN1 , brain-enriched SPOCK2 ), hepatic synthetic suppression (TTR , APOA2 , IGFBP3 ), compromised plasma antioxidant capacity (GPX3 , PON1 ), and inflammasome dis-restraint via DPP9 . PGLYRP2 and SCRIB depletion were shared by both signatures and identified a common upstream priming substrate.

Externally validated, pre-infusion proteomic panels predict severe CRS and ICANS following CAR T-cell therapy and define a coherent pre-infusion endothelial-immune priming axis (HSP90B1, MMP2, AKT) with mechanistically interpretable, druggable nodes. The dual-anatomy framework distinguishes peripheral CRS-biased from cerebrovascular ICANS-biased phenotypes downstream of a shared microbiome-host barrier priming substrate, providing a foundation for biomarker-guided risk stratification and cluster-matched prophylactic intervention to enhance the safety and outpatient feasibility of CAR T-cell therapy.