Glial Injury in Neurotoxicity After Pediatric CD19-Directed Chimeric Antigen Receptor T Cell Therapy

Abstract

Objective:

To test whether systemic cytokine release is associated with central nervous system inflammatory responses and glial injury in immune effector cell-associated neurotoxicity syndrome (ICANS) after chimeric antigen receptor (CAR)-T cell therapy in children and young adults.

Methods:

We performed a prospective cohort study of clinical manifestations as well as imaging, pathology, CSF, and blood biomarkers on 43 subjects ages 1 to 25 who received CD19-directed CAR/T cells for acute lymphoblastic leukemia (ALL).

Results:

Neurotoxicity occurred in 19 of 43 (44%) subjects. Nine subjects (21%) had CTCAE grade 3 or 4 neurological symptoms, with no neurotoxicity-related deaths. Reversible delirium, headache, decreased level of consciousness, tremor, and seizures were most commonly observed. Cornell Assessment of Pediatric Delirium (CAPD) scores ≥9 had 94% sensitivity and 33% specificity for grade ≥3 neurotoxicity, and 91% sensitivity and 72% specificity for grade ≥2 neurotoxicity. Neurotoxicity correlated with severity of cytokine release syndrome, abnormal past brain magnetic resonance imaging (MRI), and higher peak CAR-T cell numbers in blood, but not cerebrospinal fluid (CSF). CSF levels of S100 calcium-binding protein B and glial fibrillary acidic protein increased during neurotoxicity, indicating astrocyte injury. There were concomitant increases in CSF white blood cells, protein, interferon-γ (IFNγ), interleukin (IL)-6, IL-10, and granzyme B (GzB), with concurrent elevation of serum IFNγ IL-10, GzB, granulocyte macrophage colony-stimulating factor, macrophage inflammatory protein 1 alpha, and tumor necrosis factor alpha, but not IL-6. We did not find direct evidence of endothelial activation.

Interpretation:

Our data are most consistent with ICANS as a syndrome of systemic inflammation, which affects the brain through compromise of the neurovascular unit and astrocyte injury.

CD19-directed chimeric antigen receptor (CAR)-T cell therapy has shown excellent efficacy against B cell malignancies in children and adults, leading to recent US Food and Drug Administration approval of several CAR-T products.^1,2 The therapeutic principle consists of inserting a CAR transgene into T cells in vitro, expanding the cells, and reinfusing them into the patient. Binding of the CAR to the cancer surface antigen then initiates killing of the malignant cells.³ CAR-T treatment is frequently complicated by cytokine release syndrome (CRS),^4,5 which manifests with fever, hypotension, and vascular leak. In addition, neurotoxicity occurs in approximately 40% of patents, with manifestations ranging from mild delirium to fatal cerebral edema.⁶ The designation immune effector cell-associated neurotoxicity syndrome (ICANS) was recently proposed by a consensus group of the American Society of Bone Marrow Transplantation⁷ to comprise all neurological toxicities that occur with cell-based immunotherapy. There is still considerable uncertainty regarding the pathophysiology, prevention, and management of ICANS. Evidence is emerging that systemic inflammatory signaling during CAR-T cell expansion leads to disruption of the blood–brain barrier (BBB),^8,9 and that monocyte-derived cytokines are required for the development of toxicity.¹⁰ Given the intimate involvement of astrocytes in the regulation of the BBB,¹¹ we hypothesized that glial dysfunction is part of the pathophysiology of neurotoxicity. In support of this hypothesis, we now show evidence of astrocyte activation occurring in the setting of the systemic inflammatory cytokine surge that accompanies CAR-T cell expansion.

Subjects and Methods

Subject Selection, Registration, and Ethical Review

As previously described, subjects ages 1 to 26 years with relapsed or refractory B-ALL were enrolled in phase I of an open label phase I/II clinical trial (clinicaltrials.gov NCT02028455)¹² of CD19-directed, 4-1BB costimulated CAR-T cells (SCRI-CAR19v1). The study was approved by the Seattle Children’s Institutional Review Board.

Grading of Neurotoxicity

All new neurological symptoms occurring ≤28 days after CAR-T cell infusion were prospectively classified per NCI Common Terminology Criteria for Adverse Events (CTCAE), version 4.0 (ctep.cancer.gov), with signs/symptoms graded: 0, none; 1, mild; 2, moderate; 3, severe; 4, life threatening; and 5, death. Daily and overall neurotoxicity grades were determined by the most severe CTCAE grade of any neurological symptom during that period, with the following exceptions: If headaches were the only symptom above grade 2, we assigned overall grade 2; new-onset seizures were assigned grade ≥3. CRS was graded as previously described,¹² with severe CRS defined as requirement for pressors, inotropes, or respiratory support.

Clinical Data

Histories, cerebrospinal fluid (CSF), and blood samples were collected per clinical trial protocol. Infectious workup, Cornell Assessment of Pediatric Delirium (CAPD) scores, seizure treatment, neuroimaging, and electroencephalography (EEG) were performed per standard of care. The CAPD consists of eight measures of purposeful interaction with the environment, which are scored by nursing staff from 0 = always normal to 4 = always abnormal and added for the overall score. Brain magnetic resonance imaging (MRI) was obtained on a 1.5 Tesla (T) or 3T clinical scanner and reviewed by the study pediatric neuroradiologists (H.K. and G.I.).

Immunophenotyping

Isolated peripheral blood and bone marrow lymphocytes and whole CSF were stained with a viability dye and monoclonal antibodies against human CD3, CD4, CD8, CD19, and CD14. Allophycocyanin-conjugated Erbitux (BD Biosciences, San Jose, CA) was used to label cells expressing the CAR transduction marker EGFRt (truncated epidermal growth factor receptor). Cells were acquired on a LSRFortessa (BD, Franklin Lakes, NJ) and analyzed using FlowJo software (TreeStar, Ashland, OR). T cells were defined as singlets/lymphocytes/live CD3⁺ CD14⁻; CAR-T cell numbers were reported as the fraction live CD3⁺ T cells expressing EGFRt. The absolute number of CAR-T cells was calculated from blood and CSF lymphocyte counts reported by the clinical lab.

Biomarker Quantification

Serum and CSF concentrations of granulocyte macrophage colony-stimulating factor, sCD137, interferon-γ (IFNγ), sFas, sFasL, granzyme A, granzyme B (GzB), interleukin (IL)-2, IL-4, IL-5, IL-6, IL-10, IL-13, macrophage inflammatory protein (MIP)-1α, MIP-1β, tumor necrosis factor (TNF)-α, and Perforin were evaluated by Milliplex MAP Human CD8⁺ T cell Magnetic Bead Assay (Millipore, Burlington, MA), per the manufacturer’s instructions. Serum and CSF concentrations of angiopoietin (Ang)-1, Ang-2, glial fibrillary acidic protein (GFAP), S100 calcium-binding protein B (S100b), vascular endothelial growth factor (VEGF), and von Willebrand factor (VWF) were quantified using the Meso Scale Discovery platform (Meso Scale Diagnostics, Rockville, MD), per the manufacturer’s instructions.

Statistical Analysis

Comparisons of continuous variables between categories of neurotoxicity were made using nonparametric tests, or proportional odds model as appropriate, and comparisons of categorical variables were made using Fisher’s exact test. All p values reported were two-sided without adjustments for multiple comparisons, unless otherwise specified. Statistical analyses were performed using SAS (version 9.4; SAS Institute Inc., Cary, NC), JMP (version 13.0; SAS Institute), and GraphPad Prism software (version 7; GraphPad Software Inc., La Jolla, CA).

Results

Clinical Kinetics of ICANS

Forty-five subjects (median age was 12.3 years; range, 1.3–25.4) with ALL were enrolled onto the phase I portion of the study,¹² and 43 received CAR-T cell infusions. Three (6.7%) were primary refractory, 15 (33.3%) were first relapse, and 27 (60%) were in second or greater relapse. Twenty-eight patients (62%) had a history of at least one previous allogeneic transplantation. Seven had previously received CD19-directed therapies (blinatumomab, n = 6; CD19-targeted CAR-T cells, n = 1). No subjects had active CNS leukemia by CSF cytology immediately preceding the CAR-T cell infusion.

Nineteen of 43 subjects (44%) developed neurotoxicity (Table 1). Nine (21%) had severe (CTCAE grade 3–4) neurological symptoms, and six of seven dose-limiting toxicities in the cohort were attributed to neurotoxicity. No deaths were attributable to neurotoxicity. Neurological symptoms began a median of 7 days after CAR-T cell infusion (range, 0–13) and peaked in severity a median of 1 day after onset (range, 0–8; Fig 1A). Median symptom duration was 5 days (range, 1–30), with only 2 subjects remaining symptomatic past day 21.

TABLE 1.

Demographic and Treatment Factors Contributing to Neurotoxicity

Neurotoxicity (grade)		None (0) (N = 24)	Mild (1–2) (N = 10)	Severe (3–4) (N = 9)	Total (N = 43)	p a
Age, yr	Median (IQR)	12.5 (5, 16)	8.5 (6, 20)	12 (8, 16)	12 (6, 17)	0.9180
	[range]	[1, 25]	[1, 21]	[3, 23]	[1, 25]
Sex, n (%)	Male	14 (58.3)	2 (20.0)	5 (55.5)	21 (48.8)	0.1237
	Female	10 (41.7)	8 (80.0)	4 (44.4)	22 (51.2)
Race, n (%)	White	19 (79.2)	2 (20.0)	6 (66.7)	27 (62.8)	0.0049
	Nonwhite	5 (20.8)	8 (80.0)	3 (33.3)	16 (37.2)
Pre-existing neurological comorbidities, n (%)	None	9 (37.5)	4 (40.0)	1 (11.1)	14 (32.6)	0.7500b
	CNS involvement	8 (33.3)	3 (30.0)	5 (55.6)	16 (37.2)	0.5312
	Peripheral neuropathy	3 (12.5)	1 (10.0)	3 (33.3)	7 (16.3)	0.4001
	Abnormal prior MRI	1 (4.2)	0	4 (44.4)	5 (11.6)	0.0068
	Seizures	3 (12.5)	1 (10.0)	1 (11.1)	5 (11.6)	1.0000
	MTX CNS toxicity	1 (4.2)	2 (20.0)	1 (11.1)	4 (9.3)	0.2199
	Headache disorder	1 (4.2)	0	2 (22.2)	3 (7.0)	0.1485
	PRES	2 (8.3)	0	0	2 (4.7)	1.0000
Brain radiation, n (%)	None	9 (37.5)	6 (60.0)	1 (11.1)	16 (37.2)	0.1277
	TBI	13 (54.2)	3 (30.0)	5 (55.6)	21 (48.8)
	TBI + cranial boost	2 (8.3)	1 (10.0)	3 (33.3)	6 (13.9)
Past transplant, n (%)	Yes	17 (70.8)	5 (50.0)	6 (66.6)	28 (65.1)	0.4746
	No	7 (29.2)	5 (50.0)	3 (33.3)	15 (34.9)
Conditioning, n (%)	Flu/Cy	8 (33.3)	2 (20.0)	4 (44.4)	14 (32.6)	0.5547
	Non-Flu/Cy	16 (66.6)	8 (80.0)	5 (55.6)	29 (67.4)
CRS grade, n (%)	None	3 (12.5)	0	0	3 (7.0)	0.0083c
	Mild	18 (75.0)	8 (80.0)	4 (44.4)	30 (69.8)
	Severe	3 (12.5)	2 (20.0)	5 (55.6)	10 (23.2)
Treatment of complications, n (%)	Steroids	1 (4.2)	2 (20.0)	7 (77.8)	10 (23.2)	<0.0001
	Tocilizumab	8 (33.3)	3 (30.0)	6 (66.7)	17 (39.5)	0.1777
	Anakinra	0	0	3 (33.3)	3 (7.0)	n/ad

^aTwo-sided p values calculated based on Kruskal-Wallis test for continuous variables, and Fisher’s exact test for categorical variables, unless otherwise specified.

^bNone vs any.

^cTwo-sided p values calculated based on proportional odds model to account for the ordinal nature of independent variable.

^dSample size is too small to report p value.

CRS = cytokine-release syndrome; IQR = interquartile range; CNS = central nervous system; MRI = magnetic resonance imaging; MTX = methotrexate; TBI = total body irradiation; PRES = posterior reversible encephalopathy syndrome; Flu/Cy = fludarabine and cyclophosphamide.

FIGURE 1:

Kinetics and treatment of CAR-T cell treatment-related neurotoxicity (ICANS). (A) Swimmer plot showing kinetics and treatment of neurotoxicity for each of the 19 subjects who had neurotoxicity. Each lane represents 1 subject, and the colors indicate the overall neurotoxicity CTCAE grade on a given day. The daily CTCAE grade is indicated by the color code on the right. Stars indicate days when subjects received steroids, wedges indicate tocilizumab, and X indicates anakinra administration. (B) Neurotoxicity risk peaks on day 8. The y-axis indicates the mean CTCAE neurotoxicity grade averaged across all subjects in the study on any given day. (C) Early tocilizumab treatment is not associated with a change in neurotoxicity risk relative to CRS severity. Each subject who received tocilizumab is represented by one dot. NT > CRS: grade 3/4 neurotoxicity (NT) and mild CRS; NT = CRS: grade 3/4 NT and severe CRS, or grade 1/2 NT and mild CRS; CRS > NT: grade 1/2 NT and severe CRS, or no NT and mild CRS; CR≫NT: no NT and severe CRS. (D) CAPD scores for 3 individual patients over the course of ICU admission after CAR-T cell infusion. The y-axis indicates the highest CAPD score for any individual day, and the color of the dots shows the CTCAE neurotoxicity grade for that day. CAPD scores on or above the dashed line at CAPD = 9 are consistent with delirium. (E) Correlation of CAPD scores and CTCAE grades. Each dot represents the overall CTCAE score and highest CAPD score on any given patient day for the entire cohort. The dashed line at CAPD = 9 shows the delirium cutoff. Long bars indicate the median, and short bars the interquartile range. CAPD = Cornell Assessment of Pediatric Delirium; CAR = chimeric antigen receptor; CRS = cytokine-release syndrome; CTCAE = Common Terminology Criteria for Adverse Events; ICU = intensive care unit.

Signs and Symptoms of Neurotoxicity

In the 19 subjects with neurotoxicity, delirium was the most common symptom (Table 2), affecting 15 (79%). Delirium manifested depending on the individual’s developmental age as agitation, confusion, or impairment in activities of daily living. Seven of 19 (37%) had decreased level of consciousness, often in conjunction with delirium. Four of these 7 subjects developed reversible coma and required invasive ventilatory support because of depressed mental status. Tremor occurred in 8 subjects (42%) and seizures in 6 (32%), none of whom had a past history of seizures. Headache was present in 9 subjects (47%) exhibiting at least one other neurological symptom. Four additional subjects had headache without other neurological abnormalities and were not considered to have neurotoxicity.

TABLE 2.

Frequency of Neurological Adverse Events

CTCAE Grade	1	2	3	4	Total
	N =	N =	N =	N =
Delirium	0	10	5	0	15
Headache	2	5	2	—	9
Tremor	6	2	0	—	8
Decreased level of consciousness	1	2	2	2	7
Seizure	0	0	5	1	6
Hallucinations	2	3	0	—	5
Visual changes	2	3	0	0	5
Language disturbance	2	2	1	—	5
Abnormal movements	0	3	0	0	3
Ataxia	2	0	0	—	2
Focal weakness	1	0	1	0	2
Hydrocephalus	0	0	0	1	1
Intracranial hemorrhage	1	0	0	0	1

The most severe CTCAE grade for each symptom observed in each individual subject is shown. Dash indicates that the given grade for a symptom does not exist in CTCAE (ie, there is no life-threatening headache). Most subjects had more than one neurological symptom.

CTCAE = Common Terminology Criteria for Adverse Events.

Treatment With Steroids and Tocilizumab

Adverse events were treated with supportive care, as well as tocilizumab and steroids per protocol (Fig 1A). The strategy was initially to intervene only after a subject developed a dose-limiting toxicity, but this was changed subsequent to the 23rd infused subject to allow for earlier treatment. IL-6 blockade with tocilizumab was used as first-line treatment for CRS, whereas steroids were preferred first line for neurotoxicity. This distinction was made because of the concern that tocilizumab may preferentially increase central nervous system (CNS) IL-6 levels through binding of the IL-6 receptor in serum, causing a paradoxical increase in IL-6 and potentially aggravating neurotoxicity. Fourteen subjects received tocilizumab for CRS alone, 6 of whom subsequently developed neurotoxicity (Fig 1A). One received tocilizumab for neurotoxicity alone and 2 for concurrent CRS and neurotoxicity. Tocilizumab was given a median of 2 days after CRS onset (range, −1 to +5) and 1 day after neurotoxicity onset (range, −8 to +5). Steroids were given for neurotoxicity alone in 6, for CRS alone in 2, and for concurrent CRS and neurotoxicity in 2 subjects. Nine of 10 subjects who received steroids had previously received tocilizumab. Three subjects also received the IL-1 receptor antagonist, anakinra, for severe neurotoxicity (Fig 1A). No adverse events attributable to these interventions were observed.

Administration of tocilizumab early in the course of CRS was not associated with a change in neurotoxicity grade relative to the subject’s CRS grade (Fig 1C). Neurotoxicity started a median of 1 day after CRS onset, regardless of tocilizumab administration, but peaked later in subjects who received tocilizumab (median, 4 days after CRS onset; range, 0 to +10) compared to those who did not (median, 1.5 days after CRS onset; range, −3 to +4 days; p = 0.0353). Peripheral blood CAR-T cell expansion peaked after tocilizumab administration in all cases. These findings suggest that tocilizumab timing does not change the overall risk of neurotoxicity, but that it may modulate its kinetics.

CAPD Scores ≥9 Correlate With Severe Neurotoxicity

A new ICANS grading system has been devised to harmonize neurotoxicity assessments across institutions, replacing the CTCAE, which has been used by most investigators to date.⁷ In the new system, the CAPD¹³ is proposed as the principal assessment tool for pediatric patients, with a CAPD score above the validated delirium cutoff of ≥9 corresponding to an ICANS score of 3 (severe) on a range from 0 to 5.^7,14 The advantage of the CAPD, compared to other cognitive or mental status tools, is that it does not require patient cooperation or achievement of specific developmental milestones.

To validate the CAPD as a tool for detecting neurotoxicity, we compared CAPD and CTCAE scores for all patient days on which both measures were obtained (N = 100; Fig 1D,E). There was a positive correlation of CAPD and CTCAE scores (p < 0.0001; R² = 0.53 on linear regression). CAPD scores ≥9 identified grade ≥3 neurotoxicity with 94% sensitivity, but only 33% specificity. Specificity was increased to 72% by including grade ≥2 neurotoxicity, while still retaining 91% sensitivity. These findings indicate that the CAPD is a sensitive measure of neurotoxicity, but that it identifies a greater proportion of patients as having severe neurotoxicity than do CTCAE criteria.

Seizure Management

Seizure prophylaxis (levetiracetam 20mg/kg/day) was initiated when subjects developed severe CRS or new neurologic symptoms (n = 21); 2 of 21 (10%) had seizures despite levetiracetam prophylaxis. One additional subject developed seizures 2 days after seizure prophylaxis was stopped. Of the 22 subjects who did not receive seizure prophylaxis, 3 had seizures (14%).

Seizures lasted over 5 minutes in 2 subjects (8 minutes and 20–30 minutes) despite early administration of seizure rescue medications. One of the 6 subjects with acute seizures later developed epilepsy. EEG was performed in 5 of the 6 subjects with clinical seizures and captured additional seizures in 2. All EEGs demonstrated background slowing indicative of diffuse encephalopathy.

Clinical Risk Factors for Neurotoxicity

CRS severity correlated with the severity of neurotoxicity (p = 0.0083), and all subjects with neurotoxicity had CRS (Table 1). Neurotoxicity was more likely in subjects who had known brain MRI abnormalities before CAR-T cell treatment (4 of 5; 80%) than in those with a normal or no previous MRI (p = 0.0068). Of the 5 subjects with a normal MRI before treatment, only 2 (40%) developed neurotoxicity. However, not all patients had head imaging before treatment, so the true incidence of preexisting imaging abnormalities is unknown. Age at treatment and sex did not correlate with neurotoxicity. A history of other neurological risk factors, such as CNS-positive leukemia, preexisting neurological disorders, brain radiation, or bone marrow transplant, was not associated with increased likelihood or severity of neurotoxicity. Among treatment-related factors, there was no correlation of neurotoxicity with disease burden or CD19⁺ antigen burden at time of treatment, lymphodepletion regimen, or CAR-T cell dose. Presence of neurotoxicity did not correlate with response to therapy, event-free survival, or overall survival.¹²

Head Imaging Patterns

Fifteen of the 19 subjects with neurotoxicity had acute MRI imaging (median 9 days, range 5–15 days after CAR-T cells). Initial MRIs were obtained between −3 to +2 days from peak neurotoxicity and in 9 subjects within 24 hours of peak neurotoxicity. The 4 subjects with neurotoxicity who did not have an MRI all had grade ≤2 symptoms. Only 2 subjects without neurotoxicity had a brain MRI ≤21 days after treatment, and neither showed acute changes suggestive of neurotoxicity.

Of the 15 subjects with neurotoxicity who had an MRI, acute abnormalities were observed in 6. Four had a normal MRI, and the remaining 5 had chronic white matter lesions without acute abnormalities. We defined abnormalities as chronic if they were stable on preceding and/or follow-up scans, or the lesions were diffuse white matter abnormalities as commonly observed in patients exposed to chemotherapy and radiation.¹⁵ Of the 5 subjects with known MRI abnormalities before CAR-T cell treatment, 3 had new lesions in the previously affected brain regions, 1 had no new lesions, and 1 had new abnormalities attributable to CNS leukemia. Of the 5 patients with a normal baseline MRI, 2 had acute MRIs. One of these 2 subjects had neurotoxicity but no acute abnormalities on MRI; the other had no neurotoxicity, but the MRI showed pachymeningeal enhancement, brain sagging, and bilateral subdural fluid collections, consistent with intracranial hypotension after lumbar puncture.

The acute MRI abnormalities were classified into four different patterns which occurred alone or in combination. Two MRIs demonstrated patchy white matter interstitial edema (Fig 2A). Three subjects had T₂ hyperintensities of the periventricular white matter, medulla, and/or thalami (Fig 2B-D) in a pattern similar to the rare central-variant posterior reversible encephalopathy syndrome (PRES).¹⁶ Primary cerebellar involvement occurred in 2 subjects, 1 of whom had past radiation-related cerebellar injury. After CAR-T cell treatment, this subject developed acute cerebellar edema (Fig 2E,F), which led to obstruction of the cerebral aqueduct and acute hydrocephalus requiring temporary CSF diversion. Diffusion restriction in a gyral pattern was observed in 1 subject, indicating cortical cytotoxic edema (Fig 2G), which resolved on follow-up MRI 5 weeks later. This subject later developed epilepsy, and brain fluorodeoxyglucose positron emission tomography (FDG-PET) was performed for seizure workup 10 months after CAR-T treatment. This revealed volume loss and decreased glucose uptake in the previously affected region, suggesting irreversible neuronal injury (Fig 2H). There were no cases of global cerebral edema by imaging.

FIGURE 2:

Brain MRI patterns and evidence of chronic injury on brain histopathology in CAR-T cell neurotoxicity. (A) Pattern 1: acute patchy supratentorial white matter hyperintensities (arrowheads) are seen on this fluid-attenuated inversion recovery (FLAIR) image. The white matter lesions were not present on baseline imaging for this patient. (B) Additional subcortical white matter FLAIR hyperintensities (arrowhead). (C) Pattern 3: FLAIR hyperintensities in the brain stem and (D) bilateral thalami (arrowheads), with additional white matter lesions (arrow). (E,F) Pattern 3: acute edema in area of past injury. (E) FLAIR image obtained before CAR-T cell treatment shows cerebellar microhemorrhages (arrow) related to past cerebellar leukemic infiltrate and radiation therapy. (F) Acute edema of the cerebellar white matter (arrowheads) occurred during CAR-T cell treatment, as seen on this FLAIR image, as well as possible new microhemorrhages (arrow). (G,H) Pattern 4: cortical injury. (G) Diffusion restriction of the occipital cortex, right > left (arrowheads), is seen on this diffusion weighted image. (H) Ten months later, the same subject showed right occipital hypometabolism (arrowhead) on FDG-PET. (I–L) Brain histopathology from autopsy 3 years after initial CAR-T cell infusion, same subject as (G) and (H). (I) Right occipital cortex shows prominent gliosis of Chaslin (a pattern classically associated with chronic seizures), seen as a darker band of staining in the subpial region (double headed arrow), glial fibrillary acidic protein (GFAP) stain at 20× magnification. (J) Higher-power view of reactive astrocytes, GFAP stain, 60×. (K) Hemosiderin-laden macrophages in a perivascular space in the basal ganglia, hematoxylin and eosin stain, 40×. (L) Basal ganglia white matter with multiple round, darkly staining corpora amylacea suggestive of neuronal degeneration. Periodic acid Schiff stain at 20×; inset at 40×. CAR = chimeric antigen receptor; FDG-PET = fluorodeoxyglucose positron emission tomography; MRI = magnetic resonance imaging.

Chronic Neuropathology After CAR-T Cell-Related Acute Brain Injury

The subject whose imaging is shown in Figure 2G,H ultimately died from complications of ALL relapse 3 years after the initial CAR-T cell infusion. Autopsy revealed moderately severe cortical atrophy for age. On histopathology, mild-to-moderate gliosis of the gray and white matter (Fig 2I,J) was noted, most prominently in areas of previous CAR-T cell-related injury (Fig 2G,H). There was microglial activation and widening of perivascular spaces with perivascular hemosiderin accumulation suggestive of old microhemorrhage (Fig 2K), although there was no evidence of hemorrhage on imaging. Furthermore, large numbers of corpora amylacea were detected scattered areas of cortical gray and white matter, including basal forebrain and medial temporal lobe (Fig 2L). Corpora amylacea are associated with CNS damage and are a common finding in aging brains; however, in children, they are typically only observed in neurodegeneration from a variety of causes.¹⁷

Neurotoxicity Is Associated With Rise in CSF GFAP and S100b

Given our findings of acute inflammatory changes in areas of past injury and evidence of glial activation on CNS pathology, we hypothesized that neurotoxicity is a disorder of the neurovascular unit. A previously compromised BBB may be more susceptible to an acute inflammatory challenge, where systemic inflammation may lead to endothelial activation, leakiness of the blood-CSF barrier, and astrocyte activation.¹¹

To test this hypothesis, we obtained CSF samples from all subjects at baseline (median 6 days before CAR-T cell infusion; range, 4–22 days) and 21 days after CAR-T cell infusion, as well as during acute neurotoxicity and/or CRS when clinically indicated. Acute-phase CSF was obtained in 16 of 19 (84%) subjects with acute neurotoxicity (median 8 days, range 6–16 days after CAR-T cell infusion), but only 2 of 24 (8%) subjects without neurotoxicity (on days 6 and 7 after CAR-T cell infusion). There was no difference in baseline CSF protein and cell counts between those who would go on to have neurotoxicity compared to those who did not (Fig 3A,B). Among subjects with neurotoxicity, CSF protein increased from baseline median 30mg/dl (range, 11–133) to 117mg/dl during neurological symptoms (range, 11–1,127; p < 0.0001), and CSF white blood cell counts increased from median 1cell/μl (range, 0–6) to 3.5cells/μl (range, 0–89; p = 0.009). CSF protein and white blood cell counts returned to baseline levels on day 21.

FIGURE 3:

Inflammatory infiltrate and elevated GFAP and S100b levels in CSF during neurotoxicity. (A) CSF protein and (B) white blood cell (WBC) counts. (C) CSF GFAP and (D) S100b levels. In all graphs, each point represents data from an individual subject. Empty circles, no neurotoxicity; filled circles, grade 1 to 4 neurotoxicity. The number of individual subject samples in each group is given below the plots. Bars show the median (long bar) and interquartile range (short bars). “Pre”: sample obtained before CAR-T cell infusion; “Acute”: sample obtained while symptomatic with CRS and/or neurotoxicity (days 6–16); “D21”: day 21 after treatment. *p < 0.05; **p < 0.01; ***p < 0.001, Kruskal-Wallis with uncorrected Dunn’s posttest. CAR = chimeric antigen receptor; CRS = cytokine-release syndrome; CSF = cerebrospinal fluid; GFAP = glial fibrillary acidic protein.

GFAP has been well validated as a marker of astroglial injury in multiple neurological pathologies such as dementia and stroke.^18,19 We therefore quantified GFAP in the serum and CSF (Fig 3C). CSF GFAP levels before treatment were similar in patients with (median, 5,012pg/ml; range, 1,890–13,028) and without neurotoxicity (median, 5,798pg/ml; range, 1,005–12,366; p = 0.92). However, during acute neurotoxicity, median GFAP level rose significantly to a median of 8,898pg/ml (range, 2,761–41,656; p = 0.0037). GFAP levels in the serum did not differ between groups (data not shown).

S100b indicates astrocyte activation and CNS injury when measured in the CSF, whereas serum levels are more difficult to interpret because of systemic sources of S100b.²² Pretreatment CSF S100b concentrations were similar between the neurotoxicity (median, 35pg/ml; range, 0–324) and no-neurotoxicity groups (median, 57pg/ml; range, 0–213; p = 0.90). In patients with acute neurotoxicity, CSF levels of S100b rose significantly (median, 201; range, 13–828pg/ml; p = 0.0002). Together, these data provide evidence of abnormal astrocyte function during acute neurotoxicity.

Given the close relationship of astrocytes and the endothelial barrier,²⁰ we next sought evidence of coagulopathy and endothelial activation during neurotoxicity. However, we found no statistically significant association of neurotoxicity with serum levels of VEGF-A, VWF, Ang-1, Ang-2, and Ang-2/Ang-1 ratio, peak international normalized ratio and D-dimer, and nadir fibrinogen levels or platelet counts (data not shown).

CSF CAR-T Cells Skew Toward CD4⁺ in Patients With Neurotoxicity

To better understand whether T cells and other immune effector cells play an active role in modulating neurotoxicity,²¹ we next compared the composition of blood CAR-T cells with the cellular infiltrate in the CSF. Flow cytometry analysis of subject blood specimens revealed similar peak CAR-T cell expansion in subjects with and without neurotoxicity (Fig 4A). However, subjects with severe neurotoxicity (grade 3–4) had greater peak blood CAR-T cell counts (median, 518cells/μl; range, 103–10,445) than those with grade 1 to 2 neurotoxicity (median, 143cells/μl; range, 2–404; p = 0.0080) or no neurotoxicity (median, 181cells/μl; range, 0–1,451; p = 0.0103).

FIGURE 4:

Enrichment of CD4⁺ CAR-T cells in the CSF. (A) Peak and day 21 CAR-T cell counts in the blood. (B) CSF CAR-T cell counts during the acute phase and on day 21. (C) The y-axis indicates the percentage of CD3⁺ T cells that were CAR-T cells in blood and CSF samples. (D) Percentage of CAR-T cells that were CD4⁺ in blood and CSF. In all graphs, each point represents data from an individual subject. Empty circles, no neurotoxicity; filled circles, grade 1 to 4 neurotoxicity. The number of individual subject samples in each group is given below the plots. Bars show the median (long bar) and interquartile range (short bars). “EGFRt⁺”: CAR-T cells labeled with Erbitux; “Acute”: sample obtained while symptomatic with CRS and/or neurotoxicity (days 6–16); “Peak”: highest CAR-T cell density in each subject’s course of treatment; “D21”: day 21 after treatment; “No NT”: no neurotoxicity; “Grade 1-4 NT”: grade 1 to 4 neurotoxicity. The number of CAR-T cells detected in the acute CSF of patients without neurotoxicity was too low to quantify expression patterns; therefore, no data are shown for this group in (C) and (D). n.s., not significant at 95% confidence level. *p < 0.05; **p < 0.01; ***p < 0.001, Kruskal-Wallis with uncorrected Dunn’s posttest. CAR = chimeric antigen receptor; CRS = cytokine-release syndrome; CSF = cerebrospinal fluid.

Similar numbers of CAR-T cells were detected in the CSF of subjects with or without neurotoxicity on day 21 (Fig 4B). Additionally, there was no correlation between neurotoxicity grade and CSF CAR-T cell counts during neurotoxicity. In the acute period, CAR-T cells were detected in the CSF of 13 of 15 subjects with neurotoxicity, and 2 of 2 subjects without neurotoxicity. On day 21, CAR-T cells were present in the CSF of all subjects except 2, both of whom had no neurotoxicity.

The CAR-T cell fraction of total T cells in the blood and CSF was similar during acute neurotoxicity, indicating that there was no preferential trafficking of CAR-T cells into the CSF (Fig 4C). However, on day 21, the CSF contained a higher fraction of CAR-T cells compared to the blood both in subjects with and without neurotoxicity (Fig 4C). This may be attributed to persistence of a cell population in the CSF that mirrors the systemic lymphocyte composition during the acute phase of CAR-T cell expansion.

The fraction of CD4⁺ CAR-T cells was higher in the CSF compared to the blood in patients with neurotoxicity, both in the acute setting (median 49% in CSF versus 8% in blood; p = 0.0064) and on day 21 (median 39% CD4⁺ in CSF versus 6% in blood; p = 0.0050). Patients without neurotoxicity had no statistically significant difference between blood and CSF CD4⁺ fractions (day 21 CSF median CD4⁺ 22%, blood median 9%; p = 0.2700). In addition, on day 21, the CD4⁺ fraction of CAR-T cells in the CSF was higher in patients with neurotoxicity compared to those without (39% versus 22%; p = 0.0492). These data suggest preferential trafficking or survival of CD4⁺ CAR-T cells in the CSF during neurotoxicity.

IL-6, IL-10, IFN-γ, and GzB Are Key Cytokines During Neurotoxicity

Given that systemic cytokine signaling during the inflammatory state is known to interact with components of the BBB,²² including astrocytes and endothelial cells, we next measured levels of key cytokines in the blood and CSF. Concentrations of IFNγ, IL-6, IL-10, and GzB were markedly elevated in the CSF during acute neurotoxicity (Fig 5A), and we therefore focused on these markers in subsequent analyses. There were no significant changes in the remainder of biomarkers that were tested in CSF (Ang-1, Ang-2, granulocyte macrophage colony-stimulating factor, sCD137, sFas, sFasL, granzyme A, IL-2, IL-4, IL-5, IL-13, MIP-1α, MIP-1β, TNF-α, perforin, and VEGF-A). Only 2 subjects without neurotoxicity had lumbar punctures during the acute phase, thus we cannot exclude the possibility that subjects without neurotoxicity also had elevated CSF cytokines.

FIGURE 5:

Markers of CNS inflammation are elevated during acute CAR-T neurotoxicity. (A) Cytokine levels in CSF samples from subjects with neurotoxicity. (B) Serum cytokine levels on day 7 after CAR-T cell infusion, stratified by CRS grade. (C) Paired same-day CSF and serum cytokine levels. Empty circles, no neurotoxicity; filled circles, grade 1 to 4 neurotoxicity. The number of individual subject samples in each group is given below the plots. Bars show the median (long bar) and interquartile range (short bars). “Pre”: before CAR-T cell infusion; “Acute”: sample obtained while symptomatic with CRS and/or neurotoxicity (days 6–16); “D21”: on day 21 after treatment. “Gz B”: granzyme B. n.s., not significant at 95% confidence level. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001, Kruskal-Wallis with uncorrected Dunn’s posttest for (A) and (B), paired Wilcoxon signed-rank test for (C). CAR = chimeric antigen receptor; CNS = central nervous system; CRS = cytokine-release syndrome; CSF = cerebrospinal fluid; IFN = interferon; IL = interleukin.

Serum IFNγ, IL-6, and IL-10 elevations have been shown to correlate closely with CRS,^4,23 which, in turn, correlates with neurotoxicity. Indeed, in our cohort, day 7 serum concentrations of GzB, IFNγ, and IL-10 were higher in subjects with neurotoxicity compared to those without. To control for the contribution of CRS to this difference, we stratified day 7 serum samples by CRS grade (Fig 5B). IL-10 and GzB showed greater elevations in subjects with neurotoxicity within the same CRS grade compared to those without neurotoxicity, whereas there was no statistically significant difference in IL-6 and IFNγ levels.

We then compared cytokine profiles in serum and CSF to determine whether there is any evidence of cytokine production within the CNS. IFNγ was higher in the serum than in the CSF in paired acute samples from subjects with neurotoxicity (mean, 930 versus 268pg/ml; p = 0.0497), whereas GzB, IL-6, and IL-10 levels did not differ between the two compartments (Fig 5C), providing evidence against production within the CNS. Additional serum biomarkers (day 7 serum granulocyte macrophage colony-stimulating factor, MIP-1a, and TNFα) were higher in patients with neurotoxicity compared to those without, but did not show a concomitant increase in the CSF (data not shown).

Discussion

In this prospective cohort study of children and young adults undergoing CD19-directed CAR-T cell treatment, we provide new insights on ICANS pathophysiology based on comprehensive CSF analyses on all patients at predefined time points, which has not previously been performed in this age group. Our conclusions are tempered by the fact that acute CSF samples were not available from all patients without neurotoxicity, which will be needed for future confirmatory studies. The clinical phenotypes, imaging findings, and key implicated cytokines we describe are comparable to findings in other studies in adults and children.^9,24-26 This suggests that our findings may have broad validity despite the heterogeneity of age groups, CAR-T cell products, and cancer diagnoses among published studies of neurotoxicity.

Early detection of neurological deterioration is of paramount importance after CAR-T cell treatment. However, subtle changes in mental status can be easily missed in young children who may be critically ill and unable to cooperate in language-based mental status exams. We provide the first evidence that the CAPD is a sensitive measure of neurological dysfunction in this patient group, which supports its use in formal diagnostic criteria for ICANS.⁷

Prevention and management guidelines for ICANS are currently based on expert consensus, but there is intense interest in better understanding the pathophysiology of the disorder to allow development of targeted interventions. Our study contributes several important insights. First, we confirm that degree of CRS is the most robust predictor of neurotoxicity, consistent with findings in other cohorts using different CAR-T cell products for varying indications and age groups.⁶ The close relationship of CRS and neurotoxicity suggests that systemic inflammatory mediators act directly on the BBB, which may involve endothelial activation,²⁰ pericyte signaling,²⁷ and astrocyte activation.¹¹ Opening of the BBB also would allow immune effector cells to traffic into the CNS, where they can exert a localized inflammatory function.^21,28

We found that peak CAR-T cell engraftment in peripheral blood was higher in subjects with severe neurotoxicity, an association that has been noted in other studies as well.^9,29 However, the highest risk of both CRS and neurotoxicity occurs during the period of rapid CAR-T cell expansion, preceding the peak of CAR-T cell blood counts. This suggests that toxicity is primarily mediated by the inflammatory cytokine surge that accompanies CAR-T cell expansion in the marrow, rather than the CAR-T cells themselves.

The importance of systemic cytokine signaling is also illustrated by the association of elevated granulocyte macrophage colony-stimulating factor, GzB, IFNγ, IL-6, IL-10, TNFα, and MIP-1α levels with neurotoxicity. Key inflammatory cytokines showed similar elevations in the CSF and serum, suggesting that local cytokine production in the CNS is not a prominent feature of neurotoxicity. Evidence is emerging that T cells may activate monocytes through granulocyte macrophage colony-stimulating factor and that the monocytes are an important source of proinflammatory cytokines.^10,30-32 These cytokines have also been implicated in a number of other CNS diseases³³ and are known to activate the endothelial cells at the BBB,^9,34,35 supporting the theory that ICANS is initiated by a systemic cytokine surge.

However, we found no compelling evidence of endothelial activation during neurotoxicity in our pediatric cohort, given that there was no association of serum VWF, VEGF-A, Ang-1, and Ang-2 levels with neurotoxicity. This is a surprising finding given that we and others have found increased concentrations of VEGF, VWF, and Ang-2, as well as higher Ang-2/Ang-1 ratios in adult patients with severe neurotoxicity.^9,36 One possible explanation is a confounding role of CRS, which was present in almost all of our patients. It is also possible that there are differences in pathophysiology in the pediatric compared to the adult population, but this appears unlikely given the overall similar clinical presentations.

For the first time, we provide evidence of abnormal astrocyte state during neurotoxicity with increased CSF GFAP and S100b levels. Astrocytes are key osmotic regulators in the brain, and disruption of their water handling in the neurovascular unit could contribute to cerebral edema.^37,38 Given that GFAP and S100b are also elevated in other acute and chronic CNS pathologies, including stroke, trauma, infection, and neurodegenerative disorders,^39,40 animal studies will be needed to determine the precise role of glia in ICANS pathogenesis.

The cellular composition of the CSF during acute neurotoxicity does not support a primary role of infiltrating CAR-T cells in the development of ICANS, given that CAR-T cells were present in the CSF of most patients without neurotoxicity. Others have also reported efficient CNS trafficking of CAR-T cells without an increase in neurotoxicity,²⁸ and pathology studies have shown limited inflammatory infiltrates in the brain after cerebral edema during CAR-T cell treatment.^9,41 CD4⁺ CAR-T cells were over-represented in the CSF of patients with neurotoxicity, similar to findings in adults.⁹ Further study will be needed to determine what mechanisms are responsible for this bias in T cell subsets, and whether CAR-modified T cells participate in pro- or anti-inflammatory signaling in the CNS.^42,43

Although ICANS has primarily been studied in CD19-directed CAR-T cell therapy, similar neurotoxicity has been reported with blinatumomab, which targets CD19 through a bispecific T-cell engager,^44,45 and after CD22-directed CAR-T cell treatment for ALL.⁴⁶ Neurotoxicity has so far not been observed with CAR-T cells targeting non-CNS solid tumors, likely because CAR-T cell expansion and tumor killing have not been as robust.⁶ Therefore, it is possible that ICANS is related to the signaling and cellular environment that accompanies successful immunotherapy rather than the CAR-T cells themselves. Going forward, it is our hope that standardized definition and better mechanistic understanding of ICANS will allow for development of rational preventive and therapeutic approaches.