Neuroimaging Findings in Children and Young Adults With Neurotoxicity After CAR T-Cell Therapy for B-Cell Malignancies

Abstract

Background and Objectives

Neuroimaging findings in immune effector cell-associated neurotoxicity syndrome (ICANS) have not been systematically described. We created the chimeric antigen receptor (CAR) T-cell Neurotoxicity Imaging Virtual Archive Library (CARNIVAL), a centralized imaging database for children and young adults receiving CAR T-cell therapy. Objectives of this study were to (1) characterize neuroimaging findings associated with ICANS and (2) determine whether specific ICANS-related neuroimaging findings are associated with individual neurologic symptoms.

Methods

We performed a multicenter retrospective cohort study of patients ≤30 years who experienced ICANS following CAR T-cell therapy for B-cell malignancies between January 1, 12, and January 31, 23, and had a brain MRI in the first 30 days after CAR T-cell infusion. Deidentified MRIs were reviewed by a central study team of pediatric neuroradiologists with experience in ICANS neuroimaging. Imaging features were categorized and correlated with CAR product and clinical characteristics including preinfusion neurologic history, and postinfusion neurologic symptoms alongside CAR T-cell toxicities using logistic regression.

Results

Of 864 patients treated with CD19 and/or CD22-directed CAR T-cells, 343 developed ICANS. 96 of the patients with ICANS (median age 12, 43% female) had an acute brain MRI. Of these, 36% (95% CI 27%–47%) had ICANS-related MRI abnormalities, most commonly affecting the white matter (24/35, 69%), brainstem (14/35, 40%), leptomeninges (10/35, 29%), and thalami (9/35, 26%). ICANS-related white matter abnormalities were generally bilateral, symmetric, and involved the supratentorial deep white structures, including the external and extreme capsules, corticospinal tracts, centrum semiovale, and periatrial white matter. There were no significant associations between ICANS-related MRI abnormalities and baseline clinical/demographic characteristic or specific ICANS symptoms, but higher ICANS grade was positively associated with MRI abnormalities (adjusted odds ratio 3.7, p < 0.001). Among 12 patients with ICANS-related MRI abnormalities who had follow-up imaging, 10 of 12 (83%) improved and 3 of 12 fully resolved.

Discussion

ICANS-related brain MRI abnormalities demonstrate unique patterns in the cerebral white matter, brainstem and thalami; their prevalence increases with ICANS clinical grade. Because our cohort is enriched for patients with severe ICANS, it likely overestimates the incidence of ICANS-related imaging abnormalities. A better understanding of neuroimaging findings is valuable for parsing pathophysiologic mechanisms of ICANS and optimizing patient outcomes.

Introduction

Chimeric antigen receptor (CAR) T-cell therapy is highly effective in treating relapsed/refractory B-cell malignancies.^1,2 However, immune effector cell-associated neurotoxicity syndrome (ICANS) affects 25%–44% of children, adolescents, and young adults receiving CD19-targeted CAR T-cell therapy for leukemia, typically occurring within the first 7–10 days after infusion.^3,4 Core clinical symptoms include, in decreasing order of frequency: encephalopathy, language dysfunction, depressed level of consciousness, seizures, focal motor deficits, and cerebral edema.⁴ Although most symptoms resolve within 1–2 weeks, rare cases of prolonged or permanent symptoms have been reported, and severe ICANS with global cerebral edema can be fatal.^5,6

Data are sparse on neuroimaging findings in pediatric CAR T-cell recipients, limited to 2 single-center clinical trials and small case series.^7-11 To address this knowledge gap, we created the CAR T-cell Neurotoxicity Imaging Virtual Archive Library (CARNIVAL), a centralized database of neuroimaging studies obtained pre- and post-CAR T-cell infusion for pediatric and young adult patients with hematologic malignancies. The primary objective of this study was to characterize the range of neuroimaging findings that are associated with acute ICANS. The secondary objective was to determine whether distinct ICANS-related neuroimaging findings are associated with specific neurologic symptoms.

Methods

Study Design and Patient Population

We performed a retrospective multicenter cohort study of children and young adults who received CD19 and/or CD22-directed CAR T-cells for the treatment of a B-cell hematologic malignancy, across 5 US centers within the CARnation consortium (Seattle Children's Hospital, Children's Hospital of Philadelphia, St. Jude Children's Research Hospital, National Cancer Institute, and Children's Hospital Los Angeles). CAR products included second-generation CD19-targeting constructs with a 4-1BB costimulatory domain (commercially available tisagenlecleucel, NCT01683279, NCT02028455, NCT03186118, NCT03601442, NCT02435849, NCT03573700, NCT02808442, NCT02906371, NCT01626495), second-generation CD19-targeting constructs with a CD28 costimulatory domain (NCT01593696, NCT02625480), humanized CD19-targeting constructs with a 4-1BB costimulatory domain (NCT03684889, NCT02374333, NCT03792633), CD22-targeting constructs (NCT02315612, NCT02650414), and dual CD19/CD22-targeting constructs (NCT03330691, NCT03448393) (eTable 1).^12-22 Data were included for all infusions performed between January 1, 2012, and January 31, 2023, at the participating centers if the patients were age ≤30 years at the time of CAR T-cell infusion, developed ICANS, underwent MRI of the brain in the first 30 days postinfusion, and had neuroimaging studies available for review. Imaging findings for several patients included in this study were previously reported.^12,14,23,24 In addition, we reviewed all postinfusion brain MRIs for patients who met all the inclusion criteria except that they did not have ICANS.

We queried institutional databases for eligible patients. Clinical data were abstracted from local medical records or clinical trial databases, then anonymized, uploaded, and managed using a central Research Electronic Data Capture database hosted at the Institute of Translational Health Sciences, Seattle, WA.^25,26

Standard Protocol Approvals, Registrations, and Patient Consents

The study was reviewed and approved or considered exempt by each participating center's Institutional Review Board. Parents or their guardians provided written informed consent for treatment on each respective clinical trial. Anonymized data were shared between participating institutions under a data use agreement.

Neuroimaging Database

For each infusion included in this study, all brain MRIs obtained in the first 30 days post-CAR T-cell infusion were analyzed. If preinfusion and/or follow-up MRIs outside of this timeframe were available, the most recent preinfusion and the most proximal post-30-day MRIs were also analyzed. MRIs were anonymized and uploaded to a secure cloud-based imaging platform. MRIs were obtained either for clinical indications (e.g., altered mental status or seizure) or prospectively, as part of an institutional protocol. MRIs were obtained on a 1.5T or 3T magnet, with a multiplanar acquisition technique with and/or without an intravenous gadolinium-based agent. Standard sequences included diffusion-weighted imaging (DWI), T2, fluid-attenuated inversion recovery (FLAIR), susceptibility-weighted imaging (SWI), 2D or 3D precontrast and postcontrast T1 and in some cases, subtraction images. Slice thickness ranged from 2 to 5 mm on DWI, T2, FLAIR, and 2D precontrast and postcontrast T1 sequences and 1–2 mm on SWI sequences. 3D precontrast and postcontrast T1-weighted sequences were acquired with an isotropic technique and were 0.9–1 mm in slice thickness.

MRI Analysis

All images were independently interpreted by a central team of board-certified neuroradiologists with pediatric expertise (S.N.P., A.B., A.V., J.W., M.A.O., R.S., G.I.), and data were validated by the core analysis team (S.N.P., J.L.M., E.E., J.G.). Imaging features, etiology attribution, and resolution data (when available) were categorized semiquantitatively using a structured review process (eFigure 1). Imaging abnormalities were classified by both signal and location and included edema, reduced diffusion and/or enhancement in the cortex, white matter, deep gray nuclei, brainstem, cerebellum, and leptomeninges. Additional findings of volume loss, infection, evidence of malignancy, and vascular abnormalities were also captured.

Imaging abnormalities were assigned an attribution score by the core analysis team based on how likely they were to be related to ICANS. These included “definitely related,” if there was a preinfusion neuroimaging study showing absence of the finding in question and there was no alternative explanation for the acute imaging abnormality; “probably related,” if preinfusion neuroimaging was not available for comparison but there was no alternative explanation for the acute imaging abnormality; “possibly related/unknown” in cases where there was equipoise regarding attribution; “unlikely related,” if preinfusion neuroimaging was not available, but findings had an alternative explanation (e.g., volume loss, chronic leukoencephalopathy, tumor) and/or findings remained unchanged on long-term follow-up neuroimaging (when available); or “definitely unrelated,” if a preinfusion neuroimaging study was available and demonstrated that a given finding was present on prior imaging. For analysis, ICANS-related MRI abnormalities were defined as those probably or definitely related to ICANS per the above attribution schema.

When additional neuroimaging was available, the resolution of imaging abnormalities was also tracked. Follow-up imaging findings were classified as “resolved,” “improved,” “unchanged,” or “worsened” if subsequent neuroimaging was available or “unknown” if not available.

As an exploratory quantitative analysis, we examined differences in median apparent diffusion coefficient (ADC) values in 3 regions of interest between preinfusion and postinfusion imaging for patients that had comparator studies available. Pertinent MRI sequences were uploaded onto a segmentation platform (Mint Medical, Hamilton, NJ) and then the splenium of the corpus callosum, thalami, and the entire brainstem, from the cerebral peduncles in the midbrain up to the cervicomedullary junction, were segmented using a 3D tool by a trained medical student (N.S.) blinded to timing of CAR T-cell infusion and the patients' clinical status and quality controlled by a pediatric neuroradiologist (S.N.P.). Segmentation was performed on axial T1- or T2-weighted sequences, for anatomic delineation of the structures, and then copied onto the ADC maps.

Study Definitions

Toxicities, including cytokine release syndrome (CRS) and ICANS, were graded per institutional or clinical trial protocol. ICANS was graded by either American Society for Transplantation and Cellular Therapy (ASTCT) criteria or Common Terminology Criteria for Adverse Events (CTCAE).^4,27 Details on the grading schemes are provided in eFigure 2. Severe ICANS was defined as ≥grade 3 ICANS. Mild ICANS was defined as ICANS grades 1–2. “Core ICANS criteria” were defined as the criteria that are included as ICANS-defining in the ASTCT grading scheme: encephalopathy (including depressed level of consciousness, altered mental status, impaired cognition, language dysfunction), seizure, focal motor deficit, and elevated intracranial pressure/cerebral edema. We also included noncanonical neurologic symptoms, which commonly occur in the setting of ICANS but are not part of the ASTCT ICANS criteria, such as headache, tremor, vision problems (e.g., nystagmus, diplopia, new blurry vision), hallucinations, sensory changes (e.g., paresthesia, numbness) and coordination problems (e.g., dysmetria, ataxia).

Data Analysis

Continuous variables were described using medians and interquartile range (IQR) and compared between groups using the Wilcoxon rank-sum or Kruskal-Wallis tests. Categorical variables were described using counts and frequencies and compared between groups using the χ² or Fisher exact test.

Univariate and multivariate logistic regression models were built to look for independent historical and clinical associations with overall ICANS-related MRI abnormalities. Initial models were constructed using ICANS risk factors that were identified in previous studies, including age at infusion, CNS leukemic disease status in the 30 days before infusion, history of cranial radiation, history of methotrexate neurotoxicity, and history of other structural CNS injury (including stroke, intracranial hemorrhage, posterior reversible encephalopathy syndrome, or CNS infection).^28,29 Additional candidate predictor variables, including specific neurologic symptoms postinfusion, with a p value <0.2 on univariable analysis, were included. CRS severity was intentionally not included in multivariate models given the substantial collinearity with ICANS. Stepwise variable selection was then performed based on model-fit measured by Akaike Information Criteria. Variables that did not retain statistical significance in the adjusted model but whose elimination worsened overall model fit were retained.

For the exploratory quantitative analysis of ADC changes, median ADC values were automatically generated from segmented regions of interest (splenium of the corpus callosum, brainstem, thalamus) on the segmentation platform and compared between the preinfusion and postinfusion MRIs for the subset of patients who had preinfusing imaging available, using a Wilcoxon signed-rank test on the paired samples. All analyses were performed in STATA version 17.0 (Stata Corp., College Station, TX), and a p value <0.05 was considered statistically significant.

Data Availability

Anonymized data not published within this article will be made available by request from any qualified investigator.

Results

Patient Characteristics

Demographics of the overall cohort are summarized in Figure 1 and Table 1. Across the participating centers, 864 infusions with CD19 and/or CD22-directed CAR T-cells occurred during the study timeframe, of which 343 were complicated by ICANS. Among those 343, 96 infusions for 95 unique patients had an MRI performed within the first 30 days after CAR T-cell infusion and comprised the analysis cohort. One patient was infused with CAR T-cells twice during this study timeframe, contributing 2 data points over time to the analysis, although each data point will be referred to as a “patient” for readability moving forward. Of the 521 patients without ICANS, 26 had brain MRIs in the first 30 days after CAR T-cell infusion. All abnormalities seen on these MRIs were clearly attributable to non-ICANS causes, such as malignancy or infection.

Figure 1. Summary of the Cohort.

Gray boxes demonstrate patients included in the present analysis. ICANS = immune effector cell-associated neurotoxicity syndrome.

Table 1.

Demographic, Clinical Historic, and Postinfusion Characteristics of Children Who Underwent Post-CART Neuroimaging

	All (n = 96)	By any ICANS-related neuroimaging abnormality
		No (n = 61)	Yes (n = 35)	Univariate OR (95% CI), p valuea	Multivariate aOR (95% CI), p valuea
Demographics
Sex: female	41 (43)	25 (41)	16 (46)	1.2 (0.5–2.8), 0.65
Race: White	57 (59)	33 (54)	24 (69)	1.9 (0.8–4.4), 0.17
Age at infusion, y	12 (8–16)	12.4 (7–16)	12 (8–15)	1.0 (0.9–1.0), 0.58
Oncology history
CNS involvement
Never	40 (42)	26 (43)	14 (40)	Reference
In the past only	49 (51)	30 (49)	19 (54)	1.2 (0.5–2.8), 0.71
Within 30 d of infusion	7 (7)	5 (8)	2 (6)	0.7 (0.1–4.3), 0.74
Radiation history
None	59 (61)	36 (59)	23 (66)	Reference
TBI only	17 (18)	10 (16)	7 (20)	1.1 (0.4–3.3), 0.84
Cranial boost ± TBI	14 (15)	11 (18)	3 (9)	0.4 (0.1–1.7), 0.24
Neurologic history
Methotrexate neurotoxicity	10 (10)	8 (13)	2 (6)	0.4 (0.1–2.0), 0.27
Seizure	15 (16)	12 (20)	3 (9)	0.4 (0.1–1.5), 0.16
Prior structural injury (stroke, ICH, PRES, infection)	8 (8)	7 (11)	1 (3)	0.2 (0.0–1.9), 0.17	0.1 (0–1.1), 0.056
Had preinfusion (30 day) MRI brain	28 (29)	20 (33)	8 (23)	0.75 (0.2–1.7), 0.50
Preinfusion MRI abnormal	19/28 (68)	12/20 (60)	7/8 (88)	0.2 (0.0–2.1), 0.19
Preinfusion white matter changes	14/28 (50)	8/20 (40)	6/8 (75)	4.5 (0.7–28.1), 0.11
Evidence of malignancy	4/28 (14)	4/20 (20)	0/8 (0)	—
Volume loss	7/28 (25)	6/20 (30)	1/8 (13)	0.3 (0.0–3.3), 0.35
Hemorrhage	6/28 (21)	5/20 (25)	1/8 (13)	0.4 (0.0–4.4), 0.48
Postinfusion course
CRS present	92 (96)	58(95)	34 (97)	1.8 (0.2–17.6), 0.63
CRS onset (days postinfusion)	4 (2–6)	4 (2–6)	5 (3–6)	1.1 (0.9–1.2), 0.39
CRS severity (grade)	2 (2–3.5)	3 (2–4)	2 (1–3)	0.7 (0.5–1.1), 0.10
ICANS onset (days postinfusion)	7 (5–9)	7 (5–9)	7 (5–9)	1.0 (0.9–1.1), 0.79
ICANS severity (grade)	3 (2–3)	2 (2–3)	3 (3–4)	3.3 (1.9–5.9), <0.001	3.7 (2.0–6.8), <0.001
ICANS phenotype
Encephalopathy	82 (85)	52 (85)	30 (86)	1.0 (0.3–3.4), 0.95
Seizure	30 (31)	17 (28)	13 (37)	1.5 (0.6–3.7), 0.35
Focal motor deficit	7 (7)	3 (5)	4 (11)	2.5 (0.5–11.9), 0.25
Other noncanonical neurologic symptoms
Headache	32 (33)	21 (34)	11 (31)	0.9 (0.4–2.1), 0.76
Tremor	13 (14)	8 (13)	5 (14)	1.1 (0.3–3.7), 0.87
Visual problems	13 (14)	10 (16)	3 (9)	0.5 (0.1–1.9), 0.29
Hallucination	9 (9)	6 (10)	3 (9)	0.9 (0.2–3.7), 0.84
Sensory changes	5 (5)	2 (3)	3 (9)	2.8 (0.4–17.4), 0.28
Coordination problems	6 (6)	3 (5)	3 (9)	1.8 (0.3–9.5), 0.48

Abbreviations: aOR = adjusted OR; CART = CAR T-cell; CRS = cytokine release syndrome; ICANS = immune effector cell-associated neurotoxicity syndrome; ICH = intracranial hemorrhage; IQR = interquartile range; OR = odds ratio; PRES = posterior reversible encephalopathy syndrome; TBI = total body irradiation.

Categorical variables are described using n (%). Continuous variables are described using median (IQR).

^ap Values were calculated using logistic regression.

Among the 96 patients included in the study, 92% (n = 88) received CAR T-cells directed against CD19 (costimulatory domain 4-1BB, n = 78; CD28, n = 10), 4% (n = 4) against CD22, and 4% (n = 4) against both CD19 and CD22. The majority (96%) of patients had refractory or recurrent acute lymphoblastic leukemia, the remainder had non-Hodgkin lymphoma. The most common preexisting neurologic risk factors were malignant CNS involvement (58% at any time, 7% within 30 days before CAR T-cell infusion), history of cranial radiation (15%), and history of methotrexate neurotoxicity (10%).

Incidence of ICANS

Post-CAR T-cell infusion, 53 (55%) of the study patients had severe ICANS (grades 3–5), among them 1 patient had fatal cerebral edema. Ninety-one patients (95%) met the definition of the core ASTCT ICANS criteria, including encephalopathy (88%), seizure (31%), and/or focal motor deficits (7%). Five patients (5%) did not have any of the ASTCT core ICANS criteria but were still considered to have neurotoxicity because they met CTCAE neurotoxicity criteria before universal adoption of the ASTCT ICANS grading scale. Including these 5 patients, a total of 53 patients (55%) had neurologic symptoms that are not part of the ASTCT ICANS core criteria, including headache (33%), vision problems (14%), and/or tremor (14%).

To examine the possibility that patients with severe ICANS are overrepresented in our cohort, we compared age, sex, and median ICANS grade between ICANS patients with or without a brain MRI within 30 days after CAR T-cell infusion (eTable 2). Patients who underwent MRI (the 96 included in this study) had a higher median ICANS grade compared with those who did not undergo MRI (3 vs 1, p < 0.001; data available for 315/343 patients). Among the patients without a postinfusion MRI were 5 patients with fatal (grade 5) ICANS, likely due to the acuity of their presentation.

Baseline Brain MRI Findings

Twenty-eight of the 96 (29%) included patients had a baseline brain MRI in the 30 days before infusion (median day (D)-11.5; IQR D-18, D-6), either for clinical indications based on neurologic history or for routine baseline screening. Nineteen of 28 (68%) preinfusion MRIs were abnormal. The most common abnormalities included white matter changes (50%), volume loss (25%), evidence of recent or remote hemorrhage (21%), and/or evidence of malignancy (14%).

ICANS-Related MRI Abnormalities

All 96 patients had at least 1 MRI brain in the first 30 days after CAR T-cell infusion (median day +10; IQR day +7, day +13). Thirty-five patients (36%, 95% CI 27%–47%) had postinfusion ICANS-related MRI abnormalities (Table 2).

Table 2.

ICANS-Related MRI Abnormalities Within First 30 Days Post-Infusion

MRI characteristics	All (n = 96)	By CART product
		CART19/4-1BB (n = 78)	CART19/CD28 (n = 10)	CART19/22 (n = 4)	CART22 (n = 4)
Median time to first postinfusion MRI? (d)	10 (7–13)	9 (7–13)	10 (6–11)	6.5 (6–7.5)	22 (16–26)
Number of postinfusion MRIs in first 30 d?	1 (1–1)	1 (1–1)	1.5 (1–2)	1 (1–1.5)	1 (1–1)
Normal postinfusion MRI	13 (14)	9 (12)	2 (20)	1 (25)	1 (25)
Postinfusion MRI with any ICANS-related abnormality	35 (36)	27 (35)	3 (30)	3 (75)	2 (50)
Specific ICANS-related MRI abnormalities
White matter-T2 abnormalities	18 (19)	14 (18)	0	3 (75)	1 (25)
Complete resolution among those with FU imaging?	2/7	1/6	—	1/1	—
Brainstem edema	14 (15)	13 (17)	0	1 (25)	0
Complete resolution among those with FU imaging?	2/5	1/4	—	1/1	—
White matter-diffusion restriction	10 (10)	7 (9)	2 (20)	0	1 (25)
Complete resolution among those with FU imaging?	1/2	1/1	1/1	—	—
Leptomeningeal/sulcal spaces abnormal	10 (10)	8 (10)	1 (10)	0	1 (25)
Complete resolution among those with FU imaging?	2/5	1/4	1/1	—	—
Thalamic edema	8 (8)	7 (9)	0	1 (25)	0
Complete resolution among those with FU imaging?	2/3	1/2	—	—	1/1
Cortex edema	8 (8)	8 (10)	0	0	0
Complete resolution among those with FU imaging?	2/5	2/5	—	—	—
White matter-focal edema	5 (5)	4 (5)	1 (10)	0	0
Complete resolution among those with FU imaging?	—	—	—	—	—
Thalamic restricted diffusion	4 (4)	3 (4)	0	1 (25)	0
Complete resolution among those with FU imaging?	1/2	0/1	—	1/1	—
Cortex restricted diffusion	3 (3)	3 (4)	0	0	0
Complete resolution among those with FU imaging?	3/3	3/3	—	—	—
Cerebellar edema	3 (3)	3 (4)	0	0	0
Complete resolution among those with FU imaging?	1/2	1/2	—	—	—
Hemorrhage	3 (3)	3 (4)	0	0	0
Complete resolution among those with FU imaging?	0/1	0/1	—	—	—
Other deep gray edema	2 (2)	2 (3)	0	0	0
Complete resolution among those with FU imaging?	0/1	0/1	—	—	—
Other deep gray restricted diffusion	2 (2)	2 (3)	0	0	0
Complete resolution among those with FU imaging?	1/1	1/1	—	—	—
Vascular changes	2 (2)	1 (1)	0	1 (25)	0
Complete resolution among those with FU imaging?	—	—	—	—	—

Abbreviations: CART = CAR T-cell; FU = follow-up; ICANS = immune effector cell-associated neurotoxicity syndrome; IQR = interquartile range.

Categorical variables are described using n (%). Continuous variables are described using median (IQR).

ICANS-related MRI abnormalities were found predominantly in the white matter (24/35, 69%), brainstem (14/35, 40%), and thalami (9/35, 26%); patients often had more than 1 abnormal finding (Figure 2). One patient had global cerebral edema with herniation, and 23 patients had focal edema. Focal edema was often symmetric and occurred in many different locations, including white matter, cortex, deep gray nuclei, thalami, and/or brainstem. Different CD19-directed CAR T-cell products had similar rates of specific findings (eTable 3), as did patients age <18 years vs those ≥18 years (eTable 4). MRI abnormalities considered unrelated to ICANS included volume loss (n = 23), chronic leukoencephalopathy (n = 18), nonspecific subcortical and deep white matter punctate T2 hyperintensities (n = 18), evidence of malignancy (n = 11), and others.

Figure 2. Overlap of ICANS-Related MRI Features.

The Venn diagram shows the number of patients who had 1 or more of the most common ICANS-related neuroimaging phenotypes (white matter, thalamus, and brainstem abnormalities). ICANS = immune effector cell-associated neurotoxicity syndrome.

The likelihood of having ICANS-related MRI abnormalities increased with ICANS grade (Spearman ρ = 0.46, p < 0.001; Figure 3). Patients with severe ICANS (≥grade 3) commonly had abnormalities in the white matter (20/53, 38%), brainstem (13/53, 25%), and/or thalami (8/53, 15%). Patients with mild ICANS (grade 1–2) only rarely had abnormalities in the white matter (4/43, 9%), brainstem (1/43, 2%), or thalami (1/43, 2%).

Figure 3. ICANS-Related Brain MRI Abnormalities Are More Common With Severe ICANS.

The bar graphs show the proportion of patients at each ICANS grad who had (A) any ICANS-related MRI abnormality, (B) white matter abnormalities, (C) thalamus abnormalities, and (D) brainstem abnormalities. The numbers above each bar show the number of patients at each ICANS grade who had the finding in question. Many patients had more than 1 finding simultaneously. ICANS = immune effector cell-associated neurotoxicity syndrome.

A secondary objective of this study was to determine whether specific ICANS symptoms correlate with specific MRI abnormalities; for example, whether patients with seizures would be more likely to have imaging abnormalities in the cortex. We found no significant association between the location of ICANS-related abnormalities and specific ICANS clinical symptoms (eTable 5). For the 5 patients who had none of the core ICANS symptoms but had other neurologic symptoms such as tremor, no ICANS-related MRI abnormalities were seen.

On univariate analyses (Table 1), there were no significant associations between preinfusion clinical characteristics (including age at infusion, prior CNS radiation, presence of CNS disease in the 30 days before infusion, preexisting white matter changes, or neurologic history) and presence of any ICANS-related MRI abnormalities. The only postinfusion factor significantly associated with ICANS-related MRI abnormalities was ICANS grade (Table 1, Figure 3A).

Multivariate analyses examining the association between preinfusion factors and the presence of any ICANS-related MRI abnormalities mirrored the univariate findings, without significant associations. When postinfusion ICANS grade and phenotype were added to multivariate models, only ICANS grade positively, independently associated with ICANS-related MRI findings (adjusted odds ratio [aOR] 3.7, 95% CI 2.0–6.8, p < 0.001). History of CNS structural injury was associated with a lower risk of ICANS-related MRI findings, although not statistically significant (aOR 0.1, 95% CI 0–1.1, p = 0.056; Table 1).

Twenty-three of 96 (24%) patients had more than 1 postinfusion MRI within the first 30 days after CAR T-cell infusion (range 2–4 MRIs), including 12 patients with ICANS-related MRI abnormalities. Of those 12, 3 (25%) demonstrated complete resolution (eFigure 3), 7 (58%) had improvement but not complete resolution, and 2 (17%) had no improvement. Residual changes included T2/FLAIR hyperintensities in the periatrial white matter (n = 3), corticospinal tracts (n = 1), and cortical edema (n = 1). The 2 patients who had no improvement had short interval follow-up MRIs within the first 30 days postinfusion, which may have been too early to evaluate for evolution of signal changes (eFigure 4).

White Matter Findings

Most white matter signal abnormalities were supratentorial, bilaterally symmetric T2/FLAIR hyperintensities involving the external capsules, corticospinal tracts, centrum semiovale, and/or the periatrial white matter (Figure 4, A–D). The bilateral linear symmetric T2 hyperintensities of the external capsules were seen in 9 patients. This unique feature was almost universally associated with severe ICANS. Two of these 9 patients demonstrated associated restricted diffusion, 1 limited to the centrum semiovale, and 1 throughout the supratentorial deep white matter. Four of the 9 patients had additional T2/FLAIR hyperintensity in the corpus callosum, 2 of whom demonstrated associated restricted diffusion in the splenium (Figure 4, E and F). Two other patients with grade 4 ICANS had diffuse supratentorial white matter signal changes involving both the subcortical and deep white matter, either as T2/FLAIR hyperintensity (1 patient) or restricted diffusion (1 patient). More focally restricted white matter T2/FLAIR hyperintensities were seen in 2 patients (Figure 4, G and H). One other patient demonstrated T2/FLAIR hyperintensity limited to the cerebellar white matter, without associated restricted diffusion.

Figure 4. ICANS-Related White Matter Abnormalities.

The figure shows representative brain MRI images from 4 patients with different patterns of deep white matter signal abnormalities. (A, B) Patient 1 is a teenager with ALL and CNS leukemic involvement at diagnosis, presenting with grade 4 ICANS (encephalopathy and focal motor deficit) 4 days after CD19-CART infusion. The patient developed flaccid paraparesis, which did not improve despite multiple immune-based therapies.¹⁹ Axial FLAIR images show bilateral, symmetric T2/FLAIR hyperintensities in (A) the external and extreme capsules (long arrows) and along the corticospinal tracts in the posterior limbs of the internal capsules (short arrows) and (B) in the midbrain. The T2/FLAIR hyperintensity extended caudally as a holocord transverse myelitis. (C, D) Patient 2 is a child with non-Hodgkin lymphoma and CNS leukemic involvement at diagnosis, presenting with grade 4 ICANS (encephalopathy and depressed level of consciousness) 21 days after CD22-CART infusion. Axial FLAIR images show bilateral, symmetric T2/FLAIR hyperintensities in (C) the centrum semiovale, as well as (D) the periatrial white matter (short arrows) and the corticospinal tracts (long arrows). (E, F) Patient 3 is a young adult with relapsed ALL, presenting with grade 2 ICANS (encephalopathy), headache and hallucinations 13 days after CD19 CART infusion. Axial DWI (E) and ADC map (F) show new reduced diffusion in the splenium of the corpus callosum. (G, H) Patient 4 is a young adult with ALL, presenting with grade 1 ICANS (encephalopathy) 2 days after CD19-CART infusion. Axial FLAIR (G) and T2 (H) images show a new, asymmetric hyperintensity in the right occipital subcortical and periatrial white matter, extending into the overlying cortex. ADC = apparent diffusion coefficient; CART = CAR T-cell; DWI = diffusion-weighted imaging; FLAIR = fluid-attenuated inversion recovery; ICANS = immune effector cell-associated neurotoxicity syndrome.

Brainstem Findings

A characteristic pattern of T2/FLAIR hyperintensity involving the central tegmental tracts and the inferior olivary nuclei was the most common brainstem finding, seen in 8 of 11 patients with brainstem abnormalities (Figure 5, A–D). Three of these 8 patients demonstrated additional T2/FLAIR hyperintensity in the central pontine white matter and inferior cerebellar peduncles. The T2/FLAIR hyperintensity extended into the central medulla and cervical cord as a holocord transverse myelitis in 1 patient, whose neuroimaging findings have been previously reported.²⁴ There was no associated restricted diffusion or enhancement in patients with brainstem abnormalities. Interestingly, these patients did not have coexisting cerebellar signal abnormalities or abnormal enhancement, as might be seen in viral rhombencephalitis.³⁰

Figure 5. ICANS-Related MRI Abnormalities in the Brainstem and Thalami.

The figure shows representative brain MRI images from 4 different patients with brainstem and thalamic involvement. (A, B) Patient 5 is a child with ALL, presenting with grade 4 ICANS (encephalopathy and seizure) 4 days after CD19-CART infusion. (A) Axial T2 images at the level of the pons show bilateral, symmetric T2 hyperintensity in the dorsal pons, including the central tegmental tracts. (B) Axial T2 images at the level of the medulla showing bilateral, symmetric T2 hyperintensities in the inferior olivary nuclei and medial lemnisci. (C, D) Patient 6 is a child with ALL, presenting with grade 3 ICANS (encephalopathy) 6 days after CD19-CART infusion. (C) Axial T2 images at the level of the pons show hyperintensity in the central pontine white matter (short arrow), the central tegmental tracts, and mesencephalic tracts of the trigeminal nerve (long arrows). (D) Axial T2 images at the level of the medulla showing symmetric, bilateral hyperintensities in the inferior cerebellar peduncles (short arrows) and the inferior olivary nuclei (long arrows). (E, F) Patient 7 is a child with relapsed ALL, presenting with grade 2 ICANS (encephalopathy) 9 days after CD19-CART infusion. Axial T2 (E) and FLAIR (F) images show bilateral, symmetric hyperintensities in the thalami (short arrows). Note that this patient also had symmetric hyperintensities in bilateral external capsules (“double-smiley sign,” long arrows). (G, H) Patient 8 is a young adult with non-Hodgkin lymphoma, presenting with grade 4 ICANS (depressed level of consciousness, encephalopathy, and seizure) 8 days after CD19 CAR T infusion. Axial DWI (G) and ADC map (H) show reduced diffusion in the ventrolateral thalami. ADC = apparent diffusion coefficient; CART = CAR T-cell; DWI = diffusion-weighted imaging; FLAIR = fluid-attenuated inversion recovery; ICANS = immune effector cell-associated neurotoxicity syndrome.

Thalami Findings

Among the patients with thalamic abnormalities, all had bilateral, symmetric T2/FLAIR hyperintensities with associated expansile change suggestive of edema (Figure 5, E–H). Two patients had additional restricted diffusion in the bilateral thalami, but none demonstrated enhancement.

ADC Measurement

Restricted diffusion was noted in the supratentorial deep white matter in 5 of 35 (14%) patients with ICANS-related MRI abnormalities, and in the bilateral thalami in 2 of 35 (6%) patients with ICANS-related MRI abnormalities. Based on this finding, we hypothesized that ICANS may be associated with a quantitative decrease in ADC in the deep white matter, thalami and brainstem, even when not apparent on clinical review. To test this hypothesis, we measured ADC changes in the splenium, bilateral thalami, and the brainstem in the subset of patients who had pre-CAR T baseline imaging and an acute MRI in the first 14 days post-CAR T-cell infusion (n = 17). The splenium of the corpus callosum was selected as a representative of the supratentorial deep white matter, because of its transverse orientation of fibers, and reliability of ADC measurement. Information on the variation in scan parameters such as repetition time, echo time, and diffusion gradients (b values) was not available on the central imaging database. When comparing baseline and acute ADC values, we found no statistically significant differences in any of the regions of interest, with ADC values increased during ICANS in some patients and decreased in others (eFigure 5).

Discussion

Cancer immunotherapy has created novel syndromes of neurologic toxicity whose mechanisms remain incompletely understood. A thorough assessment of the spectrum of associated neuroimaging findings has been hampered by lack of comprehensive data sets. To address this, we formed a collaboration of pediatric CAR T-cell clinical trial institutions to collect all available brain MRIs in our cohorts of patients who have received investigational and commercial CAR T-cell therapies for hematologic malignancies. We found characteristic patterns of acute white matter, brainstem, and thalamic changes. Furthermore, although specific ICANS-related neuroimaging findings were not associated with individual neurologic symptoms, the presence of any ICANS-related imaging abnormalities was positively associated with higher ICANS grade.

This study captures the majority of patients enrolled in pivotal pediatric CAR T-cell clinical trials for leukemia/lymphoma in the United States and can be considered representative of this population at large. Of 864 CAR T-cell infusions in the participating centers within the study timeframe, 40% were complicated by grade ≥1 ICANS. This proportion is consistent with published studies in children and adults.¹² To be included in this study, patients had to have both ICANS and an MRI of the brain in the first 30 days after CAR T-cell infusion, which was only true for 96 of 343 (28%) patients. In clinical practice, the threshold to obtain an MRI for acute clinical evaluation varies by institution. Imaging may be omitted if ICANS symptoms are mild and/or transient, and MRI is often not possible in patients with fulminant cerebral edema. Our neuroimaging cohort is biased toward patients with more severe neurotoxicity (eTable 2). Thus, the point estimate for the proportion of ICANS patients with ICANS-related neuroimaging abnormalities (36%) is likely higher than the proportion we would have found if all patients with ICANS underwent neuroimaging. Abnormal MRI findings of the brain have been previously reported in 10%–40% of pediatric or adult patients with ICANS in smaller cohorts.^7-11

We demonstrated that ICANS is associated with characteristic patterns of brain MRI abnormalities. The most common ICANS-related MRI abnormalities in our cohort included symmetric T2/FLAIR white matter abnormalities, often with involvement of the external and extreme capsule, and edema of the brainstem and thalami. In some cases, all these findings were present simultaneously, but most patients only had subsets of the findings. It remains uncertain whether global cerebral edema represents the same pattern in its most severe form. Involvement of the external and extreme capsule appears to be a finding that is specific to ICANS and otherwise not seen in the setting of cancer therapy and its complications. Of course, the radiographic differential diagnosis must take the clinical setting into account. Involvement of the external and extreme capsules can be seen in rare conditions such as eastern equine encephalitis and other encephalitides, as well as genetic and metabolic disorders such as CADASIL, urea cycle disorders, certain acidurias, and hepatic encephalopathies.^31-33 In patients with T2/FLAIR hyperintensities limited to the centrum semiovale and/or periatrial white matter, methotrexate-induced leukoencephalopathy is an important consideration.³⁴ In patients with parieto-occipital patterns of deep white matter involvement, fludarabine-related leukoencephalopathy is on the differential, especially if there is associated restricted diffusion. However, fludarabine-related leukoencephalopathy typically has a delayed (>6 weeks) onset, a progressive course, and is uncommon with current lower dosing regimens.^35,36 The bithalamic pattern of T2/FLAIR hyperintensities was another distinctive feature of ICANS in our cohort that has not been described in chemotherapy-related neurotoxicities. However, the thalami are frequently involved in other pathologies. In the correct clinical setting, differential considerations also include acute disseminated encephalomyelitis (ADEM), genetic disorders, metabolic abnormalities such as hypoxic ischemic encephalopathy, severe hypoglycemia, and a range of viral infections such as Japanese and West Nile encephalitides.^37-42 Finally, several of our patients had leptomeningeal enhancement attributed to ICANS. Leptomeningeal changes related to ICANS, as opposed to leukemic involvement, were distinguished by timing (new onset during ICANS and resolution on short interval follow-up), associated ICANS-related parenchymal changes, and absence of leukemic blasts on CSF analysis.

Although ADC values often changed from preinfusion to postinfusion, the direction of change was not consistent. Changes in diffusivity are likely dynamic over the course of ICANS, with restricted diffusion sometimes evolving after the onset of clinical ICANS and resolving over time. We likely sampled different time points in this complex temporal evolution. In addition, vasogenic edema can cause elevated ADC values, confounding ADC analysis. Further quantitative neuroimaging approaches could determine whether there are microstructural changes which are not apparent on clinical review.

It remains unresolved whether ICANS represents a common pathophysiology related to nonspecific immune activation or whether there is on-target off-tumor injury to normal brain structures expressing the antigen that is targeted by the CAR.^43,44 In our cohort, different CD19-directed CAR T-cell products had similar proportions of specific neuroimaging findings (eTable 3). Notably however, none of the patients who received CD22-directed CAR T-cell products had the ICANS-specific patterns of symmetric white matter changes involving the external capsule, brainstem, and/or thalami. The reason for this may be that CD22-CAR T-cells are associated with lower rates and lower severity of ICANS, thus reducing the likelihood of abnormal imaging findings.^20,45 It is also possible that CD22-CAR T-cell related neurotoxicities have distinct imaging patterns. Given the small number of patients in our study who received CD22-CAR T-cells, further study is required to answer this question conclusively.

We had hypothesized that specific imaging findings might correlate with specific ICANS symptoms, but did not find statistically significant associations. For example, language dysfunction is a prominent feature of ICANS, but this was not associated with lesions in the cortical language areas. Instead, subcortical network dysfunction is more likely to be responsible. In addition, different pathophysiologies may result in a common phenotype such as depressed consciousness or seizures. Notably though, ICANS grade strongly correlated with the presence of ICANS-related neuroimaging findings; patients with mild ICANS only rarely had associated neuroimaging abnormalities. Although our study was not designed to determine the relationship of ICANS-related neuroimaging findings with longer-term neurocognitive and survival outcomes, this key question in the field is under active investigation.

A limitation of our study is the retrospective nature of the analysis, with many patients lacking baseline or follow-up imaging. Because we aimed to be conservative in our ICANS attribution, some real ICANS-associated neuroimaging abnormalities may have been excluded. For example, small nonspecific white matter changes were assigned as “unlikely related” unless comparison images were available. Prospective studies with universal baseline, acute, and follow-up neuroimaging are required to understand the true incidence and resolution of ICANS-related imaging abnormalities, but such studies have not been practical in this medically fragile population. ICANS is managed clinically, and imaging often does not change management. However, screening baseline neuroimaging before CAR T-cell infusion has been shown to be feasible by pediatric and adult studies and could be implemented more broadly to help determine whether acute abnormalities are attributable to ICANS.^45,46

This study may guide clinical decision making in several ways. Our results suggest that neuroimaging has the highest yield in patients with more severe ICANS. In these patients, typical ICANS-related neuroimaging findings may improve diagnostic certainty and help curtail a more extensive symptom investigation. Radiographic involvement of the brainstem was not infrequent, typically in the absence of overt clinical signs of brainstem dysfunction. Given that brainstem involvement can extend into the spinal cord, this finding should prompt careful clinical surveillance and a low threshold for spinal cord imaging considering recent reports of myelitis as an emerging ICANS phenotype.²⁴

In summary, CARNIVAL provides a comprehensive overview of ICANS-related neuroimaging findings post-CAR T-cell therapy for pediatric and young adults with hematologic malignancies, laying a foundation for future prospective neuroimaging research and establishing characteristic ICANS-related imaging findings. These data will become even more important as we move toward understanding late effects and delayed toxicities of CAR T-cell therapy. Our study highlights the importance of prospectively obtaining baseline MRIs in all patients to help determine which neuroimaging findings are acute and of obtaining acute and follow-up MRIs in all clinically symptomatic patients to better understand the spectrum and trajectory of neuroimaging abnormalities. CARNIVAL developed a multisite collaborative model for combining imaging and clinical correlations. This approach could be useful for other rare pediatric neurologic disorders. CARNIVAL is open to additional institutions worldwide, and future analysis using advanced methodologies may uncover new mechanistic insights into ICANS.

Glossary

ADC

apparent diffusion coefficient

aOR

adjusted odds ratio

ASTCT

American Society for Transplantation and Cellular Therapy

CAR

chimeric antigen receptor

CARNIVAL

CAR T-cell Neurotoxicity Imaging Virtual Archive Library

CRS

cytokine release syndrome

CTCAE

Common Terminology Criteria for Adverse Events

DWI

diffusion-weighted imaging

FLAIR

fluid-attenuated inversion recovery

ICANS

immune effector cell-associated neurotoxicity syndrome

IQR

interquartile range

SWI

susceptibility-weighted imaging

Author Contributions

J.L. McGuire: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. S. Pinto: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. E.N. Erdogan: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data. Y. Li: analysis or interpretation of data. A. Bhatia: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. M.A. Oztek: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. A. Vossough: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. J.N. Wright: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. R. Shah: analysis or interpretation of data. N.T. Carapia: major role in the acquisition of data. N. Shams: analysis or interpretation of data. C. Westermann: analysis or interpretation of data. A. Taraseviciute: drafting/revision of the manuscript for content, including medical writing for content; study concept or design. B. Yates: major role in the acquisition of data. S. Naik: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. R. Gardner: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. C. Annesley: major role in the acquisition of data. E. Hsieh: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. C. Diorio: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. R. Myers: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. R. Epperly: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. A. Talleur: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. H. Shalabi: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design. N. Shah: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. J. Gust: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data.

Study Funding

This work was funded by the CAR T neurocognitive gift fund (J.G.), the Kelly Cancer Immunotherapy Fund (J.L.M.), Alex's Lemonade Stand Fund “A” award (C.D.), and NIH K08CA286762 (C.D.). The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government.

Disclosure

E. Hsieh serves as a consultant for Novartis. N. Shah receives research funding from Lentigen, VOR Bio, and CARGO Therapeutics; receives royalites from CARGO; and has attended advisory board meetings (no honoraria) for VOR, ImmunoACT, and Sobi. All other authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.