Development and characterization of a low-affinity humanized CD19 chimeric antigen receptor for B-cell malignancies

The adoptive transfer of CD19-targeted chimeric antigen receptor (CD19-CAR) T cells to treat patients with B-cell malignancies has resulted in impressive clinical response. Currently, 4 anti–CD19-CAR T cell products have been approved by the US Food and Drug Administration for clinical use to treat patients with B-cell malignancies.^1-3 These products are second-generation CARs containing an anti-CD19 single-chain variable fragment (scFv) derived from a murine monoclonal antibody (FMC63). Despite the initial success of various CD19-CAR T cell products, 50% of patients with acute lymphoblastic leukemia (ALL) relapsed within a year.^4,5 One mechanism that contributes to decreased CAR T cell persistence after infusion is the development of humoral and/or cellular immune responses against the CAR. After infusion, a patient’s immune system may mount an anti-CAR response against the mouse-derived scFv and remove the CAR T cells from the patient’s body,^6,7 increasing the likelihood of tumor relapse. To overcome these problems, several investigators have redesigned the CD19-specific scFv to decrease the anti-CAR immune response. Thus far, 4 humanized CD19-CAR constructs have been developed. Clinical trials assessing these humanized CD19-CAR T cells demonstrated safety and efficacy in treating ALL and non-Hodgkin lymphoma (NHL).^8-12 However, the properties of the new humanized scFvs such as affinity, stability, the spectrum of targeting, and toxicity in comparison with the most successful FMC63 CAR have not been fully studied. Our laboratory at City of Hope previously developed an FMC63-based, CD19-CAR T cell platform. Across several clinical trials, this CD19-CAR T cell platform is safe and effective against NHL and ALL.^13-15 However, the durability of remission remains a challenge. As part of further translational efforts with this CD19-CAR T cell platform, we undertook humanization of the FMC63 scFv to create a clinically viable humanized CAR. This CAR is anticipated to perform on par with the FMC63 CAR while mitigating potential anti-mouse immunogenic responses and enhancing persistence in patients with NHL or ALL. After humanization, we characterized the stability and binding affinity of our humanized scFv. We performed a preclinical evaluation of 2 huCD19 CAR candidates and found that the lead candidate, huCD19R (VH4vk1), had equivalent in vitro effector function and in vivo antitumor activity to that of the parental CD19-CAR against tumors with CD19 expression varying on a clinically relevant range.

Raji wild-type (WT) cells (CCL-86), Nalm6 (CRL-3273), and KG-1a (CCL-246.1) were purchased from American Type Culture Collection. The CD19 gene was edited by CRISPR–associated protein 9 technology on Raji-WT cells. Raji CD19 knockout (KO) cells were established by fluorescence-activated cell sorting the CD19⁻ population. The low-antigen cell line was produced by expressing CD19 using a weak 3-phosphoglycerate kinase (PGK) promoter (PGK100p) on Raji KO cells by lentiviral transduction. To achieve clonal similarity, these low-antigen–expressing cells were seeded at a density of 1 cell per well. Four single-cell clones were examined for CD19 expression by flow cytometry and clone 4 was used as the Raji CD19-low cell line for the follow-up experiments. Raji/Nalm6 cells were transduced with a lentiviral vector encoding an enhanced green fluorescent protein and firefly luciferase. OKT3-expressing LCL (OKT3-LCL) cells were generated as previously described.^16,17 

CD3 (BD, Clone UCHT1), epidermal growth factor receptor (EGFR; Biolegend, Clone AY13), CD8 (BD, Clone SK1), CD4 (BD, Clone SK3), CD19 (Beckman Coulter, Clone J3-119) and (BD, Clone SJ25C1), CD27 (BD, Clone M-T271), CD80 (BD, Clone L307.4), CD86 (BD, Clone 2331), CD107a (BD, Clone H4A3), interferon gamma (IFN-γ; BD, Clone B27), tumor necrosis factor α (BD, Clone MAb11), CD45RA (BD, Clone HI100), CCR7 (Biolegend, Clone G043H7), CD28 (Biolegend, Clone CD28.2), programmed cell death protein 1 (Invitrogen, Clone MIH4), T-cell immunoglobulin and mucin-domain containing-3 (TIM3; Invitrogen, Clone F38-2E2), lymphocyte-activation gene 3 (LAG-3; Invitrogen, Clone 3DS223H), and programmed death ligand 1 (PD-L1; Invitrogen, Clone MIH1). A Quantum Simply Cellular kit (Bangs Laboratories, 815) was used to determine CD19 expression using an anti-CD19 antibody (Beckman Coulter, Clone J3-119) following the manufacturer’s protocol. Briefly, beads with 4 different levels of immobilized anti-mouse antibody and negative control beads were labeled using anti-CD19 antibody. Fluorescence was assessed using flow cytometry, and median fluorescence intensity for each bead population was used to generate a standard curve using a lot-specific quantification spreadsheet from the manufacturer. Cell lines were labeled using the same anti-CD19 antibody and analyzed by flow cytometry. Median fluorescence was correlated to CD19 expression through the generated standard curve. A similar protocol was followed for measuring CD80/86, and PD-L1 density using appropriate antibodies. Data acquisition was performed on a MACSquant (Miltenyi) and analyzed using FCS Express version 7 software.

The FMC63 scFv was humanized using 2 separate complementary determining region (CDR) grafting approaches: (1) CDR grafting onto a stable variable region framework,¹⁸ and (2) structure-guided homology modeling onto a closely related human variable region framework.¹⁹ For the first approach, CDRs H1, H3, L1, L2, and L3 were grafted onto the 4D5 framework, with CDR H2 containing the first 10 residues from FMC63 and the final 6 residues left as the sequence native to that framework. In the second approach, a homology model for the FMC63 variable heavy and variable light chains was built using the PIGS server for later structural comparison. Next, the closest human homologs to the heavy and light chains were identified using the IMGT DomainGapAlign tool, yielding IGHVR-30-4∗01 and IGKV1-33∗01 frameworks with the highest homology to the parental murine domains. The FMC63 CDRs were then grafted onto these human frameworks using TabHu,²⁰ with 2 back mutations made in the light chain based on the probability of contacting antigen. The resulting homology model of huCD19R(VH4vκ1) was then subjected to energy minimization using the GROMOS96 force field in Swiss PDB Viewer, demonstrating that no further back mutations were necessary to correct steric clashes or voids in the humanized framework.

Yeast culture and display were performed using previously described methods.²¹ Plasmids encoding for the FMC63 scFv or the humanized variants fused N-terminally to agglutinin 2 and a polypeptide linker with flanking HA (amino acid sequence of N-Tyr-Pro-Tyr-Asp-Val-Pro-Asp-Tyr-Ala-C) and c-Myc epitope tags were transformed into EBY100 Saccharomyces cerevisiae yeast using a Frozen EZ-Yeast Transformation II kit following the manufacturer’s protocol (Zymo Research). Yeast harboring plasmids of interest were grown to an optical density between 0.5 and 1 in SD-CAA medium (synthetic defined medium that is used to grow yeast cells) at 30°C with shaking. Protein expression was induced by media change to SRG-CAA medium followed by overnight growth (8-24 hours) at 30°C with shaking.

Yeast displaying scFvs of interest was resuspended in PBSA (phosphate-buffered saline containing 0.1% bovine serum albumin) containing defined concentrations of biotinylated recombinant CD19 extracellular domain. Yeast displaying the FMC63 scFv was incubated with 0, 0.56, 1.67, 5, 15, 45, and 135 nM CD19. Yeast displaying humanized variants were incubated with 0, 27, 81, 243, and 500 nM CD19. After incubation, yeast was resuspended with goat anti–c-Myc fluorescein isothiocyanate antibody (Bethyl Laboratories, Inc) and streptavidin Alexa Fluor-647 conjugate. Fluorescence was analyzed with a flow cytometer. Median fluorescence in the streptavidin Alexa Fluor-647 channel was fit to a sigmoidal curve with fluorescence minimum and maximum calculated by the fit.

Yeast displaying scFvs of interest were heated to defined temperatures (room temperature [RT] or 40°C-90°C) for 30 minutes, followed by 10-minute incubation on ice. Yeast cells were then resuspended in 20 μL PBSA containing a fold-specific binding reagent (50 nM recombinant biotinylated CD19 for FMC63, biotinylated protein L for humanized variants) for 20 minutes at RT. Yeast were then incubated with goat anti–c-Myc fluorescein isothiocyanate antibody and streptavidin Alexa Fluor-647 for 20 minutes at RT and analyzed with a flow cytometer. Median fluorescence in the streptavidin Alexa Fluor-647 channel was fit to a 2-state unfolding model with maximum fluorescence calculated from the RT sample and minimum fluorescence calculated using a secondary reagent only control.

The variable fragment heavy chain (V_H) and variable fragment light chain (V_L) gene segments of the CD19-specific FMC63 or humanized CD19 scFvs (huCD19R VH4vκ1 or 4D5) were cloned into a second-generation CAR lentiviral vector consisting of immunoglobulin G4 hinge with CH3 connected by a 10-amino-acid glycine and serine linker (ch2Δ), CD4 transmembrane domain, cytoplasmic 4-1BB costimulatory domain, and cytoplasmic CD3ζ stimulatory domain. A non-signaling truncated epidermal growth factor receptor (EGFRt) is separated from the CAR by a T2A ribosome skip sequence for tracking transduction efficiency. T-naïve and -memory (Tn/mem) cells defined as CD62L⁺CD14⁻CD25⁻ were used for CAR engineering as in our clinical platform.¹³ 

Murine and humanized CD19 CAR T cells were cocultured with tumor targets at an effector-to-target (E:T) ratio of 2:1 for 5 hours followed by CD107a flow cytometry analysis. CAR T cells at an E:T ratio of 1:1 were cocultured overnight followed by IFN-γ cytokine staining. Raji-WT tumor rechallenge assay was performed using an E:T ratio of 2:1. Fold expansion of effector CAR T cells followed by each rechallenge was analyzed compared with day 0. At the end of the assay on day 12, exhaustion markers were analyzed by flow cytometry.

All animal experiments were performed under protocols approved by the City of Hope institutional animal care and use committee. Six- to 10-week-old immunodeficient NOD/Scid IL2RᵞC^null (NSG)mice were used for in vivo experiments. Mice were given 0.5 × 10⁶ Raji/Nalm6 WT or Raji CD19-low tumor cells intravenously. For initial sorting, mice were imaged in random groups using the same scale for luminescence followed by sorting based on average flux signal to limit the standard deviation between each group. A total of 1 × 10⁶ CAR T cells were intravenously injected 5 days after tumor engraftment. Untreated and mock groups were used as controls. Firefly luciferase–positive tumor growth was monitored by bioluminescence imaging using Lago X-spectral Instruments once every week. Mice were monitored daily for survival analysis.

Cytokine analysis was carried out for macrophage activation as previously described.²² Briefly, CD14⁺/CD16⁺ monocytes were enriched from healthy donor peripheral blood mononuclear cells using a STEMCELL enrichment kit (19058) and differentiated to macrophages by treating with 5 ng/mL human granulocyte-macrophage colony-stimulating factor (GM-CSF; Biolegend, 572903) for 7 days. Meanwhile, NSG mice were engrafted with 0.5 × 10⁶ Raji WT followed by treatment with 1 × 10⁶ CAR T cells (n = 4). Three days later, serum was collected from these mice and either frozen for baseline cytokines or directly added to the donor-matched activated macrophages in culture; 48 hours after serum addition, the supernatant was collected from macrophages and cytokine analysis was carried out by Luminex cytokine assay.

Major histocompatibility complex 1 (MHC-1) binding predictions for FMC63, huCD19R(VH4Vκ1), and huCD19R(4D5) were made on 26 January 2024 using the Immune Epitope Database (IEDB) analysis resource Consensus tool,²³ which combines predictions from ANN, also known as NetMHC (4.0);^24-26 SMM;²⁷ and Comblib.²⁸ The analyses involved inputting the complete scFv sequences (including linkers) individually and predicting peptide epitopes of any length likely to be presented in HLA-A∗02:01, HLA-A∗01:01, HLA-A∗03:01, and HLA-A∗24:02, 4 abundant MHC alleles.²⁹ From these predictions, the number of peptides with positive total prediction scores, MHC 50% inhibitory concentration (IC50) of <10 nM, and MHC IC50 of <100 nM were quantified. The top 10 peptide epitopes for each scFv, as determined by total prediction scores, were mapped in each scFv to determine which regions were likely to be the most immunogenic.

Data are presented as mean ± standard error of the mean. The number of samples in figure legends represent biological replicates of different healthy donor-derived CAR T cells. All in vitro assays were performed using 2 technical replicates for each sample. In vitro, functional assays were analyzed using the nonparametric Kruskal-Wallis test followed by the Mann-Whitney U test. In vivo assays were carried out using 5 to 10 technical replicates per group. Kaplan-Meier survival analysis for in vivo experiments was carried out using the log-rank (Mantel-Cox) test. All data are analyzed using GraphPad Prism 8. ns, P > .05; ∗P < .05; ∗∗P < .01; and ∗∗∗P < .001.

Ethics approval and consent to participate: all human tissues and blood samples were obtained under protocols approved by the institutional review board (University Health Network Research Ethics Board). Surgical specimens were obtained from the UHN (University Health Network)Biospecimen Program. Written informed consent was obtained from all donors.

Antibody humanization often involves grafting CDRs from an antibody of nonhuman origin onto human variable region frameworks, with design iteration needed to preserve CDR geometries that yield functional binding. We used 2 established protein engineering approaches to humanize the FMC63 variable regions (Figure 1A). In 1 approach, we used CDR grafting onto a stable antibody framework known to be capable of accepting a variety of nonnative CDRs,¹⁸ yielding the huCD19R(4D5) clone. In a second approach, we used structure-guided homology modeling to graft the FMC63 CDRs onto the closest related human framework^19,20 in an attempt to keep the CDRs in their most native-like molecular environment, yielding the huCD19R(VH4vκ1) clone. The parental murine FMC63 and the humanized variants were displayed on the surface of the yeast (Figure 1B). scFv expression was assessed by flow cytometry using an anti–c-Myc antibody, demonstrating robust display (Figure 1C). Affinity titration of yeast displayed scFvs with recombinant CD19 extracellular domain demonstrated remarkable affinity reduction (∼2000-fold and 10 000-fold for huCD19R(VH4vκ1) [K_D at least 8.3 μM] and huCD19R(4D5) [K_D at least 45 μM], respectively) of the humanized variants relative to parental (K_D = 4.8 ± 0.7 nM; Figure 1D). Thermal denaturation of yeast that displayed scFvs demonstrated improved thermal stability of both huCD19R(VH4Vκ1) (61 ± 0.9°C) and huCD19R(4D5) (62 ± 0.6°C) relative to the WT (55 ± 0.7°C; Figure 1E). Immunogenicity analysis of the humanized scFvs predicts lower immunogenicity through MHC-I presentation relative to FMC63 (supplemental Tables 1 and 2). FMC63 contains 22 peptide epitopes with positive prediction scores for MHC-I presentation and 3 epitopes with MHC-I IC50 of <10 nM, which is predictive of eradication of the cells presenting such epitopes.³⁰ In contract, huCD19R(VH4Vκ1) and huCD19R(4D5) contain 20 and 15 peptide epitopes with positive prediction scores, respectively. Importantly, neither humanized scFv showed any epitopes predicted to have MHC-I IC50 of <10 nM, suggesting that these scFvs are less likely to elicit an ablative immune response.

Humanized CD19 CAR T cells exhibit similar transduction efficiency and phenotype to FMC63 CAR T cells

Tn/mem cells isolated from healthy donors were activated and lentivirally transduced³¹ with huCD19R(VH4vκ1), huCD19R(4D5), and FMC63 CAR with the same backbone as a control (Figure 2A) and expanded for 14 days. We were able to successfully transduce both humanized CARs with similar transduction efficiency to FMC63 CAR as assessed by EGFR staining (Figure 2B). Phenotypic analysis of the same donor FMC63 and humanized CARs show similar expression of memory and stem memory markers such as CD27 and CCR7 and CD45RA and CD28, respectively, and low exhaustion markers such as PD1, TIM3, and LAG3 (Figure 2C).

The expansion of T cells harboring the humanized CARs was similar in comparison with those harboring the FMC63 CAR (Figure 3A). To evaluate the function of the humanized CD19 CARs, B-cell lines that represent different B-cell tumor types were used. To note, CD19 expression based on molecules per cell was similar on both ALL Nalm6 and Burkitt lymphoma Raji cells whereas CD80/86 expression was only found on the Raji cell line (Figure 3B), which results in different responses to CAR T cell therapy.³² CD19⁻ acute myeloid leukemia KG-1a was used as a negative control. Stimulation of huCD19R(VH4vκ1) and huCD19R(4D5) CARs produced an antigen-dependent expression of CD107a and production of IFN-γ that was equivalent to levels expressed by FMC63 CAR T cells (Figure 3C-D).

We further tested the long-term effector function of humanized CAR T cells by performing a tumor rechallenge assay in which huCD19R(VH4vκ1), huCD19R(4D5), and FMC63 CAR T cells were cocultured with Raji WT cells at an E:T ratio of 2:1. CAR T cells were repetitively challenged with Raji WT cells on days 3, 6, and 9 (supplemental Figure 1A). We observed significant killing of Raji and Nalm-6 tumor by huCD19R(VH4vκ1) CAR T cells on days 9 and 12 (Figure 3E; supplemental Figure 1B) as with FMC63 CAR T cells. However, huCD19R(4D5) CAR T cells essentially failed to proliferate upon tumor rechallenge (Figure 3F; supplemental Figure 1C). At the end of the 12-day assay, we found comparable levels of PD1 and LAG3 on huCD19R(VH4vκ1), huCD19R(4D5), and FMC63 CAR T cells (Figure 3G, supplemental Figure 1D). Together, these data suggest that both humanized CAR T cells mediate similar in vitro effector functions. However, huCD19R(VH4vκ1) but not huCD19R(4D5) CAR T cells were able to kill and proliferate upon repetitive antigen rechallenge.

To determine the antitumor efficacy of the 2 humanized CARs in vivo in comparison with FMC63 CAR, we tested CAR T cells against the established CD19⁺ Raji model (Raji-WT; Figure 4A). We observed that huCD19R(VH4vκ1) CAR T cells and FMC63 CAR T cells mediated significantly better antitumor activity compared with huCD19R(4D5) CAR T cells (P < .001; Figure 4B-D). Furthermore, huCD19R(VH4vκ1) CAR T cells significantly improved the survival of mice in Nalm6 model compared with huCD19R(4D5) CAR T cells (P < .05; supplemental Figure 2).

CD19 antigen expression is heterogenous because of disease type, stage, or antigen downregulation.³³ As we expected, CD19 expression in primary ALL samples ranged from 10⁵ to 10⁴ molecules per cell measured by Quantum Simply Cellular assay (Figure 5A). To determine the efficacy of humanized CD19 CARs against low antigens, we created a Raji cell line with low CD19 density. First, CD19 was CRISPR-edited out of Raji-WT cells to produce Raji-KO cells. CD19 was then reintroduced into these Raji-KO cells by lentiviral transduction under the regulation of a PGK100 promoter (mutant PGK promoter driving weak expression),³⁴ to produce Raji CD19-low cells. Raji CD19-low clones were generated by limiting dilution³³ and expanded for functional assays (supplemental Figure 3). Flow cytometry analysis was used to determine the levels of CD19 expression on Raji-WT, Raji-KO, and 4 Raji CD19-low clones, with Raji CD19-low clone 4 matching the level of lower CD19 expression in certain patient samples (Figure 5A-B). Hence, Raji WT and Raji CD19-low clone 4, representing the range of high to low antigen expression in primary tumor cells from patients, were used as targets to assess CAR functionality. Notably, with Raji CD19-low tumor in vitro, huCD19R(4D5) CAR-T cells showed slightly lower levels of degranulation and IFN-γ production compared with huCD19R(VH4vκ1) and FMC63 CAR T cells (Figure 5C-D).

When testing humanized CARs against Raji CD19-low tumor (Figure 6A; supplemental Figure 4A), we found that huCD19R(VH4vκ1) CAR T cells effectively controlled the tumor growth (Figure 6B-C). Importantly, treatment with huCD19R(4D5) CAR failed to improve survival of mice compared with the huCD19R(VH4vκ1) treated group (Figure 6D; P < .001), despite the similar CAR T cell expansion/persistence at the end points (Figure 6E), suggesting that the affinity of huCD19R(4D5) CAR is below the threshold required for CAR function in this context.³⁵ Antitumor efficacy of huCD19R(VH4vκ1) CAR T cells was similar when compared with the FMC63 CAR T cell group. Tumor antigen expression (CD19 surface density) was assessed at the end of the experiments and found to have maintained low CD19 expression (supplemental Figure 4B), demonstrating that antigen loss is not the mechanism for continued tumor growth. Together these results imply that the huCD19R(4D5) CAR is not efficacious for in vivo tumor treatment with low levels of antigen. Therefore, huCD19R(VH4vκ1) CAR was the selected candidate for further study and clinical translation.

Cytokine release syndrome (CRS) is a common adverse event in CAR T cell therapies. However, there is a lack of an optimal model for evaluating or predicting the risk of CRS. To mimic the sequence of events leading to CRS, we compared cytokine profiles of the lead candidate huCD19R(VH4vκ1) CAR and FMC63 CAR by adopting a previously reported hybrid in vitro/in vivo model.²² Briefly, serum from mice engrafted with Raji-WT and treated with huCD19R(VH4vκ1) CAR T cells, FMC63 CAR T cells, or mock T cells were added to donor-matched GM-CSF–activated macrophages. The supernatant was collected after 48 hours and analyzed using Luminex (Figure 7A). This model allows us to analyze the trigger cytokines from CAR T cell–tumor engagement as well as secondary inflammatory cytokines derived from bystander immune cells such as macrophages. We found that baseline serum collected from huCD19R(VH4vκ1) CAR–treated mice had higher INF-γ, compared with the FMC63 CAR group (Figure 7B). Importantly, when activated macrophages were exposed to this serum, no statistically significant differences in hallmark cytokines of CRS such as IL-6 and GM-CSF were observed between huCD19R(VH4vκ1) and FMC63 CAR groups (P > .05). We did, however, observe higher levels of monokine induced by IFN-γ (MIG) in the huCD19R(VH4vκ1) CAR group (Figure 7C). These data are consistent with higher levels of IFN-γ in serum that, in turn, upregulate MIG as an IFN-γ–induced cytokine.³⁶ These data demonstrate that huCD19R(VH4vk1) CAR and FMC63 CAR exhibit similar cytokine safety profiles, making this humanized CAR ideal to translate to the clinic.

Built on the preclinical characterization and evaluation, we conducted studies with guanosine 3′,5′-cyclic monophosphate–grade lentivirus encoding huCD19R(VH4vκ1) CAR to conduct Investigational New Drug application (IND) enabling studies with T cells from 3 different healthy donors following the manufacturing standard operating procedures. We observed consistent transduction efficiency (supplemental Figure 5A) and expansion (supplemental Figure 5B). The CAR T cell products exhibited consistent memory phenotypical characteristics (supplemental Figure 5C) and effector function upon stimulation with CD19⁺ target (supplemental Figure 5D). Lastly, the CAR T cell products met all release criteria for infusion (supplemental Table 3).

In this study, 2 candidate humanized CD19 scFvs were derived from the murine scFv FMC63, readily displayed on the yeast surface but demonstrated differences in binding affinity and thermal stability relative to the parental murine scFv. Both humanized variants had a drastic reduction in binding affinity, which is a relatively common outcome from the initial humanization process.³⁷ Depending on the CAR design and the antigens in tumors, different binding affinities are required for the proper function of CAR T cells. Some CAR T cells with T cell receptor–level affinities (weak, μM) can work efficiently in certain conditions but not in all.^38-41 Therefore, each CAR and target combination must be carefully validated. The ability of weakly binding CARs to mediate T cell activation is likely because of the high avidity between CAR T cells and antigen-expressing tumor cells, which can overcome the weak individual interactions (ie, low affinity) because of the strength of combined interaction (ie, high avidity). In contrast, the humanized scFvs both demonstrated improved thermal stability relative to the parental scFv. It has been shown previously that the stability of a folded state influences the immunogenicity of proteins of interest, with improving fold stability leading to diminished T cell epitope availability.⁴² This is because of well-folded proteins exposing fewer linear epitopes to proteolytic degradation, which protects from MHC presentation and subsequent immunogenicity. An improvement in fold stability alongside diminished quality of T cell epitope presentation in MHC-I, which was demonstrated through computational efforts, could improve the safety and efficacy of CD19-targeted CAR T cells through diminished prevalence and availability of T cell epitopes. This is especially important because a previous CD19-CAR T cell trial noted treatment rejection due to MHC-centric eradication of therapeutic cells.⁴³ 

Despite the differing biophysical properties of the humanized scFv variants, all CAR T cell groups demonstrated comparable growth potential and phenotypic characteristics as compared with FMC63 CAR T cells. Both humanized CAR T cells were capable of effector functions such as degranulation and IFN-γ production while targeting different CD19-expressing tumors. Interestingly, the tumor rechallenge assay revealed differential proliferation dependent on binding affinity. huCD19R(VH4vκ1) but not huCD19R(4D5) CAR was able to mediate robust proliferation after repetitive antigen stimulation without increased exhaustion, indicating that there is a threshold of affinity for proper CAR function, an observation that has been made in prior work as well.^39,44 Consistently, we observed significantly better antitumor activity and prolonged mouse survival when treated with huCD19R(VH4vκ1) as compared with huCD19R(4D5) CAR. The in vivo results using huCD19R(VH4vκ1) CAR T cells were equivalent to FMC63 CAR T cells.

CD19 antigen expression levels on tumor cells could be heterogeneous because of disease type, stage, and targeted treatment.^45,46 The density of antigen expression can affect all components of an antitumor CAR T cell response.^47,48 Although our humanized CD19 CARs have a low affinity, we observed a more efficient effector function of huCD19R(VH4vκ1) and FMC63 CD19 CAR T cells against Raji CD19-low tumor than huCD19R(4D5) CAR T cells. When Raji CD19-low tumor-bearing mice were treated with humanized CAR T cells, huCD19R(VH4vκ1) outperformed huCD19R(4D5) CAR T cells, resulting in significantly prolonged mouse survival. Again, a similar ability to target CD19-low tumors was observed between huCD19R(VH4vκ1) and FMC63 CAR T cells. Data also suggested that differences in the antitumor activity are indeed the result of the unfavorable function of low-affinity huCD19R(4D5) CAR rather than CAR T-cell persistence.³⁵ 

To mimic the sequence of events leading to CRS, we compared cytokine profiles of huCD19R(VH4vκ1) CAR T cells and FMC63 CAR T cells by adopting a hybrid in vitro/in vivo model.²² There was higher IFN-γ secreted by huCD19R(VH4vκ1) CAR T cells than FMC63 CAR T cells. However, the increased IFN-γ did not induce higher levels of IL-6, GM-CSF, IL-1, and Macrophage Inflammatory Protein (MIP), which are the hallmarks of CRS.⁴⁹ 

Based on preclinical studies, the huCD19R(VH4vκ1) T cells have received US Food and Drug Administration approval for a phase 1 clinical trial (IND 30031, IRB 23328). This trial aims to investigate the safety and feasibility of using huCD19 CAR T cells for the treatment of relapsed/refractory ALL.

This study was financially supported by Lymphoma SPORE (P50CA107399; S.J.F. and X.W.), the Borstein Family Foundation (X.W), and a postdoctoral fellowship award from the Tobacco-Related Disease Research Program of the Universities of California, 28FT-0072 (L.A.S.)

Contribution: X.W., C.E.B., L.A.S., J.C.W., and S.J.F. contributed to the study concept and design, and data interpretation; V.V., L.A.S., L.L., C.H., R.U., R.E., Z.W., and M.S.T. performed experiments and generated figures; and X.W., V.V., and L.A.S. wrote the manuscript.

Conflict-of-interest disclosure: L.A.S., J.C.W., S.J.F. X.W., C.E.B. are the inventors of the huCD19CAR patent application. The remaining authors declare no competing financial interests.

The current affiliation for L.A.S. is the Department of Chemical, Biological, and Materials Engineering, University of South Florida, Tampa, FL

Correspondence: Xiuli Wang, Department of Hematology and Hematopoietic Cell Transplantation, City of Hope National Medical Center, 1500 E Duarte Rd, Duarte, CA 91010; email: xiuwang@coh.org.