CD4⁺ T cells mediate CAR-T cell-associated immune-related adverse events after BCMA CAR-T cell therapy

Abstract

B cell maturation antigen (BCMA)-targeted chimeric antigen receptor (CAR) T cell therapy has revolutionized the treatment of multiple myeloma but can cause unique toxicities, including cranial nerve palsy, parkinsonism and enterocolitis, which we refer to collectively as CAR T cell therapy-associated immune-related adverse events (CirAEs). Among 198 patients treated with ciltacabtagene autoleucel or idecabtagene vicleucel (June 2021–December 2024), 27 (13.6%) developed CirAEs. This included one remarkable case with three distinct CirAEs in association with an extreme CD4⁺ CAR T cell expansion (peak lymphocytes: 197 × 10³ per microliter), which was abrogated in vitro by CCR5 inhibition. CirAEs were associated with significantly higher non-relapse mortality (hazard ratio = 5.2, P = 0.006), and independent risk factors included ciltacabtagene autoleucel (odds ratio = 4.5, P = 0.058), peak absolute lymphocyte count ≥ 2.4 × 10³ per microliter in the first 14 days post-infusion (odds ratio = 4.3, P < 0.001) and apheresis CD4:CD8 ratio > 1 (odds ratio = 2.6, P = 0.048). We identified marked CD4⁺ CAR T cell infiltration in all available CirAE tissues, including cerebrospinal fluid during neurologic CirAEs, implicating CD4⁺ CAR T cell therapy as a key mediator of these toxicities.

BCMA-targeted CAR T cell therapy is now standard of care for relapsed/refractory multiple myeloma (RRMM). In the United States, two BCMA-targeted CAR T cell therapy products, idecabtagene vicleucel (ide-cel) and ciltacabtagene autoleucel (cilta-cel), are approved¹. These therapies yield deep, durable responses in RRMM^2–8 and are being investigated in the front-line and high-risk smoldering multiple myeloma settings^9–12. However, longer follow-up has revealed a spectrum of delayed immune-related toxicities that are distinct from acute CAR T cell toxicities such as cytokine release syndrome (CRS), immune effector cell (IEC)-associated neurotoxicity syndrome (ICANS), IEC-associated hemophagocytic lymphohistiocytosis-like syndrome (IEC-HS) and IEC-associated hematologic toxicity (ICAHT)^13,14. Notably, approximately 10% and 2% of patients treated with cilta-cel have been reported to develop delayed neurotoxicities (for example, parkinsonism and cranial nerve palsy)^7,15–17 and IEC-associated enterocolitis (IEC-enterocolitis), respectively^7,18. These toxicities remain poorly characterized due to their heterogenous presentations. To reflect their distinct features, and their resemblance to immune-related adverse events (irAEs) observed with checkpoint inhibitors¹⁹, we propose the term ‘CAR T cell-associated immune-related adverse events’ (CirAEs).

As BCMA-targeted CAR T cell therapy moves into earlier lines of therapy, understanding CirAEs is increasingly important, particularly given emerging data suggesting that CirAEs may significantly contribute to non-relapse mortality (NRM) (for example, IEC-enterocolitis exhibited a 36% real-world mortality rate¹⁸). Accordingly, there is a crucial need to (1) better define CirAEs, (2) identify predictors to guide preemptive management and (3) elucidate mechanisms to inform targeted therapies. Here we report the incidence, spectrum, risk factors and outcomes of CirAEs in a large BCMA-targeted CAR T cell therapy cohort and provide mechanistic insights into CirAE pathogenesis. Notably, unlike CRS, ICANS, IEC-HS and ICAHT, which are primarily driven by uncontrolled immune activation, we show that CirAEs arise from direct CAR T cell-mediated attacks on normal tissues. Finally, we describe an illustrative case of extreme, CD4-skewed polyclonal cilta-cel expansion that uncovered an IL-15/CCR5-driven mechanism. The same patient developed three different CirAE manifestations, underscoring unrestrained CD4+ CAR T cell expansion as a key risk factor for CirAEs.

Results

Real-world outcomes comparison of cilta-cel and ide-cel BCMA-CART therapy

Between 1 June 2021 and 31 December 2024, 198 patients received commercial BCMA-targeted CAR T cell therapy at our center (cilta-cel: 125, ide-cel: 73; Extended Data Fig. 1a). At the 31 May 2025 cutoff, median follow-up for the entire cohort was 20.5 months, with ide-cel patients having significantly longer follow-up (30.6 months versus 14.1 months; P < 0.001). Patient characteristics by product and CirAE status are summarized in Table 1 and Supplementary Tables 1 and 2.

Table 1 |.

Overview of clinical characteristics stratified by product and CirAE

Characteristics	Entire cohort (n=198)	Stratified by product			Stratified by CirAE
		Ide-cel (n=73)	Cilta-cel (n=125)	P value	No CirAE (n=171)	CirAE (n=27)	P value
Median follow-up, months (IQR)	20.5 (8.8–31.2)	30.6 (23.8–41.3)	14.1 (6.6–22.3)	<0.001	20.5 (8.8–31.9)	16.3 (7.7–23.8)	0.31
Median age, years (IQR)	65 (59–70)	65 (61–71)	65 (56–69)	0.18	65 (58–70)	67 (61–73)	0.10
Male, n (%)	114 (57.5%)	42 (57.5%)	72 (57.6%)	1	104 (60.8%)	10 (37.0%)	0.02
Race, n(%)
White	159 (80.3%)	60 (82.2%)	99 (81.1%)	0.98	135 (78.9%)	24 (88.8%)	0.82
Black	33 (16.6%)	12 (16.4%)	21 (17.2%)		30 (17.5%)	3 (11.1%)
Asian	3 (1.5%)	1 (1.4%)	2 (1.6%)		3 (2.3%)	0 (0%)
Missing	3 (1.5%)	0 (0%)	3 (2.4%)		3 (1.8%)	0 (0%)
Ethnicity, n (%)
Hispanic	4 (2%)	0 (0%)	4 (3.2%)	0.22	3 (1.7%)	1 (3.7%)	0.52
Non-Hispanic	193 (97.5%)	73 (100%)	120 (96%)		167 (97.6%)	26 (96.3%)
Missing	1 (0.5%)	0 (0%)	1 (0.8%)		1 (0.7%)	0 (0%)
R-ISS at diagnosis, n (%)
1	35 (17.7%)	14 (19.2%)	21 (16.8%)	0.39	30 (17.5%)	5 (18.5%)	0.88
2	70 (35.4%)	26 (35.6%)	44 (35.2%)		59 (34.5%)	11 (40.7%)
3	33 (16.6%)	8 (10.9%)	25 (20%)		30 (17.5%)	3 (11.1%)
Missing	60 (30.3%)	25 (34.2%)	35 (28%)		52 (30.4%)	8 (29.6%)
Cytogenetic risk, n (%)
Standard risk	120 (60.6%)	43 (58.9%)	77 (61.6%)	0.66	103 (60.2%)	17 (63%)	1
High risk	37 (18.7%)	16 (21.9%)	21 (16.8%)		32 (18.7%)	5 (18.5%)
Missing	41 (20.7%)	14 (19.1%)	27 (21.6%)		36 (21.1%)	5 (18.5%)
History of immune disease, n (%)	13 (6.5%)	3 (4.1%)	10 (8%)	0.38	10 (5.8%)	3 (11.1%)	0.08
Prior lines of therapy, median (IQR)	5 (4–7)	6 (5–8)	5 (4–6)	<0.001	6 (4–7)	5 (4–6)	0.01
Prior autologous SCT, n (%)	170 (85.9%)	68 (93.2%)	102 (81.6%)	0.03	151 (85.4%)	19 (70.4%)	0.03
Prior allogeneic SCT, n (%)	5 (2.5%)	3 (4.1%)	2 (1.6%)	0.36	5 (2.9%)	0 (0%)	1
Prior BCMA therapy, n (%)	35 (17.8%)	16 (21.9%)	19 (15.2%)	0.25	32(18.8%)	3 (11.1%)	0.43
Prior CAR T cell therapy, n (%)	6 (3.0%)	1 (1.4%)	5 (3.9%)	0.66	6 (3.5%)	0 (0%)	1
Triple-class refractory, n (%)	146 (73.7%)	63 (86.3%)	83 (66.4%)	0.002	126 (73.5%)	20 (74.1%)	1
Penta-drug refractory, n (%)	49 (25%)	24 (32.9%)	25 (20.2%)	0.06	45 (26.6%)	4 (14.8%)	0.24
Bridging received, n (%)	186 (93.9%)	67 (91.8%)	119(95.2%)	0.36	159 (93%)	27 (100%)	0.38
Extramedullary MM CAR T cell therapy, n (%)	31/181(17.1%)	12/64 (18.2%)	19/117 (16.2%)	0.68	28/156 (17.9%)	3/25 (12%)	0.58
Plasma cell leukemia at infusion, n (%)	5 (2.5%)	3(4.1%)	2 (1.6%)	0.36	4 (2.3%)	1 (3.7%)	0.41
M-spike at infusion (mgdl−1), median (IQR)	0.7 (0.2–1.7)	0.7 (0.2–1.8)	0.7 (0.2–1.6)	0.59	0.8 (0.2–1.8)	0.5 (0.1–1.5)	0.37
dFLC at infusion (mgl−1), median (IQR)	79.5 (13.3–410)	134 (35–642)	60 (9–352)	0.02	79 (13–459)	80(14–353)	0.87
ECOG at infusion>3, n (%)	6/197 (3.0%)	4/73 (5.5%)	2/124 (1.6%)	0.19	5/170 (2.9%)	1/27 (3.7%)	0.59
Lymphodepletion regimen, n (%)
Flu/Cy	182 (91.9%)	62 (84.9%)	120 (96%)	0.01	155 (90.6%)	27 (100%)	0.33
Cy only	15 (7.5%)	10 (13.7%)	5 (4%)		15 (8.8%)	0 (0%)
Bendamustine	1 (0.5%)	1 (1.4%)	0 (0%)		1(0.6%)	0 (0%)
Infusion product
Ide-cel, n (%)	73 (36.8%)	-			71 (41.5%)	2 (7.4%)	0.001
CiLta-cel, n (%)	125 (63.1%)				100 (58.5%)	25 (92.6%)
Product out of specification, n (%)	11 (5.5%)	0 (0%)	11 (8.8%)	<0.001	9 (5.2%)	2 (7.4%)	0.64

Cy, cyclophosphamide; dFLC, difference between involved and uninvolved serum free light chains; ECOG, Eastern Cooperative Oncology Group; Flu/Cy, fludarabine/cyclophosphamide; High-risk cytogenetics include (1) del17p and/or TP53 mutation, (2) biallelic del1p, (3) t(4;14), t(14;16) or t(14;20) plus either gain/amp1q or del1p and (4) gain/amp1q plus del1p; IQR, interquartile range; MM, multiple myeloma; Prior BCMA therapy includes BCMA bispecific antibodies or BCMA antibody–drug conjugate; Penta-drug refractory, MM refractory to bortezomib, carfilzomib, lenalidomide, pomalidomide and anti-CD38 monoclonal antibody; R-ISS, Revised International Staging System; Triple-class refractory disease, MM refractory to a proteasome inhibitor, an immunomodulatory drug and anti-CD38 monoclonal antibody.

The overall response rate (ORR) for the entire cohort was 85.4%, with deeper responses from cilta-cel (92% versus 74%; P < 0.001; Extended Data Fig. 1b). The median progression-free survival (PFS) was 21.1 months overall and longer with cilta-cel (24.4 months versus 9.9 months; P = 0.0014). Twelve-month overall survival was 83.3% and was similar between the two products (Extended Data Fig. 1c–f). CRS occurred in 82.2% (6.6% grade ≥3) and ICANS in 12.6% (2% grade ≥3), with no major product differences (Extended Data Fig. 1g,h). We used peak absolute lymphocyte count (ALC^peak) within the first 14 or 21 days post-infusion as a surrogate for CAR T cell expansion²⁰. Cilta-cel patients had significantly higher day 14 ALC^peak (median: 1.4 versus 0.5 × 10³ cells per microliter, P < 0.001) and day 21 ALC^peak (median: 1.5 versus 0.6 × 10³ cells per microliter, P < 0.001), suggesting greater expansion (Extended Data Fig. 1i,j). Although CD4:CD8 ratios were similar at apheresis, they trended higher with cilta-cel at day 7 and month 1 (Extended Data Fig. 1k).

By data cutoff, 12 patients died without multiple myeloma progression (1-year NRM: 4.8%; Extended Data Fig. 1l). One-year NRM was higher after cilta-cel than after ide-cel (7.3% versus 1.4%; P = 0.13; Extended Data Fig. 1m). CirAEs contributed disproportionately: five of 12 NRM cases (41.7%) had CirAEs (Supplementary Table 3), and four deaths (33.3%) were directly attributable to them, prompting a detailed investigation of CirAE frequency, spectrum and outcomes.

CirAEs after commercial BCMA-targeted CAR T cell therapy

CirAEs were significantly more frequent with cilta-cel (20%) versus ide-cel (2.7%) (P < 0.001; Fig. 1a) and could be classified into four groups: (1) rheumatologic (n = 2), (2) neurologic (n = 19), (3) gastrointestinal (n = 7) and (4) pulmonary (n = 2) (Fig. 1a and Extended Data Table 1). Among cilta-cel patients, the most common CirAEs were cranial nerve palsies (8%), enterocolitis (5.6%), parkinsonism (4.8%) and arthritis (1.6%). Less frequent toxicities included delayed ICANS, encephalopathy and peripheral neuropathy (0.8% each). Notably, the only CirAE observed with ide-cel was pneumonitis (Fig. 1a). CirAE median onset was 28 days (Fig. 1b and Extended Data Table 1). Two patients had multiple CirAEs: cilta-cel patient 4 experienced facial palsy, delayed ICANS and IEC-enterocolitis, and cilta-cel patient 25 developed facial nerve palsy and parkinsonism (Fig. 1b,c). At last follow-up, 59.3% of patients with CirAEs had persistent symptoms or died from CirAEs.

Fig. 1 |. CirAEs after BCMA-targeted CAR T cell therapy.

a, Frequency (top) and types (bottom) of CirAEs in patients with MM treated with either of the two commercially available BCMA-targeted CAR T cell therapy products: cilta-cel and ide-cel. CirAEs were significantly more frequent in patients receiving cilta-cel compared to ide-cel (P = 0.000414). b, Time to development of CirAEs after BCMA-targeted CAR T cell therapy. c, Swimmer’s plot of the 27 patients who developed CirAEs. NRM was defined as death not in the setting of progressive MM. Cilta-cel patient 14 had an asymptomatic local plasmacytoma recurrence in the right scapula found incidentally on routine restaging PET/CT scan 3 months after cilta-cel infusion. His M-spike continued to decline, and serum free light chain levels had normalized. He was treated with radiation therapy to the right scapula and maintained a serological partial remission until NRM. Cilta-cel patient 23 died from respiratory complications associated with CAR T cell therapy-related parkinsonism amid MM progression and was not categorized as NRM. The cumulative dose of steroids (dexamethasone-equivalent) is indicated beneath each patient ID. d, NRM by CirAE status. Survival distributions were compared with the log-rank test. e, Cumulative dexamethasone dose in NRM patients stratified by CirAE status. Twelve patients experienced NRM, including seven without CirAEs and five with CirAEs. Data are presented as median values ± interquartile range. Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. f, Cilta-cel patient 4 duodenal biopsy light microscopy H&E staining shows infiltration of lymphocytes in the lamina propria. CD138 IHC stains positive within the epithelium but not within the lamina propria, and MUM1 was able to identify rare plasma cells in the lamina propria (circled). Multiplex RNAscope images revealed the co-localization of CAR (red) and BCMA (green) staining within the lamina propria. Representative micrographs from one of two independent experiments with similar results are shown. BCMA-CART, BCMA-targeted CAR T cell therapy; CART, CAR T cell; CI, confidence interval; Dex, dexamethasone; HR, hazard ratio; MM, multiple myeloma; MUM1, multiple myeloma oncogene 1; n.a., not available; PET/CT, positron emission tomography/computed tomography; PR, partial remission; RT, radiation therapy; sCR, stringent complete remission; SD, stable disease; Tal, talquetamab; Tec, teclistamab; VGPR, very good partial remission.

Notably, unlike checkpoint inhibitor-induced irAEs²¹, CirAEs were not associated with improved outcomes (Supplementary Fig. 1a–c). Instead, there was a trend toward inferior overall survival in patients with CirAEs (1-year overall survival: 74.4% versus 84.8%; P = 0.13, log-rank; Supplementary Fig. 1c), driven by significantly higher NRM in the CirAE group (1-year NRM: 17% versus 2.7%; P = 0.0021; Fig. 1d). Sixty percent of patients with CirAEs who experienced NRM died of infections compared to 14.3% in the non-CirAE group. This disparity was likely driven by significantly higher cumulative steroid exposure in patients with CirAEs (median dexamethasone-equivalent: 337 mg versus 49.5 mg; P = 0.048; Fig. 1e). Other causes of NRM included non-infectious CirAE complications, CRS/ICANS, IEC-HS, therapy-associated myeloid neoplasms and hemorrhagic shock (Supplementary Table 3). Because 25 of 27 CirAE cases (92.6%) and all six CirAE-related deaths (4.8% of cilta-cel-treated patients) occurred in the cilta-cel group (Fig. 1a,c and Extended Data Table 1), we performed a cilta-cel-only analysis, yielding similar conclusions (Extended Data Fig. 2).

CirAEs have heterogeneous clinical presentations and outcomes

Two patients developed immune-mediated polyarthritis on day 7 and day 28 after cilta-cel infusion, presenting with severe pain and stiffness in multiple joints, despite no prior rheumatologic disease. Both were managed with non-steroidal antiinflammatory drugs, analgesics and corticosteroids. The arthritis of cilta-cel patient 7 resolved within 4 months, whereas cilta-cel patient 2 remained symptomatic more than 27 months later.

Neurologic CirAEs (neuro-CirAEs) occurred in 17 cilta-cel-treated patients (13.4%). Cranial nerve palsies included lower facial nerve palsy (n = 9) and optic neuritis (n = 1). Parkinsonism (n = 6) typically presented with tremor, bradykinesia and hypomimia. Single cases included delayed ICANS, encephalopathy and peripheral sensorimotor neuropathy. Median onset of neuro-CirAEs was 27 days after infusion, and symptoms lasted a median of 93 days. Magnetic resonance imaging (MRI) findings (Extended Data Fig. 3a) supported immune-mediated injury, with examples including left orbital apex enhancement in optic neuritis (cilta-cel patient 21), a rim-enhancing medial thalamic lesion in encephalopathy (cilta-cel patient 14) and enhancement of the facial nerve in facial palsy (cilta-cel patient 24). Cerebrospinal fluid (CSF) analysis confirmed CAR T cell presence in all patients who underwent lumbar puncture (n = 6) (Supplementary Table 4). Median cumulative corticosteroid exposure was 280 mg dexamethasone-equivalent; some also received intravenous immunoglobulin (IVIG) and/or ruxolitinib (Extended Data Table 1). Outcomes were variable: six patients (35.2%) died from neuro-CirAEs (median 5.4 months from onset, with four of six in multiple myeloma remission); five patients (29.4%) had persistent neurologic deficits (median 8.4 months from onset); and six patients fully recovered (median 11 months from onset).

IEC-enterocolitis occurred in seven cilta-cel-treated patients (5.6%), at a median of 3.5 months after infusion. Symptoms included intractable diarrhea (often more than 10 per day), abdominal pain and nutritional compromise, lasting a median of 7 months. Cilta-cel patient 4 had Clostridioides difficile detected early, but symptoms persisted despite treatment. Most received systemic steroids (median dexamethasone-equivalent: 243.5 mg) and, in refractory cases, IVIG, ruxolitinib or infliximab (Extended Data Table 1). Three patients (42.9%) had unresolved symptoms (median 7.3 months) at last follow-up. Endoscopic biopsies (n = 5) showed celiac/graft-versus-host/common variable immunodeficiency-like features with mucosal ulceration, crypt apoptosis, villous blunting and a paucity of plasma cells in the lamina propria. RNAscope and immunohistochemistry (IHC) analysis confirmed abundant CAR T cells co-localizing with rare residual BCMA⁺ plasma cells in the inflamed mucosa (Fig. 1f and Extended Data Fig. 3b,c), suggesting an on-target/off-tumor effect leading to mucosal injury. Notably, cilta-cel patient 10 demonstrated persistent tissue residency of CAR T cells in the colon 9 months after onset, despite prolonged enterocolitis-directed treatment.

Two ide-cel patients developed immune-mediated pneumonitis (approximately 11–12 months after infusion). In one case (ide-cel patient 1), pneumonitis arose after teclistamab initiation. Both presented with cough, dyspnea, hypoxemia and multifocal ground-glass opacities on chest computed tomography (Extended Data Fig. 3a), with negative infectious workups including bronchoscopy. In ide-cel patient 1, pneumonitis resolved with steroids after holding teclistamab. Ide-cel patient 2 continued to be steroid dependent nearly 1 year after onset. No lung biopsies were obtained, so a direct causal link to ide-cel is unproven; nevertheless, these cases are included for completeness.

Factors associated with the development of CirAEs after BCMA-targeted CAR T cell therapy

Pre-infusion multiple myeloma burden and CRS did not predict CirAEs (Fig. 2a–c). By contrast, ICANS was strongly associated with CirAE development (odds ratio = 3.8, P = 0.0066), particularly neuro-CirAEs (odds ratio = 4.7, P = 0.0063) (Fig. 2d,g–j). Complete blood count and serum inflammatory markers are summarized in Supplementary Fig. 1d. Median day 14 ALC^Peak (highest lymphocyte count within 14 days post-infusion) was 3–4-fold higher in patients who developed CirAEs (3.3 versus 0.9 × 10³ per microliter; P < 0.001), a pattern that persisted at day 21 ALC^peak (3.4 versus 1.0 × 10³ per microliter; P < 0.001) and was correlated with higher circulating day 14 CAR T cell counts (Fig. 2e; P = 0.032). Indeed, a day 14 ALC^Peak ≥ 2.4 × 10³ per microliter was associated with elevated CirAE risk (odds ratio = 7.8; P < 0.001; Fig. 2g–j). Patients with CirAEs had persistently higher CD4⁺ T cell proportions from apheresis (P < 0.001) through day 7 (P = 0.045) and months 1, 3 and 6 (P = 0.029) after infusion (Fig. 2f). Accordingly, apheresis CD4:CD8 ratio > 1 was associated with more than three-fold higher odds of CirAE (P = 0.0081; Fig. 2g–j). Both day 14 ALC^Peak ≥ 2.4 × 10³ per microliter and apheresis CD4:CD8 > 1 remained independent predictors of CirAE in multivariate analyses adjusted for ICANS or product.

Fig. 2 |. Factors linked to the development of CirAEs after BCMA-targeted CAR T cell therapy.

a, Pre-infusion M-spike levels determined by serum protein electrophoresis, stratified by CirAE status. Data are presented as median values ± interquartile range. Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. b, Pre-infusion differential free light chains (dFLC) by CirAE status. Data are presented as median values ± interquartile range. Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. CRS (c) and ICANS (d) grade by CirAE status. Categorical variables were compared with Fisher’s exact test. e, Left, ALC^peak (median with interquartile range) at apheresis, infusion (day 0) and days 7, 14 and 21; right: absolute CAR T cell count (%CAR⁺ × ALC) at day 14 by CirAE status. Data are presented as median values ± interquartile range. Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. Exact P values are provided in Supplementary Table 9. f, Top, CD4⁺ and CD8⁺ proportion of CD3⁺ T cells at apheresis, days 7 and 28 and months 3 and 6 after BCMA-targeted CAR T cell therapy, stratified by CirAE. Bar graphs show the median, and the error bars denote the 25th and 75th percentiles. Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. Exact P values are provided in Supplementary Table 9. Bottom, area under the curve (AUC) for CD4⁺ T cells from apheresis to month 6, stratified by CirAE. Data are shown as median values, with error bars representing the 25th and 75th percentiles. Shaded regions indicate the AUC for each group. g, Univariable logistic regression was performed using a binomial generalized linear model to assess associations between clinical factors and CirAE development. Statistical significance of model coefficients was assessed using two-tailed Wald tests. h, Univariable logistic regression was performed using a binomial generalized linear model to assess associations between clinical factors and neuro-CirAE development. Patients with only non-neuro-CirAEs were excluded from analysis. Statistical significance of model coefficients was assessed using two-tailed Wald tests. i, Univariable logistic regression was performed using a binomial generalized linear model to assess associations between clinical factors and non-neuro-CirAE development. Patients with only neuro-CirAEs were excluded from analysis. Statistical significance of model coefficients was assessed using two-tailed Wald tests. j, Multivariate analysis was performed using logistic regression to assess the independent association between clinical factors and CirAE development. Statistical significance of model coefficients was assessed using two-tailed Wald tests. P value for day 14 ALC^Peak ≥ 2.4: 0.000996. CI, confidence interval.

A unique case of multiple CirAEs, extreme hyperleukocytosis from polyclonal CAR T cell hyperexpansion and CD4 predominance

Cilta-cel patient 4, a 74-year-old male with t(11;14) λ light chain multiple myeloma, developed marked CD4-skewed hyperleukocytosis (ALC^Peak: 197.5 × 10³ cells per microliter; CD4:CD8 ratio: 12.6) at day 13 and three distinct CirAEs (Fig. 3a and Supplementary Fig. 2). He had received eight prior lines of therapy and was bridged with selinexor, bortezomib and dexamethasone. At lymphodepletion, he had stable disease (λ light chain: 1,320 mg l⁻¹) and received 0.6 × 10⁶ CAR⁺ cells per kilogram. He developed grade 2 CRS on day 5, treated with tocilizumab and dexamethasone. After CRS, white blood cells (WBCs) and ALC rose sharply, peaking on day 13 (CAR T cell count: 193.5 × 10³ cells per microliter). He received cyclophosphamide 500 mg and dexamethasone for cytoreduction and was discharged on day 15. On day 31, he developed left-sided facial palsy, which resolved with dexamethasone. Two days after steroid discontinuation, he presented with lethargy, disorientation and disequilibrium, requiring admission on day 57 (ALC: 22.4 × 10³ cells per microliter). MRI and electroencephalography were unremarkable. CSF showed 4,120 WBCs per microliter (94% T cells; Supplementary Table 4). He was diagnosed with delayed ICANS (grade 1) and recovered fully with IVIG and dexamethasone. On day 76, 4 days after steroid taper, he developed bloating, nausea, anorexia and profuse diarrhea (up to 20 times per day), with 13 kg (18%) weight loss. Stool on day 92 was positive for Clostridium difficile but did not improve with vancomycin. Endoscopy (day 105) revealed duodenal ulceration, crypt apoptosis, mucin depletion, chronic mucosal injury and markedly reduced lamina propria plasma cells (Fig. 1f), consistent with IEC-enterocolitis. He received budesonide (day 107) and prednisone (day 115) without improvement, necessitating infliximab (day 122). Total parenteral nutrition (TPN) was started (day 137), followed by two more infliximab doses (days 140 and 171), leading to clinical improvement. On day 148, the patient reached minimal residual disease (MRD)-negative stringent complete response. Loose stools were managed with loperamide and continued to improve, resolving completely by day 777. On day 777, he relapsed biochemically, with bone marrow biopsy showing 15% λ-restricted, BCMA-negative multiple myeloma cells and cilta-cel persistence, comprising 21% of bone marrow-infiltrating T cells.

Fig. 3 |. Hyperleukocytosis in cilta-cel patient 4: evidence of polyclonal CD4⁺ CAR T cell expansion.

a, Longitudinal trajectory of circulating CAR T cell counts, overlaid with key clinical events in cilta-cel patient 4. b, CAR transgene copy numbers in peripheral blood from cilta-cel patient 4 compared to the reported mean levels extracted from the CARTITUDE-1 (CART-1) trial using PlotDigitizer. c, TCR Vβ repertoire of CAR T cells from cilta-cel patient 4 demonstrating polyclonal CAR T cell expansion. d, TCR Vβ repertoire changes over time in cilta-cel patient 4. e, Lentiviral integration site analysis in cilta-cel patient 4 at days 13, 21 and 31 as well as months 11 and 15. f, CD4:CD8 ratio extracted from the electronic health records comparing cilta-cel patient 4 to other BCMA-targeted CAR T cell therapy-treated patients. Data are presented as median values ± interquartile range. g, CD4:CD8 ratio from correlative flow cytometry analysis of research samples, comparing cilta-cel patient 4 to other cilta-cel-treated patients. Data are presented as median values ± interquartile range. h,UMAP visualization of the immunophenotypic changes of cilta-cel patient 4’s CD3⁺ T cells analyzed by flow cytometry from pre-lymphodepletion (Pre-LD) timepoint to month 15 after CAR T cell infusion. Act/Pro T_EM, activated proliferating effector memory T cell; Act T_EM, activated effector memory T cell; ANC, absolute neutrophil count; BID, twice a day; BSI, bloodstream infection; C, cellulitis; C. Diff, C. difficile; CART, CAR T cell; D, day; PET, positron emission tomography; QID, four times a day; sCR, stringent complete response; T_EMRA, terminally differentiated effector memory T cell.

Cilta-cel patient 4 exhibited the highest BCMA-targeted CAR T cell therapy expansion reported to date²², exceeding CARTITUDE-1 (ref. 5) peak levels and persisting at unusually high levels (Fig. 3b). T cell receptor (TCR) Vβ repertoire remained diverse from pre-lymphodepletion through peak expansion to day 21 (Fig. 3c,d), and vector integration site analysis showed no clonal dominance (top clone: 0.12%; Fig. 3e). Whole-exome sequencing (WES) on peripheral blood (day 13), saliva (germline) and bone marrow (day 148) found no pathogenic or clonal hematopoiesis of indeterminate potential (CHIP) mutations in a panel of 252 cancer-associated genes (Supplementary Table 5), confirming polyclonal, non-malignant hyperexpansion.

Cilta-cel patient 4 also had the greatest CD4 bias in our cohort, evident from apheresis onwards (Fig. 3f,g). Uniform manifold approximation and projection (UMAP) illustrated the immunophenotypic evolution from predominantly naive CD4-like T cells at pre-lymphodepletion to activated, proliferating effector memory CD4-like CAR T cells at peak expansion (day 13) to exhausted, terminally differentiated CD4-like CAR T cells during delayed ICANS (day 57), before reverting to a more naive-like state at months 11 and 15 (Fig. 3h and Supplementary Figs. 3 and 4a,b). Functionally, cilta-cel’s CAR T cells showed superior antigen-specific degranulation, cytotoxicity and production of granzyme B, tumor necrosis factor (TNF), granulocyte-macrophage colony-stimulating factor (GM-CSF) and IL-17a compared to CAR T cells from cilta-cel patient 22 or normal donors (Extended Data Fig. 4a–c).

IL-15 drives polyclonal CART expansion in cilta-cel patient 4 via the CCL5–CCR5 axis

We investigated the cytokine milieu preceding cilta-cel patient 4’s peak CAR T cell expansion, as the absence of exaggerated secondary CAR T cell expansions during subsequent CirAEs or infections, combined with a negative WES for pathogenic variants, argued against a T cell intrinsic monogenic predisposition to hyperproliferation. Day 7 serum proteomics revealed marked upregulation of lymphopoietic cytokines IL-2, IL-7 and IL-15, with chemokines CCL2, CCL5, CXCL9 and CXCL10 exceeding the upper limit of quantification (Extended Data Fig. 4d). These cytokines/chemokines were markedly higher compared to two other patients with CirAEs (cilta-cel patient 8 and cilta-cel patient 12) and a non-CirAE patient (cilta-cel patient 26) (Fig. 4a). When equal aliquots (10%) of serum were added to culture, cilta-cel patient 4’s day 7 serum significantly enhanced the survival of healthy-donor-derived CAR T cells, as compared to sera from other patients (Fig. 4b). Cilta-cel patient 4’s CAR T cells at peak expansion (day 13) were cultured with IL-2 (10 ng ml⁻¹), IL-7 (20 ng ml⁻¹) and/or IL-15 (20 ng ml⁻¹)—doses eliciting similar levels of proliferation normal donor-derived CAR T cells (Extended Data Fig. 4e). Cilta-cel patient 4’s CAR T cells showed robust proliferation and pSTAT5 activation in response to IL-15, exceeding responses to IL-2 or IL-7 (Fig. 4c), consistent with elevated IL-15RA expression relative to IL-2RA and IL-7RA in cilta-cel patient 4’s CAR T cells (Extended Data Fig. 4f). IL-15 induced CCL5 secretion in CAR T cells from both cilta-cel patient 4 and normal donors (Fig. 4d), but only cilta-cel patient 4’s CAR T cells upregulated CCR5, whereas normal donor CAR T cells downregulated CCR5 after cytokine stimulation (Extended Data Fig. 4g).

Fig. 4 |. Maraviroc abrogates cilta-cel patient 4 CAR T cell proliferation in vitro.

a, Serum cytokine profiling was performed for four cilta-cel-treated patients (numbers 4, 8, 12 and 26) at day 7 post-infusion and around peak expansion (days 13 or 14). A heatmap shows the z-scores of detected cytokines. Cilta-cel patient 26 did not experience a CirAE, whereas cilta-cel patients 4, 8 and 12 developed CirAEs. ‘^’ represents cytokine concentration above the upper limit of quantification for LUMINEX assay. b, Absolute counts of cilta-cel CAR T cells manufactured from a normal donor in a 3-day proliferation assay using RPMI supplemented with 10% serum from four different cilta-cel-treated patients collected at days 7, 13 or 14. Cilta-cel patient 26 did not experience a CirAE, whereas cilta-cel patients 4, 8 and 12 developed CirAEs. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. Exact P values are provided in Supplementary Table 9. c, Proliferation assay of cytokine-stimulated (IL-2: 10 ng ml⁻¹, IL-7: 20 ng ml⁻¹ or IL-15: 20 ng ml⁻¹) cilta-cel patient 4’s CAR T cells labeled with CTV. A representative flow cytometry histogram from one of three independent replicates with similar results is shown, illustrating dye dilution corresponding to cell division (left panel). Percentage of phospho-STAT5 (pSTAT5)-positive cells after 15 minutes of stimulation with IL-2, IL-7 or IL-15. Data are presented as mean values ± s.e.m. from n = 3 biological replicates (right panel). Statistical comparisons between groups were performed using the two-tailed Student’s t-test. Exact P values are provided in Supplementary Table 9. d, Left, cytokine secretome analysis of supernatant obtained from a 5-day cytokine (IL-2: 10 ng ml⁻¹, IL-7: 20 ng ml⁻¹ or IL-15: 20 ng ml⁻¹) stimulation of cilta-cel CAR T cells manufactured from three normal donors and cilta-cel patient 4. The cytokine profiles are shown as z-scores of the log₂ fold change of the mean cytokine concentration stimulated versus unstimulated conditions. Statistical significance was determined using two-tailed Mann–Whitney tests. P^unadj indicates unadjusted P values (without correction for multiple comparisons); P^adj indicates adjusted P values (with Holm–Sidak correction for multiple comparisons); *P < 0.05. Right, absolute CCL5 concentrations (in pg ml⁻¹) are shown. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. e, Maraviroc dose-finding experiment to determine the optimal drug concentration that abrogates CD3/CD28-dependent T cell proliferation without compromising T cell viability. The x axis shows the fold change in viability relative to CD3/CD28 bead-stimulated T cells without maraviroc; the y axis shows the fold change in proliferation relative to CD3/CD28 bead-stimulated T cells without maraviroc. Data from three donors, each with n = 3 biological replicates, are presented. f,g, 40 μM maraviroc reduces CAR T cell (cilta-cel and Penn’s in-house BCMA CAR) proliferation in vitro. f, Normal donor-derived cilta-cel CAR T cells cultured for 3 days with RPMI supplemented with 25% cilta-cel patient 4 day 7 serum exhibited reduced absolute CAR T cell counts upon treatment with 40 μM maraviroc compared to DMSO control. RPMI supplemented with 25% commercial human AB serum served as an additional control. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. Exact P values are provided in Supplementary Table 9. g, Normal donor-derived cilta-cel CAR T cells cultured for 3 days with IL-15 (20 ng ml⁻¹) or irradiated MM.1S exhibited reduced absolute CAR T cell counts upon treatment with 40 μM maraviroc compared to DMSO control. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. Exact P values are provided in Supplementary Table 9. h,i, Maraviroc does not impair cytotoxic function of cilta-cel CAR T cells. h, Cilta-cel CAR T cells maintained cytotoxicity against RPMI-8226 cells after 48 hours of co-culture in the presence of 40 μM maraviroc, with no impairment compared to DMSO control. A 2:1 effector:tumor ratio was used. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. Exact P values are provided in Supplementary Table 9. i, Cilta-cel CAR T cells maintained functional degranulation capacity, as evidenced by preserved CD107a and granzyme B (GZMB) expression after 72 hours of co-culture with irradiated MM.1S in the presence of 40 μM maraviroc. A 1:1 effector:tumor ratio was used. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. Exact P values are provided in Supplementary Table 9. j, Absolute counts of cilta-cel CAR T cells after 72 hours of co-culture with irradiated MM.1S cells (1:1 effector:tumor ratio), comparing CCR5 wild-type (CCR5^WT) and CCR5^KO CAR T cells in the presence of 40 μM maraviroc or DMSO. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. Exact P values are provided in Supplementary Table 9. k, α-CCR5 antibody abrogates the proliferation of IL-15 (20 ng ml⁻¹)-stimulated cilta-cel patient 4 CAR T cells. Clockwise: light microscopy images of the proliferation assay, CTV tracking of proliferating generations and absolute CAR T cell counts after 72 hours of IL-15 stimulation. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. Exact P values are provided in Supplementary Table 9. α-CCR5, antibody against CCR5; BCMA-CART, BCMA-targeted CAR T cell therapy; CART, CAR T cells; Cilta 4, cilta-cel patient 4; D, day; E:T ratio, effector:target ratio; FC, fold change; H, hours; MVC, maraviroc; ND, normal donor; NS, not significant.

Given that (1) IL-15 promotes CCL5 secretion by follicular dendritic cells, macrophages and T cells, supporting the survival and proliferation of CCR5⁺CD4⁺ T cells^23–26 and that (2) cilta-cel patient 4ʼs CAR T cells showed heightened IL-15 sensitivity, pronounced CCR5 upregulation upon IL-15 stimulation and marked CD4 skewing, we hypothesized that cilta-cel patient 4ʼs CD4-predominant hyperproliferation was driven by an IL-15–CCL5–CCR5 loop. We further posited that disrupting this axis would mitigate hyperproliferation without affecting CAR T cell viability or cytotoxicity. To test this, we cultured CD3/CD28-activated T cells in the presence of increasing concentrations of maraviroc. At 40 μM, maraviroc suppressed proliferation with limited impact on T cell viability (Fig. 4e and Extended Data Fig. 4h). We then activated BCMA-targeted CAR T cells manufactured from normal donors using (1) media supplemented with 25% day 7 serum from cilta-cel patient 4 (Fig. 4f) or (2) IL-15 or MM.1S, in the presence of 40 μM maraviroc. Maraviroc significantly abrogated both antigen-independent and antigen-dependent activation of BCMA-targeted CAR T cells (Fig. 4f,g) without compromising anti-multiple myeloma cytotoxicity (Fig. 4h,i). To confirm target specificity, we generated CCR5-knockout (CCR5^KO) cilta-cel CAR T cells (Supplementary Fig. 5). CCR5^KO impaired antigen-dependent CAR T cell proliferation but rendered cells resistant to maraviroc (Fig. 4j). Anti-CCR5 antibody (20 μg ml⁻¹) also inhibited IL-15-stimulated proliferation in cilta-cel patient 4’s CAR T cells (Fig. 4k).

Cilta-cel and ide-cel have similar BCMA avidity but distinct signaling and metabolic profiles

We hypothesized that differences in CAR design between cilta-cel and ide-cel (Supplementary Table 6)^27,28 may influence BCMA-binding avidity or downstream signaling, contributing to higher CirAE incidence with cilta-cel. To test BCMA-binding avidity, we expressed each CAR on SUP-T1 cells, measured force of engagement with MM.1S and did not observe a significance difference in BMCA-binding avidity (Extended Data Fig. 5a,b).

We next evaluated downstream signaling using Jurkat reporter cells²⁹ expressing either CAR. Cilta-cel showed higher baseline (tonic) NFAT/NF-κB activity (Extended Data Fig. 5c–e) and metabolic output (Extended Data Fig. 5f–k), suggesting a more preactivated, metabolically engaged state without antigen³⁰. Upon MM.1S stimulation, however, the pattern flipped: ide-cel exhibited greater inducible activation with stronger upregulation of NFAT/NF-κB activity and larger increases in oxygen consumption, ATP production and spare respiratory capacity (Extended Data Fig. 5e–k). Thus, ide-cel appeared less active at baseline but more ‘elastic’ or responsive to antigen, whereas cilta-cel was more active and metabolically saturated at baseline.

Longitudinal immunophenotypic profiling of cilta-cel-treated patients

At pre-lymphodepletion, the T cells of CirAE patients were enriched for CD4⁺ central memory and naive-like cells, whereas non-CirAE patients had more CD8⁺ terminally differentiated-like cells (Fig. 5a and Supplementary Figs. 6 and 7a–c). Consistently, patients with CirAEs had significantly higher proportions of total CD4⁺ T cells. By contrast, there were no baseline differences in the expression of inhibitory checkpoints or in non-T-cell populations between the two groups (Supplementary Fig. 7d,e).

Fig. 5 |. Cilta-cel-associated CirAEs are mediated by CD4⁺ T cells.

a, Left, UMAP visualization of flow cytometry characterization of CD3⁺ T cells from 24 patients at pre-lymphodepletion (Pre-LD) revealed distinct T cell phenotypes associated with the development of CirAEs. Middle, FlowSOM clustering revealed a CD4⁺ central memory-like T cell population (cluster 2) enriched in patients who developed CirAEs, whereas patients who did not develop CirAEs showed a T cell phenotype resembling CD8⁺ T effector memory cells reexpressing CD45RA (T_EMRA-like, cluster 8). Violin plots show the mirrored probability density of the data; dashed line indicates the median; and dotted lines denote the 25th and 75th percentiles. Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. P^unadj indicates unadjusted P values (without correction for multiple comparisons); P^adj indicates adjusted P values (with Holm–Sidak correction for multiple comparisons). Right, proportions of CD4, CD8 and double-negative T cell subsets by CirAE. Bar graphs show the median, and the error bars denote the 25th and 75th percentiles. Violin plots show the mirrored probability density of the data; dashed line indicates the median; and dotted lines denote the 25th and 75th percentiles. Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. b, Immunohistochemistry analysis of CD3, CD4 and CD8 on biopsies obtained during immune effector cell-associated enterocolitis revealed marked infiltration of CD4⁺ T cells in the lamina propria. The H-score was calculated by multiplying the percentage of positive cells by the staining intensity (graded on a scale from 1 to 300). Representative micrographs from a single experiment; imaging was performed once and not independently replicated due to limited available tissue. c, Proportions of CD4⁺ and CD8⁺ CAR T cell clusters in paired peripheral blood mononuclear cell (PBMC) and CSF samples obtained at the time of neuro-CirAE, analyzed by single-cell RNA sequencing. CSF samples showed enrichment of CD4⁺ CAR T cell clusters, including an almost CSF-exclusive interferon-stimulated tissue-resident memory-like (T_RM-like) CAR T cell population (blue). d, Left, CD4:CD8 ratio of CAR T cells in the matched CSF and peripheral blood; right, CD4:CD8 ratio of CD3⁺ T cells in matched enterocolitis tissues and peripheral blood during CirAE. Data are presented as median values ± interquartile range. Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. e, UMAP visualization of CAR T cells in matched peripheral blood and CSF, colored by T cell clusters. f, Density plot of AUCell enrichment score for the Gene Ontology Biological Process (GOBP): leukocyte-mediated cytotoxicity showing high enrichment in CD4⁺ T_RM-like CSF-specific CAR T cells. g, Density plot of representative cytotoxic gene expression GNLY (encoding granulysin) and GZMK (encoding granzyme K) in CSF CAR T cells. h, LUMINEX proteomics analysis of matched CSF and serum samples at the time of neuro-CirAE showed an enrichment of CXCL10, CCL2, CCL3 and IL-8 in the CSF of four patients with neuro-CirAEs. Box plots show the median (center line), interquartile range (bounds of box; 25th–75th percentiles) and minimum and maximum values (whiskers). Statistical comparisons between groups were performed using the two-tailed Wilcoxon test. P^unadj indicates unadjusted P values (without correction for multiple comparisons); P^adj indicates adjusted P values (with Holm–Sidak correction for multiple comparisons). i, Transwell migration assays were performed to evaluate CAR T cell chemotaxis using matched CSF and serum from a patient with neuro-CirAE (cilta-cel patient 16). CD4⁺ CAR T cell migration significantly increased when the lower chamber contained RPMI supplemented with 80% CSF from cilta-cel patient 16, suggesting specific CD4⁺ CAR T cell recruitment driven by a CSF-specific chemotactic gradient. Top, representative flow cytometry density plot from one of two independent experiments with similar results is shown; bottom, data are presented as mean values ± s.e.m. from n = 2 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. j, Binding avidity of CD4⁺ versus CD8⁺ CAR T cells in both BCMA-targeted (top) and CD19-targeted (bottom) CAR T cell therapy models showed higher BCMA-binding avidity of CD4⁺ versus CD8⁺ CAR T cells. Data are presented as mean values ± s.e.m. from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s t-test. CART, CAR T cells; DN, double-negative CD4⁻, CD8⁻ T-cell population; FASLG, Fas ligand; PB, peripheral blood; scRNA-seq, single-cell RNA sequencing; T_CM, central memory T cell; T_EMRA, terminally differentiated effector memory T cell; T_RM-like, tissue-resident memory-like T cell.

At peak expansion, the CAR⁺ population of patients with CirAEs was predominantly CD4⁺ effector memory (EM)-like CAR T cells, whereas non-CirAE patients had balanced CD4⁺ and CD8⁺ EM-like CAR T cells (Extended Data Fig. 6a and Supplementary Figs. 8 and 9a), reflecting a non-significant trend toward a higher CD4 CAR T cell proportion in CirAE. Overall, CAR expression was similar between groups, but cilta-cel patient 4 had a CAR median fluorescence intensity (MFI) that was >2.5 s.d. above the mean at peak expansion (Extended Data Fig. 6a and Supplementary Fig. 4d), further increasing and peaking during CirAE toxicities (month 11; Supplementary Fig. 4e). Notably, cilta-cel patient 4 was not an outlier in vector copy number among CAR T cell therapy patients at our institution (Supplementary Fig. 4f)²². Inhibitory receptor expression and proportions of non-T-cell populations were similar between the two groups (Supplementary Fig. 9b,c).

Comparison of 10 patients during CirAE and six non-CirAE patients at matched timepoints did not reveal a significant difference in peripheral blood CAR T cell counts between CirAE and non-CirAE patients (P = 0.22; Extended Data Fig. 6b). Interestingly, the proportion of CD4⁺ T cells expressing CAR was significantly higher in CirAE patients (P = 0.042; Extended Data Fig. 6b). When comparing CD4 trends in paired samples between pre-lymphodepletion and during CirAE/matched timepoints, there was a significant decrease in the CirAE group (P = 0.049), whereas no significant change was observed in patients without CirAE (P = 0.4) (Extended Data Fig. 6c), supporting the hypothesis that CD4⁺ CAR T cells preferentially traffic out of peripheral blood into CirAE-affected tissues.

CD4⁺ CAR T cells mediate CirAE after cilta-cel treatment

IHC of all available enterocolitis biopsies (n = 5) revealed marked CD4 skewing of tissue-infiltrating CAR T cells (median CD4 H-score: 270; Fig. 5b,d). Single-cell RNA sequencing analysis of all available matched peripheral blood and CSF samples in neuro-CirAE patients (n = 4) also showed CD4 skewing in CSF CAR T cells (Fig. 5c,d and Extended Data Fig. 7a), with a highly enriched interferon-stimulated, resident memory-like (T_RM) CD4 CAR T cell population in the CSF (17% in CSF versus 0.06% in peripheral blood; Fig. 5e). CSF-infiltrating CAR T cells also showed greater enrichment for leukocyte-mediated cytotoxicity gene signatures and expressed more granzyme B than peripheral blood CAR T cells (Extended Data Figs. 6d and 7b).

Within the CSF, CD4⁺ CAR T cells, and especially the T_RM subset, had greater enrichment for cytotoxicity-associated gene signatures (Fig. 5f) and expressed higher levels of effector genes (for example, GNLY, GZMK, IFNG, PRF1 and CD107a) compared to CD8⁺ CAR T cells (Fig. 5g and Extended Data Fig. 7c). Supporting this, BCMA-dependent in vitro activation of CSF-derived CD4⁺ CAR T cells from cilta-cel patient 4 produced more CD107a, granzyme B, IL-17a, IL-2 and GM-CSF than CD8⁺ CAR T cells (Extended Data Fig. 4a).

Compared to matched peripheral blood, CD4:CD8 ratios were significantly higher in intestinal biopsies (P = 0.0016) and trended higher in the CSF (P = 0.089; Fig. 5d), supporting selective CD4⁺ CAR T cell tissue infiltration rather than passive redistribution from the circulation. Proteomic analysis of matched serum and CSF revealed distinct profiles, with the CSF enriched in chemokines such as CXCL10, CCL2, CCL3 and IL-8 (Fig. 5h). Supporting our hypothesis of chemokine-driven CD4⁺ CAR T cell trafficking into the CSF, a Transwell assay using cilta-cel patient 16 serum (upper chamber) and CSF (lower chamber) obtained during neuro-CirAE preferentially promoted CD4⁺ CAR T cell migration (Fig. 5i). Finally, antigen-specific avidity assays with BCMA⁺ RPMI-8226 and CD19⁺ HBL-1 cells demonstrated greater antigen-binding avidity of CD4⁺ versus CD8⁺ CAR T cells in both cilta-cel and CD19-targeted CAR T cell (CART19) models (Fig. 5j).

An early surge in proinflammatory cytokines is associated with the development of CirAE

Proteomic profiling of sera from 31 cilta-cel patients (Supplementary Tables 7 and 8) revealed significantly elevated EGF, CCL4 and CCL5 at pre-lymphodepletion in patients who later developed CirAE (Extended Data Fig. 8a,b). These cytokines are implicated in lupus, rheumatoid arthritis, Sjögren’s syndrome and inflammatory bowel disease^31–34. By day 7, CirAE patients had higher levels of ICANS-associated cytokines, including GM-CSF, IL-13, IL-5, IL-6 and CXCL10, further supporting ICANS as a risk factor for CirAE. Longitudinal analysis revealed three proinflammatory cytokine kinetic patterns in CirAE patients: (1) progressive increases (G-CSF, IL-6 and IL-8), (2) day 7 peak (IL-2, IL-5, IL-13 and IFNγ) and day 14 peak (IL-4 and CXCL10) (Extended Data Fig. 8c). By contrast, antiinflammatory cytokines (IL-12p40, IFNα2 and IL-10) were consistently lower before and during CirAE (Extended Data Fig. 8d). These findings highlight a potential therapeutic window for preemptive interventions to reduce the risk of CirAE.

Discussion

In this study, we characterized the incidence, clinical spectrum and outcomes of CirAEs across a large real-world cohort of patients and identified risk factors and mechanisms. Notably, efficacy (ORR, PFS and overall survival) and overall safety (NRM, CRS and ICANS) mirror other recent real-world studies of cilta-cel and ide-cel^4,7, supporting the generalizability of our findings. We found that CirAEs substantially impact survival after BCMA-targeted CAR T cell therapy. At approximately 14 months of median follow-up, one-third of patients who developed CirAEs had died, two-thirds from CirAEs, corresponding to CirAE-related mortality rate of 4.8% among cilta-cel-treated patients. One-year NRM in patients with CirAEs was approximately 17%, significantly higher than approximately 3% in patients without CirAEs. Most CirAE-related deaths were from infections (60% versus 14% of deaths in non-CirAE patients), likely from intensive immunosuppression, especially high-dose steroids, underscoring the need to balance CirAE control with infection prophylaxis. Consistent with this, from real-world cilta-cel cohorts, infections were also identified as the leading cause of NRM⁷.

In contrast to checkpoint inhibitor-associated irAEs in solid cancers (which correlate with improved tumor control and survival³⁵), CirAEs conferred no survival benefit in our study. This likely reflects fundamental mechanistic differences: checkpoint blockade invigorates a broad repertoire of antitumor T cells, sometimes augmenting cancer clearance, whereas BCMA-targeted CAR T cells target a single antigen, making CirAE a form of collateral damage (on-target, off-tumor toxicity) without added anti-multiple myeloma efficacy^16,36.

A unique aspect of our study is the in-depth investigation of an extreme case (cilta-cel patient 4) that uncovered a novel mechanism (that is, an IL-15–CCL5–CCR5 positive feedback loop) of polyclonal CAR T cell hyperexpansion. IL-15, which surges after cyclophosphamide/fludarabine lymphodepletion, promotes T cell expansion while maintaining stemness^37,38. In cilta-cel patient 4, exceptionally high IL-15 levels preceded peak expansion, and his CAR T cells had heightened IL-15 sensitivity (via high IL-15Rα), triggering a self-reinforcing loop: IL-15 induced CCR5 upregulation and CCL5 secretion by CAR T cells, with CCL5 feeding back through CCR5 to further amplify CAR T cell proliferation and survival. This mirrors a previously reported case of IL-15-driven CAR T cell hyperexpansion (in a GD2 CAR-NKT cell therapy)³⁹, where, once a critical cell mass was reached, IL-15 secretion fueled a runaway expansion. We demonstrated that blocking this axis with the CCR5 antagonist maraviroc dramatically curtailed CAR T cell proliferation in vitro in both antigen-dependent and IL-15-driven contexts, without compromising T cell viability or cytotoxicity. These results suggest a possible strategy for managing unrestrained CAR T cell expansion. Given its favorable safety profile and established clinical use, including as prophylaxis for graft-versus-host disease^40,41, maraviroc may warrant further study as a strategy to mitigate hyperexpansion while preserving anti-multiple myeloma cytotoxicity.

Beyond the single-case insights, our study identified two measurable risk factors for CirAE: early post-infusion ALC expansion and baseline CD4 bias. Both remained independent predictors in multivariable analysis, supporting their use in risk stratification models. A day 14 ALC^Peak ≥ 2.4 × 10³ per microliter was strongly associated with CirAE development, aligning with reports linking high CAR T cell expansion to delayed neurotoxicity in cilta-cel^17,42. Similarly, patients with CD4-skewed (CD4:CD8 > 1) apheresis products were significantly more likely to develop CirAEs. This novel observation implicates the patient’s intrinsic immune profile or CAR T cell product composition in toxicity risk, a finding not highlighted in previous studies.

Longitudinal analyses further support the central role of CD4⁺ T cells in CirAEs. CirAE patients consistently exhibited higher peripheral blood CD4⁺ T cell proportions than non-CirAE patients through 6 months after CAR T cell infusion (the latest time point evaluated). Notably, this difference narrowed at months 1 and 3, coinciding with the typical timing of CirAEs, suggesting selective trafficking of CD4⁺ CAR T cells from the circulation into affected tissues. Indeed, in all eight patients with available samples (CSF or gastrointestinal biopsy), infiltrating CAR T cells were overwhelmingly CD4⁺, corroborating recent findings by Atanackovic et al.³⁰, who identified CD4⁺ CAR T cells in the CSF of three neurotoxicity cases. Our comprehensive tissue profiling extends that observation, showing that CD4⁺ CAR T cells in CirAE lesions are not bystanders but, rather, potent effectors that express cytotoxic molecules.

Using single-cell RNA sequencing, we identified a CSF-enriched cluster of CD4⁺ tissue-resident memory-like CAR T cells characterized by elevated expression of cytotoxic gene signatures, suggesting that central nervous system (CNS) infiltration of CAR T cells reflects selective trafficking rather than passive leakage through a disrupted blood–brain barrier⁴³. In vitro functional assays confirmed that CSF CAR T cells maintain anti-myeloma cytotoxicity, with CD4⁺ CAR T cells exhibiting greater degranulation than their CD8⁺ counterparts. This aligns with previous studies implicating CD4⁺ CAR T cells in toxicities across other disease indications. For instance, Haas et al.⁴⁴ described prominent CD4⁺ CAR T cell infiltration in the lungs of two patients with fatal pulmonary CirAEs after mesothelin-directed CAR T cell therapy, and Boulch et al.⁴⁵ demonstrated that CD4⁺ CAR T cells were more potent than CD8⁺ CAR T cells in driving CRS in a murine model of CD19-targeted CAR T cell therapy. Notably, a recent biophysical study of the T cell immunologic synapse demonstrated that the CD4 co-receptor can enhance antigen-specific TCR avidity by up to 5.5-fold compared to CD8 and that CD4 T cells exhibit lower antigen-specific TCR activation thresholds compared to CD8 T cells⁴⁶. These findings mirror the increased BCMA-binding avidity and greater cytotoxic capacity of CD4⁺ CAR T cells in our study. By demonstrating preferential trafficking, retention and heightened cytotoxicity of CD4⁺ CAR T cells in CirAEs, we establish a mechanistic framework wherein CD4⁺ T cells are the central drivers of CirAE pathology.

The question remains why CirAEs are more common with cilta-cel than with ide-cel. Head-to-head product comparisons indicated that the dual-epitope binder in cilta-cel does not increase BCMA-binding avidity relative to ide-cel’s single-chain variable fragment (scFv) construct. However, cilta-cel had higher tonic signaling and baseline metabolic activity, whereas ide-cel mounted stronger responses upon antigen stimulation. Notably, in multivariate analysis, product type was not an independent predictor of CirAEs when controlling for expansion metrics and CD4 skewing. In other words, cilta-cel’s higher rate of CirAEs could be driven by greater product expansion.

Our cytokine analysis offers further insights into CirAE susceptibility. By day 7 after infusion, patients who later developed CirAEs had significantly higher levels of proinflammatory cytokines (for example, GM-CSF and IL-6), independent of CRS or ICANS severity. Conversely, CirAE patients had lower levels of IL-12p40, a cytokine whose deficiency has been linked to unchecked inflammation in animal models of colitis, multiple sclerosis and arthritis^47–49, paralleling impaired immune braking inferred in CirAE patients. These immune signatures suggest that patients predisposed to CirAEs begin CAR T cell therapy with a preexisting proinflammatory bias and diminished antiinflammatory regulation.

These findings have direct clinical implications. In parallel with ongoing validation of identified predictive cutoffs, our institution has implemented risk-adapted strategies to mitigate CirAEs, including preemptive corticosteroids in cilta-cel-treated patients based on ICANS (all grades), CD4:CD8 ratio > 1 at apheresis and unexpectedly high ALC within the first 2–3 weeks after infusion. The goal is to curb excessive CD4⁺ CAR T cell expansion before overt toxicity occurs. Notably, preliminary results from another center reported that prophylactic dexamethasone in cilta-cel patients with ALC > 5 × 10³ per microliter may improve early survival⁵⁰ without compromising efficacy⁵¹.

Our study has limitations inherent to its retrospective design. Not all patients had research samples at all timepoints, which may introduce bias. Additionally, one ide-cel-treated patient developed pneumonitis after starting teclistamab, clouding attribution. Nonetheless, by unifying this heterogeneous group of immune-related toxicities with a shared immunologic signature under the CirAE framework, we highlight a significant and previously underrecognized cause of morbidity and mortality after BCMA-targeted CAR T cell therapy. Our integrated clinical and correlative approach provides a comprehensive understanding of CirAE incidence, predictors and mechanisms. These insights lay the groundwork for proactive strategies, such as identifying high-risk patients for closer follow-up, engineering CAR T cell products with balanced T cell subsets or targeted interventions such as CCR5 blockade, toward the goal of mitigating CirAEs without undermining the potent anti-multiple myeloma activity of BCMA-targeted CAR T cell therapy.

Methods

Patient characteristics and clinical outcomes

This study was conducted in accordance with the Declaration of Helsinki and received approval from the University of Pennsylvania institutional review board. Written consent was obtained to review patients’ medical history and conduct research analysis on collected biospecimens. The research cohort comprised 198 individuals diagnosed with RRMM who underwent treatment with either of the two commercially available BCMA-targeted CAR T cell therapies: ide-cel or cilta-cel. Summary of clinical characteristics of the entire can be found in Table 1. CirAE patients are detailed in Supplementary Table 3. Infusions were administered between 1 June 2021 and 30 December 2024. Cutoff for clinical follow-up was 31 May 2025. High-risk cytogenetic abnormalities were defined according to the Mayo Clinic stratification for myeloma and risk-adapted therapy (mSMART) 4.0: (1) del17p and/or TP53 mutation; (2) biallelic del1p; (3) t(4;14), t(14;16) or t(14;20) plus either gain/amp1q or del1p; and (4) gain/amp1q plus del1p⁵². Extramedullary multiple myeloma was defined as visceral or soft tissue plasmacytomas not contiguous with bone. Triple-class refractory disease was defined as multiple myeloma refractory to a proteasome inhibitor, an immunomodulatory drug and a monoclonal antibody⁵³. Penta-drug refractory disease was defined as multiple myeloma refractory to bortezomib, carfilzomib, lenalidomide, pomalidomide and anti-CD38 monoclonal antibody⁵³. The key clinical outcomes included response rates, PFS and overall survival. Patients were evaluated for best response according to the International Myeloma Working Group 2016 consensus criteria⁵⁴, with the addition of unconfirmed complete response (uCR)—defined as undetectable serum and urine immunofixation, along with the disappearance of any soft tissue plasmacytomas, in the absence of a bone marrow biopsy. PFS was calculated as the time between the date of BCMA-targeted CAR T cell therapy and infusion and the event date (progression or death) or censored at the last follow-up visit or the start of maintenance therapy. Overall survival was calculated as the time between the date of BCMA-targeted CAR T cell infusion and the event date (death) or censored at last follow-up. NRM was defined as death not in the setting of progressive multiple myeloma. CRS and ICANS were classified according to the American Society for Transplantation and Cellular Therapy criteria or Common Terminology Criteria for Adverse Events (version 5.0). CirAEs were defined as toxicities arising from CAR T cell activation and loss of self-tolerance that may involve any organ system but are distinct from other well-characterized syndromes occurring in the acute phase, such as CRS, ICANS, IEC-HS and ICAHT. CirAEs were graded according to the grading system used for immune checkpoint inhibitor irAEs⁵⁵. The electronic health records were accessed to obtain relevant clinical information, including complete blood count with differential (ALC, absolute monocyte count and absolute neutrophil count), ferritin, C-reactive protein, M-spike, serum free light chain levels, imaging reports, endoscopy reports, biopsy results and bone marrow biopsy results, including cytogenetics and next-generation WES. Whole blood from the patients was processed using a Ficoll gradient.

IHC

After sectioning of 5-μm slides from formalin-fixed, paraffin-embedded (FFPE) blocks, the tissue was subjected to deparaffinization in xylene and subsequent rehydration through a graded series of ethanol solutions, with decreasing ethanol concentrations. Hematoxylin and eosin (H&E)-stained sections, processed using the Sakura Tissue-Tek Prisma Plus automated slide stainer, were reviewed by the same pathologist to assess the overall tissue morphology. For plasma cell visualization, IHC was performed using CD138 (1:200 dilution; Dako/Agilent Technologies, clone M115), MUM1 (1:100 dilution; Dako/Agilent Technologies, clone MUM1p), CD3 (pre-dilute; Leica Biosystems, clone LN10), CD4 (pre-dilute; Biocare Medical, clone EP204) and CD8 (1:40 dilution; Dako/Agilent Technologies, clone C8/144B). Antibodies were applied on select cases using the Leica BOND-PRIME IHC staining platform for optimal antigen detection.

RNAscope in situ hybridization

FFPE slides were prepared by mounting 5-μm sections on slides following NanoString guidelines. FFPE slides were pretreated starting with a 60-minute bake at 60 °C. Tissue samples were deparaffinized using xylene and ethanol washes before undergoing target retrieval with a 1× RNAscope target retrieval reagent for 15 minutes at 99 °C in a steamer. After target retrieval, hydrogen peroxide treatment was applied to the slides to block endogenous peroxidases, followed by protease plus incubation for tissue permeabilization. For hybridization, probes for cilta-cel CAR T cells (hCAR-TNFRSF17-C3) and BCMA (hTNFRSF17-C1) (ACDBio, 585791) were prepared and hybridized onto the tissue for 2 hours at 40 °C. After hybridization, the slides underwent multiple rounds of amplification (AMP1, AMP2 and AMP3) with specific tyramide signal amplification (TSA) dye labels, followed by signal development using HRP-C1, HRP-C2 or HRP-C3, depending on the probe used. Finally, slides were stained with nuclear dye, SYTO83 (Thermo Fisher Scientific, S11364) and antibodies targeting human CD3e. After multiple washes in 2× SSC, tissues were imaged on the GeoMx platform (software version 3.1.0.222).

Cilta-cel quantitative PCR

Genomic DNA was isolated from PBMCs, and qRT–PCR analysis was performed using ABI TaqMan technology to detect the integrated CAR transgene sequence, using triplicates of 200 ng of genomic DNA per timepoint for patient samples. To determine copy number per unit DNA, a nine-point standard curve was generated using increasing numbers of lentivirus plasmid copies from five copies to 1 × 10⁶ copies spiked into 200 ng of non-transduced control genomic DNA. The number of copies of plasmid present in the standard curve was verified using digital qRT–PCR with the same primer/probe set and performed on a QIAcuity digital PCR instrument (Qiagen). For quality control checks, each data point (sample and standard curve) was evaluated in triplicate with a positive C_t value in three of three replicates. Additionally, the acceptable percent coefficient of variation was less than 0.95% for all quantifiable values. To control for the quality of interrogated DNA, we performed a parallel amplification reaction using 10 ng of genomic DNA and a primer/probe combination specific for a non-transcribed genomic sequence upstream of the CDKN1Ap21 gene. These amplification reactions generated a correction factor to adjust for calculated versus actual DNA input. Copies of transgene per microgram of DNA were calculated according to the following formula: copies per microgram of genomic DNA = (copies calculated from CAR T standard curve) × correction factor / (amount DNA evaluated in nanograms) × 1,000 ng.

Data extraction of CAR transgene level from CARTITUDE-1 trial

We used PlotDigitizer⁵⁶ to extract data from ‘Supplementary Fig. 2: CAR-T expansion and persistence in peripheral blood’ from the published CARTITUDE-1 study⁵. We began by determining the x and y coordinates and the scale of the axes by selecting and inputting values for the origin, the maximum x axis value and the maximum y axis value. Next, we extracted the values of all data points from each session in the analysis. To capture each data point, we clicked on its center, which generated the corresponding x and y coordinates. The extracted data were then compiled into a Microsoft Excel spreadsheet using PlotDigitizer.

Integration site sequencing

Integration site isolation and sequencing were carried out using ligation-mediated PCR as previously described⁵⁷. In brief, genomic DNA (400–1,000 ng) was sheared using a Covaris M220 ultrasonicator to achieve a fragment size of 600–900 bp, followed by bead purification and linker ligation. Nested PCR was performed to amplify from the linker to the vector long terminal repeat (LTR). Amplified samples were pooled, purified, quantified and sequenced on an Illumina MiniSeq sequencer with a library loading concentration of 1.2 pM with a 40% PhiX spike-in. Sequence reads were quality controlled, aligned to the human genome draft hg38 and analyzed as described^58,59. Integration site seq oligos are listed in Supplementary Table 10. All integration site analysis data are available at the National Center for Biotechnology Information (NCBI) Sequence Read Archive (SRA) under accession number PRJNA1235822. Integration site analysis was carried out at the Viral Molecular High Density Sequencing Core at the University of Pennsylvania under the guidance of Frederic Bushman (RRID: SCR_022433).

Cilta-cel patient 4 flow cytometry assessment of TCR Vβ repertoires

The IOTest Beta Mark TCR Vβ Repertoire Kit (Beckman Coulter) was used to characterize TCR Vβ chains. The kit is composed of eight vials containing mixtures of conjugated TCR Vβ antibodies corresponding to 24 different specificities. A total of 24 TCR Vβ families were stained across eight separate wells, with anti-TCR Vβ antibodies labeled using PE, FITC and PE+FITC. BCMA-targeted CAR T cell expression was assessed using a bis-biotinylated huBCMA-Fc recombinant protein (iPROT ID 100536) as the primary antibody, followed by Streptavidin-AF647 (BioLegend) as the secondary antibody. To ensure accurate detection, fluorescence minus one (FMO) controls were performed for BCMA-targeted CAR T cell detection. Dead cells were excluded using the LIVE/DEAD Fixable Aqua Dead Cell Kit (Life Technologies) for 20 minutes at room temperature. Cells were then blocked with 10% human albumin (GeminiBio) for 15 minutes at room temperature, followed by incubation with 1 μg ml⁻¹ BCMA-targeted CAR T cell primary antibody for 20 minutes at room temperature. After four washes with flow buffer (PBS with 1% FBS and 0.02% sodium azide), surface staining was performed using specific surface marker antibodies, Streptavidin-AF647 (BioLegend) for BCMA-targeted CAR T cell detection and anti-TCR Vβ antibodies in BD Horizon Brilliant Stain Buffer (BD Biosciences) for 20 minutes at room temperature. Finally, cells were washed with flow buffer, resuspended in PBS and acquired using a BD Fortessa flow cytometer (BD Biosciences) equipped with violet (405 nm), blue (488 nm), green (560 nm) and red (640 nm) lasers. Compensation values were determined using eBioscience UltraComp eBeads (eBioscience) and BD FACSDiva software (BD Biosciences). Data analysis was performed using FlowJo (BD Life Sciences). The surface staining panel included anti-CD8 APC-H7 (clone SK1), anti-CD4 PerCP-Cy5.5 (clone RPA-T4), anti-CD14 V500 (clone MφP9), anti-CD19 V500 (clone HIB19) (all from BD Biosciences), anti-CD3 BV605 (clone OKT3) and Streptavidin-AF647 (both from Biolegend) and anti-TCR Vβ antibodies labeled with PE, FITC and PE+FITC.

Multiparameter flow cytometry of patient samples

Cryopreserved PBMCs and/or CSF from patients were thawed and counted. The dead cells were excluded using the LIVE/DEAD Fixable Blue Dead Cell Kit (Life Technologies) for 20 minutes at room temperature. Cells were first blocked with 10% human albumin (GeminiBio) for 15 minutes at room temperature and then incubated with 1 μg ml⁻¹ BCMA-targeted CAR T cell primary antibody for 20 minutes at room temperature, followed by four washes with flow buffer (PBS with 1% FBS and 0.02% sodium azide). Further blocking was performed using Human TruStain FcX (BioLegend) for 10 minutes at room temperature before staining with specific surface marker antibodies and the secondary antibody Streptavidin-AF647 (BioLegend) for BCMA-targeted CAR T cell detection in BD Horizon Brilliant Stain Buffer Plus (BD Biosciences) and True-Stain Monocyte Blocker (BioLegend) for 20 minutes at room temperature. Finally, cells were washed, fixed and permeabilized using the FOXP3 buffer set (Invitrogen) and stained with intracellular antibodies in permeabilization buffer and BD Horizon Brilliant Stain Buffer Plus for 20 minutes at room temperature. Then, cells were acquired using a Cytek Aurora flow cytometer (Cytek) equipped with a UV laser (355 nm), a violet (405 nm), a blue (488 nm), a green (560 nm) and a red (640 nm) laser and ensuring proper spectral unmixing and compensation. Compensation values (or reference controls) were established using paired normal donor PBMCs and activated PBMCs and cells with CAR BCMA expression. Data were analyzed using FlowJo (BD Life Sciences). Surface staining antibodies were as follows: anti-CD45RA BUV395 (clone HI100), anti-CD16 BUV496 (clone 3G8), anti-TIM3 BUV615 (clone 7D3), anti-CD28 BUV737 (clone CD28.2), anti-CD8 BUV805 (clone SK1), anti-CD19 BV480 (HIB19), anti-CD3 BV750 (clone SK7), anti-CD95 BB700 (clone DX2) and anti-CCR7 PE-CF594 (clone 150503) (all from BD Biosciences); anti-PD-1 BV421 (clone EH12.2H7), anti-CD27 BV510 (clone O323), anti-CD33 BV570 (clone WM53), anti-CD56 BV605 (clone HCD56), anti-HLA-DR BV711 (clone L243), anti-CD4 BV785 (clone OKT4) and antiCD45 APC-Fire810 (clone HI30) (all from BioLegend); and anti-CD123 SB436 (clone 6H6) and anti-LAG3 PE-Cy7 (clone 3DS223H) (both from Invitrogen). Intracellular staining antibodies were as follows: anti-CTLA-4 PE-Cy5 (clone BNI3) and anti-Ki67 AF700 (clone B56) (both from BD Biosciences); anti-granzyme B APC-Fire750 (clone QA16A02) (from BioLegend); anti-EOMES FITC (clone WD1928) (from Life Technologies); and anti-TOX PE (clone REA473) (from Miltenyi Biotec).

Serum and CSF cytokine profiling

Cryopreserved serum and CSF samples stored at −80 °C were thawed and analyzed for the following 30 human analytes using either the human cytokine custom 32-plex panel (SPRCUS1673, EMD Millipore) or the 48-plex panel (HCYTA-60K-PX48, EMD Millipore): EGF, FGF-2, Eotaxin, G-CSF, GM-CSF, IFNα2, IFNγ, IL-10, IL-12P40, IL-12P70, IL-13, IL-15, IL-17A, IL-1RA, IL-1β, CXCL9/MIG, IL-2, IL-4, IL-5, IL-6, IL-7, CXCL8/IL-8, CXCL10/IP-10, CCL2/MCP-1, CCL3/MIP-1α, CCL4/MIP-1β, RANTES, TNF, VEGF and IL-18. The assay was set up following the manufacturer’s protocol. Assay plates were measured using a FLEXMAP 3D instrument, and data acquisition and analysis were done using xPONENT (version 4.2) software. Data quality was examined based on the following criteria. The standard curve for each analyte has a five-parameter R² > 0.95 with or without minor fitting using xPONENT software. To pass assay technical quality control, the results for two controls in the kit needed to be within the 95% confidence interval provided by the vendor for more than 25 of the tested analytes. No further tests were done on samples with results out of range low (<OOR). Samples with results that were out of range high (>OOR) or greater than the standard curve maximum value (SC max) were not tested at higher dilutions without further request.

Cilta-cel and ide-cel CAR constructs

The cilta-cel CAR construct was kindly provided by Neil Sheppard (University of Pennsylvania), and its sequence was confirmed using the referenced patient document²⁸. The ide-cel CAR sequence was obtained from the corresponding patent document²⁸. Sequences of both CAR transgenes are listed in Supplementary Table 6. Cilta-cel and ide-cel CAR constructs were cloned into a pTRPE vector containing a P2A-linked tNGFR backbone.

Lentivirus production

Third-generation, replication-defective lentiviral vectors were generated using HEK293T cells (American Type Culture Collection, ACS-4500). Transfection was performed using Lipofectamine 2000 (Invitrogen) along with the packaging plasmids pMDG.1, pRSV.rev and pMDLg/p.RRE as well as plasmid-containing CAR construct (Supplementary Table 5). At 24 hours and 48 hours after transfection, the culture supernatant was collected, filtered and concentrated via high-speed ultracentrifugation (8,000g for 16 hours or 25,000g for 2.5 hours). The resulting lentiviral concentrate was stored at −80 °C.

CAR T cell manufacturing

Healthy donor T cells were isolated from PBMCs using the Pan T Cell Isolation Kit according to the manufacturer’s instructions (Miltenyi Biotec). T cells were activated with anti-CD3/CD28 antibody-coated Dynabeads (Thermo Fisher Scientific) at 3:1 bead:cell ratio in T cell medium (OpTmizer CTS SFM medium (Gibco) supplemented with 5% human AB serum and human IL-2 (100 μm ml⁻¹)). After a 24-hour incubation, lentivirus encoding the cilta-cel CAR was added to the culture at a multiplicity of infection (MOI) of 2.5. CAR T cell expansion was carried out for 9 days.

In vitro cytokine-stimulated proliferation assays

Healthy donor CAR T cells and PBMCs from cilta-cel patient 4 were thawed. For cilta-cel patient 4, T cells were isolated using the Pan T Cell Isolation Kit (Miltenyi Biotec). BCMA-targeted CAR T cells were resuspended in PBS with CellTrace Violet (CTV) stain (1:1,000 dilution, Thermo Fisher Scientific), at a concentration of 1 × 10⁶ per milliliter, and incubated for 15 minutes at 37 °C. T cells were washed with 10 ml of R10 medium and stimulated with IL-2 (10 ng ml⁻¹, PeproTech), IL-7 (20 ng ml⁻¹, PeproTech) or IL-15 (20 ng ml⁻¹, PeproTech) and all respective permutations. After 5 days, proliferation via CTV, BMCA-CAR (MonoRab Rabbit Anti-Camelid VHH Cocktail, GenScript), phosphorylated STAT5 (clone 47/Stat5 (pY694) BD Biosciences), IL-2RA (clone BC96, BioLegend), IL-7RA (clone A019D5, BioLegend) and IL-15RA (clone JM7A4, BD Biosciences) expression was assessed via flow cytometry.

Degranulation and cytokine production assay

Healthy donor CAR T cells and PBMCs from cilta-cel patient 4 were thawed and incubated with MM.1S for 72 hours. For cilta-cel patient 4, T cells were isolated using the Pan T Cell Isolation Kit (Miltenyi Biotec) prior to incubation. Then, to assess intracellular cytokine production and cytotoxic granule content, cells were cultured in the presence of BD GolgiPlug (brefeldin A), BD GolgiStop (monensin) and BD Horizon BUV395 mouse anti-human CD107a antibody (clone 565113, BD Biosciences) for 5 hours. After incubation, cells were harvested and stained with CD8 (clone RPA-T8, BD Biosciences), CD3 (clone UCHT3, BioLegend), CD4 (clone OKT4, BD Biosciences), CCR7 (clone G043H7, BioLegend), CD45RA (clone HI100, BioLegend), BMCA-CAR (MonoRab Rabbit Anti-Camelid VHH Cocktail, GenScript) and BD Horizon Fixable Viability Stain 780 (BD Biosciences) for 20 minutes on ice. Cells were then fixed and permeabilized using the FOXP3 buffer set (Invitrogen) and stained in permeabilization buffer with intracellular antibodies IL-2 (clone MQ1–17H12, BioLegend), IL-17A (clone BL168, BioLegend), IFNγ (clone 4S.B3, BioLegend), TNF (clone MAb11, BioLegend), GM-CSF (clone BVD2–21C11, BioLegend), perforin (clone 8G9, BD Biosciences) and granzyme B (clone GB11, BD Biosciences). Then, samples were acquired using a Cytek Aurora flow cytometer (Cytek).

Proliferation assays of CRISPR–Cas9-mediated knockout of CCR5

On day 3 after healthy donor CAR T cells manufacturing (as described above), CD3/CD28 Dynabeads were removed using magnetic separation. CAR T cells were electroporated using the P3 Primary Cell 4D-Nucleofector Kit (Lonza). A total of 2 × 10⁶ CAR T cells were resuspended in P3 solution and preincubated with 6 μg of TrueCut Streptococcus pyogenes Cas9 (Invitrogen) and 0.1 nmol of chemically modified trans-activating CRISPR RNA (tracrRNA) and CRISPR RNA (crRNA) (Integrated DNA Technologies). Electroporation was performed with the 4D-Nucleofector X Unit (program EO-115, Lonza), and cells were cultured as previously described. The crRNA sequences used were as follows: AAVS1: 5′-ACCTCTAAGGTTTGCTT-3′ and CCR5: 5′-CCTGCCTCCGCTCTACTCAC-3′. Target loci were PCR amplified using the following primers: CCR5-F1: 5′-TGCTTGGCCAAAAAGAGAGTTA-3′ and CCR5-R1: 5′-TCCATGCTGTGTTTGCTTTAAA-3′. The frequency of CRISPR–Cas9-induced insertion–deletion (indel) mutations was quantified using Inference of CRISPR Edits (ICE) analysis. The CAR T cells (CCR5 KO or AAVS1 KO) were resuspended in PBS with CTV stain (1:1,000 dilution, Thermo Fisher Scientific) at a concentration of 1 × 10⁶ per milliliter and incubated for 15 minutes at 37 °C. After stimulation with MM.1S for 72 hours, the T cells were stained using CD8 (clone RPA-T8, BD Biosciences), CD3 (clone UCHT3, BioLegend), CD4 (clone OKT4, BD Biosciences) and BD Horizon Fixable Viability Stain 780 (BD Biosciences) for 20 minutes on ice. Then, samples were acquired with 123count eBeads Counting Beads (Invitrogen) using a Cytek Aurora flow cytometer (Cytek).

CCR5 inhibition assay with maraviroc

To find the optimal conditions with regards to viability, T cells were stimulated with CD3/CD28 Dynabeads and different concentrations of maraviroc (Selleck Chemicals) ranging from 0 μM to 40 μM for 72 hours. Healthy donor CAR T cells were then incubated at the optimal maraviroc concentration (40 μM) with either 25% of cilta-cel patient 4 serum (days 7 and 14) or IL-15 (20 ng ml⁻¹, PeproTech). After 72 hours, T cells were stained using CD8 (clone RPA-T8, BD Biosciences), CD3 (clone UCHT3, BioLegend), CD4 (clone OKT4, BD Biosciences) and BD Horizon Fixable Viability Stain 780 (BD Biosciences) for 20 minutes on ice. Then, samples were acquired with 123count eBeads Counting Beads (Invitrogen) using a Cytek Aurora flow cytometer (Cytek).

Antibody-mediated CCR5 inhibition

T cells from cilta-cel patient 4 were resuspended in PBS containing CTV (1:1,000 dilution, Thermo Fisher Scientific) at a concentration of 1 × 10⁶ cells per milliliter and incubated for 15 minutes at 37 °C. After a PBS wash, cells were incubated with IL-15 (20 ng ml⁻¹, PeproTech) and CCR5 antibody (clone J418F1, BioLegend), either alone or in combination, depending on the experimental condition. Surface staining was then performed on ice for 20 minutes using CD8 (clone RPA-T8, BD Biosciences), CD3 (clone UCHT3, BioLegend), CD4 (clone OKT4, BD Biosciences) and BD Horizon Fixable Viability Stain 780 (BD Biosciences). Samples were acquired on a Cytek Aurora flow cytometer (Cytek) with the inclusion of 123count eBeads Counting Beads (Invitrogen) for absolute quantification.

Triple-reporter Jurkat T cell activation assay ide-cel versus cilta-cel

To directly test whether ectodomain and/or linker differences contribute to CirAE risk, we generated cilta-cel and ide-cel CAR constructs using sequences reported in their respective patents. Both were cloned into a common pTPRE plasmid backbone and included a truncated rat nerve growth factor receptor (tNGFR) via T2A as a control reporter for transduction efficiency. We transduced the SUPT1 T cell line and the Jurkat triple-parameter-reporter (TPR) T cell line²⁹ with cilta-cel and ide-cel constructs, normalizing for transduction efficacy using fluorescence-activated cell sorting based on tNGFR MFI. We then conducted a series of in vitro mechanistic studies to assess (1) BCMA-specific binding avidity and (2) downstream activation of NFAT and NF-κB signaling pathways in the presence of BCMA-expressing multiple myeloma cell lines.

Chemotaxis Transwell assay

T cell migration was assessed using Corning HTS Transwell-96 permeable supports with 5.0-μm polycarbonate membranes (Corning). Cells were serum starved overnight in RPMI supplemented with 1% FBS. The next day, 250,000–500,000 cells in 75 μl of serum-free medium were seeded into the upper chamber, and 225 μl of complete medium containing the chemoattractant was added to the lower chamber. After overnight incubation, migrated cells were collected from the bottom well and stained with CD8 (clone RPA-T8, BD Biosciences), CD3 (clone UCHT3, BioLegend), CD4 (clone OKT4, BD Biosciences) and BD Horizon Fixable Viability Stain 780 (BD Biosciences). Samples were analyzed by flow cytometry.

Luciferase cell killing assay

RPMI 8266 and MM.1S cell lines were engineered to co-express CBG and GFP. Tumor cell survival was assessed via bioluminescent quantification. D-luciferin potassium salt (Gold Biotechnology, 115144-35-9) was added to cultures at a final concentration of 15 μg ml⁻¹ and incubated at 37 °C for 10 minutes. Bioluminescence was then measured using a BioTek Synergy H4 plate reader and analyzed with Gen5 software. To assess tumor clearance, luminescence values were normalized to appropriate controls (either CAR-negative T cell co-cultures or tumor-only wells).

Metabolic assessment of ide-cel and cilta-cel Jurkat TPR cells

Mitochondrial function was evaluated using the Agilent/Seahorse Bioscience extracellular flux analyzer. Individual wells of an XF96 cell culture microplate were coated with Cell-Tak following the manufacturer’s protocol. The matrix was adsorbed overnight at 37 °C and then aspirated, air dried and stored at 4 °C until use. For mitochondrial functional assays, cells were centrifuged at 500g for 5 minutes and washed in 1× PBS. The cell pellets were resuspended in XF assay medium and seeded at a density of 2 × 10⁵ cells per well. During instrument calibration, the microplate was centrifuged at 1,000g for 2 minutes and incubated in a CO₂-free environment at 37 °C for 15 minutes. Oligomycin A, BAM15 and rotenone were used as the inhibitors for this assay. Cellular oxygen consumption rates (OCRs) and extracellular acidification rates (ECARs) were recorded under basal conditions and after sequential treatments with 1.5 μM oligomycin A, 2.5 μM BAM15 and 0.5 μM rotenone + 0.5 μM antimycin A (Seahorse XF T Cell Metabolic Profiling Kit, Agilent Technologies). Dynabead-activated murine T cells or human T cells (from patient CART19 expansions or healthy volunteers trial) were resuspended in XF assay medium supplemented with 10 mM carnosine and 30 mM glycylsarcosine (Gly-Sar) from Sigma-Aldrich. In all instances, the pH was adjusted to 7.4 before the assay. MitoATP production is the rate of ATP production (expressed in pmol ATP per minute) specific to the mitochondrial oxidative phosphorylation. GlycoATP production is the rate of ATP production (expressed in pmol ATP per minute) correlated with the glycolytic pathway.

Analysis of flow cytometry data

Analysis was performed using FlowJo version 10.10 software (BD Life Sciences). SPICE plots were generated from single-gated markers, grouped using Boolean combinatorial gates and plotted using SPICE 6.1 software (https://niaid.github.io/spice/). For high-dimensional analysis, cells were gated on viable, singlet lymphocytes (CD3⁺, CD19⁻, CD14⁻, CD16⁻) and further pre-gated for CAR⁺ in all samples, except for those acquired pre-lymphodepletion (Supplementary Fig. 5). The pre-gated populations were exported as CSV files and further processed in R version 4.3 using the tidyverse package. Data were downsampled to 100,000 cells per sample and transformed using asinh. The data were then processed into a Seurat object using the Seurat package for dimensionality reduction and UMAP clustering with a nearest neighbors of 20. Visualization was performed using built-in Seurat functions and the ggplot2 package. FlowSOM clustering was conducted using the FlowSOM package, identifying 10 metaclusters. Heatmaps were generated using the pheatmap package. The following markers were used for UMAP and FlowSOM analysis: CD34, granzyme B, CD45, BMCA-CAR, Ki67, CD95, CD45RA, TIM3, CD28, CD8, PD1, CD27, CD33, CD56, HLA-DR, CD4, EOMES, TOX, CCR7, CTLA4, LAG3 and CD123.

Single-cell RNA sequencing and analysis

Single-cell RNA sequencing was performed using the 10x Genomics Chromium GEM-X Single Cell 3’ v4 Gene Expression assay following the manufacturer’s protocol (10x Genomics). In brief, single-cell suspensions derived from PBMCs and CSF were evaluated for viability and concentration prior to loading onto the Chromium X instrument for gel beads-in-emulsion (GEM) generation, enabling the barcoding of individual cells and their transcripts. Reverse transcription and cDNA amplification were carried out within the GEMs, followed by library construction, using the Chromium GEM-X Single Cell 3’ Kit (PN-1000691), the Chromium GEM-X Single Cell 3’ Chip Kit (PN-1000690) and the Dual Index Kit TT Set A (PN-1000215). Final libraries were quantified, pooled and sequenced on an Illumina NovaSeq X system using the 10B Reagent Kit (100 cycles) to a depth of 100,000 reads per cell. Reads were aligned to the human reference genome (GRCh38), augmented with a custom reference to detect the cilta-cel transgene, using Cell Ranger version 8.0.0. Quality control and downstream analyses were performed in Seurat version 5.0. Cells with fewer than 200 or more than 10,000 detected genes, or more than 10% mitochondrial gene content, were excluded. CD3⁺, CD4⁺, CD8⁺ and cilta-cel⁺ cells were retained, yielding 42,123 cells. Gene–cell matrices from eight samples were integrated using SelectIntegrationFeatures, FindIntegrationAnchors and IntegrateData. Dimensionality reduction was performed via UMAP on the top 15 principal components, using 15 neighbors and two output dimensions. Clustering was conducted using FindNeighbors and FindClusters, with differential gene expression analyzed using the FindMarkers function. Gene expression densities were visualized with Nebulosa version 1.12, and single-cell pathway enrichment scores were calculated using AUCell version 1.24.

Cell-to-cell avidity assay

Avidity was measured using the LUMICKS Z-Movi assay. Tumor cells were incubated on poly-L-lysine-coated Z-Movi-compatible acoustofluidic chips for 3 hours to generate a stable monolayer. CellTrace Far Red-labeled CAR T cells were then added and allowed to interact with the tumor monolayer for 10 minutes, after which a ramping acoustic force was applied. Cell detachment was analyzed using ImageJ and R. Avidity was normalized to the percentage of cells remaining bound at 30 pN. All experiments were performed according to the manufacturer’s (LUMICKS) instructions.

Statistical analysis

Categorical variables were evaluated with Fisher’s exact or Kruskal–Wallis tests as appropriate. Unpaired continuous variables were evaluated by Mann–Whitney test. Paired continuous variables were evaluated by the Wilcoxon matched-pairs test. Survival analyses were conducted using the Kaplan–Meier method and compared with the log-rank test. Univariate tests were conducted using logistic regression. All tests were performed using a nominal significance threshold of P < 0.05. Analyses were performed using R version 4.4.2 and GraphPad Prism version 10 (GraphPad Software).

Extended Data

Extended Data Fig. 1 |. Stratified analysis of ciltacabtagene autoleucel (cilta-cel) vs idecabtagene vicleucel (ide-cel).

(a) Cilta-cel and ide-cel infusion dates at our institution. (b) Response rate of the entire BCMA-CART cohort, and by cilta-cel and ide-cel. Categorical variables were compared with Fisher’s exact test. (c) Progression-free survival analysis (PFS) of the entire cohort. (d) Overall survival (OS) analysis of the entire cohort. (e) Progression-free survival analysis stratified by product. Survival distributions were compared with the log-rank test. (f) Overall survival analysis stratified by product. Survival distributions were compared with the log-rank test. (g) Cytokine release syndrome (CRS) grade stratified by product. Categorical variables were compared with Fisher’s exact test. (h) Immune effector cell-associated neurotoxicity syndrome (ICANS) grade stratified by product. Categorical variables were compared with Fisher’s exact test. (i) Absolute lymphocyte count (ALC; median with interquartile range) within the first 24 days following BCMA-CART, and at Day 28, Month 3, and Month 6, stratified by product received. Data are presented as mean values ± 95% confidence interval. Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. Exact P-values are provided in Supplementary Table S9. (j) Peak absolute lymphocyte count (ALC^Peak) at apheresis, Day 0, Day 7, Day 14, and Day 21 stratified by product. Data are presented as median values ± interquartile range. Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. Exact P-values are provided in Supplementary Table S9. (k) CD4⁺ and CD8⁺ proportion of T-cells at apheresis, Days 7 and 28, Months 3 and 6 post-BCMA-CART, stratified by product. Bar graphs show the median and the error bars denote the 25^th and 75^th percentiles (top). Area under the curve (AUC) analysis of CD4⁺ T-cell proportions from apheresis to Month 6 in the cilta-cel cohort, stratified by product. Data are shown as median values with error bars representing the 25th and 75th percentiles. Shaded regions indicate the area under the curve (AUC) for each group (bottom). Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. * Indicates P < 0.05, ** indicates P < 0.01, *** indicates P < 0.001. (l) Non-relapse mortality of the entire cohort. (m) Non-relapse mortality analysis stratified by product. Survival distributions were compared with the log-rank test. ALC^Peak: peak absolute lymphocyte count; AUC: area under curve; CART: CAR T-cell; cilta-cel: ciltacabtagene-autoleucel; CI: confidence interval; CirAE: CART-associated immune-related adverse events; CRS: cytokine release syndrome; ICANS: immune effector cell-associated neurotoxicity syndrome; NRM: non-relapse mortality; ORR: overall response rate; OS: overall survival; ide-cel: idacabtagene-vicleucel; PFS: progression free survival; ≥ VGPR: very good partial response or better.

Extended Data Fig. 2 |. CART-associated immune-related adverse events (CirAE) Clinical correlates in the cilta-cel cohort.

(a) Response rate in the cilta-cel cohort stratified by CART-associated immune-related adverse events, CirAE. Categorical variables were compared with Fisher’s exact test. (b) Progression-free survival (PFS) analysis of the cilta-cel cohort, stratified by CirAE. Survival distributions were compared with the log-rank test. (c) Overall survival (OS) analysis of the cilta-cel cohort, stratified by CirAE. Survival distributions were compared with the log-rank test. (d) Non-relapse mortality (NRM) analysis of the cilta-cel cohort, stratified by CirAE. Survival distributions were compared with the log-rank test. (e) Pre-infusion M-spike levels of the cilta-cel cohort, stratified by CirAE. Data are shown as median values with error bars representing the 25th and 75th percentiles. Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. (f) Pre-infusion differential free light chain (dFLC) levels of the cilta-cel cohort, stratified by CirAE. Data are shown as median values with error bars representing the 25th and 75th percentiles. Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. (g) Cytokine release syndrome (CRS) grade of the cilta-cel cohort, stratified by CirAE. Categorical variables were compared with Fisher’s exact test. (h) Immune effector cell-associated neurotoxicity (ICANS) grade of the cilta-cel cohort, stratified by CirAE. Categorical variables were compared with Fisher’s exact test. (i) Top: absolute lymphocyte count (ALC; median with interquartile range) within the 28 days following BCMA-CART, and at Day 28, Month 3 and Month 6, in the cilta-cel cohort, stratified by CirAE. Data are presented as mean values ± 95% confidence interval. Bottom: peak absolute lymphocyte count (ALC^Peak) at apheresis, Day 0, Day 7, Day 14, and Day 21 of the cilta-cel cohort, stratified by CirAE (bottom). Data are presented as median values ± interquartile range. Day 14 ALC^Peak of 2.9 falls within the third quartile of the cilta-cel non-CirAE cohort distribution. Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. * Indicates P < 0.05, ** indicates P < 0.01, *** indicates P < 0.001. Exact P-values are provided in Supplementary Table S9. (j) Top: CD4⁺ and CD8⁺ proportion of T-cells in the cilta-cel cohort, stratified by CirAE. Bar graphs show the median and the error bars denote the 25^th and 75^th percentiles. Bottom: area under the curve (AUC) analysis of CD4⁺ T-cell proportions from apheresis to Month 6 in the cilta-cel cohort, stratified by CirAE. Data are shown as median values with error bars representing the 25th and 75th percentiles. Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. Exact P-values are provided in Supplementary Table S9. (k) Univariable logistic regression was performed using a binomial generalized linear model to assess associations between clinical factors and CirAE (left), neurologic CirAE (neuro-CirAE, middle), and non-neuro CirAE (right). Statistical significance of model coefficients was assessed using two-tailed Wald tests. (l) Multivariable logistic regression analysis of risk factors for CirAE development in the cilta-cel cohort. Statistical significance of model coefficients was assessed using two-tailed Wald tests. ALC^Peak: peak absolute lymphocyte count; AUC: area under curve; CART: CAR T-cell; cilta-cel: ciltacabtagene-autoleucel; CI: confidence interval; CirAE: CART-associated immune-related adverse events; CRS: cytokine release syndrome; ICANS: immune effector cell-associated neurotoxicity syndrome; NRM: non-relapse mortality; ORR: overall response rate; OS: overall survival; ide-cel: idacabtagene-vicleucel; PFS: progression free survival; ≥ VGPR: very good partial response or better.

Extended Data Fig. 3 |. Clinical imaging of CART-associated immune-related adverse events (CirAE) and colon biopsy results in patients who developed immune effector cell-associated enterocolitis.

(a) Clinical imaging of post-BCMA-CART CirAE: Cilta-cel#21: Magnetic resonance imaging (MRI) of the head and orbits illustrates enhancement within the left orbital apex extending to the left superior orbital fissure. There is thickening and edema in the left prechiasmatic optic nerve and left optic chiasm and enhancement within and along the left intraorbital, intracanalicular, and prechiasmatic optic nerve/sheath complex, consistent with optic neuritis. Chimeric antigen receptor (CAR) transgene was detectable at 59596 copies/μg gDNA in the cerebrospinal fluid (CSF). Cilta-cel#14: MRI of the head shows a rim-enhancing lesion in bilateral medial thalami extending to the hypothalamus measuring approximately 2.8 ×2.7 ×2.2 cm, with surrounding FLAIR signal abnormality extending bilaterally into the surrounding thalami, gangliocapsular regions, mammillary bodies, left greater than right midbrain, and optic chiasm. Ide-cel#1: CT chest showing multifocal tree-in-bud ground-glass opacities consistent with pneumonitis. Cilta-cel#4: endoscopy revealed many non-bleeding, cratered duodenal ulcers, and erosions without stigmata of bleeding beyond the second portion of the duodenum. Colonoscopy did not show any areas of ulcerations or erosions within the terminal ileum and colon. (b) H&E shows duodenal mucosa with graft-versus-host and Celiac disease-like patterns of injury, changes compatible with treatment (ciltacabtagene autoleucel) effects, and mild acute inflammation. Circle highlights villous atrophy and crypt hyperplasia. Arrow highlights prominent crypt apoptosis. RNAscope shows CART infiltration. BCMA was not detected. Representative micrographs from a single experiment; imaging was performed once and not independently replicated due to limited available tissue. (c) H&E shows colonic mucosa with mild crypt architectural distortion and markedly reduced plasma cells in the lamina propria. RNAscope shows CART infiltration. BCMA was not detected. Circle highlights a paucity of plasma cells in the lamina propria, resembling histologic features observed in common variable immunodeficiency. Representative micrographs from a single experiment; imaging was performed once and not independently replicated due to limited available tissue. BCMA: B-cell maturation antigen; CART: chimeric antigen receptor T-cell; Cy5: cyanine 5; FITC: fluorescein isothiocyanate; MRI: magnetic resonance imaging.

Extended Data Fig. 4 |. in vitro functionality analysis of Cilta-cel#4’s CARTs.

(a) Heatmap shows z-scores of the log₂ fold changes for each marker across each row. Left: antigen-specific degranulation and cytokine production in CD4⁺ and CD8⁺ CARTs from peripheral blood (PB) and cerebrospinal fluid (CSF) from Cilta-cel#4 following coculture with MM.1S, compared to PB CARTs from three normal donors and Cilta-cel#22. Right: fold change of CD4⁺/CD8⁺ CART degranulation and cytokine production in the same samples. (b) In vitro cytotoxicity assay comparing Cilta-cel#4 PB CARTs at Day 13 with CARTs manufactured from three normal donors, using MM.1S as target. An effector: tumor ratio of 1:1 was used. Data are shown as mean values with error bars representing the standard error of the mean (SEM) from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s T-test. * Indicates P < 0.05, ** indicates P < 0.01. Exact P-values are provided in Supplementary Table S9. (c) In vitro cytotoxicity assay comparing Cilta-cel#4 PB and CSF CARTs at Month 2 (M2) with to Cilta-cel#22 PB CARTs, using RPMI-8226 as target. A range of effector: tumor ratios between 2:1 and 1:2 was used. Data are shown as mean values with error bars representing the standard error of the mean (SEM) from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s T-test. Exact P-values are provided in Supplementary Table S9. (d) Serum cytokine profiling of Cilta-cel#4 from pre-lymphodepletion (Pre-LD), Day 7, peak expansion (Day 13), to Month 2. Left: heatmap shows z-score normalized cytokine concentration. Right, top: temporal kinetics of cytokines associated with T-cell proliferation (IL-1b, IL-2, IL-4, IL-7, IL-15, IL-18). Right, bottom: Temporal kinetics of cytokines implicated in Cytokine release syndrome (CRS; IL-6, TNF-α, GM-CSF, IL-8, IL-10). ^ represents cytokine concentration above upper limit of quantification of the LUMINEX assay. (e) CellTrace Violet-stained CARTs from three normal donors were seeded into 96-well plates, stimulated with IL-2: 10 ng/mL, IL-7: 20 g/mL, or IL-15: 20 ng/mL, and incubated for five days. Left: proliferation was assessed by flow cytometry analysis of CellTrace Violet dilution. Right: phosphorylated STAT5 (pSTAT5⁺) activation following 15 min of activation with IL-2 (10 ng/mL), IL-7 (20 ng/mL), or IL-15 (20 ng/mL) in 3 normal donors. Data are shown as mean values with error bars representing the standard error of the mean (SEM) from n = 3 biological replicates. (f) Baseline IL2, IL7, and IL15 receptor-alpha (RA) chain expression in Cilta-cel#4 compared to 3 normal donors before cytokine stimulation. Data are shown as mean values with error bars representing the standard error of the mean (SEM) from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s T-test. * Indicates P < 0.05, ** indicates P < 0.01, *** indicates P < 0.001. Exact P-values are provided in Supplementary Table S9. (g) CCR5 median fluorescence intensity (MFI) in three normal donors and Cilta-cel#4 following cytokine stimulation with IL-2 (10 ng/mL), IL-7 (20 ng/mL), or IL-15 (20 ng/mL). Data are shown as mean values with error bars representing the standard error of the mean (SEM) from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s T-test. * Indicates P < 0.05, ** indicates P < 0.01, *** indicates P < 0.001. Exact P-values are provided in Supplementary Table S9. (h) Absolute count (left) and viability (right) of CD3⁺ T-cells from 3 normal donors after 3 days of CD3/28 stimulation in the presence of increasing maraviroc concentrations (2.5 μM, 5 μM, 10 μM, 20 μM, 40 μM). Data are shown as mean values with error bars representing the standard error of the mean (SEM) from 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s T-test. * Indicates P < 0.05, ** indicates P < 0.01, *** indicates P < 0.001. CSF: cerebrospinal fluid; E:T: effector: tumor ratio.

Extended Data Fig. 5 |. Cilta-cel and Ide-cel in vitro functional differences.

(a) Overlay histogram of tNGFR and BCMA-CAR expression in SUP-T1 cell line stably transduced with ide-cel CAR (ide-cel), cilta-cel CAR (cilta-cel), mesothelin-directed CAR (M5) or untransduced control cells (UTD). (b) Comparison of BCMA-binding avidity between ide-cel vs cilta-cel, with M5 and UTD as controls. Left: Percentage of MM.1S cells bound across a range of applied forces. Right: Percentage of MM.1S binding at 1000 pN force; data are presented as mean values ± the standard error of the mean (SEM) from 2–5 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s T-test. (c) Overlay histogram of tNGFR and VHH-CAR expression in Jurkat triple parameter reporter (Jurkat_TPR) cell line stably transduced with the ScFv-based ide-cel CAR (ide-cel) or VHH-based cilta-cel CAR (cilta-cel). (d) Baseline (tonic) NFAT and NKFB activation status in Jurkat_TPR cells expressing ide-cel and cilta-cel in the absence of antigen stimulation; data are presented as mean values ± the standard error of the mean (SEM) from n = 3 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s T-test. (e) Left: fold-change from unstimulated in NFAT and NKFB % positive cells (top left) and median fluorescence intensity (MFI, bottom left) in Jurkat_TPR expressing ide-cel or cilta-cel following 18 and 24 h of stimulation with MM.1S with a range of effector: tumor ratios between 5:1 to 1:5. Right: fold-change from unstimulated in NFAT and NKFB % positive cells (top right) and median fluorescence intensity (MFI, bottom right) in Jurkat_TPR expressing ide-cel or cilta-cel following 1:1 stimulation with MM.1S across multiple timepoints ranging from 3 to 72 h. Data are shown as mean values with error bars representing the standard error of the mean (SEM) from 3 biological replicates. (f) Jurkat_TPR expressing ide-cel or cilta-cel were co-cultured 1:1 with irradiated MM.1S for 24 to 48 h, then separated from MM.1S using tNGFR-PE magnetic bead (positive) selection. The contour plot confirms high post-sort purity of Jurkat_TPR cells used for subsequent Seahorse metabolic analysis. (g) Mitochondrial respiration, (h) Maximal and spare respiratory capacity, (i) Glycolytic rate, (j) Total ATP production (bioenergetic capacity), and (k) ATP production rate in Jurkat_TPR expressing ide-cel or cilta-cel assessed at baseline (unstimulated), and after 24 and 48 h of stimulation with MM.1S. Data are presented as mean values ± the standard error of the mean (SEM) from 3–6 biological replicates. Statistical comparisons between groups were performed using the two-tailed Student’s T-test. APC: allophycocyanin; ATP: adenosine triphosphate; ECAR: extracellular Acidification Rate; MFI: median fluorescence intensity; NFAT: nuclear factor of activated T-cells; NFkB: nuclear factor kappa-light-chain-enhancer of activated B-cells; NS: not significant; OCR: oxygen consumption Rate; PE: phycoerythrin; ScFV: single-chain variable fragment; SRC: spare respiratory capacity; tNGFR: truncated nerve growth factor receptor; VHH: camelid single-domain antibody/nanobody.

Extended Data Fig. 6 |. T-Cell immunophenotype of CirAE patients Analyzed by Flow Cytometry.

(a) Left: uniform Manifold Approximation and Projection (UMAP) visualization of flow cytometry characterization of CD3⁺ T-cells from 9 patients at peak of expansion, stratified by CirAE. Middle: CAR percentage and median fluorescence intensity (MFI) in CD3⁺ T-cells at peak expansion, stratified by CirAE. Violin plots show the mirrored probability density of the data; dashed line indicates the median, and dotted lines denote the 25^th and 75^th percentiles. Right: proportions of CD4, CD8, and double-negative CART subsets at peak expansion, stratified by CirAE. Bar graphs show the median and the error bars denote the 25^th and 75^th percentiles. Violin plots show the mirrored probability density of the data; dashed line indicates the median, and dotted lines denote the 25^th and 75^th percentiles. Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. (b) Flow cytometry analysis of peripheral blood mononuclear cells from 10 patients during CirAE and 6 non-CirAE patients at matched timepoints. CAR⁺ percentage (top) and MFI (bottom) of CART during CirAE in CD3⁺ (left), CD3⁺CD4⁺ (middle), CD3⁺CD8⁺ (right) T-cells. The top rightmost panel depicts absolute CART count (%CAR⁺ x ALC), stratified by CirAE. Violin plots show the mirrored probability density of the data; dashed line indicates the median, and dotted lines denote the 25^th and 75^th percentiles. Statistical comparisons between groups were performed using the two-tailed Wilcoxin test. (c) Comparison of peripheral blood (PB) CD4⁺ T-cell proportions between CirAE patients at the time of CirAE and time-matched non-CirAE controls. Left: unpaired comparison of CD4⁺ T-cell proportions at pre-lymphodepletion and CirAE/matched timepoints, stratified by CirAE; bar graphs show the median and the error bars denote the 25^th and 75^th percentiles. Right: comparison of CD4⁺ T-cell proportions at pre-lymphodepletion and CirAE/matched timepoints in paired longitudinal samples, stratified by CirAE. Statistical comparisons between groups were performed using the two-tailed paired-sample T-test. (d) Flow cytometry immunophenotyping of T-cells from matched PB and cerebrospinal fluid (CSF) samples during Cilta-cel#4’s delayed ICANS episode.

Extended Data Fig. 7 |. Single-cell analysis of matched peripheral blood and CSF samples in patients with neuro-CirAE (n = 4).

(a) Uniform Manifold Approximation and Projection (UMAP) visualization of CART cells in peripheral blood (PB) and cerebrospinal fluid (CSF), colored by T-cell clusters in each of the samples. (b) AUCell enrichment score for Gene Ontology Biological process (GOBP): leukocyte-mediated cytotoxicity score in each of the clusters. Box plots show the median (center line), interquartile range (bounds of box; 25^th–75^th percentiles), and minimum and maximum values (whiskers). Statistical comparisons between groups were performed using two-tailed Wilcoxin test. (c) Density plot of representative genes involved in T-cell cytotoxic function (Granzyme K,B,A: GZMK,GZMB,GZMA; Granulysin: GNLY; interferon gamma: IFNG; perforin 1: PRF1; CD107a) in PBMC and CSF samples.

Extended Data Fig. 8 |. Early pro-inflammatory cytokines are associated with CirAE development.

(a) Serum cytokine profile of CirAE compared to non-CirAE. Heatmap shows the log₂ (median CirAE/ median no CirAE) for pre-lymphodepletion (Pre-LD), Days 0, 7, 14, during CirAE. Statistical significance was determined using two-tailed Mann–Whitney tests. P^unadj indicates unadjusted P values (without correction for multiple comparisons); P^adj indicates adjusted P values (with Holm–Šidák correction for multiple comparisons); * indicates p-values < 0.05. (b) Serum proteins with significantly different concentrations at each time point (Pre-LD, days 0, 7, 14, and during CirAE) in patients who developed CirAE compared with those who did not. Box plots show the median (center line), interquartile range (bounds of box; 25^th–75^th percentile), and minimum and maximum values (whiskers). Statistical significance was determined using two-tailed Mann–Whitney tests. P^unadj indicates unadjusted P values (without correction for multiple comparisons); P^adj indicates adjusted P values (with Holm–Šidák correction for multiple comparisons); * indicates p-values < 0.05. (c-d) Line graphs show the dynamic changes of pro-inflammatory and anti-inflammatory cytokines in CirAE patients from Pre-LD to the time of CirAE.

Extended Data Table 1 |.

Detailed descriptions of patient characteristics and CirAE presentation, management and course

ID	Age	Sex	Type of CirAE	Diagnostic Test	CirAE Grade (Based on ICI)	Onset (Days)	Presentation	Management	Steroid dose (Dex equivalent)	Duration of CirAE (Days)	Resolution of CirAE?
Cilta-cel#1	75	M	Facial nerve palsy	Head CT: negative	2	19	Left sided facial droop at Day+19; bilateral facial droop at Day+69	Multiple courses of oral steroids	280 mg	90	Yes
Cilta-cel#2	67	M	Inflammatory arthritis	Labs: RF, CCP, HLA-B27 negative; Knee X-ray: bilateral knee arthritis & effusion	2	28	Pain and swelling of fingers, elbows, shoulders, knees, ankles	Steroids, Hydroxychloroquine	147 mg	825	Ongoing
Cilta-cel#3	53	F	Parkinsonism	Head CT: negative	2	39	Shuffled gait, rigidity, and right-sided head tilt	Steroid IV, IVIG, trihexyphenidyl and amantadine	1020 mg	690	Ongoing
Cilta-cel#4	74	M	Facial nerve palsy	Clinical diagnosis	2	31	Left sided facial weakness	Oral steroids	560 mg	29	Yes
			Delayed ICANS	Brain MRI/CT: negative CSF: negative workup for infections or malignancy: CAR transgene detected	2	57	Altered mental status, slurred speech, disorientation	Steroid IV, IVIG		7	Yes
			Enterocolitis	Endoscopy: immune-mediated enterocolitis	3	76	Diarrhea, weight loss	TPN. Refractory to budesonide and oral steroid, responsive to infliximab		765	Yes
Cilta-cel#5	66	F	Enterocolitis	Endoscopy: immune-mediated enterocolitis	2	154	Diarrhea, nausea	Supportive (Loperamide, Cholestyramine)	37.5 mg	97	Yes
Cilta-cel#6	65	F	Enterocolitis	CRP and ESR negative	1	25	Diarrhea, nausea	Loperamide, supportive treatment	150 mg	152	Yes
Cilta-cel#7	60	F	Inflammatory arthritis	Clinical diagnosis; ESR elevated, SSA borderline	2	7	Bilateral arm, knee, and ankle pain with hand swelling and nocturnal stiffness.	Oral steroids, physical therapy	32 mg	127	Yes
Cilta-cel#8	68	M	Enterocolitis	Negative C.diff and enteric pathogen testing; Endoscopy: immune-mediated enterocolitis	1	123	Diarrhea, weight loss	Mesalamine, loperamide and budesonide	0 mg	178	Yes
Cilta-cel#9	54	M	Neuropathy (Parsonage-turner-like)	MRI brachial plexus: normal; EMG: neuropathy and myopathic findings	2	9	Bilateral shoulder and neck pain, left upper extremity weakness	Oral steroids, physical therapy, with improvement	332 mg	438	Ongoing
Cilta-cel#10	67	F	Enterocolitis	Endoscopy: Immune-mediated enterocolitis; no CMV or viral inclusion, C.diff negative	4	86	Severe diarrhea, abdominal pain, pneumatosis intestinalis	Loperamide, steroids, IVIG, ruxolitinib, TPN	337 mg	357	No, CirAE related mortality
Cilta-cel#11	67	F	Parkinsonism	Clinical diagnosis; Brain MRI: negative	2	297	Fatigue, hypokinesis	Carbidopa-Levodopa	0 mg	144	Ongoing
Cilta-cel#12	70	M	Parkinsonism	Clinical diagnosis; Brain MRI: unremarkable	1	36	Blunted affect, micrographia	Oral steroids	116 mg	66	Yes
Cilta-cel#13	77	F	Enterocolitis	Clinical diagnosis	1	118	Diarrhea	Loperamide, oral steroids	N/A	223	Ongoing
Cilta-cel#14	69	M	Encephalopathy	MRI: bilateral thalamic diffusion restriction with peripheral enhancement CSF: elevated protein, CAR detected, JC virus positive (no clinical/ radiologic correlate)	5	263	Altered mental status, confusion, generalized fatigue, fever	Steroids IV	937.5 mg	18	No, CirAE related mortality
Cilta-cel#15	65	F	Parkinsonism	Head CT: negative; MRI: pachymeningeal thickening/enhancement. CSF: elevated protein, lymphocytic. CAR transcript detected.	3	28	Tremors, bradykinesia, altered mental status	Steroids, intrathecal chemotherapy, ruxolitinib	585 mg	225	Ongoing
Cilta-cel#16	58	F	Facial nerve palsy	MRI/CT brain: normal CSF: normal, CAR transcript detected	2	17	Right-sided facial droop at Day+17; left-sided facial droop at Day +44	IVIG, Oral and IV steroids	268 mg	30	Yes
Cilta-cel#17	67	M	Facial nerve palsy	Head CT: Negative	3	16	Left sided facial droop/weakness	Oral, IV steroids	108 mg	210	ongoing
Cilta-cel#18	67	F	Facial nerve palsy	Clinical diagnosis	2	18	Left sided facial droop/weakness	Steroids oral	200 mg	90	Yes
Cilta-cel#19	53	F	Facial nerve palsy	Head CT: Negative	2	20	Right sided facial droop/weakness	Steroids oral	78 mg	33	Yes
Cilta-cel#20	71	M	Enterocolitis	Negative C.diff Bowel CT: Extensive pneumatosis, moderate pneumoperitoneum Endoscopy: immune-mediated colitis	4	139	Failure to thrive, Diarrhea, Severe Weight Loss	Steroids oral and IV, ruxolitinib, TPN	460 mg	45	ongoing
Cilta-cel#21	61	F	Optic neuritis	MRI: left optic neuritis/perineuritis; CSF: negative for infections/malignancy; CAR transgene detected	5	62	Headache, left eye swelling, ptosis, left field visual defects	Steroids oral and IV, tocilizumab without Improvement	5100 mg	108	No, CirAE related mortality
Cilta-cel#22	55	F	Facial nerve palsy	Head CT: Negative, CSF: CAR transcript detected	2	22	Left sided facial droop, ageusia, and headache	Steroids oral and IV	444 mg	135	ongoing
Cilta-cel#23	77	M	Parkinsonism	Clinical diagnosis	3	57	Unsteady gait apraxla, dizziness	Steroids IV, IVIG with improvement but not resolution	562.5 mg	99	No, CirAE related mortality
Cilta-cel#24	74	F	Facial nerve palsy	MRI: Right facial nerve enhancement (labyrinthine segment)	2	42	Right facial droop, blurry vision	Steroids oral	73,5 mg	50	No, CirAE related mortality
Cilta-cel#25	75	F	Facial nerve palsy	Head CT/Brain MRI: negative	5	37	Left facial weakness	Steroids IV without improvement	232 mg	31	No, CirAE related mortality
			Parkinsonism	Clinical diagnosis	5	41	Masked facies, upper extremity rigidity, and hypophonia.	Steroids IV, Carbidopa-Levodopa without Improvement		27
Ida-cel#1	61	F	Pneumonitis	Chest CT: bilateral lower lobe predominant, peri-bronchovascular groundglass opacities	1	333	Dyspnea	Supportive, temporary suspension of teclistamab	0 mg	27	Yes
Ide-cel#2	82	F	Pneumonitis	CT: diffuse bilateral ground-glass nodules; Bronchoscopy: negative for infection.	3	361	Cough. Hypoxemia, dyspnea	Steroids, ruxolitinib	650 mg	354	Ongoing

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41591-025-04121-8.

Data availability

All requests for raw and analyzed data and materials are reviewed within 4 weeks by the University of Pennsylvania and the corresponding authors, to determine whether they are subject to intellectual property or confidentiality obligations. Patient-related data not included in the paper may be subject to patient confidentiality. The email addresses for the corresponding authors are as follows: mruella@upenn.edu, adam.cohen@pennmedicine.upenn.edu and jfrai@upenn.edu. Any data and materials that can be shared will be released via a material transfer agreement. Single-cell RNA sequencing data for matched peripheral blood and CSF are available at the NCBI Gene Expression Omnibus under accession code GSE309403. The integration site sequencing data have been deposited in the SRA hosted by the NCBI. The accession number for the SRA dataset is PRJNA1235822. Source data are provided with this paper.

Code availability

All analyses were performed with RStudio version 4.1.1. Packages used in analyses included the following: AUCell (1.25.2), celldex (1.14.0), cowplot (1.1.3), dplyr (1.1.4), enrichR (3.4), flowCore (2.16.0), flowViz (1.68.0), flowVS (1.36.0), flowWorkspace (4.16.0), FlowSOM (2.12.0), ggplot2 (3.5.2), limma (3.60.6), Matrix (1.7.3), msigdbr (24.1.0), Nebulosa (1.14.0), patchwork (1.3.0), pheatmap (1.0.12), RColorBrewer(1.1–3), Seurat (5.0.3), SingleR (2.6.0), tidyverse (2.0.0), VennDiagram (1.7.3), viridis (0.6.5) and writexl (1.5.4). All code needed for reproduction of analyses is available via GitHub at https://github.com/jhnoll/BMCA_CAR_irAE_NatMed.git.