AUTHORS
Anastasia Radko, MSc, Tatiana Vasileva, MS2, Diana Lupova, MSc3, Ivan Zubarev, MSc4, Elizaveta Khotlubei, MSc5, Anastasiia Bolshakova, MSc6, Svetlana Bezlepkina, MSc7, Susanna Kirakosyan, MD8, Elina Ohanjanyan, BSc9, Olga Lukonina, MSc10, Ilya Fedin, MSc11, Ksenia Markova, PhD12, Michael Goldberg, PhD13
AFFILIATIONS
1 BostonGene Corporation, Waltham, Massachusetts, United States
2 BostonGene Corporation, Waltham, Massachusetts, United States
3 BostonGene Corporation, Waltham, Massachusetts, United States
4 BostonGene Corporation, Waltham, Massachusetts, United States
5 BostonGene Corporation, Waltham, Massachusetts, United States
6 BostonGene Corporation, Waltham, Massachusetts, United States
7 BostonGene Corporation, Waltham, Massachusetts, United States
8 BostonGene Corporation, Waltham, Massachusetts, United States
9 BostonGene Corporation, Waltham, Massachusetts, United States
10 BostonGene Corporation, Waltham, Massachusetts, United States
11 BostonGene Corporation, Waltham, Massachusetts, United States
12 BostonGene Corporation, Waltham, Massachusetts, United States
13 BostonGene Corporation, Waltham, Massachusetts, United States
ABSTRACT
Introduction:
Autologous chimeric antigen receptor (CAR) T cells have transformed treatment for relapsed and refractory (r/r) hematologic malignancies, but severe toxicities, including cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS), limit their earlier deployment. Driven by patient T-cell heterogeneity, construct design, and manufacturing variations, toxicity risks can vary drastically across products. Here, we used integrative RNA-seq analysis of CD19-directed CAR-T cohorts to uncover key product- and patient-specific immune profiles associated with ICANS and CRS risk.
Methods:
Infusion-product transcriptomic profiles were compiled from three B-cell lymphoma cohorts receiving CD19-directed CAR T-cell therapy: C1 (N=41; axi-cel n=30, tisa-cel n=11), C2 (N=24; axi-cel), and C3 (N=6; axi-cel). Single-cell RNA-seq datasets were aggregated into sample-level pseudobulk profiles. Patients were categorized by toxicity severity (CRS-high: grade ≥2; ICANS-high: grade ≥3; low: below thresholds). Group-level signature scores were derived using GSEA against publicly curated gene sets to obtain normalized enrichment scores (NESs), while patient-level scores depicting specific immune states were derived using ssGSEA for previously validated immune signatures. Signature scores were compared between high- and low-toxicity groups within each dataset. Shared toxicity-associated signatures were derived from all datasets, while product-specific signatures were assessed only within mixed-product cohorts and then evaluated for overlap with CRS/ICANS features for each product.
Results:
CAR-T products showed distinct transcriptional profiles independent of CRS/ICANS status. Axi-cel samples showed higher T-cell activation and exhaustion, senescence-like (median difference 364.16 to 1,765.82, p<0.05), and myeloid-associated signature scores (median difference 1,818.17, p<0.001) relative to tisa-cel. Interestingly, proliferation-associated signature scores were highest in tisa-cel.
CRS-high axi-cel samples consistently showed elevated hypoxia (NES 2.49, FDR<0.001), TNF/NFKB-associated signaling (NES 2.23, FDR<0.001), and glycolysis (NES 1.62, FDR=0.02). Conversely, CRS-high tisa-cel samples exhibited lower but broader inflammatory signaling (NES −1.92 to −1.66, FDR<0.01) paired with oxidative phosphorylation/proliferative metabolic signatures (NES 2.10 to 2.52, FDR<0.05). Across products, CRS-high samples shared attenuated antitumor cytokine and monocyte-chemokine (CCL4) signaling (median difference −2,079.28 to −209.61, p<0.05).
In aggregate and single product analyses, ICANS-high samples shared a unique inflammatory and senescence-like profile: elevated TNF, IL6, and interferon signaling (NES 1.46 to 2.67, FDR<0.05), myeloid/monocyte activity (median difference 328.97 to 3,539.12, p<0.05), and senescence/SASP programs (median difference 426.74 to 1,147.64, p<0.05) paired with reduced cell proliferation (median difference −289.93, p=0.03).
Summary/Conclusion:
Our results establish a two-axis risk model of CAR-T toxicity. First, a product-intrinsic axis driven by baseline transcriptional divergence and product-specific CRS signatures highlights the critical impact of construct design and manufacturing processes. Second, a patient-intrinsic axis is driven by conserved, product-agnostic CRS and ICANS profiles. This framework offers an actionable roadmap for risk mitigation: product-dependent profiles can guide manufacturing optimization and quality control, while patient-dependent profiles can be engineered into generalizable pre-infusion biomarkers for clinical stratification and targeted intervention.
DISCLOSURES
Michael Goldberg, PhD: BostonGene Corporation - Current holder of stock options in a privately-held company; BostonGene Corporation - Current equity holder in private company; BostonGene Corporation - Patents & Royalties; BostonGene Corporation - Current Employment