Differential Role of Th1 and Th2 Cytokines in Autotoxicity Driven by CD19-Specific Second-Generation Chimeric Antigen Receptor T Cells in a Mouse Model

T cells engrafted with Ab-based chimeric AgRs (CARs) recognize tumor cells through the HLA-independent recognition of cell surface protein Ags (1–3). B cell malignancies are being targeted extensively by CD19-specific CAR T cells, with recent reports of objective clinical responses in chronic lymphocytic leukemia underlying the potential of this therapy (4–8). Toxicity associated with this therapy in patients has been acute, generally as a result of high serum cytokine levels (6, 9). Toxicity resulting from CD19-specific CAR T cell activity has not been reported in preclinical models. However, we report in this study acute and chronic toxicities that were observed in a syngeneic mouse model testing T cells engrafted with mouse CD19-specific CARs. Both toxicities appear likely to be the result of cytokines produced by the CAR T cells; however, chronic toxicity was related to prolonged expression of Th2-type cytokines.

Recent clinical trials targeting B cell tumors used second-generation CARs comprising a CD19-specific single chain variable fragment (scFv) Ab domain fused to a transmembrane region and two intracellular signaling domains; the first is derived from costimulatory receptors, including CD28 or CD137 (4-1BB), and the second is from CD3ζ, a part of the TCR/CD3 complex. The inclusion of the costimulatory receptor domain provides signals that facilitate a full activation of the effector functions of the CAR T cell over and above that of first-generation CARs consisting of the signaling domain of CD3ζ or similar alone (10). These include the upregulation of antiapoptotic genes, IL-2 secretion, and Ag-specific proliferation of CAR T cells (11–13). This approach was shown to lead to greater in vivo expansion and persistence in a small-scale direct comparison in patients (14).

To investigate the potential advantages of second-generation CARs, we sought to compare and contrast the activity of CD19-specific second-generation CARs in a syngeneic model system previously used to examine the functionality of a first-generation CAR. In this previous study (15), mouse T cells bearing a mouse CD19-specific CAR (CD19.ζ) eradicated established A20 B cell lymphoma cells in the immunocompetent BALB/c mouse but failed to persist in vivo for extended periods of time. T cells armed with CD28-based second-generation CARs were tested in this model system; they effectively eradicated established tumor, persisted for extended periods, and induced B cell aplasia. However, a short-term toxicity likely due to high levels of secreted Th1 cytokines was observed with one of the receptors; a longer-term, chronic toxicity, observed in animals who received either of the CD28-containing second-generation CARs, appeared to be driven by Th2-type cytokines and was characterized by a major expansion of CD4⁺CAR⁺ T cells and myeloid-derived suppressor cells (MDSCs). Although these observations were mouse model specific, they support the view that monitoring the Th2 bias of the cytokine-secretion profile of adoptively transferred CAR T cells should be considered in the design of future CAR T cell trials.

Hamster anti-mouse CD3ε (clone 145-2C11) and hamster anti-mouse CD28 (clone 37.51) were obtained from BD Pharmingen (Cowley, U.K.). Human IL-2 (Proleukin) was obtained from Novartis (Camberley, U.K.), and recombinant murine IL-7 was obtained from R&D Systems (Abingdon, U.K.).

The pMP71.tCD34.2A.CD19.ζ and pMP71.tCD34.2A.CD19.mtm retroviral vectors were generated as described previously (15). The carcinoembryonic Ag (CEA)-specific scFv MFE23 was subcloned into the pMP71.tCD34.2A.mtm/pMP71.tCD34.2A.CD3ζ vectors as a ClaI, NotI fragment. pMP71.tCD34.2A.CD19.TOPζ and pMP71.tCD34.2A.CD19.IEVζ retroviral vectors were generated as follows. Full-length human CD28(TOP) (bp 61–660, aa 21–220) and truncated CD28(IEV) (bp 336–660, aa 114–220) sequences (12) were generated by PCR, fused to human CD3ζ using overlapping primers and annealing PCR, and subcloned into pMP71.tCD34.2A.CD19/pMP71.tCD34.2A.MFE as a NotI, HindIII fragment.

High-titer (>10⁶ CFU/ml) GP+e86 producer cell lines were generated by transduction with amphotropic retroviral supernatant generated by transient 293T transfection and sorted for CD34 expression (Miltenyi Biotec, Bisley, U.K.) (10).

Mouse T cells were generated from BALB/c splenocytes, as previously described (15).

All in vivo studies were carried out under the auspices of the Animals (Scientific Procedures) Act 1986 and under United Kingdom Coordinating Committee on Cancer Research guidelines, with the approval of the local ethical review committee. Animals were housed under specific pathogen–free conditions. Six- to eight-week-old BALB/c mice (Harlan, Bicester, U.K.) received 200 mg/kg cyclophosphamide (Cytoxan; Baxter, Newbury, U.K.) 24 h prior to i.v. injection of saline or CAR⁺ T cells at varying doses and time points, as detailed in the figure legends. Heparinized blood from the tail vein was collected at various time points after T cell infusion and analyzed for the presence of CD4⁺, CD8⁺, and CD34⁺ T cells and CD19⁺ cells by flow cytometry. Live cells were identified based upon forward scatter/side scatter (FSC/SSC) profile and gated as described above. The number of cells/ml blood was enumerated by the addition of CountBright beads during the acquisition process, using the assumption that 1 g blood is equivalent to 1 ml blood. Mice were monitored and weighed regularly and culled if weight loss was >20–25% or if they developed symptoms such as swollen, distended abdomens, labored breathing, piloerection, or other signs of ill-health. Organs were removed upon necropsy, photographed, and fixed in 4% formalin, or splenocytes were isolated, and RBCs were lysed (BD Pharm Lyse) and cryopreserved in 90% FCS, 10% DMSO in liquid nitrogen for later analysis by flow cytometry.

A total of 10⁵ T cells (6 × 10⁴ CAR⁺) was cultured with 10⁵ tumor cells (A20) for 24 h at 37°C, and supernatant was analyzed for IFN-γ and IL-2 by ELISA [Diaclone IFNγ Elipairs (Gen-Probe, Manchester, U.K.)] and BD Biosciences IL-2 paired Abs). T cells were labeled with 2 μM CFDA-SE (Molecular Probes, Invitrogen, Paisley, U.K.), and 10⁶ T cells (6 × 10⁵ CAR⁺) were cultured with 5 × 10⁵ A20 cells in complete RPMI 1640 in 24-well plates in triplicate. After 4–5 d, cells were collected and stained with Abs specific for CD4 or CD8 PE-Cy7, CD34 allophycocyanin, or CD19 PE. CountBright beads were added, and samples were acquired on a FACSCalibur (Becton-Dickinson, Cowley, U.K.) and analyzed as described below. The number of CD4⁺CAR⁺ and CD8⁺CAR⁺ T cells that had undergone more than one round of cell division (determined by loss of CFSE dye) were enumerated by flow cytometry with the aid of CountBright counting beads. The fold increase in the number of CAR⁺ cells that had undergone more than one round of cell division was calculated relative to CD19mtm CAR⁺ T cells for each signaling CAR. Cells were gated on the live cells (FSC/SSC profile) and CD4⁺CD34⁺ or CD8⁺CD34⁺. The number of CFSE⁺ and CFSE⁻ (>1 division) cells was calculated.

Linear amplification–mediated PCR was performed as described previously (16), except that pMP vector–specific LTR primers were used for the linear PCR steps (LTR-I 5′-ATCCTGTTTGGCCCATATTCTC-3′ and LTR-II 5′-TCCTAACCTTGATCTGAACTTCTC-3′, each used at 0.25 pmol). PCR products were electrophoresed on a Spreadex EL 1200 mini gel (VH Bio, Gateshead, U.K.). Excised bands were cloned into the TOPO TA cloning vector (Life Technologies, Paisley, U.K.) prior to sequencing. Insertion site sequences were aligned to the murine genome using the National Center for Biotechnology Information BlastN search tool.

Spleens were taken from normal BALB/c mice or mice that received cyclophosphamide preconditioning and CD19.TOPζ T cells. RNA was either isolated directly from splenocytes or, where appropriate, splenocytes from four CD19.TOPζ CAR-treated mice were pooled, stained with CD4-PE, CD34-PE-Cy5, CD11b-allophycocyanin, and Gr-1 FITC, and sorted into CD4⁺CD34⁺, CD11b⁺Gr-1^Hi, CD11b⁺Gr-1^Lo, and CD4⁻CD34⁻CD11b⁻Gr-1⁻ populations. Each sorted population was collected into RNA Save solution prior to RNA isolation using a RNeasy Mini kit (QIAGEN, Hilden, Germany). A total of 500 ng RNA from each sample was used to generate cDNA using a RNA-to-cDNA kit (LifeTech, Warrington, U.K.), and duplicate samples were run on a 96-gene TaqMan Mouse Immune Array (LifeTech) using a 7900HT Real-Time PCR machine. The relative expression of 90 immune-related genes was compared with either normal BALB/c spleen or the CD4⁻CD34⁻CD11b⁻Gr-1⁻ population using SDS2.4 and RQ Manager software (LifeTech). Microarray data have been deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus under accession number GSE54696 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE54696).

Formalin-fixed tissue samples were embedded in paraffin wax and cut into 4-μm sections. For both the CD34 and arginase immune-staining, Ag retrieval was performed on hydrated sections by immersing in 1 l 10 mM citric acid (pH 6) and microwaving at full power (800 W) for 25 min, followed by 15 min of cooling. Endogenous peroxidase was blocked using 0.03% H₂O₂ for 10 min, followed by a nonspecific blocking step using 10% goat serum for 20 min prior to Ab incubation. The CD34 Ab (M7165; Dako, Ely, U.K.) and arginase-1 Ab (610708; BD Transduction Laboratories), or appropriate mouse IgG isotype control were used at a dilution of 3.7 and 2.5 μg/ml, respectively with the Animal Research Kit (K3954; Dako). Incubation with the biotinylated Ab for 15 min was followed by detection with streptavidin peroxidase and 3,3′-diaminobenzidine. Nuclei were counterstained with Gill 1 hematoxylin (Thermo-Electron, Runcorn, U.K.). Ki-67 staining was performed as above, except that the primary Ab (rat anti-mouse Ki-67 [M7249; Dako]) was used at 148 μg/ml, followed by a biotinylated rabbit anti-rat biotin (E0468; Dako) at 1:400. H&E analysis was performed on 4 μm sections using the Leica Autostainer XL and Gill 2 Hematoxylin. Collagen was stained using a mixture of Light green, Ponceau Fuchsin, and phosphomolybdic acid. Slides were scanned on a Zeiss Mirax scanner (internal magnification ×10), and images were acquired using Mirax Viewer Software.

Statistical analysis was performed using GraphPad Prism Software (GraphPad). Data were deemed not to be distributed normally, so nonparametric analyses were performed (Mann–Whitney U test, Kruskal–Wallis, or ANOVA). Survival data were analyzed by Kaplan–Meyer and log-rank tests.

To determine whether second-generation CAR conferred improved T cell persistence and sustained in vivo functionality, two mouse CD19-specific, CD28-containing CARs were tested in the A20 tumor model (15). The CD19.TOPζ CAR consists of the scFv fused to the full-length extracellular, transmembrane and cytoplasmic CD28 receptor coupled to the CD3ζ-signaling domain (Fig. 1A). The CD19.IEVζ receptor contains the scFv fused to a truncated CD28 receptor lacking 113 aa of the extracellular domain fused to the CD3ζ signaling domain (17) and was shown to lack binding to the B7 costimulatory ligands, despite the presence of the MYPPPY motif (18) (Fig. 1A). Mouse T cells engrafted with either the CD19.TOPζ or CD19.IEVζ CAR produced IFN-γ at levels comparable to T cells harboring the CD19.ζ CAR when cultured with A20 B cell lymphoma cells (Fig. 1B) and, with the exception of IL-2 secretion by CD19.IEVζ T cells, comparable IL-2 secretion (Fig. 1C) and proliferation (Fig. 1D), most likely indicative of the high levels of costimulatory ligands present on the A20 B cell line (19). No statistically significant differences in cytokine production were seen between the CD19.IEVζ and CD19.TOPζ CAR constructs; however, when cytokine production was examined in purified T cell subsets, it was apparent that, although both CD4 and CD8 subsets produced IFN-γ, only the CD4 subset secreted IL-2. Using these pure populations, significantly higher levels of IL-2 were produced by CD19.IEVζ CAR T cells than CD19.TOPζ CAR T cells when cultured with A20 tumor cells (Supplemental Fig. 1A). Moreover, cytokine production was independent of MHC class I (Supplemental Fig. 1A).

In a therapy setting, T cells bearing either first- or second-generation CARs effectively eradicated systemic A20 tumors in BALB/c mice, whereas mice treated with nonsignaling CAR T cells (CD19.mtm) or vehicle control succumbed to tumor 30 d post-T cell transfer (Fig. 1E, 1F). However, the six animals receiving T cells bearing the CD19.TOPζ CAR developed swollen, distended abdomens and labored breathing between 50 and 107 d after T cell transfer (Fig. 1G). Upon necropsy, these mice possessed severely enlarged spleens and livers, with multiple white areas visible to the naked eye (Fig. 2A). In contrast, animals that received the CD19.IEVζ or CD19.ζ CAR T cells remained asymptomatic until the end of experiment and, upon necropsy, presented with normal spleen and liver (Fig. 1H).

To investigate this phenomenon further, experiments were performed with batches of T cells transduced to varying levels (24–72%) that were adoptively transferred to cyclophosphamide-preconditioned, tumor-free, BALB/c mice. Of the 44 mice receiving >5 × 10⁶ CD19.TOPζ CAR T cells, 75% (33) were culled 2–4 d after T cell transfer as the result of significant weight loss (20–25%) and further adverse symptoms, including pilo-erect coats, a lack of movement, and hunched appearance (Fig. 2B). Toxicity was not dependent upon cyclophosphamide, because animals preconditioned with 5 Gy total body irradiation prior to the adoptive transfer of T cells engrafted with the CD19.TOPζ CAR also developed this toxicity (data not shown). Cohorts of animals receiving T cell populations with <5 × 10⁶ CD19.TOPζ T cells (range: 8.5 × 10⁵–3.6 × 10⁶) displayed similar, but less severe, symptoms and eventually regained weight and returned to a healthy appearance within 10–15 d (Fig. 2B, 2C). Provision of mash and hydropacks was essential to prevent further weight loss, suggesting that dehydration and anorexia were the primary causes of weight loss. Animals receiving T cells engrafted with all other CARs or saline had reduced weight loss and improved general symptoms (Fig. 2B, 2C).

Such a rapid onset of symptoms strongly implicated a cytokine storm effect. Indeed, 4 h after the adoptive transfer of 9 × 10⁶ CD19.TOPζ T cells, serum IFN-γ and TNF-α were elevated compared with CD19.ζ and control animals (Fig. 2D), suggesting that increased levels of Th1-type cytokines may be the underlying cause of this short-term toxicity.

BALB/c mice that received CD19.TOPζ CAR T cells (3 × 10⁶) showed prolonged B cell depletion consistent with the reported enhanced activity of T cells bearing second-generation CARs (12, 17, 20–22), with systemic B cell depletion seen through day 54 after adoptive transfer. In contrast, B cells returned to normal levels by day 35 in animals receiving CD19.ζ CAR T cells (data not shown). However, starting at 2 mo after adoptive transfer, significant numbers of animals infused with CD19.TOPζ (8.5 × 10⁵–5 × 10⁶) or CD19.IEVζ (2 × 10⁶–10⁷) CAR T cells developed swollen, distended abdomens and labored breathing, resulting in a reduced overall survival compared with saline control– or CD19ζ T cell–treated animals (median survival: CD19.TOPζ, 95 d; CD19.IEVζ, 170 d; saline and CD19.ζ > 220 d, Fig. 2E). This difference in survival was CD19-scFv specific, because animals receiving the anti-CEA version of each receptor (MFE.IEVζ, 3 × 10⁶; MFE.TOPζ, 3-3.1 × 10⁶) did not demonstrate similarly reduced survival (Fig. 2F).

Upon necropsy, 66% (18 of 27; Table I) of CD19.TOPζ CAR T cell–treated animals had severely enlarged spleens and livers, with multiple white areas visible to the naked eye (Fig. 2A). In some instances, mice also presented with enlarged lymph nodes (including mesenteric and mediastinal lymph nodes), whereas all other organs appeared normal (data not shown). Similarly, 50% (14 of 28; Table I) of CD19.IEVζ-treated animals displayed a similar abnormal splenic and liver architecture (Supplemental Fig. 1B–D). In contrast, 0 of 14 animals receiving CD19.ζ CAR T cells had abnormal spleen, liver, or lymph nodes (data not shown). Furthermore, although CEA-specific second-generation CAR T cells did not appear to impact upon overall survival, 20% (3 of 15) of animals receiving MFE.TOPζ CAR T cells and 6% (1 of 17) receiving MFE.IEVζ CAR T cells presented with abnormal spleen and liver, suggesting that nontargeted second-generation CARs also possessed the potential to drive this chronic toxicity, albeit with much delayed kinetics.

Analysis of splenocytes from the animals described above at the time of cull revealed a difference in the relative frequency of CAR T cells between cohorts. Animals receiving either CD19.TOPζ or CD19.IEVζ CAR T cells had significantly greater frequencies of CD4⁺CAR⁺ T cells compared with animals receiving T cells bearing the CD19.ζ CAR or any of the CEA-specific CAR (Fig. 3A), whereas very few CD4⁺CAR⁻ T cells were present in the spleens of animals receiving CD19.TOPζ CAR T cells (Fig. 3B). In contrast, there was little difference in the frequency of CD8⁺ T cells, whether transduced or not (Fig. 3C, 3D). A molecular analysis of the engrafted CAR⁺ T cells did not reveal a restricted range of retroviral vector–integration sites (Supplemental Fig. 1E), nor could the CAR⁺ T cells thrive in culture without the addition of cytokines (data not shown), suggesting that retroviral vector insertion and transformation were not major causes of CAR⁺ T cell proliferation in this model.

CD19⁺B220⁺ B cell depletion was profound in animals receiving either CD19.TOPζ or CD19.IEVζ CAR T cells (Fig. 3E). However, a notable feature was the significantly increased frequency of CD11b⁺Gr-1⁺ cells in the splenocytes of CD19.TOPζ CAR–treated animals compared with all other groups (Fig. 3F). Relative increases in the frequency of this cellular compartment were present in animals treated with CD19.IEVζ CAR T cells and, to a lesser extent, in animals receiving MFE.TOPζ CAR T cells (Fig. 3F). The slightly reduced frequency of CD11b⁺Gr-1⁺ cells in these mice compared with CD19.TOPζ CAR–treated animals may reflect the different times at which the animals were culled due to the onset of symptoms (CD19.TOPζ, 20.1 ± 8.1 d; CD19.IEVζ, 164.3 ± 64.2 d; MFE.TOPζ, 189.1 ± 35.7; all other groups > 200 d).

Immunohistochemical analysis of spleens taken from animals receiving CD19.IEVζ CAR T cells demonstrated extensive infiltration of CAR⁺ T cells within the white pulp (Fig. 4A, Supplemental Fig. 2A), with evidence of collagen deposition (Supplemental Fig. 2B). Enlarged lymph nodes (particularly the mesenteric and mediastinal nodes; Fig. 4B) and livers (Fig. 4C, Supplemental Fig. 2G) were also heavily infiltrated with CAR⁺ T cells. Cells within the areas surrounding the CAR⁺ T cell regions in both the spleens and lymph nodes were arginase-1⁺ (Fig. 4A, 4B, Supplemental Fig. 2E) and, in some instances, large numbers of arginase-1–producing multinucleated Langerhans-type cells were seen surrounding areas of CAR⁺ T cells (Fig. 4A). A similar phenotype was seen in mice receiving CD19.TOPζ CAR T cells (Supplemental Fig. 2). Extensive Ki-67 staining was seen in the red pulp areas and germinal centers of spleens of CD19.ζ-treated mice (Supplemental Fig. 3D), as well as the red pulp areas of the abnormal spleens from CD19.IEVζ-treated mice (Supplemental Fig. 3B) and CD19.TOPζ-treated mice (data not shown). However, only limited Ki-67 staining was apparent in CAR⁺ T cell areas, suggesting that the CAR⁺ T cells were not undergoing rapid proliferation at the time of necropsy. Spleens from animals receiving other CAR T cells essentially appeared normal (Supplemental Fig. 2).

Within 144 h of receiving 2.9 × 10⁶ CD19.TOPζ CAR T cells, CD4⁺ and CD8⁺ CAR T cell engraftment was evident in recipient cyclophosphamide-preconditioned BALB/c mice (Fig. 5A, 5B), by which point spleens were enlarged, with abnormal white areas visible to the naked eye (Fig. 5L, 5M), and abnormal architecture was observed by immunohistochemistry (data not shown). A concomitant decrease in CD19⁺B220⁺ B cells was also evident (Fig. 5C). Interestingly, 4 h after T cell transfer and preconditioning, the relative levels of CD11b⁺Gr-1⁺ myeloid cells increased before decreasing rapidly by 48 h and increasing steadily to the 2-wk end point (Fig. 5D). Systemic IFN-γ and TNF-α cytokine levels increased shortly after adoptive CAR T cell transfer, yet they decreased within 48 h (Fig. 5E, 5F). Conversely, IL-4 and IL-5 levels peaked at ∼144 h (Fig. 5G, 5H), whereas systemic levels of IL-6 remained consistently higher than baseline after T cell transfer (Fig. 5I). Interestingly, both IL-10 and IL-13 showed relative increases in cytokine levels during the 2-wk period (Fig. 5J, 5K). By day 14, all cytokines, with the exception of TNF-α, were at significantly higher levels in the serum of CD19.TOPζ T cell–treated mice than in saline controls (Supplemental Fig. 3E).

Furthermore, comparison of immune-related gene signatures between an abnormal spleen culled 56 d after transfer of CD19.TOPζ T cells and a normal BALB/c spleen showed significant upregulation of Th2-type cytokine genes (Fig. 6A). To elucidate the cells responsible for driving the production of cytokines at these time points, 35–55 d after the adoptive transfer of 3.2 × 10⁶ CD19.TOPζ T cells, pooled splenocytes from four mice were sorted to enrich CD4⁺CAR⁺ T cells, CD11b⁺Gr-1^Hi cells, and CD11b⁺Gr-1^Lo cells (Fig. 6B). mRNA also was isolated from cells that were negative for each of these markers, and each cell population was analyzed for its cytokine gene profile, with the relative gene expression of each population determined against this quadruple-negative CD4⁻CAR⁻CD11b⁻Gr-1⁻ cell population. From this analysis, it was clear that mRNA encoding for the cytokines IL-2, IL-4, IL-5, IL-9, IL-10, and IL-13 was restricted to the CD4⁺CAR⁺ T cell population, indicating that these cells, and not the myeloid cell compartment, were the source of Th2 cytokines in the spleens of affected animals (Fig. 6C). The presence of CD4⁺ and other relevant T cell markers only in the CD4⁺CAR⁺ cell population (Fig. 6D) confirmed the phenotype of the sorted CAR cell population.

The enhanced in vitro and in vivo functionality of T cells engrafted with second-generation CARs strongly supports their development over early first-generation CARs. In this study, the in vivo antitumor activity of T cells bearing first- and second-generation CARs appeared equivalent. However, prolonged B cell depletion was only associated with second-generation CAR-expressing T cells, indicative of the enhanced functionality of this class of CARs.

In the BALB/c model, the adoptive transfer of T cells bearing second-generation CD28-containing CARs into cyclophosphamide-preconditioned mice drove a short- and long-term toxicity that appeared to be related, in part, to the targeting moiety and, in part, to the CD28 extracellular domain. Mice receiving T cells bearing the CAR with full-length CD28 (CD19.TOPζ) developed an acute toxicity (at doses > 2.5 × 10⁸/kg) and a chronic toxicity, whereas mice receiving T cells bearing a CAR with a truncated extracellular domain of CD28, which lacks B7 ligand binding potential and is currently in trial (7) (CD19.IEVζ), only developed chronic toxicity. These observations suggest that the potential binding of CD28 to its natural ligands may play a role in the severity of the observed short-term toxicity but appeared less likely to be a major driver of the intensity of late toxicity; although limited long-term toxicity seen with control CEA-binding CAR suggests some role for CD28 ligand binding. However, it should be noted that both CD19.IEVζ and CD19.TOPζ CAR T cells can bind to natural B7-1/B7-2 ligands on normal B cells through native CD28 expressed on the T cell surface. This suggests that CAR binding to B7 ligands may deliver a costimulatory signal of increased potency compared with that achieved with native CD28-B7 binding, potentially explaining the increased toxicity seen with the CD19.TOPζ CAR and also suggests that B7 target cell expression alone is not the sole driver of this toxicity.

Serum cytokine analysis shortly after CD19.TOPζ CAR T cell transfer confirmed that increased levels of IFN-γ and TNF-α were present compared with the levels seen after first-generation CD19.ζ CAR T cell adoptive transfer. Increased levels of both cytokines were reported in four of eight patients with advanced B cell malignancy receiving second-generation CD28-containing CAR T cells who developed acute toxicities within 8 d of CAR transfer (6). These observations in patients support the view that the short-term toxicity observed in this model with CD19.TOPζ CAR T cells is related to the production of IFN-γ and TNF-α. Interestingly, there was no obvious elevation of IL-6 in animals treated with CD19.TOPζ T cells. Elevated IL-6 serum levels were found in patients receiving T cells bearing a CD19-specific 4-1BBζ CAR, with cytokine storm symptoms alleviated by treatment with an inhibitor of IL-6 (23). Moreover, IL-6 was implicated in a cytokine-release syndrome associated with CD19/CD3 bispecific T cell engaging (BiTE) Ab treatment in B-precursor acute lymphoblastic leukemia (24). The absence of high levels of IL-6 in this model suggests that the T cells armed with CD28-based CARs fail to induce significant IL-6. Clearly, investigating whether T cells bearing a 4-1BB second-generation CAR induce high levels of IL-6 and, thereby, resemble the clinical situation is warranted. Moreover, the adoptive transfer of T cells bearing the CD19.IEVζ CAR did not result in the same severity of short-term toxicity as observed with the CD19.TOPζ CAR. However, a significant number of animals developed long-term toxicity that presented in an essentially identical manner to that observed in animals receiving T cells bearing the CD19.TOPζ CAR.

A key pathological feature associated with the late toxicity was the observation of granuloma-like features predominantly present within spleen and liver of affected animals. Despite the relatively high frequency of CD4⁺CAR⁺ T cells resident within these organs, there was little direct evidence of mutagenic-driven proliferation of these T cells, suggesting that the granulomas were not a direct effect of CAR⁺ T cell hyperproliferation. The parallel expansion of CD11b⁺Gr-1⁺ and arginase-1⁺–expressing cell populations was suggestive that these cells were a population of MDSCs. MDSCs are potent suppressors of T cell activity (25), and in a rat kidney allograft model, anti-CD28 treatment drives the expansion of MDSCs (26). The expression of arginase-1 in mouse macrophages and MDSCs depletes extracellular l-arginine, arresting the proliferation of activated T cells (reviewed in Ref. 27). There is evidence to suggest that the depletion of extracellular arginine has a greater impact on CD8⁺ T cells than on CD4⁺ T cells (28), potentially reflecting the absence of CD8⁺ T cells and the low level of T cell–specific proliferation identified by Ki-67 staining within mice treated with either CD19.TOPζ or CD19.IEVζ CAR T cells.

Interestingly, there are strong parallels between the granuloma formation observed in CAR⁺ T cell–treated BALB/c mice and granuloma formation during immune responses against the eggs of the helminth parasite Schistosoma mansoni. In this system, the development of granulomas is linked strongly to the presence of Th2-type (IL-4, IL-10, IL-13) cytokines (29–31). Moreover, Th2-type cytokines are known to induce arginase-1 enzyme activity as a key part of the downregulation of cellular immune responses (32), and, in the S. mansoni egg model, Th2 cytokines play a pivotal role in driving granuloma formation through the modulation of arginine metabolism via the inhibition of NO synthase-2 and promotion of arginase-1 enzyme activity (29). In the CD28-containing CAR⁺ T cell–treated animals, increasing levels of IL-10 and IL-13 cytokines at later time points after adoptive transfer and high frequencies of arginase-1⁺ cells present within the spleens and liver of animals affected by late toxicity strongly implicated Th2 cytokines in these CAR T cell–driven granuloma formation, with mRNA analysis strongly implicating CD4⁺ CAR⁺ T cells as the source of Th2 cytokines.

The Th2-cytokine bias appears to be critical for development of CD19-specific CAR pathology. Although the adoptive transfer of CD19.IEVζ CAR T cells resulted in a high frequency of long-term toxicity in BALB/c mice [Th2 bias (33–35)], no evidence of similar toxicity was apparent in 21 C57bl/6 or C3H mice [Th1 bias (33–35)] receiving CD19.IEVζ T cells (data not shown). This is consistent with recent studies (36) that observed no toxicity in a syngeneic model in which C3H mice received CAR T cells expressing the murine equivalent of the CD19.IEVζ CAR. A Th2 bias in humans was associated with aging (37), zinc deficiency (38, 39), allergies (40, 41), and lymphoma (42), and polymorphisms in Th2 genes, including the IL-10 promoter, were linked with an increased risk for developing non-Hodgkin lymphoma (43, 44), suggesting that there are subgroups of people who may be more at risk for Th-2 cytokine–driven CAR T cell proliferation than others.

Moreover, there was no apparent effect of different preconditioning regimens (total body irradiation against cyclophosphamide; Supplemental Fig. 4A), with CD19.TOPζ suggesting that the observed pathology was not dependent upon a specific preconditioning regimen. There also was no apparent correlation between T cell dose and survival, suggesting that dose alone is unlikely to be an effective predictor of toxicity onset in this model (Supplemental Fig. 4B, 4C).

To summarize, the adoptive transfer of T cells bearing CD19-specific, CD28-containing CARs to BALB/c mice can result in rapid weight loss shortly after transfer, potentially caused by elevated Th1-biased cytokine levels. For animals that survive this weight loss, the continual stimulation of CAR⁺ T cells by B cells emerging from the bone marrow, combined with proliferative and antiapoptotic signals mediated by the CD28 cytoplasmic domain, leads to a massive expansion of the CD4⁺CAR⁺ population and prolonged production of Th2 cytokines. These cytokines induce arginase-1 enzyme expression in an expanded myeloid cell compartment, resulting in long-term granuloma formation, splenomegaly, lymph node enlargement, and disruption of normal liver architecture, necessitating sacrifice of the animal. The observation of chronic toxicity in this CD19 model has implications for other CAR T cell therapies in which the target Ag of choice is expressed on normal cells that can be regenerated from stem cells after CAR T cell depletion. Ags that are expressed on cancer cells and normal cells that are dispensable for survival (e.g., normal B cells) are attractive targets for CAR T cell therapy, but chronic stimulation of CAR T cells by emerging cells from the bone marrow suggests that additional safety measures, such as suicide genes, may be necessary.

Although high levels of cytokines have been implicated as mediators of short-term toxicity in patients receiving T cells bearing a CD19-specific second-generation CAR (6, 9, 23), and rapid onset hepato-splenomegaly and macrophage activation syndrome were reported in a CD19 CAR T cell trial (9), no long-term toxicity resembling that described in this model study has been reported. This likely represents that fact that these studies were performed using in-bred animals with polarized immune-response profiles compared with clinical studies focusing upon the broad outbred patient population. However, in the situation in which a patient receiving CAR T cells displays a prolonged, polarized Th2 cytokine response after adoptive transfer, this would justify the importance of including control systems into the design of the expression vector, such as suicide gene technology to specifically deplete the hyperactive CAR T cell (45–47). Importantly, this model provides a system that could be used to investigate the potency of such suicide gene systems to deplete CAR T cells and to assess whether such interventions can be given when symptoms first appear to control and reverse the symptoms of late toxicity.

We thank the staff at the Biological Resources Unit, Cancer Research UK Manchester Institute, for assistance with in vivo experiments. We also thank the service units at the Cancer Research UK Manchester Institute, in particular Caron Abbey (Histology), Achille Dunne (Advanced Imaging Facility), the staff of the Molecular Biology Core Facility, and the Flow Cytometry Service. We thank Prof. Doug Fearon (University of Cambridge, Cambridge, U.K.) for providing the 1D3 hybridoma and Hayley Batha, Allison O’Neill, and Jennifer Loconto (Clinical and Experimental Immunotherapy Group) for experimental assistance.

R.E.H. and D.E.G. are cofounders of Cellular Therapeutics Ltd. All other authors have no financial conflicts of interest to disclose.