Chimeric Antigen Receptor T Cell–Mediated Neurotoxicity in Nonhuman Primates

CD19-specific CAR T-cell immunotherapy can mediate the regression of relapsed and refractory leukemias and lymphomas (1–4). However, the side effects and toxicities of this therapy, particularly in patients with acute lymphoblastic leukemia (ALL) with high tumor burden, are significant and can be life-threatening (5–7). The most common toxicities include cytokine release syndrome (CRS) and neurotoxicity, which have resulted in a number of patient deaths in multiple CAR T-cell trials (5–11). The neurologic toxicities span a wide spectrum, ranging from transient word-finding difficulties, visual and auditory hallucinations, delirium, and movement disorders, to encephalopathy and seizures. Rapid onset cerebral edema, unresponsive to medical measures, is a rare but lethal event reported in numerous clinical trials (12). Corticosteroids and the anti-IL6 receptor (IL6R) monoclonal antibody tocilizumab can ameliorate CRS; however, medical interventions to reduce the incidence and severity of neurotoxicity are currently lacking (7, 10, 11).

At present, the pathobiology of severe neurotoxicity associated with CAR T-cell activation and expansion has not been defined, and animal models are needed to derive mechanistic insights and test therapeutic interventions. To address these limitations, we have developed the first nonhuman primate (NHP) model of CRS and neurotoxicity, using CD20 CAR T cells in rhesus macaques (RM). This model recapitulates the key aspects of CAR T-cell on-target effects such as CAR T-cell expansion with induction of B-cell aplasia as well as CRS and neurotoxicity, and clarifies that toxicities are neither (i) tumor-dependent, (ii) specific CAR T cell–dependent, nor (iii) CD19 antigen-specific. Moreover, it demonstrates that neurotoxicity is associated with a complex program of immune activation that results in asymmetrically high levels of proinflammatory cytokines and chemokines in the cerebrospinal fluid (CSF) and widespread infiltration of the brain parenchyma with both EGFRt⁺ (CD20 CAR) and EGFRt⁻ (non-CAR) T cells in equal proportions to those present in the circulation.

RMs express multiple tripartite motif-containing 5α (TRIM5α) isoforms that confer resistance to retroviral infection, resulting in poor transduction efficiency with HIV-1–based lentiviruses due to HIV-1 capsid degradation (13). In order to overcome this restriction, and to improve transduction efficiency, we utilized two previously developed lentiviruses, χHIV (HIV-1–based vector genome packaged with an SIV capsid) and SIV (SIV vector genome packaged with an SIV capsid) for RM T-cell transductions (14, 15). The CD20 CAR χHIV or SIV lentivirus vectors housed a huEF1p or a CAG promoter, respectively, a human/RM cross-reactive CD20-specific second-generation 4-1BB:zeta CAR, a T2A ribosomal skip sequence, and a truncated human EGFR (EGFRt) marker (Supplementary Fig. S1A; refs. 16, 17). Following anti-CD3/anti-CD28 bead activation, immunomagnetically purified RM T cells were amenable to transduction with GFP-χHIV (R.301), GFP-χHIV and CD20 CAR/EGFRt-χHIV (R.302), CD20 CAR/EGFRt-χHIV (R.301, R.303), and CD20 CAR/EGFRt-SIV (R.304) lentiviruses, at multiplicities of infection (MOI) ranging from 1 to 5, resulting in transduction efficiencies of 22% to 75% (Fig. 1A)

Following transduction, CD20 CAR T-cell products were successfully expanded for all recipients (R.301–R.304, n = 4) by addition of IL2 (50 U/mL) to X-Vivo 15 cell culture medium to achieve the targeted cell dose of 1 × 10⁷ CD20 CAR T cells/kg. Ex vivo T-cell expansion ranged from 5- to 36-fold after 8 to 17 days of culture (Supplementary Table S1). The final GFP⁺ and EGFRt⁺ (CD20 CAR) T-cell products at the end of ex vivo culture were used for adoptive transfer experiments and consisted of a ∼2:1 CD8⁺:CD4⁺ ratio (60.6% ± 1.7%: 28.4% ± 3.6%; n = 5; Fig. 1A). We further assessed the CD20 CAR T-cell products with respect to the relative proportions of CD28⁺/CD95⁺ (NHP central memory phenotype), CD28⁻/CD95⁺ (NHP effector memory phenotype), and CD28⁺/CD95⁻ (NHP naïve phenotype) T cells in both EGFRt⁺ (CD20 CAR) and EGFRt⁻ (non-CAR) populations at the end of ex vivo expansion (Fig. 1B). In the four CD20 CAR T-cell products, the majority of EGFRt⁺ T cells displayed a CD28⁺/CD95⁺, central memory phenotype, for both CD4⁺ (85% ± 8.4%) and CD8⁺ cells (67.6% ± 8.1%; Fig. 1B, top). Similarly, the majority of EGFRt⁻ T cells also displayed a CD28⁺/CD95⁺ phenotype, for both CD4⁺ (78.5% ± 9%) and CD8⁺ cells (56.7% ± 7.5%; Fig. 1B, bottom). A smaller proportion of EGFRt⁺ T cells displayed a CD28⁻/CD95⁺ effector memory phenotype, for both CD4⁺ (12.7% ± 8.2%) and CD8⁺ cells (29.2% ± 8.7%), whereas very few cells displayed a CD28⁺/CD95⁻ naïve phenotype, for both CD4⁺ (1% ± 0.8%) and CD8⁺ cells (1% ± 0.5%; Fig. 1B, top). The same phenotype was observed in the EGFRt⁻ T cells in the products (CD28⁻/CD95⁺ CD4⁺: 18.1% ± 9.6% CD4⁺ and CD8⁺: 41.9% ± 8.2% cells; CD28⁺/CD95⁻ CD4⁺: 1.9% ± 1.4% and CD8⁺: 0.8% ± 0.5% cells; Fig. 1B, bottom). These data demonstrate that the majority of T cells in the infused products comprised a CD28⁺/CD95⁺, central memory phenotype, a T-cell subset that has been shown to mediate enhanced T-cell activity in murine models, and increased persistence following adoptive transfer in NHP studies (18, 19). Furthermore, although the infused cell products contained both EGFRt⁺ (CD20 CAR) and EGFRt⁻ (non-CAR) T cells, their CD4⁺ and CD8⁺ T-cell phenotypes were similar irrespective of CD20 CAR expression. We subsequently monitored these T-cell phenotypes longitudinally in vivo following CD20 CAR T-cell adoptive transfer.

We assessed the in vitro cytolytic activity of each of the CD20 CAR-transduced T-cell products in a chromium release assay by their ability to mediate CD20-specific cytolysis of both human (CD20-K562) and RM (RM B-LCL) cells with a wide range of CD20 antigen expression (Fig. 1C). CD20-specific cytotoxicity was demonstrated by the lack of CD20 CAR T cell–mediated cytolysis of human K562 cells, which do not express CD20 (Fig. 1C).

The schema used for adoptive transfer of CD20 CAR T cells into RMs is outlined in Fig. 2A. For the first recipient, R.301, 48 days prior to the CD20 CAR T-cell infusion, we administered autologous GFP T cells, following lymphodepletion, to assess the impact of adoptive transfer of T cells lacking the CD20 CAR construct. The GFP T-cell infusion was well tolerated, without any observed clinical toxicities (Fig. 3A and B and Fig. 4A, R.301 GFP). As expected, GFP T cells failed to undergo expansion (peak level of 28 GFP⁺ T cells/μL) and did not induce B-cell aplasia (Fig. 2B, R.301 GFP). These cells were not measurable in the peripheral blood after day 14 following infusion (Fig. 2B, R.301 GFP). These findings were similar to a previous report of CAR T cells targeting the solid tumor antigen ROR1 in RMs, which demonstrated limited expansion and persistence of ROR1 CAR T cells (20). In contrast, upon transfer of CD20 CAR T cells to all four recipients, R.301 to R.304, the CD20 CAR T cells underwent significant expansion, with peak levels of 272 to 4,450 (mean level, 1,987 ± 987) EGFRt⁺ T cells/μL in the peripheral blood, which occurred on days 7 to 8 after infusion (Fig. 2B). This EGFRt⁺ T-cell expansion was accompanied by a concomitant B-cell aplasia in all four recipients, which occurred on days 5 to 7 following CD20 CAR T-cell infusion (Fig. 2B). Thus, the CD20 antigen load prior to lymphodepletion was 700 ± 394 cells/μL, and after lymphodepletion, but prior to CAR T-cell infusion, 135 ± 32 cells/μL in blood, and these cells became unmeasurable concomitant with CAR T-cell expansion. In two recipients who were followed long term, R.301 and R.302, B-cell aplasia persisted through day 32 (R.301) and day 43 (R.302; Fig. 2B).

In two pilot experiments, R.301 and R.302 were also evaluated for the impact of an infusion of T cells engineered to express the CD20 antigen (RhCD20 T-APC) on CD20-CAR T-cell expansion and persistence. A single infusion of RhCD20 T-APCs on day 47 had no effect on CD20 CAR T-cell numbers in R.301, likely due to the fact that the T-APCs were administered when the CD20 CAR T cells were no longer detectable and native B cells had reconstituted in the peripheral blood (Fig. 2B). In contrast, a single infusion of RhCD20 T-APCs administered on day 15 after CD20 CAR T-cell transfer in R.302 resulted in a 10-fold increase in circulating EGFRt⁺ T-cell numbers, from 9 to 93 EGFRt⁺ cells/μL, measured 5 days following T-APC administration. These data suggest that engineered RhCD20 T-APCs are potentially capable of activating and expanding circulating CD20 CAR T cells and represents an important area for future investigations.

Following adoptive transfer of CD20 CAR T cells, activation and expansion of both EGFRt⁺ (CD20 CAR) and, to a lesser extent, EGFRt⁻ (non-CAR) T-cell populations were observed (Fig. 2B and E; Supplementary Fig. S2). This expansion favored CD8⁺ T cells, which resulted in an increased CD8⁺:CD4⁺ EGFRt⁺ ratio of 9.1 ± 2 in all four recipients (R.301–R.304) during peak expansion following CD20 CAR T-cell infusion, compared with a ratio of 2.2 ± 0.3 in the starting infused products (P = 0.03, Fig. 2D). Although R.303 displayed a more modest increase in the CD8⁺:CD4⁺ EGFRt⁺ ratio compared with other recipients, in this recipient we observed a significant expansion of CD8⁺CD4⁺ double-positive EGFRt⁺ T cells (previously shown to represent an effector population in RMs; ref. 21), which were not captured in the CD8⁺:CD4⁺ ratio. Although EGFRt⁻ (non-CAR) T cells also underwent proliferation (Fig. 2E, top two plots), their CD8⁺:CD4⁺ ratio during peak expansion was minimally skewed to CD8⁺ T cells, 1.6 ± 0.8, which was significantly different from the CD8⁺:CD4⁺ ratio of EGFRt⁺ T cells at the same time point (P = 0.01), yet unchanged from the day 0 (pre-CD20 CAR T-cell infusion) CD8⁺:CD4⁺ EGFRt⁻ ratio of 0.6 ± 0.05 (P = 0.2, Fig. 2D). In addition to enumerating the expansion of EGFRt⁺ (Fig. 2B) and EGFRt⁻ T cells (Supplementary Fig. S2), we also measured their proliferation status by Ki67 expression (Fig. 2E). Both EGFRt⁺ CD4⁺ and CD8⁺ cells were highly proliferative, with Ki67 levels reaching 89.8% ± 4.9% in CD4⁺ and 94% ± 4.4% in CD8⁺ cells in all four recipients (R.301–R.304) during peak expansion, which approximated the proliferation in EGFRt⁺ cells in the starting infused products: 84% ± 5% in CD4⁺ and 90.3% ± 4.4% in CD8⁺ cells. Although both CD4⁺ and CD8⁺ EGFRt⁻ (non-CAR) cells also increased concomitantly with CD20 CAR T-cell expansion, their level of proliferation was significantly less compared with EGFRt⁺ cells, with Ki67 expression of 38.5% ± 7% in CD4⁺ and 51.5% ± 13% in CD8⁺ cells (P = 0.02 for CD4⁺, P = 0.02 for CD8⁺; Fig. 2E, top two plots). The EGFRt⁺ cells displayed a trend toward higher levels of CD25 in comparison with EGFRt⁻ cells, suggesting that they were more activated during proliferation; however, this difference in CD25 expression was not statistically significant [77.3% ± 12% vs. 45.8% ± 18% for CD4⁺ (P = 0.09) and 77.8% ± 13% vs. 47.8% ± 17% for CD8⁺ (P = 0.15) Fig. 2E]. In addition to proliferation and activation, we also monitored the longitudinal dynamics of memory, effector, and naïve T-cell phenotypes of CD20 CAR T cells after adoptive transfer. Within both CD4⁺ and CD8⁺ EGFRt⁺ (CD20 CAR) T cells, we observed the preservation of a similar proportion of the CD28⁺/CD95⁺ central memory subset during peak expansion compared with infused product [78.3% ± 6.2% vs. 85% ± 8.4% for CD4⁺ (P = 0.5) and 61.9% ± 8.6% vs. 67.6% ± 8.1% for CD8⁺ (P = 0.2)]. The same was true for the CD28⁻/CD95⁺ effector memory subset [15.5% ± 6.6% vs. 12.7% ± 8.2% for CD4⁺ (P = 0.8) and 24.8% ± 10% vs. 29.2% ± 8.7% for CD8⁺ (P = 0.7)], and the CD28⁺/CD95⁻ naïve subset [3.6% ± 2.8% vs. 1% ± 0.77% CD4⁺ (P = 0.5) and 8.3% ± 7.7% versus 1% ± 0.5% CD8⁺ (P = 0.4)] when EGFRt⁺ T-cell subsets were compared at peak expansion with infused product (Fig. 2F). These data suggest that the majority of EGFRt⁺ (CD20 CAR) T cells maintain a CD28⁺/CD95⁺ central memory phenotype in vivo, during peak expansion, at similar proportions to those present in the infused CD20 CAR T-cell products. It should be noted that differences that were observed in the proportion of EGFRt⁺ CD28⁺/CD95⁺ CD8⁺ cells in R.301 and R.302 after T-APC infusion (Fig. 2F) may have been affected by the T-APC infusion, and thus conclusions about the natural history of CAR T-cell phenotypes in these animals cannot be made.

After adoptive cell transfer, we closely monitored the clinical status of each recipient for specific signs and symptoms of CRS and neurotoxicity. All of the recipients of CD20 CAR T-cell infusions developed clinical signs of CRS and exhibited abnormal neurologic manifestations (described in detail below), which was in contrast to R.301 after the control GFP T-cell infusion. Weight loss exceeding 10% occurred in 2 of 4 animals, R.301 and R.302, whereas temperature instability, characterized by fever (>101F), occurred in all 4 animals (R.301–R.304), and hypothermia (<98F) occurred in 1 animal (R.301) after CD20 CAR T-cell infusion (Fig. 3A, right). C-reactive protein (CRP) was elevated to peak levels of 2.8- to 4.2-fold above the upper limit of normal (ULN) in all recipients of CD20 CAR T cells and accompanied CAR T-cell expansion, but remained within normal limits in R.301 following non-CAR, GFP T-cell infusion (Fig. 3B). Ferritin and lactate dehydrogenase (LDH) levels were more heterogeneous: in two recipients, R.301 and R.302, peak ferritin was elevated 2.1- and 4.2-fold above the ULN (180 ng/mL), whereas peak LDH was elevated 6.7- and 9.6-fold above the ULN (210 U/L), respectively, following CD20 CAR T-cell infusion (Fig. 3B). In the remaining 2 recipients, R.303 and R.304, we observed normal ferritin levels and modestly elevated peak LDH levels (2.1- and 1.3-fold above ULN, respectively) following CD20 CAR T-cell infusion (Fig. 3B). To examine changes in cytokine levels following CD20 CAR infusion, we serially measured the serum levels of 29 NHP cytokines and chemokines. We found that levels of proinflammatory cytokines IL6, IL8, and IL1RA and chemokines MIG (CXCL9) and I-TAC (CXCL11) were statistically significantly elevated in all 4 recipients (R.301–R.304) following CD20 CAR T-cell infusion compared with preinfusion levels (Fig. 3C). In CD20 CAR T-cell recipients, the peak serum levels of IL6 were elevated 12-fold to 14.7 ± 3.5 pg/mL (P = 0.008), IL8 levels were elevated 3.3-fold to 127.8 ± 26 pg/mL (P = 0.02), IL1RA levels were elevated 10-fold to 2,694.3 ± 633 pg/mL (P = 0.009), MIG levels were elevated 79-fold to 1,268.7 ± 457.9 pg/mL (P = 0.03), and I-TAC levels were elevated 23-fold to 2,151 ± 448.2 pg/mL (P = 0.004), compared with preinfusion levels (peak levels were observed on days 2–10 following CAR T-cell infusion and preinfusion levels were measured on day −7 or day −8 prior to infusion; Fig. 3C. The full panel of cytokines measured is shown in Supplementary Fig. S3).

To quantify the neurologic impact of CD20 CAR T-cell therapy, we developed a clinical scoring system for neurotoxicity (described in Methods and Supplementary Table S2). Neurotoxicity occurred in all recipients receiving CD20 CAR T cells (R.301–R.304; n = 4), with the most severe neurotoxicity observed between days 7 and 14 following infusion (Fig. 4A). The most common neurotoxicity symptoms included tremor (ranging from minimal to severe), decreased activity or lethargy, slow movements, and ataxic gait (Fig. 4A; Supplementary Table S2). Tremors and decreased activity level as well as lethargy contributed the most to the neurotoxicity score, whereas slow movement, ataxia, and seizures had a smaller contribution to the total score (Fig. 4A). These neurologic symptoms have also been reported in some patients following CD19 CAR T-cell therapy, where tremors, decreased level of consciousness, and ataxia/other abnormal movements were reported in 19%, 25%, and 10% of patients with neurologic toxicity, respectively (12). Three of the 4 recipients (R.301–R.303) were treated with levetiracetam (15 mg/kg b.i.d. or t.i.d.), which was started after neurologic symptoms manifested. Levetiracetam therapy was maintained for 8 to 11 days following full resolution of all neurologic symptoms (through day 22 for R.301 and day 29 for R.302) or until study end point (day 12 for R.303; Fig. 4A, arrows). R.304 underwent a planned terminal analysis on day 8 after CD20 CAR T-cell infusion, prior to the start of levetiracetam. For R.301 and R.302, who were followed long term, neurotoxicity symptoms waned and resolved at 14 and 18 days after CD20 CAR T-cell infusion, respectively, concurrent with CD20 CAR T-cell contraction (Fig. 4A).

Given that in human clinical trials of CD19 CAR T-cell therapy, high numbers of CD19 CAR T cells were found in the CSF and correlated with neurotoxicity (22), we sought to determine whether CD20 CAR T cells also accumulated in the CSF in the RM model. In three recipients (R.302–R.304), the CSF was examined longitudinally and revealed trafficking of T cells into the CSF after adoptive CD20 CAR T-cell transfer, coincident with CAR T-cell expansion in the blood and the onset of clinical neurotoxicity (Fig. 4B; Supplementary Fig. S4B). Importantly, the T-cell accumulation in the CSF included both EGFRt⁺ (CD20 CAR) and EGFRt− (non-CAR) cells, with the proportion of each closely mirroring that in the peripheral blood (Fig. 4B). Notably, when cytokine levels were measured in the CSF and compared with time-matched levels in the serum, these were markedly different for several cytokines, demonstrating increased concentrations of multiple proinflammatory cytokines in the CSF relative to serum (Fig. 4C). This included increased concentrations of IL6 (478-fold increase, mean peak CSF level, 5,314.7 ± 175 pg/mL, mean serum level 11.1 ± 3.2 pg/mL, P = 0.000007), IL2 (10-fold increase, mean peak CSF level 118 ± 21.1 pg/mL, mean serum level 11.9 ± 11.9 pg/mL, P = 0.012), GM-CSF (8.2-fold increase, mean peak CSF level 1.96 ± 0.09 pg/mL, mean serum level 0.24 ± 0.16 pg/mL, P = 0.00077) and VEGF (6.3-fold increase, mean peak CSF level 6.95 ± 1.8 pg/mL, mean serum level 1.1± 0.26 pg/mL, P = 0.03; Fig. 4C). IL1β, IL1RA, MCP-1, and IP-10 levels were also asymmetrically elevated in the CSF; however, the difference between CSF and serum did not reach statistical significance. IL1β was elevated 7.8-fold (mean peak CSF level, 25.4 ± 13.4 pg/mL, mean serum level, 3.27 ± 0.5 pg/mL, P = 0.17), IL1RA was elevated 6.3-fold (mean peak CSF level 13,338 ± 7,048.6 pg/mL, mean serum level 2,129 ± 1,273 pg/mL, P = 0.19), MCP-1 was elevated 13.7-fold (mean peak CSF level 3,405 ± 1,603.7 pg/mL, mean serum level 248 ± 81.1 pg/mL, P =0.12) and IP-10 was elevated 97-fold (mean peak CSF level 849 ± 439.8 pg/mL, mean serum level 8.7 ± 7.8 pg/mL, P = 0.13) in the CSF compared with paired serum samples (Fig. 4C; the full panel of serum cytokines is included in Supplementary Fig. S3 and CSF cytokines in Supplementary Fig. S5).

To date, limited data are available as to the numbers, composition, and activation state of T cells within the brain parenchyma of human subjects experiencing CAR T-cell neurotoxicity. To rigorously determine the anatomic correlates of neurotoxicity and trafficking of EGFRt⁺ (CD20 CAR) and EGFRt⁻ (non-CAR) T cells to the central nervous system (CNS), we performed timed necropsies on recipients R.303 [on day 12, during peak neurotoxicity symptoms, which included moderate tremors and lethargy, slow movements, and ataxia (with ataxia manifested by an animal's inability to maintain balance on a perch and/or the need for support to maintain balance while sitting or standing)] and R.304 (on day 8, during peak CD20 CAR T-cell expansion, at the onset of neurotoxicity symptoms, which included mild tremor and decreased activity on day 7), in order to assess the degree of CSF and brain infiltration at predetermined times after T-cell infusion. We focused on specific regions of the brain, including the frontal, parietal, temporal, and occipital lobes, cerebellum, hippocampus, basal ganglia, brain stem, choroid plexus, meninges, and thalamus. These experiments demonstrated that, for R.304, during peak CAR T-cell expansion (on day 8 after CD20 CAR T-cell transfer), histopathologic changes were present in all regions of the brain examined, including in both the gray and white matter, and were exemplified by mild multifocal meningitis and mild to moderate, multifocal perivascular T-cell cuffing, as demonstrated by hematoxylin and eosin (H&E) staining in the basal ganglia and the parietal lobe (Fig. 5A, bottom). In R.303, the recipient that underwent planned terminal analysis during peak neurotoxicity (on day 12 after CD20 CAR T-cell transfer), minimal histopathologic cerebral abnormalities were demonstrated on H&E staining, with most of the brain appearing normal. However, rare abnormalities were observed in R.303, as demonstrated by focal perivascular edema and focal perivascular inflammation (Fig. 5A, top). Although there was marked histopathologic heterogeneity observed in H&E brain sections from R.303 and R.304, immunofluorescence (IF) analysis for CD3-expressing T cells in the brain demonstrated a pattern of widespread intraparenchymal cellular infiltration in both animals, which was observed in all the brain regions analyzed, including the frontal lobe, parietal lobe, occipital lobe, temporal lobe, cerebellum, and thalamus. Despite differences in brain histopathology in R.303 and R.304, we did not observe any consistent variation in cytokine or biochemical markers of CRS or neurologic symptoms that were predictive of these differences.

T-cell infiltration into the brain parenchyma, without predilection for specific brain regions, revealed increased CD3 cell infiltration in the cerebellum, parietal lobe, and thalamus in R.304 (day 8 after CAR T-cell infusion) compared with R.303 (day 12 after CAR T-cell infusion) and compared with normal RM brain (Fig. 5B). Notably, the perivascular T-cell infiltration observed in R.304 was analogous to key histopathologic changes in the brain of an adult patient who developed severe CRS and neurotoxicity, characterized by delirium that progressed to coma, who died 13 days following CD19 CAR T-cell infusion secondary to cerebral edema (Fig. 5C; ref. 12). T-cell infiltration of the brain did not appear to be antigen dependent, given the lack of CD20 expression in normal RM brain (Supplementary Fig. S5A) as measured by IF. Moreover, we found that brain T-cell infiltration was comprised of both EGFRt⁺ (CD20 CAR) and EGFRt⁻ (non-CAR) T cells, in similar proportions to those observed in the CSF and in the blood (Fig. 5D). Thus, on day 8 after adoptive transfer of CD20 CAR T cells in R.304, when 80.5% of CD3 cells were EGFRt⁺ in the blood and 97% in the CSF, the brain also contained a similar proportion, 90.2%, EGFRt⁺ T cells (Fig. 5D). Likewise, when analyzed on day 12 after CD20 CAR T-cell transfer in R.303, following CD20 CAR T-cell contraction, there was a predominance of EGFRt⁻ (non-CAR) T cells, with only 3% of CD3 cells remaining that were EGFRt⁺ in the blood, 3.7% in the CSF, and 5.9% EGFRt⁺ T cells in the brain (Fig. 5D). The CD8⁺:CD4⁺ T-cell balance was also interrogated in the brains of R.303 and R.304 and compared with the CD8⁺:CD4⁺ balance in the blood and in the CSF. In R.304, there was a predominance of CD8⁺ EGFRt⁺ T-cell infiltration in the brain (94%), which closely mirrored the blood (92%) and CSF (93%), on day 8. R.303 also showed a predominance of CD8⁺ EGFRt⁺ T cells in the brain (84%), which was slightly higher than the balance in the CSF (72%) and significantly higher than the CD8⁺:CD4⁺ balance observed in the blood (59%). The EGFRt⁻ (non-CAR) T cells infiltrating in the brain were also more highly skewed toward CD8⁺ EGFRt⁻ T cells compared with the peripheral blood in both R.303 [brain (90%), CSF (73%), and blood (45%)] and R.304 [brain (88%), CSF (71%), and blood (39%); Fig. 5D].

We further examined the phenotypic characteristics of EGFRt⁺ and EGFRt⁻ T cells in the CSF (R.302–R.304; n = 3) and the brain (R.304; n = 1) during peak expansion compared with T cells in the blood (normal controls; n = 5), prior to CAR T-cell infusion (d0) or at peak CAR T-cell expansion (days 7 to 8; R.301–R.304; n = 4). We measured T-cell proliferation (by Ki67 expression) and activation (by CD25 and CD69 expression). Both CD4⁺ and CD8⁺ EGFRt⁺ T cells were highly proliferative in the blood when compared with CD4⁺ and CD8⁺ EGFRt⁻ (non-CAR) T cells on days 7 to 8 (CD4⁺: 91.5% ± 2% vs. 20.1% ± 2%, P = 0.003; CD8⁺: 92.5% ± 2.7% vs. 40.9% ± 9.6%, P = 0.01; Fig. 5E) or compared with day 0 or to normal controls (CD4⁺: 24.8% ± 5%, CD8⁺: 10.4% ± 1.3%, day 0; CD4⁺:14.5% ± 3.8%, CD8⁺: 2.8% ± 0.4%, normal controls; Fig. 5E). Substantial levels of Ki67 expression were observed in both CD4⁺ and CD8⁺ EGFRt⁺ (CD20 CAR) and EGFRt⁻ (non-CAR) T cells in the CSF (CD4⁺: 73.3% ± 1.5% and 45.3% ± 12.9%, respectively, P = 0.07; CD8⁺: 84.4% ± 3% and 70.8 ± 6.4, respectively, P = 0.19; Fig. 5E). Although CD25 expression could be measured on both EGFRt⁺ and EGFRt⁻ CD4⁺ and CD8⁺ T cells in the blood, CSF, and brain following CD20 CAR T-cell infusion, no significant differences in CD25 expression levels were observed when compared with day 0 preinfusion and normal control blood samples (Fig. 5E). Similarly, although CD69 expression could be measured in both CD4⁺ and CD8⁺ EGFRt⁺ and EGFRt⁻ T cells in the blood, CSF and brain compartments following CD20 CAR T-cell infusion (CD4⁺: 51.7% ± 8.5%, 66.7% ± 13.2%, 97.6% vs. 18.6% ± 3.4%, 47.1% ± 17.5%, 93.5% EGFRt⁺ vs. EGFRt⁻; CD8⁺: 52.8% ± 7.6%, 66.9% ± 20.2%, 96.3% vs. 34.7% ± 0.9%, 54.7% ± 17.2%, and 95.8% EGFRt⁺ vs. EGFRt⁻), these were not significantly different from CD69 levels in the blood on day 0 or in normal controls (CD4⁺: 26.5% ± 8.1%, 26.5% ± 5.2%; CD8⁺: 34.6% ± 5.7%, 41.7% ± 3.3%; Fig. 5E).

We also examined one of the key integrins involved in T-cell migration through the blood–brain barrier (BBB), the α4β1 integrin (also known as VLA-4), which mediates T-cell adhesion to cerebral endothelial cells through interaction with VCAM-1 (23). We examined changes in α4 and β1 integrin expression in EGFRt⁺ (CD20 CAR) and EGFRt⁻ (non-CAR) T cells to determine if there were any associations between integrin expression and CAR-T expansion and the observed infiltration of these cells into the CNS (Supplementary Fig. S5C). As expected, α4β1 integrin was expressed in CD4⁺ and CD8⁺ T cells in the blood of normal control RMs (CD4⁺: 49.2% ± 3.6%, CD8⁺: 78.5% ± 3.6%) and in the blood of R.301–R.304 on day 0, prior to CAR T-cell infusion (CD4⁺: 48.9% ± 8.4%, CD8⁺: 41.3% ± 8.1%; Supplementary Fig. S5C). During peak CD20 CAR T-cell expansion, α4β1 integrin levels were significantly increased in the blood in both CD4⁺ and CD8⁺ EGFRt⁺ T cells compared with EGFRt⁻ T cells (CD4⁺: 67.3 ± 3.1% vs. 42.9 ± 9.8%, P = 0.00004; CD8⁺: 78.8% ± 2.6% vs. 47.4% ± 2%, P = 0.03; Supplementary Fig. S5C). For CD8⁺ EGFRt⁺ cells, this expression level was similar to that found in normal controls. For CD4⁺ EGFRt⁺ cells, this represented an increase compared with normal controls (P = 0.0009; Supplementary Fig S5C). Although preliminary, these data suggest that upregulation of integrins may facilitate trafficking of T cells into the brain after CAR T-cell infusion.

CAR T cells represent a breakthrough therapy, capable of inducing remission in patients with relapsed or refractory B-cell ALL, non-Hodgkin lymphoma, and chronic lymphocytic leukemia, many of whom have failed all other available therapies (1–4). However, this therapy has also been associated with significant toxicities, including severe CRS and fatal neurotoxicity, the cause of which has not yet been determined (5–7). Here, we describe the first NHP model of CAR T-cell targeting of B cells that also recapitulates CRS and neurotoxicity.

This model has identified two major components of CAR T cell–mediated neurotoxicity that point to a complex interplay between soluble factors and cell-mediated processes: (i) the selective increase of multiple proinflammatory cytokines in the CSF compared with serum during peak CD20 CAR T-cell expansion and (ii) the concurrent development of encephalitis, in which both CAR and non-CAR T cells accumulate in the CSF and the brain. These observations point to a program of cytokine release following CAR T-cell activation, resulting in systemic and CSF inflammation, disruption of the BBB and subsequent CAR and non-CAR T-cell infiltration of the CSF and CNS.

The pathobiology of CAR T cell–associated neurotoxicity remains obscure and is limited by the availability of CSF and tissue samples from patients participating in CAR T-cell trials. However, the hypothesis that neurotoxicity in patients is caused by a unique effector mechanism elicited by CD19 CAR-specific CAR T cells in response to CD19 antigen in the brain (24), based on either tumor seeding or physiologic CD19 expression within the CNS, appears to be less likely based on the following observations from this study: (i) CD20 CAR T cells caused neurotoxicity in the absence of identifiable CD20 expression in the normal RM brain or occult B-cell tumor cells in the CNS; and (ii) both CAR and non-CAR T cells infiltrate the brain. These observations, in addition to lack of neurotoxicity reported in antibody therapies targeting CD20 (rituximab, ofatumumab, and obinutuzumab; ref. 25), are more consistent with an antigen-independent encephalitis during neurotoxicity resulting from CAR and non-CAR T-cell activation outside the CNS. However, although we demonstrated that normal RM brain lacks CD20 expression via IF (Supplementary Fig. S4A), it is possible that more sensitive, molecular approaches may be able to detect this antigen in the brain. This possibility is supported by the recently described CNS lymphatic system (26) and the data from the NIH Blueprint NHP Atlas using molecular probes for CD20 (MS4A1 transcript; http://www.blueprintnhpatlas.org; ref. 27). The role that antigen specificity and/or antigen load plays in neurotoxicity therefore remains an open question and a critical area for future studies.

We demonstrated a disproportionate increase in proinflammatory cytokine and chemokine levels in the CSF compared with serum, during neurotoxicity following CD20 CAR T-cell infusion, which provides clues as to the mechanism underlying neurotoxicity. First, it points to the possibility of selective cytokine accumulation in the CSF, either due to increased influx from the blood to the CSF, or due to decreased efflux from the CSF into the blood. Indeed, in studies of systemic inflammation, increased cytokine influx of TNFα and decreased efflux of prostaglandin E₂ have been identified as key mechanisms that result in cytokine retention in the CSF. Alternatively, or in addition, it suggests that high levels of CSF cytokines could represent cytokine production in situ, either by T cells that have trafficked to the brain or by resident CNS cells, such as microglia, astrocytes, or activated endothelial cells (12, 28, 29). In experimental autoimmune encephalitis (EAE), an animal model for neuroinflammation in multiple sclerosis, activated microglia produce proinflammatory cytokines including IFNγ, TNFα, and IL1β, as well as the chemokine MCP-1 (29). In patients with relapsing remitting multiple sclerosis (MS), CSF levels of IFNγ have been found to be elevated (30). Furthermore, CSF levels of IL6, IL8, IP-10, RANTES, MCP-1, and MIG were also elevated in patients with neuropsychiatric systemic lupus erythematosus (SLE), an inflammatory autoimmune disorder that includes symptoms such as headache, mood disorder, seizure, and psychosis (31). A more recent study of adult patients with neurotoxicity following CD19 CAR T-cell therapy has identified endothelial activation and vascular disruption as key features associated with neurotoxicity and highlighted the important role of brain vascular pericytes, which produce IL6 and VEGF in response to IFNγ (12). The extent to which these different mechanisms drive neurotoxicity awaits further investigation in the NHP model, with the ultimate goal of identifying pharmacologically targetable pathways in order to ameliorate this toxicity.

The complexity of the inflammatory response and neurotoxicity that occurs with CD19 and CD20 CAR T cells is unusual compared with other cellular adoptive therapies, including donor lymphocyte infusions as well as CAR T cells directed at other antigens, such as the nonmature B-cell solid tumor antigen ROR1 (20). However, given the fact that no postmortem examination of the brain was performed in either the previous ROR1 CAR experiments (20) or in the GFP control experiment described here, it remains possible that subclinical CNS abnormalities could occur. As such, further detailed studies of the CNS after infusion of CAR T cells that do not cause overt neurotoxicity would be beneficial to the field. Nevertheless, there is precedence in the literature for neurotoxicity with other immune agonists (including IL2, IFNα, the CD3-CD19 bispecific T-cell engager blinatumomab, as well as the CD28 agonist TGN1412), each of which has been associated with a similarly complex spectrum of neurologic abnormalities (32–34). As with CAR T cells, the precise mechanistic pathways leading to the perturbation of CNS homeostasis with these agents have yet to be fully dissected. Although the pathways are still undetermined, there is also a commonality of each of these immune activation syndromes with what has been termed “sepsis-associated encephalopathy” or “septic encephalopathy,” in which cerebral dysfunction, characterized by delirium or coma, occurs in the setting of sepsis in the absence of direct cerebral infection (35). Although multiple upstream inciting factors exist, the common final pathway involves a significant breach of the integrity of the BBB neurovascular unit, as well as astrocyte dysfunction (36). This BBB breach is clearly evident in RM recipients of CD20 CAR T cells, given the accumulation of both cytokines and T cells in the CNS in these animals. The extent to which systemic cytokine release ignites the breakdown of the BBB is at this point unclear, but there is commonality in the cytokines upregulated during sepsis and those during CAR T cell–mediated CRS (prominently including IL1β, IL6, IL8, and others), suggesting that finding and neutralizing the common cellular source of these cytokines, rather than targeting individual molecules, could be more successful in preventing the breach of the BBB and resultant neurotoxicity (35).

Our data (and that from patients) also provide a potential explanation for the ineffectiveness of currently available therapies, such as IL6R blockade with tocilizumab, to ameliorate neurotoxicity. Treatment with tocilizumab has been shown to improve signs of CRS (reduction in fever and hypotension); however, it has not been shown to prevent or improve neurotoxicity. With our novel NHP model, we now have the opportunity to test alternative strategies to ameliorate neurotoxicity, which include (i) blocking CAR and non-CAR T-cell infiltration into the CNS by the use of natalizumab, an anti-α4 integrin antibody used clinically in patients with MS to block leukocyte infiltration into the CNS, (ii) modulating cytokine production by T cells through CAR engineering approaches, or (iii) depleting the source of CNS cytokine production by targeting activated endothelial cells or microglia.

In summary, the experiments presented here describe the first NHP model of CAR T cell–mediated CRS and neurotoxicity. They point to a complex program of systemic and local neuroinflammation leading to this clinical complication, which includes the accumulation of proinflammatory cytokines in the CSF and significant expansion and activation of both CAR and non-CAR T cells as well as their migration into the CSF and brain parenchyma. This model represents a platform for detailed interrogation into the mechanisms driving neurotoxicity after adoptive cellular therapy and for preclinical testing of therapeutic strategies to eliminate this deadly complication of adoptive immunotherapy.

GFP-χHIV lentivirus was generated using a SIN-HIV-1–based lentiviral vector expressing enhanced GFP-firefly luciferase (EGFP-ffluc), under the control of the human elongation factor 1 alpha promoter (huEF1p). RhCD20-χHIV lentivirus was generated using a codon-optimized Rhesus CD20 (RhCD20) sequence synthesized by Genewiz, based on the coding region of the predicted cDNA sequence of RM CD20 (RefSeq: XM_001086364.1), and insertion, following NheI/NotI restriction enzyme digestion, into a SIN-HIV-1–based lentiviral vector, under the control of the huEF1p. The second-generation 4-1BB:zeta CD20CAR construct encoded a Leu16 (murine anti-human CD20) single-chain variable fragment fused to a human IgG4 C_H2–C_H3 hinge, including two mutations in the C_H2 region (L235D and N297Q, both of which reduce binding to FcγR; ref. 37), a CD28 transmembrane domain, a 4-1BB (CD137) costimulatory domain and a CD3ζ domain. The CD20CAR-4-1BB:zeta construct also contained a transduction and selection marker downstream of a Thoseaasigna virus 2A (T2A) peptide sequence, comprising an EGFRt domain (Supplementary Fig. S1; refs. 16, 17). This expression cassette was incorporated into either a SIN-HIV-1– or a SIV-based lentiviral vector, under the control of the huEF1p or the CMV early enhancer/chicken beta actin (CAG) promoter, respectively, and used to generate either χHIV- or SIV-based lentiviruses (14, 15). GFP, RhCD20, or CD20 CAR-expressing χHIV lentiviruses were generated by four plasmid (the χHIV Gag/Pol plasmid, HIV-1 Rev/Tat plasmid, VSV-G envelope plasmid, and either GFP, RhCD20, or CD20 CAR-expressing SIN-HIV-1–vector plasmid) cotransfections of 293T cells (ATCC), using Lipofectamine 2000 (Invitrogen). The CD20 CAR-expressing SIV lentivirus was similarly generated by four plasmid (the SIV Gag/Pol plasmid, SIV Rev/Tat plasmid, VSV-G envelope plasmid, and the CD20 CAR-expressing SIV-vector plasmid) cotransfections of 293T cells, using Lipofectamine 2000. Lentivirus-containing supernatant was collected 72 to 96 hours after transfection and concentrated by ultracentrifugation at 24,500 rpm for 1 hour 34 minutes at 4°C and the virus was titered using human H9 T cells (ATCC).

Peripheral blood mononuclear cells (PBMC) were isolated from adult RM peripheral blood by standard protocols, using Ficoll-Paque PLUS (GE Healthcare Bio-Sciences). Total T cells were isolated from PBMCs using an NHP pan–T-cell isolation kit per manufacturer's instructions (Miltenyi Biotec; Supplementary Table S3). Polyclonal T cells were activated with NHP anti-CD2/anti-CD3/anti-CD28 stimulation beads at a 1:2 bead to T-cell ratio (Miltenyi Biotec) in X-Vivo 15 medium (Lonza) supplemented with 10% FBS (HyClone), 1% penicillin/streptomycin/l-glutamine (Invitrogen), 50 μmol/L beta-mercaptoethanol (Sigma) and recombinant human IL2 (rhIL2, 50 U/mL; R&D Systems). Lentiviral transduction with spinoculation was performed on day 2 or 3 using polybrene (8 μg/mL; Sigma) and χHIV-GFP, χHIV-RhCD20, χHIV-, or SIV-CD20 CAR lentiviruses (MOI range, 1–5). Stimulation beads were removed between days 6 and 8 of culture. T-cell cultures were supplemented with IL2 (50 U/mL) three times per week. At the end of the stimulation cycle, after 8 to 17 days of culture, T cells were washed in 1× PBS and infused as fresh cells, for R.301, or cryopreserved into cell bank aliquots, for R.302 to R.304, and subsequently thawed, washed in 1× PBS and resuspended in 1× PBS supplemented with 10% autologous plasma immediately prior to infusion. These culture conditions yielded a final product with a ∼2:1 ratio of CD8⁺:CD4⁺T cells (Fig. 1A). In this study, we did not select a 1:1 CD8⁺:CD4⁺ GFP⁺ or EGFRt⁺ T-cell ratio for infusion, as the RM experiments were designed to parallel the original CD19 CAR T-cell clinical trial at our institution that did not use a defined CD8⁺:CD4⁺ T-cell ratio (ClinicalTrials.gov #NCT01683279).

The human K562 chronic myelogenous leukemia cell line was obtained from the European Collection of Cell Cultures through Sigma-Aldrich in 2012. K562 cells were authenticated by STR profiling matched to the DSMZ database by the University of Arizona Genetics Core on October 12, 2015. Dr. Amitinder Kaur produced the RM B-lymphoblastoid cell lines (B-LCL), Mm211.98 (B-LCL1) and Mm309.98 (B-LCL2), by herpesvirus transformation of allogeneic RM B cells acquired in 2014 (38). B-LCLs were not authenticated. The CD19-transgene and CD20-transgene expressing K562 cell lines (CD19-K562 and CD20-K562) were previously described and obtained from Stanley Riddell in 2012 (39). CD19-K562 and CD20-K562 were not authenticated. K562, CD19-K562, and CD20-K562 cell lines were cultured in RPMI (Invitrogen) supplemented with 10% FBS (HyClone) and 1% l-glutamine (Invitrogen). B-LCL1 and B-LCL2 cell lines were cultured in RPMI supplemented with 20% FBS and 1% penicillin/streptomycin/l-glutamine (Invitrogen). All cell lines were negative for Mycoplasma using qPCR-based testing (Sigma Lookout mycoplasma kit or Promokine PCR mycoplasma kit). All cell lines used in these experiments were between passages 1 and 18.

Human target cells K562, CD19-K562, and CD20-K562 and RM target cells B-LCL1 and B-LCL2 were labeled with ⁵¹Cr (PerkinElmer), washed, and incubated in triplicate in 96-well round-bottom plates with human or RM effector T cells (mock-, CD19 CAR-, and CD20 CAR-transduced T cells) at various effector-to-target (E:T) ratios. Supernatants were collected after 4-hour incubation for γ-counting using TopCount NTX (PerkinElmer), and percent specific lysis was calculated as previously described (40).

Immunophenotyping was performed using fluorophore-conjugated mAbs: anti-CD3, anti-CD4, anti-CD8, anti-CD19, anti-CD20, anti-CD25, anti-CD28, anti-CD29, anti-CD49d, anti-CD95, and anti-Ki67 (Supplementary Table S3). Cell surface expression of EGFRt was detected using an APC-conjugated anti-EGFR mAb, cetuximab (Supplementary Table S3). Anti-CD28 and anti-CD95 antibodies were used to assess T-cell phenotypes (central memory, effector memory, and naïve) in the products and longitudinally in vivo (41). Flow cytometric analysis was performed on the LSRFortessa (BD Biosciences), and data were analyzed using FlowJo software (TreeStar).

NHP experiments were performed according to the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, which were approved by the University of Washington Institutional Animal Care and Use Committee. Each recipient received cyclophosphamide (Baxter) as lymphodepleting chemotherapy. R.301 received 30 mg/kg cyclophosphamide/dose administered on 2 consecutive days, starting on day −7 prior to GFP-expressing control cell infusion. Because R.301 tolerated the cyclophosphamide infusion well, and to optimize lymphodepletion and dose similarity to clinical practice, all subsequent CD20 CAR T-cell infusions were performed after 2 doses of 40 mg/kg/dose cyclophosphamide. R.301 received cyclophosphamide on days −6 and −5 prior to CAR T-cell infusion. R.302 and R.304 received cyclophosphamide on days −7 and day −6 prior to CAR T-cell infusion. R.303 received cyclophosphamide on day −4 and day −3 (cyclophosphamide lymphodepletion was delayed by 3 days in R.303 due to pulmonary complications in the recipient on day −7, following central catheter placement, necessitating clinical observation prior to being deemed eligible to safely proceed by the veterinarian on day −4) prior to CAR T-cell infusion. Mesna (Sagent Pharmaceuticals) was administered to each recipient as a bladder protectant. The total mesna dose was equal to the cyclophosphamide dose (30–40 mg/kg/day), divided into 4 doses (7.5–10 mg/kg/dose), administered intravenously (i.v.), 30 minutes prior, and 3, 6, and 8 hours after cyclophosphamide infusion. GFP control (R.301 only), a combination of GFP and CD20 CAR T cells (1:1 ratio, R.302 only) or CD20 CAR T cells (R.303 and R.304) were infused i.v. at doses ranging from 0.6 × 10⁷ to 1.2 × 10⁷ GFP⁺ or EGFRt⁺ T cells/kg. All recipients received antibiotic prophylaxis with ceftazidime and vancomycin, antiviral prophylaxis with acyclovir and antifungal prophylaxis with fluconazole; additional antibiotics included enrofloxazin, administered to R.301–R.303 during neutropenia at the discretion of the veterinarian. Recipients underwent clinical and neurologic monitoring according to the scoring system in Supplementary Table S2. In addition, blood, lymph nodes, and bone marrow aspirates were collected longitudinally in all recipients.

CD20 CAR T-cell engraftment was monitored longitudinally via flow cytometry by detection of EGFRt⁺ T cells in peripheral blood, bone marrow, and lymph node specimens. The accumulation of both EGFRt⁺ (CD20 CAR) and EGFRt⁻ (non-CAR) T cells was also tracked longitudinally in the CSF and in brain specimens obtained at necropsy. For flow-cytometric analysis, the cells were prepared as follows: peripheral blood erythrocytes in whole blood, PBMC, and bone marrow samples were lysed using BD Pharm Lyse solution (BD Biosciences). Following lysis, Fc block was performed (for all CD20 CAR T-cell recipients) in 1× PBS containing 20% autologous RM plasma. Cell surface epitopes were stained with directly labeled antibodies in Fc block followed by cytofix/cytoperm fixation and permeabilization (BD Biosciences) and intracellular staining with directly labeled antibodies. Lymph nodes were dissociated through wire mesh and collected cells were washed in 1× PBS prior to antibody incubation. CSF collections were performed via lumbar or foramen magnum puncture, CSF supernatant was processed for cytokine analysis, and CSF cells were collected via centrifugation for flow-cytometric evaluation.

Blood samples were collected longitudinally before and after CD20 CAR T-cell infusion and analyzed at the University of Washington clinical laboratory for CRP, ferritin, and LDH.

A neurologic monitoring strategy was developed, which allowed us to assess the animal's activity level, movement speed, presence of tremors, gait abnormalities, and seizure before and after CD20 CAR T-cell transfer (Supplementary Table S2). This resulted in a calculated neurotoxicity score. Recipients were scored for their neurologic symptoms twice daily using the scoring system, and raw scores were reported (Supplementary Table S2).

Board-certified veterinary pathologists (J.M. Snyder, A. Baldessari, and H.D. Liggitt) analyzed neuropathologic features in H&E-stained brain sections, and images were acquired using the NIS-Elements Basic Research v3.2 for 64-bit imaging software (Nikon Instruments) and plated in Adobe Photoshop Elements (Adobe Systems). Image brightness and contrast were adjusted using autowhite balance and/or autocontrast corrections that were applied to the entire image.

For CD3 and CD20 IHC peroxidase staining, slides containing brain sections embedded in paraffin were baked and deparaffinized on the Leica Bond Automated Immunostainer (Leica Biosystems). Antigen retrieval was performed using citrate, followed by blocking with normal goat serum (10% in TBS). Slides were incubated with primary antibodies CD3 (clone LN10, 1:250 dilution; Leica Biosystems), CD20 (Clone L26, 1:500 dilution; Leica Biosystems), or mouse IgG control, in Leica Primary Antibody Diluent. The Leica Bond Post Primary was used followed by Leica Bond Polymer. For substrate detection, slides were incubated with the Leica Bond Mixed Refine (DAB) substrate detection kit, followed by hematoxylin (Leica Biosystems) counterstain. Sections were cleared to xylene and mounted with synthetic resin mounting medium and cover slips.

For IF staining, paraffin sections were placed on silanized slides, and slides were deparaffinized in CitriSolv (Thermo Fisher Scientific), followed by 100% ethanol and water. Heat-induced antigen retrieval was performed by placing slides into a decloaking chamber (Biocare Medical). Universal blocking was performed using Background Sniper (Biocare Medical). Slides were incubated with anti-CD3 (Bio-Rad) and anti-smooth muscle actin antibodies and washed in 1× PBS (Gibco). Secondary antibody incubation was performed with goat anti-rat AF 488 and goat anti-mouse IgG2a AF568 (Invitrogen), and slides were washed in 1× PBS followed by water. Hoechst dye (Invitrogen) was used for nuclear counter stain. Slides were mounted using ProLong antifade gold + DAPI mounting medium (Invitrogen), and images were captured using the Nuance microscope (PerkinElmer).

Serum and CSF supernatant samples were stored at −80°C prior to cytokine analysis. Samples were thawed and clarified by centrifugation (16,000 × g at 4°C). Undiluted samples were incubated with Monkey Cytokine Magnetic 29-Plex antibody beads (Invitrogen). For analyte detection, the Biotinylated Detector Antibody was used, followed by Streptavidin-RPE, per manufacturer's instructions. Levels of cytokines and chemokines were measured using the Bioplex-200 system (Bio-Rad).

RM brains were harvested at necropsy after perfusion with 3 to 5 L of room temperature 1× PBS until organ pallor and clear fluid was visually confirmed at the perfusion exit site. A veterinary pathologist (A. Baldessari) identified and sectioned multiple sections of the brain (frontal lobe, parietal lobe, temporal lobe, occipital lobe, cerebellum, hippocampus, basal ganglia, brainstem, choroid plexus, meninges, and thalamus). The brain tissue was enzymatically dissociated using the human tumor dissociation kit (Miltenyi Biotec), washed with 1× PBS and passed through a 70-μm strainer. Debris removal solution (Miltenyi Biotec) was used followed by immunophenotypic analysis on isolated cells using flow cytometry as described above.

Unless otherwise indicated, data are expressed as means ± SEM. Differences between values were examined using the parametric two-tailed unpaired Student t test, and differences were considered to be statistically significant when P < 0.05.

C.J. Turtle reports receiving commercial research grants from Juno Therapeutics and Nektar Therapeutics, has ownership interest (including patents) in Precision Biosciences, and is a consultant/advisory board member for Juno Therapeutics, Precision Biosciences, Celgene, and Gilead. S. Riddell reports receiving a commercial research grant from Juno Therapeutics, has ownership interest (including patents) in Juno Therapeutics, and is a consultant/advisory board member for Juno Therapeutics, Adaptive Biotechnologies, NOHLA Therapeutics, and Cell Medica. L.S. Kean is a consultant for Kymab Ltd., Bristol-Myers Squibb, Equillium, Calimmune, and Alpine Sciences; reports receiving commercial research grants from Bristol-Myers Squibb and Regeneron, and is a consultant/advisory board member for Jazz. M.C. Jensen reports receiving a commercial research grant from Juno Therapeutics, Inc., and is a consultant/advisory board member for the same. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Taraseviciute, A. Yu, S.R. Riddell, R. Gardner, L.S. Kean, M.C. Jensen

Development of methodology: A. Taraseviciute, A. Yu, S.N. Furlan, O. Finney, C. Berger, L.S. Kean, M.C. Jensen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Taraseviciute, C.J. Turtle, H.D. Liggitt, A. Baldessari, C. English, A. Yu, D.J. Hunt, V. Hoglund, H. Brakke, L.S. KeanAnalysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Taraseviciute, V. Tkachev, R. Ponce, H.D. Liggitt, D. Myerson, A. Baldessari, O. Finney, B.R. Blazar, L.S. Kean, M.C. Jensen

Writing, review, and/or revision of the manuscript: A. Taraseviciute, V. Tkachev, C.J. Turtle, J.M. Snyder, H.D. Liggitt, D. Myerson, L. Gonzalez-Cuyar, A. Baldessari, S.N. Furlan, O. Finney, B.R. Blazar, C. Berger, S.R. Riddell, R. Gardner, L.S. Kean, M.C. Jensen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Taraseviciute, V. Tkachev, R. Ponce, J.M. Snyder, H.D. Liggitt, A. Yu, H. Zheng, H. Brakke, L.S. Kean, M.C. Jensen

Study supervision: A. Taraseviciute, R. Gardner, L.S. Kean, M.C. Jensen

Other (pathology): D. Myerson

This work was supported by the NIH (grants P01CA065493, to B.R. Blazar; 1R01HL095791, 2U19AI051731, 1UM1AI126617, U19AI117945, and R33AI116184, to L.S. Kean; and 2R01CA136551, to M.C. Jensen) and the Fred Hutchinson Cancer Research Center/University of Washington Cancer Consortium Cancer Center Support grant (to M.C. Jensen). A. Taraseviciute was supported by a departmental Ruth L. Kirschstein NRSA training grant (T32CA009351). Additional support to A. Taraseviciute was provided by the St. Baldrick's Fellowship, ASCO Conquer Cancer Young Investigator Award, and ASBMT New Investigator Award.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.