Robust anti-tumor activity and low cytokine production by novel humanized anti-CD19 CAR-T cells

Abstract

Recent studies have described the remarkable clinical outcome of anti-CD19 chimeric antigen receptor (CAR) T cells in treating B-cell malignancies. However, over 50% of patients develop life-threatening toxicities associated with cytokine release syndrome (CRS) which may limit its utilization in low-resource settings. To mitigate the toxicity, we designed a novel humanized anti-CD19 CAR-T cells by humanizing the framework region of scFv derived from a murine FMC63 mAb and combining it with CD8a transmembrane domain, 4–1BB costimulatory domain and CD3ζ signalling domain (h1CAR19–8BBζ). Docking studies followed by molecular dynamics (MD) simulation revealed that the humanized anti-CD19 scFv (h1CAR19) establishes higher binding affinity and has a flexible molecular structure with CD19 antigen compared to murine scFv (mCAR19). Ex vivo studies with CAR-T cells generated from healthy donors and patients with relapsed/refractory (r/r) B-cell acute lymphoblastic leukemia (B-ALL) expressing either h1CAR19 or mCAR19 showed comparable anti-tumor activity and proliferation. More importantly, h1CAR19–8BBζ-T cells produced lower levels of cytokines (IFN-γ, TNF-α) upon antigen encounter and reduced the induction of IL-6 cytokine from monocytes than mCAR19–8BBζ-T cells. There was a comparable proliferation of h1CAR19–8BBζ-T cells and mCAR19–8BBζ-T cells upon repeated antigen encounter. Finally, h1CAR19–8BBζ-T cells efficiently eliminated NALM6 tumor cells in a preclinical model. In conclusion, the distinct structural modification in CAR design confers the novel humanized anti-CD19 CAR with a favorable balance of efficacy to toxicity providing a rationale to test this construct in a phase I trial.

Introduction

Chimeric Antigen Receptor T cell (CAR-T cell) therapy has shown remarkable clinical response in CD19 positive hematological malignancies [1]. Despite remarkable remission rates, limitations include life-threatening toxicities, poor T cell persistence, immunogenic reactions, tumor escape, and technological challenges inherent to complex manufacturing processes [2–6]. Moreover, the prohibitive cost of commercially available CAR-T cell therapy and the additional costs of ancillary services associated with intensive-care unit (ICU) utilization make CAR-T therapy inaccessible to patients in countries with limited resources [7–9]. Altogether, the severity of the toxicity profile along with scientific and socio-economic challenges hinder the therapeutic index and accessibility of CAR-T cells in a majority of the patients who could benefit from this therapy worldwide.

Many groups reported life-threatening toxicities such as cytokine release syndrome (CRS) and neurotoxicity in large numbers of patients treated with CD19 CAR-T cell therapy [4, 10]. CRS is characterized by the elevated immune response of the body that appears as high fever, inflammation, hemodynamic and respiratory compromise, macrophage activation syndrome, and vascular instability. Neurotoxicity is generally manifested by symptoms of aphasia, encephalopathy, and seizures. The onset of these toxicities has been attributed to both intrinsic and extrinsic factors. Intrinsic factors such as high disease burden during the CAR-T cell infusion are positively correlated with the severity of CRS and neurotoxicity [2]. Extrinsic factors such as the design of the CAR construct play a decisive role in inducing toxicity [11]. One of the most accepted plausible mechanisms for induction of toxicity is a release of a plethora of cytokines by CAR-T cells upon interaction with tumor cells leading to hyperimmune activation [10]. Various preclinical as well as clinical studies have demonstrated the positive correlation of severe CRS and neurotoxicity with the early elevation of pro-inflammatory cytokines such as IFN‐γ, TNF-α, IL-2, IL-6, IL-8, IL-10 post CAR-T cell infusion [12–14]. While IL-6 produced by activated macrophages and by endothelial cells is believed to be the major cause of toxicities, the precise mechanisms of neurotoxicity and CRS remains unclear due to unavailability of suitable experimental models [15–17]. Very recently, CD19 expression is reported on the brain mural cells, and binding of anti-CD19 CAR-T cells to CD19 positive mural cells might cause neurotoxicity [17]. However mouse mural cells express negligible levels of CD19 and thus limiting the utility of the murine model in assessing the in vivo neurotoxicity. Therefore, the development of CAR-T cells with a more favorable toxicity profile would extend its applicability, particularly in low resource settings.

The attributes of cytokine secretion and anti-tumor efficacy of CAR-T cells are largely dependent upon the anatomy and design of the CAR. We and others have highlighted that each domain of the CAR construct has a regulatory role in determining the functional efficacy and safety of the CAR-T cell therapy [1, 11, 18, 19]. Very recently, just a single amino acid residue mutation in the CD28 co-stimulatory domain of a CAR has been shown to ameliorate CAR-T cell persistence and durability of the anti-tumor function compared to the native CD28 co-stimulatory domain [20]. Another study highlighted the role of CD8α hinge and transmembrane domain of the CAR in inducing feeble T cell activation and less cytokine secretion and thereby alleviating neurotoxicity in the patients [13]. These studies indicate that even a minor alteration in CAR design strongly influences the anti-tumor reactivity and safety of CAR-T cells.

Here, we postulated that the development and further optimization of a humanized anti-CD19 CAR could improve upon the efficacy to toxicity balance. We developed a novel humanized anti-CD19 CAR (h1CAR19–8BBζ) and demonstrated that h1CAR19 scFv interacts with CD19 antigen with higher binding affinity and more flexible complex (h1CAR19 scFv-CD19 antigen) using structural modeling and molecular dynamics simulation. Functionally, ex vivo and preclinical in vivo assessment demonstrated the robust anti-tumor activity of h1CAR19–8BBζ-T cells with two unique features compared to the murine counterpart: 1) very low levels of cytokine production, and 2) equal distribution of CD4+ and CD8+ T cells in the final product. Further, lower cytokine production had no detrimental effect on anti-tumor activity and proliferation ability of CAR-T cells.

Results

Humanized h1CAR19–8BBζ-T cells exhibit potent anti-tumor activity with lower cytokine secretion, and equal CD4:CD8+ T cell product expansion in contrast to mCAR19–8BBζ-T cells

The modifications to the CAR design, especially in the regulatory domains, affect the molecular structure and are critical in improving the efficacy to toxicity ratio. We designed several unique humanized scFv derived from the murine FMC63 monoclonal antibody (mAb) by selecting the framework of a heavy chain, VH4–34 and light chain, VK1-O18 from VBASE2 database based on the best fit similarity [21]. Further, amino acid residues proximal to the complementarity determining regions (CDRs), which are critical for optimal binding with CD19 antigen, were identified, and four murine residues (S²⁵, I⁶⁹, K⁷⁰, F⁷⁸) were conserved in the humanized scFv. Both the heavy and the light chain were joined by a flexible linker (G4S)_3. Two second-generation humanized anti-CD19 CAR constructs; h1CAR19–8BBζ and h2CAR19–8BBζ, comprise of humanized scFv, hinge and transmembrane domains derived from human CD8α, human 4–1BB co-stimulatory domain, and human CD3ζ signaling domain were synthesized (Figure 1A, Supplementary Table 2). The only difference between the scFv of two humanized CARs is that h1CAR19–8BBζ contains CDRs of previously published humanized anti-CD19 CAR and h2CAR19–8BBζ contains CDRs from the FMC63 mAb [22, 23].

Figure 1: Humanized h1CAR19–8BBζ-T cells exhibit potent anti-tumor activity with lower cytokine secretion in an antigen dependent manner.

(A) Schematic representation of three CAR constructs that differ in their scFv sequences. The mCAR19–8BBζ contains murine CD19 scFv while h1CAR19–8BBζ and h2CAR19–8BBζ CARs contain humanized scFv sequences. All three CARs have hinge (H) and transmembrane (TM) domain of CD8α, human 4–1BB co-stimulatory domain and human CD3ζ signaling domain. (B) Schematic representation of CAR-T cell generation process. (C) Percent transduction efficiency (% TE) of all three CAR constructs was determined using protein-L staining by flow cytometry (n=4–5). (D) CAR-T cells were co-cultured with tumor cells (NALM6, Raji, K562CD19+ and K562CD19-ve) for 24 hours and antigen specific cytotoxicity was determined by flowcytometry (n=3). (E) CAR-T cells were co-cultured with tumor cells for 24 hours and IFN-γ, IL-2 and TNF-α were quantified in cell-free supernatants by ELISA (n=3). (F) CAR-T cell proliferation kinetics is depicted till day 8 (n=6–8). (G) Distribution of CD4+ and CD8+ T cell numbers on day 0 (initiation of the culture) and on day 14 (final CAR-T cell product) is plotted (n=3). Numbers of experiments are represented as n.

Next, we explored the impact of CAR design on the anti-tumor potential of h1CAR19–8BBζ and h2CAR19–8BBζ-T cells and compared it with murine anti-CD19 CAR (mCAR19–8BBζ) [24]. CD3+ T cells were isolated from PBMCs of healthy donors and were transduced with CAR encoding lentivirus, and the expression of CAR on the T cells surface was analyzed by flow cytometry (Figure 1B). All the constructs stably expressed the CAR on the T cell surface (percent transduction efficiency (Mean ± SEM)); mCAR19–8BBζ: 28.6 ± 3.8%, h1CAR19–8BBζ: 27.2 ± 2.4% and h2CAR19–8BBζ: 30 ± 5.90%) (Figure 1C). Further, antigen-dependent cytotoxicity of the h1CAR19–8BBζ and h2CAR19–8BBζ-T cells was determined by co-culture assay using CD19 expressing tumor cell lines (NALM6, Raji, and transgenic K562CD19+ cells). The methodology of K562CD19+ cells generation is described in supplementary information. Wild type K562 cell line (CD19-ve) was used as a negative control. All the CAR-T cells showed remarkable antigen-dependent cytotoxicity as quantified by flow cytometry (Figure 1D).

Further, we quantified cytokine profile, especially IFN-γ, IL-2, and TNF-α secreted by CAR-T cells upon antigen encounter by ELISA. Surprisingly, the h1CAR19–8BBζ-T cells produced several-fold lower levels of IFN-γ and decreased TNF-α upon tumor cell exposure compared to the other two CARs (mCAR19–8BBζ and h2CAR19–8BBζ) in a highly reproducible manner. IL-2 production was comparable and varied based on interaction with the type of tumor cell lines. The h2CAR19–8BBζ-T cells showed a cytokine profile similar to mCAR19–8BBζ-T cells (Figure 1E). Despite producing low levels of cytokines, the h1CAR19–8BBζ-T cells exhibited comparable in vitro expansion as well as expansion kinetics to mCAR19–8BBζ-T cells (Figure 1F).

Since the therapeutic efficacy of CAR-T cell products critically depends on the subset distribution of the final product [25], we investigated the CD4+ and CD8+ T cell distribution in the h1CAR19–8BBζ and mCAR19–8BBζ-T cells products manufactured from T cells of the same donors. Interestingly, upon flow cytometry analysis of the expanded CAR-T cell product, there was equal proliferation of both CD4+ and CD8+ T cells (CD4:CD8+ T cell ratio (Mean ± SEM); 0.74 ± 0.035) in h1CAR19–8BBζ-T cells. However, mCAR19–8BBζ-T cells showed a more skewed proliferation of CD8+ T cells (CD4:CD8+ T cell ratio (Mean ± SEM); 0.52 ± 0.024, Figure 1G). Thus, h1CAR19–8BBζ-T cells show a more equal CD4:CD8+ T cell distribution compared to mCAR19–8BBζ-T cells.

Humanized h1CAR19–8BBζ-T cells induce lower amount of IL-6 production by monocytes compared to mCAR19–8BBζ-T cells

Next, we examined the impact of h1CAR19-8BBζ-T cells on IL-6 production by monocytes using a previously reported co-culture assay system, an in vitro model for CRS [15, 26, 27]. The h1CAR19–8BBζ-T cells or mCAR19–8BBζ-T cells from multiple donors (n=9) were co-cultured with CD19+ Raji tumor cells and the supernatant was collected after 24 hours. These cell-free supernatants were used to stimulate the monocytes isolated from a healthy donor. The supernatant derived from h1CAR19-8BBζ-T cells induced a significantly lower amount of IL-6 by the monocytes compared to mCAR19-8BBζ-T cells. However, there were comparable levels of IL-1β production by monocytes (Figure 2A).

Figure 2: Humanized h1CAR19–8BBζ-T cells induce lower amount of IL-6 production by monocytes compared to mCAR19–8BBζ-T cells and possess robust proliferation potential and superior anti-tumor T cell subsets in the expanded product.

(A) In a co-culture assay, CAR-T cells (0.5×10⁵) from healthy donors (n=9) were cultured with Raji cells (0.5×10⁵) for 24h and cell-free supernatant was collected. Monocytes (1×10⁴) from a healthy donor were isolated and stimulated with cell-free supernatant of co-culture assay for 24 hours. IL-6 and IL-1β release by monocytes is quantified by ELISA (n= 9). (B) In a repeat antigen stimulus stress test, h1CAR19–8BBζ-T cells were stimulated with CD19+ Raji cells on day 0 and 3. Cells were counted before and after 1^st and 2^nd stimulation (n= 6). (C-E) Multi-parametric phenotypic analysis was performed on expanded CAR-T cells (day 14) by flowcytometry for the quantification of the various T cell subsets. Representative dot plots depicting gating strategies for T stem cell memory (T_SCM, CD3+CD62L+CD45RO-CCR7+CD95+), T central memory (T_CM: CD3+CD62L+CD45RO+CCR7+), T effector memory (T_EM: CD3+CD62L-CD45RO+CCR7+), and T regulatory cells (T_reg: CD3+CD4+CD25+FoxP3+) in CD4+ T cells are shown (C). Both CD4+ and CD8+ T cell subsets were analyzed using identical gating strategy. The cumulative data from four experiments (n=4) depicting the distribution of the T cell subsets in the final product is plotted (D). Differentiation status of CAR-T cells in expanded product is examined based on CD45RO and CCR7 expression by flowcytometry and cumulative data from three independent experiments is shown (E).

In order to examine the robustness of the h1CAR19–8BBζ-T cells in vitro, we performed a repeat antigen stimulus stress test. The h1CAR19–8BBζ-T cells were stimulated with CD19+ Raji cells with repeat stimulation after three days without any exogenous IL-2 support. The h1CAR19–8BBζ-T cells showed similar proliferation, and were not exhausted after repeat tumor-specific antigen encounter similar to mCAR19–8BBζ-T cells (Figure 2B). These data suggest that lower secretion of cytokines like IFN-γ had no detrimental impact on anti-tumor cytotoxicity and proliferation of h1CAR19–8BBζ-T cells.

Further we examined the presence of various T cell subsets and differentiation status of T cells in CART-cells after 14 days of expansion by multiparametric flow cytometry analysis as reported previously [28, 29]. The distribution of various T cell subsets (T_SCM; CD3+CD62L+CD45RO-CCR7+CD95+), T_CM; CD3+CD62L+CD45RO+CCR7+, and T_EM; CD3+CD62L-CD45RO+CCR7+) and differentiation status of T cells (based on CD45RO and CCR7 expression) in both h1CAR19–8BBζ-T and mCAR19–8BBζ-T cells were comparable (Figure 2C-E), advocating that humanization preserved the stemness, memory and effector phenotypes in both the CD4+ as well as the CD8+ T cells.

Moreover, in order to rule out the malignant transformation of h1CAR19–8BBζ-T cells in long-term expansion, h1CAR19–8BBζ-T cells were injected (subcutaneous) into immunodeficient NOD/SCID mice, and tumor growth was monitored until four weeks. No tumor growth at the injection site was observed in mice (Supplementary Figure 2A-B).

Humanized h1CAR19 scFv exhibits higher binding affinity and flexibility upon binding to CD19 antigen in contrast to the h2CAR19 as well as mCAR19 scFv

Next, we probed the impact of CAR design on binding characteristics of humanized scFv to the CD19 target antigen. The three-dimensional structures of scFvs and CD19 antigen (PDB ID: 6AL5) were obtained using the molecular modeling as described in detail in the method section and in supplementary information (Supplementary Figure 3-5 and Supplementary Table 3-6). While molecular docking study revealed that all three scFvs prefer similar binding mode with the CD19 antigen, the lowest energy docked conformation of h1CAR19 (−21510.4 kcal/mol) compared to mCAR19 (−16684.8 kcal/mol) and h2CAR19 (−15189.2 kcal/mol) scFv with CD19 antigen (Figure 3A). The docking analysis further indicated that h1CAR19 scFv has the lowest binding energy and establishes the maximum number of bonding interactions with the CD19 antigen compare to other scFvs and CD19 complexes (Supplementary Table 3-6). Next, molecular dynamics (MD) simulation analysis was performed as described in detail in the supplementary information, indicated that all the CD19-CAR scFv complexes reached their stability after 50ns (Figure 3B). Interestingly, h1CAR19 scFv shows higher RMS fluctuations compared to other scFvs (Figure 3C, Supplementary Video 1–3) due to the residue composition variations in the h1CAR19 scFv, which resulted in the conformational flexibility of the heavy chain and in the linker region (Figure 3C, D). The analysis of hydrogen bonding interactions in the MD simulated end structure is listed in Supplementary Table 3-5. Furthermore, the binding energy of CD19 antigen and scFv complexes were decreased in the order of CD19-h1CAR19 scFv (−254.57 kcal/mol) > CD19-h2CAR19 scFv (−249.18 kcal/mol) > CD19-mCAR19 scFv (−188.71 kcal/mol) (Table 1), indicating h1CAR19 scFv has a higher binding affinity towards the CD19 antigen compared to mCAR19. In the CD19-h1CAR19 scFv complex, electrostatic interaction energy makes favorable contributions compared to mCAR19 and h2CAR19 complexes. Overall our molecular modeling study confirms the higher binding affinity of h1CAR19 towards the CD19 antigen.

Figure 3: Humanized h1CAR19 scFv exhibits higher binding affinity and flexibility upon binding to CD19 antigen in contrast to the h2CAR19 and mCAR19 scFv.

(A) Overlapped docked structures of CD19 and scFvs (CD19-mCAR19: cyan, CD19-h1CAR19: purple, and CD19-h2CAR: orange (B) Root mean square deviations (RMSD) of CD19-scFv complexes. The CD19-mCAR19-scFv: black color, CD19-h1CAR19-scFv: blue color, and CD19-h2CAR19-scFv complex: red color. The RMSD of CD19-mCAR19 scFv and CD19-h2CAR19 scFv is stabilized within 3–4 Å, whereas the CD19-h1CAR19 scFv shows higher fluctuations. (C) Root mean square fluctuations (RMSF) of Cα atoms of all three scFvs (mCAR19 scFv: black color, h1CAR19 scFv: blue color and h2CAR19 scFv: red color) in complexes with CD19 antigen. (D) RMSF of the CD19 protein only (color codes are similar to C) in complex with all three scFvs. Higher RMSF value indicates the high flexibility while the constrained regions show low RMSF value.

Table 1:

Binding energy of scFv proteins with CD19 based on the MM-PBSA method with standard error of mean

CD19-scFv complexes	Van der Waals energy kcal/mol	Electrostatic energy kcal/mol	Polar solvation energy kcal/mol	SASA energy kcal/mol	Total binding energy kcal/mol
CD19-mCAR19-scFv	−103.63 ± 1.10	−218.39 ± 4.59	144.99 ± 2.82	−11.68 ±0.12	−188.71 ±3.48
CD19-h1CAR19-scFv	−103.78 ±2.09	−351.14 ±4.06	212.11 ±3.09	−11.76 ±0.18	−254.57 ±4.05
CD19-h2CAR19-scFv	−103.45 ±1.01	−326.54 ±3.60	190.57 ±2.95	−11.75 ±0.12	−249.18 ±2.79

In vivo preclinical studies demonstrate robust anti-tumor efficacy of h1CAR19–8BBζ-T cells

Given the unique profile of h1CAR19–8BBζ-T cells (potent in vitro anti-tumor activity and very low levels of cytokine production), next, we focused on examining the anti-tumor activity of these cells extensively in in vivo model. To validate the in vivo functional efficacy, NALM6 cells (CD19+ acute lymphoblastic leukemia) bearing immunodeficient NOD/SCID mice were either treated with h1CAR19–8BBζ-T cells or mCAR19–8BBζ-T cells or with untransduced T cells (UT) (Figure 4A). Tumor burden was measured weekly post tumor engraftment by bioluminescence imaging. Mice treated with h1CAR19–8BBζ-T cells or mCAR19–8BBζ-T cells had negligible tumor burden on day 12 compared to control groups (untreated as well as mice treated with UT cells) (Figure 4B-C). The mice treated with h1CAR19–8BBζ-T cells showed significant survival benefit compared to control groups (Figure 4D). Mice treated with repeat dose of h1CAR19–8BBζ-T cells (5×10⁶/mouse on day 1 and day 2 post tumor cell injection) had very negligible tumor burden on day 14 compared to control groups (Figure 4E-F). These data indicate that mice treated with h1CAR19–8BBζ-T cells (either single dose or repeat dose) showed excellent efficacy and extended survival.

Figure 4: Preclinical efficacy of h1CAR19–8BBζ-T cells in in vivo B-ALL model.

(A) The schematic representation of preclinical animal model is shown. The NOD/SCID mice were sub-lethally irradiated. Both F-Luc⁺NALM6 cells (0.5×10⁶ cells/mouse) and CAR-T cells (5.0×10⁶ cells/mouse) were administered through intravenous tail vein injection as indicated in the figure. (B-C) The representative bioluminescence images of tumor burden (total flux in photons/sec unit) on day 12 (B) and cumulative analysis of day 12 data (C) are depicted from one experiment. Each group had minimum of three mice. Two independent experiments resulted into similar data. (D) Kaplan-Meier survival curves for each group are shown. Log-rank (Mantel-Cox) analysis was performed for calculating survival statistics. Two independent experiments resulted into similar data. (E-F) Bioluminescence images on day 14 (E) and cumulative data analysis (F) of sub-lethally irradiated tumor bearing NOD/SCID mice treated with h1CAR19–8BBζ-T cell (serial dosing of 5.0×10⁶ cells/mouse at day 1 and 2) are shown. Each group had minimum of four mice.

Further, in vivo safety profile and toxicity of h1CAR19–8BBζ-T cells was determined by serum biochemistry, blood hematology, and histopathology studies of vital organs and corroborated no abnormal deviation or disease pathology in h1CAR19–8BBζ-T cells treated mice (Supplementary Figure 2C-F). Upon visual inspection, the control mice depicted splenomegaly in comparison to the h1CAR19–8BBζ-T cells treated mice group (Supplementary Figure 2D).

Ex vivo studies of h1CAR19–8BBζ-T cells manufactured from r/r B-ALL patients and development of patient-scale manufacturing process for conducting Phase I clinical trial

Next, our goal was to assess the feasibility and efficacy of h1CAR19–8BBζ-T cells generated from r/r B-ALL patients in ex vivo settings. Table 3 describes the detailed characteristics of the recruited patients. In brief, r/r B-ALL patients were defined by medullary CD19+ B-cell relapse after an initial remission (relapse), or by persistent bone-marrow blasts after three or more cycles of intensive chemotherapy detected by 10-color flow cytometry with minimal residual disease (MRD) positivity of >0.01% (refractory). CD3+ T cells from the patients were isolated, activated, and transduced with h1CAR19–8BBζ construct, and their functional characterization was performed, similar to healthy donors. The transduced T cells showed the effective killing of CD19+ leukemia and lymphoma cells (NALM6 and Raji cells) and cytokine production (IFN-γ and IL-2) in an antigen-specific manner (Figure 5A-B). No cytokine secretion and cytotoxicity was observed in CAR-T cells cultured with wild type K562 cells (CD19-ve).

Table 3:

Characteristics of the r/r B-ALL patients recruited in the study

Case No.	Patient ID	Age (Yrs)	Lymphoma / Leukemia type	Number of prior lines of therapy
CH26365	5	8	Relapsed	First line chemotherapy followed by High-risk relapse
CP23182	8	12	Relapsed	First line chemotherapy followed by High-risk relapse and salvage chemotherapy
CP42698	10	9	Refractory	First-line chemotherapy- refractory, and given 2nd-line chemotherapy
CR11792	13	21	Refractory	First-line chemotherapy- refractory, and given 2nd-line chemotherapy
CN16436	18	7	Relapsed	First line chemotherapy followed by High-risk relapse and salvage chemotherapy

Figure 5: Ex vivo studies of h1CAR19–8BBζ-T cells from r/r B-ALL patients and development and validation of patient-scale manufacturing process.

(A) Antigen specific cytotoxicity of h1CAR19–8BBζ-T cells from r/r B-ALL patients were determined ex-vivo by co-culturing them with CD19+ (NALM6 and Raji cells) and CD19-ve (K562) cell lines (n=4). (B) CAR-T cells were co-cultured with tumor cells for 24 hours and IFN-γ, IL-2 and TNF-α were quantified in cell-free supernatants by ELISA (n=3). (C) Schematic representation of different steps of patient scale CAR-T cell manufacturing as per cGMP guidelines. (D) Percent transduction efficiency (% TE) in six batches of h1CAR19–8BBζ-T cells is depicted. (E) Expansion (fold proliferation) of six batches of h1CAR19–8BBζ-T cells till day 9 (n=6) is shown.

Next, we optimized and validated the patient-scale CAR-T manufacturing under current good manufacturing practices (cGMP) for future clinical use. Multiple batches of h1CAR19–8BBζ-T cells were manufactured in a clean environment following GMP guidelines using a process outline in Figure 5C. In brief, CD3+ T cells of healthy individuals were isolated and activated using CTS^™ Dynabeads^™ CD3/CD28 magnetic beads in gas-permeable cell culture bags. Next, cells were transduced with h1CAR19–8BBζ encoding lentivirus and expanded for nine days to achieve a clinically significant numbers. These batches of h1CAR19–8BBζ-T cells showed CAR surface expression with transduction efficiency of 33.88 ± 3.26% (Mean ± SEM) and consistent expansion (average 70 to 80-fold) within nine days (Figure 5D-E). Quality control assays for safety, identity, potency and purity were performed for each batch to validate the process as described in supplementary information. Table-2 demonstrates the three representative batches of h1CAR19–8BBζ-T cells qualified as per defined acceptance criteria of release testing.

Table 2:

Quality Control assessment of three batches of humanized CD19 CART-cells manufactured in cGMP compliant settings

Quality Control of Humanized CD19 CART-cells
Criteria	Parameter	Method	Acceptance criteria	Batch 1	Batch 2	Batch 3
Safety	RCL test	VSV-G gene qPCR: Taqman based detection	Negative for VSV-G gene	Negative	Negative	Negative
	Sterility	BACTEC and culture based assay	Negative	Negative	Negative	Negative
	Mycoplasma	PCR based	Negative	Negative	Negative	Negative
Identity	Appearance	Visual inspection	Yellowish, Milky, no aggregates in cell suspension	Milky, no aggregates	Yellowish, no aggregates	Yellowish, no aggregates
	CAR surface expression	% protein L positive of viable CD3+ T cells	Transduction efficiency (TE)- ≥ 10%	23%	33%	30%
	CAR copy number (copies/ 100ng DNA)	Taqman based CAR gene qPCR method	CAR gene detection with > 5 copies	100 copies	100 copies	100 copies
Purity	Cell viability	Trypan blue exclusion assay	Cell viability > 70%	91%	96.3%	94%
	CD19 + viable cell detection	CD19 protein surface detection by Flow cytometry	Negative for the presence of CD19+ cells	Negative	Negative	Negative
	Detection of magnetic beads	Microscopic detection of CD3/CD28 magnetic beads	Presence of residual beads (<100 beads/3×10^6 cells)	No beads	57 beads/3×10^6 cells	16 beads/3×10^6 cells
Potency	Cytokine detection	IFN-γ ELISA assay	IFN-γ release (>50pg/ml)	Detected	Detected	Detected
	Cytotoxicity	Flow cytometry based	>50% killing of CD19+ tumor cells	Achieved	Achieved	Achieved

Discussion

CD19 CAR-T cell therapy is associated with life-threatening toxicities, which include CRS and neurotoxicity. Recent data suggest that even minor changes in CAR design can play a major role in the efficacy and safety of CARs in preclinical and clinical settings [1, 11, 13, 20, 27, 30, 31]. Here we demonstrate how our fine-tuning of the molecular structure of h1CAR19–8BBζ resulted in a favorable efficacy and toxicity profile, providing support for clinical testing of this novel construct in a low-resource setting where limited toxicity profiles will be needed to make CAR-T cell strategies more feasible.

Extensive preclinical in vivo and ex vivo data indicate that the design of the h1CAR19–8BBζ construct demonstrates the potential for a favorable efficacy. Although all three tested CAR-T cell constructs showed potent anti-tumor activity, there were striking differences in their structural and functional characteristics; 1) the scFv of h1CAR19–8BBζ construct formed a stable and flexible binding complex with CD19 antigen compared to the scFv of h2CAR19–8BBζ and mCAR19–8BBζ, 2) the h1CAR19–8BBζ-T cells produced significantly lower levels of cytokines (IFN-γ and TNF-α) compared to h2CAR19–8BBζ and mCAR19–8BBζ-T cells, 3) the monocytes produced lower levels of IL-6 upon stimulation with cell-free supernatant derived from co-culture of h1CAR19–8BBζ-T cells with CD19+tumor cells compared to mCAR19–8BBζ-T cells, and 4) the final product of h1CAR19–8BBζ-T cells possess a more even CD4+ and CD8+ T cell distribution compared to mCAR19–8BBζ-T cells.

We believe these structural and functional differences are due to the unique design of the h1CAR19–8BBζ CAR construct, which incorporate a combination of framework as well as CDRs together. Because a previous study with a humanized anti-CD19 CAR containing identical CDRs like h1CAR19 scFv showed a cytokine profile as well as a binding pattern, similar to murine mCAR19 scFv [23]. Moreover, h2CAR19 scFv containing a framework identical to h1CAR19 scFv showed a binding pattern and cytokine production similar to mCAR19 scFv. However, few studies reported that CAR scFvs with a strong affinity to tumor antigen resulted in increased “on-target off-tumor” toxicity and reduced tendency of serial killing [30, 32]. Our h1CAR19 scFv forms a stable but flexible binding complex, unlike the rigid binding structure of mCAR19 and h2CAR19 scFv. These binding features of h1CAR19 scFv might be beneficial in dissociation from tumor cells and augmenting serial killing of tumor cells.

Functionally, h1CAR19–8BBζ-T cells produced several folds lower IFN-γ compared to h2CAR19–8BBζ as well as mCAR19–8BBζ-T cells with potent in vivo and ex vivo anti-tumor activity. Interestingly, there was lower levels of IL-6 production by monocyte stimulated with cell free supernatants of h1CAR19–8BBζ-T cells compared to murine counterparts in a co-culture assay, an in vitro model of CRS [10, 15, 16]. Although, there are some limitations of this in-vitro system due to lack of endothelial cells and other stromal elements, however, there are no suitable preclinical models to assess the CRS in in vivo settings. These data indicate that the design of the h1CAR19–8BBζ constructs favorably impacted the efficacy to toxicity balance. Initially, low levels of IFN-γ production by h1CAR19–8BBζ-T cells was concerning as few studies reported the positive correlation between higher cytokine production by CAR-T cells with potent anti-tumor efficacy and positive clinical outcome [33, 34]. However, similar to our observations, a very recent study showed that IFN-γ deficient CAR-T cells demonstrate potent anti-tumor activity in a preclinical model [35]. Further, recent studies demonstrated that despite low cytokine rerlease by anti-CD19 CAR-T cells and by anti-CD19 antibody TCR platform (AbTCR), these cells showed excellent anti-tumor activity in a clinical trial and in in vivo models respectively [13, 27]. Few other studies also reported the positive correlation of increased cytokine levels especially IFN-γ as one of the major factors responsible for CRS as well as neurotoxicity [10, 12, 14, 36]. CAR-T cell treatment-related toxicities are a major barrier to the widespread use of CAR-T cell therapy and highlight the appropriate intervention in this direction [2, 3, 37, 38]. Although various immunosuppressive agents including tocilizumab (anti-IL-6R monoclonal antibody), corticosteroids and IL-1 antagonists like anakinra have been shown to be effective in lowering the toxicity [10, 39, 40], the impact of these approaches on anti-tumor efficacy, and particularly persistence, are largely unknown and may increase the likelihood of tumor relapse and or complications from immunosuppression. Moreover, such strategies to limit toxicity may be cost-prohivitive in low resource settings, therefore development of a low toxicity CAR is critical.

Another interesting observation of our study was that h1CAR19–8BBζ-T cells could support the proliferation of both CD4+ and CD8+ T cell subsets depicting an equal distribution of CD4+ and CD8+ T cells in contrast to mCAR19–8BBζ-T cells that supports primarily proliferation of CD8+ T cell subsets. Earlier studies in preclinical as well as clinical models have established that therapeutic efficacy of CAR-T cell products critically depends on the cell phenotype, and products with defined CD4:CD8+ T cell ratio demonstrate superior clinical efficacy [12, 25].

There are two additional indirect benefits of this study. First, we are developing an infrastructure for training and establishing the manpower to treat patients with CD19 CAR-T cell therapy in low-resource settings like India, where the CD19 positive malignancy burden exceeds over 30,000 patients per year (numbers are based on our experience on many large-cancer care centers). With this study, we developed a patient-scale CAR-T cell manufacturing process and established release criteria assays of identity, safety, potency and purity required for quality control. Second, humanized anti-CD19 CAR-T cells will help in reducing the immunogenicity. Clinical studies of CAR-T cells with murine scFv, including Kymriah^™ and Yescarta^™ reported immunogenicity as a major concern in subsets of patients [41–43]. Patients treated with murine CAR-T cells do not respond well to repeat infusion due to immunogenicity [12, 44, 45]. However, infusion of humanized anti-CD19 CAR-T cells has shown remissions in 64% of the patients who were relapsed /unresponsive to murine CAR-T cells [44]. In line with these studies, we believe that humanized h1CAR19–8BBζ-T cells will reduce the risk of immunogenicity in case of multiple dosing is required in clinical settings.

Overall, our study provides strong preclinical and ex vivo evidence for the development of a novel humanized anti-CD19 CAR with potent anti-tumor activity and low cytokine production that confers a favorable efficacy and toxicity profile. Several clinical trials utilizing the h1CAR19–8BBζ-T cells for CD19 positive malignancies, including r/r B-ALL and DLBCL, are planned. In addition, the capacity building (infrastructure and trained personnel) in low resource settings like India with over 30,000 patients of CD19 positive malignancies will be beneficial in bringing CAR-T cell therapy to the majority of the patients where the toxicity of CAR-T therapy limits its applicability.

Methods

Recruitment of healthy donors and r/r B-ALL patients

This study was approved by the institute ethics committee (IEC) as well as by the institute biosafety committee (IBSCs) of the Indian Institute of Technology, Bombay (IIT-B) and Tata Memorial Centre (TMC), Mumbai. The project is approved from the review committee on genetic manipulation (RCGM) of department of biotechnology (DBT). The blood sample from healthy donors was collected at IIT-B. The r/r B-ALL patients (pediatric and young adults) were consented and recruited at TMC Mumbai upon obtaining informed written consent. The patient details are described in Table 3.

Process of humanization of a murine anti-CD19 scFV, design of two humanized scFv (h1CAR19 and h2CAR19) and CAR constructs

The humanization of two scFvs (h1CAR19 and h2CAR19) against CD19 antigen from a murine anti-CD19 scFv (FMC63 clone) was performed using the complementarity determining region (CDR) grafting method [22, 46]. In brief, the amino acid sequences of the FMC63 clone were subjected to Kabat numbering for the identification of CDRs. Next, the suitable acceptor humanized framework sequences were obtained using the VBASE2 database and multiple sequence alignment [21]. Based on the best fit similarity with the parent antibody, the VH4–34 and VK1-O18 were identified as suitable acceptor framework sequences for the heavy chain and light chain of scFv, respectively. Further, the donor CDRs residues were grafted in the acceptor framework followed by modeling with I-TASSER server. The amino acid residues in the proximity of the CDRs were identified using Pymol software and Vernier zone identification by Kabat numbering [47]. The four murine residues (S²⁵, I⁶⁹, K⁷⁰, F⁷⁸) were identified and conserved in the humanized framework regions (Supplementary Table 2). Further, the heavy chain of scFv was linked to the light chain using a flexible (G4S)3 linker. Using this unique framework, two humanized scFvs, h1CAR19 scFv, which contains CDRs of previously published humanized CD19 CAR, and h2CAR19 scFv, which contains CDRs from the FMC63 mAb, were designed [22, 23]. Next, two second-generation humanized anti-CD19 CAR constructs were synthesized; h1CAR19–8BBζ and h2CAR19–8BBζ, both consisting of hinge and transmembrane domains derived from human CD8α, human 4–1BB co-stimulatory domain, and human CD3ζ signaling domain were chemically synthesized by Gene Art (Germany) and cloned in E1-T 3^rd generation lentiviral transfer vector under EF-1α promotor.

Production of high titer lentivirus and manufacturing of CAR-T cells for in vitro and ex vivo studies

Lentivirus production, concentration, and purification processes have been described in detail in the supplementary information and in Supplementary Figure 1. Quality control assays were performed with all batches of lentivirus (Supplementary Table 1). CD3+ T cells were purified by EasySep^™ Direct Human T cell Isolation Kit (Stem Cell Technologies, USA) from the blood of the healthy donors and r/r B-ALL patients. Purified CD3+ T cells were activated with CD3/CD28 dynabeads (Thermo Fisher Scientific, USA) and transduced with CAR encoding lentiviruses at 1000xg for two hours at 32⁰C on two consecutive days. The percent transduction efficiency (%TE) was determined by flow cytometry using Protein-L staining. The CAR-T cells were expanded till fourteen days post-transduction, maintaining cell density of 0.3–0.5×10⁶/ml. The phenotypic analysis of expanded cells (day 14) was performed by multiparametric flow cytometry using the staining protocol and similar gating strategies as previously reported [28, 29, 48]. Data were acquired on BD FACSVerse^™ flow cytometer and analyzed using BD FACSuite^™ Software.

In vitro cytotoxicity and cytokine release assay

Humanized anti-CD19 CAR-T cells (effector cells) were co-cultured with GFP expressing CD19+ lymphoma or leukemia cells as target cells (Raji, NALM6, K562CD19+) for 24 hours at multiple effector to target (E:T) ratios. The cell-free supernatant was collected after 24 hours for cytokines quantification, and the cell pellet was used to measure the antigen-specific killing of CD19+ tumor cells by flow cytometry. The cell-free supernatant was used to quantify the levels of IL-2, IFN-γ and TNF-α by ELISA as per manufacturer instructions (Thermo Fisher Scientific, USA).

Monocytes derived IL-6 and IL-1β release assay

The mCAR19–8BBζ-T cells and h1CAR19–8BBζ-T cells generated from healthy donors (n=9) were co-cultured with Raji target cells (E:T ratio; 1:1) for 24 hours. The cell free supernatant was collected. The monocytes were isolated from a healthy donor and 1×10⁴ monocytes were stimulated with supernanat derived from co-cultures of all nine donors individually for 18–24 hours. The cell-free supernatant from monocyte was collected and quantified for IL-6 and IL-1β by ELISA as per manufacturer instruction (Thermo Fisher Scientific, USA)

Repeat antigen stimulus stress test

The h1CAR19–8BBζ-T cells were subjected to a repeated antigen stimulus stress test to evaluate in vitro CAR-T cell proliferation ability. The h1CAR19–8BBζ-T cells were co-cultured with CD19+ Raji cells (E:T ratio; 1:1). After an interval of three days, the cells were again stimulated with CD19+ Raji cells. At each interval, the absence of target cells was confirmed by flow cytometry, and the cells were counted using the trypan blue dye exclusion method.

Molecular Docking and MD simulation study of CD19 and scFv domain:

To explore the binding mode of mCAR19, h1CAR19, and h2CAR19 scFv domains with CD19 antigen, we first performed molecular docking using the ClusPro 2.0 server with ‘Antibody mode’ followed by the HADDOCK [49]. This search was used to find out the interaction and orientation between the two molecules to determine the correct binding between the CD19 and scFv domain. The docked complexes of CD19 and scFv domains were further used for the molecular dynamics simulation study as a starting structure. MD simulation approach (described in the supplementary information) was employed to investigate the refined binding mode and interaction between CD19 and scFv domain. Here, production MD simulations for 100ns were performed for all the systems such as CD19-mCAR19-scFv, CD19-h1CAR19-scFv, and CD19-h2CAR19-scFv using the GROMACS 18.1 software. Further, analysis and visualization of MD simulation trajectories were done by using the discovery studio visualizer, chimera, and PyMolsoftware. The visual molecular dynamics (VMD) were used similar to the preceding study to make MD simulation movies [50].

In vivo functional efficacy and safety study

Animal studies were performed upon approval of the institutional animal ethics committee (IAEC) of TMC, Mumbai. The immunocompromised NOD/SCID female mice (6–8 weeks old) were used for the study. Total body irradiation (TBI) of 2.5Gy dose was given one day prior to tumor injection (day −1). The following day (day 0), F-Luc⁺NALM6 cells (0.5×10⁶/mouse) expressing firefly luciferase fusion protein were administered by tail vein injection. The next day (day 1), mice were either treated with h1CAR19–8BBζ-T cells (2.5 ×10⁶/mouse, 5 ×10⁶/mouse or serial doses of 5 ×10⁶ /mouse on day 1 and day 2) or control groups (untreated controls or treated with un-transduced (UT) cells (5 ×10⁶/mouse)) through tail vein injection. Mice treated with mCAR19–8BBζ-T cells (5 ×10⁶/mouse) were used as benchmark control. The tumor burden was monitored by injecting 150 mg/kg D-luciferin (SRL, India) 5–6 min prior to bioluminescence imaging (Perkin Elmer IVIS 100 Imaging System). The bioluminescence flux was quantified by Live-Image software (Caliper Life Science, Waltham, MA, USA). Mice were imaged under 2% isoflurane anesthesia and 2L/min O₂. Mice were followed-up till day 40 post-tumor injection. For toxicity and safety studies, blood, bone marrow, and other vital organs such as heart, lungs, liver, lymph node, spleen, and kidneys were collected and immunohistochemistry was performed by H&E staining and blood hematology profile and serum biochemistry was also examined.

For tumorigenicity studies, mice were injected (subcutaneous) with 5×10⁶ h1CAR19–8BBζ-T cells and monitored till four weeks for any tumor development.

Manufacturing of patient scale CAR-T cells

PBMCs were isolated from the blood of healthy donors by ficoll-hypaque density centrifugation method. CD3+ T cells were enriched and activated by CTS^™ Dynabeads^™ CD3/CD28 as per the manufacturer protocol using CTS^™ DynaMag^™ magnet (Thermo Fisher Scientific, USA) in CTS^™ AIM V^™ media (Thermo Fisher Scientific, USA) supplemented with 5% heat inactivated human AB serum (Valley Biomedical, USA) containing rhIL-2 (50ng/ml, CTS^™ Recombinant Human Protein (Thermo Fisher Scientific, USA ) at 37°C and 5% CO₂ humidified incubator for 48 hrs in permalife cell culture bags (Origen, USA). T cells were transduced with multiplicity of infection (MOI) five of lentivirus in presence of protamine sulphate (1μg/ml) for two consecutive days. Percent transduction efficiency (% TE) was determined by flow cytometery. The CAR-T cells were expanded for nine days by maintaining the cell density of 0.3–0.5×10⁶/ml in expansion bag (Charter Medical, USA). On day 9, CAR-T cells were counted and aliquoted for quality control assays. The methodology of quality control assays is described in supplementary information in detail.

Statistical analysis

Data were analyzed using GraphPad Prism version 8 and presented in the form of descriptive statistics. Error bars in each graph represent Mean ± SEM. The p-values were determined by unpaired t-test except in Figure 1G (paired t-test). The p value <0.05, 0.01 and 0.001 are represented as *, ** and *** respectively and p value >0.05 is considered as non-significant (ns).