Modifications to the Framework Regions Eliminate Chimeric Antigen Receptor Tonic Signaling

Chimeric antigen receptors (CAR), in their original conception, are fusion proteins in which the variable regions of the heavy chain (V_H) and light chain (V_L) of a mAb are assembled with a noncleavable flexible linker to form an antibody with a single-chain variable fragment (scFv), which is fused with signaling molecules of the T-cell receptor (TCR) and costimulatory endodomains (1–3). Remarkable and sustained antitumor activity has been achieved using CD19-specific CAR-T cells, and these cells have been infused in pediatric patients with acute B-cell leukemia (4).

CAR engagement with the targeted antigen expressed by tumor cells promotes rapid activation of the T cells, which is characterized by cytolysis, cytokine secretion, and proliferation (5). However, it has emerged that CARs can cause tonic signaling in T cells, which denotes sustained antigen-independent activation leading to rapid T-cell exhaustion and impaired antitumor activity (6). The first report describing tonic signaling in CAR-T cells attributed this effect to the propensity of certain scFvs to self-aggregate, causing cell surface CAR clustering and consequent signaling (6).

It is well known that V_H and V_L domains coupled to form scFvs have the tendency to unfold, leading to scFv oligomerization (7). A variety of strategies have been developed in the effort to stabilize scFvs which include engineering disulfide bonds between the V_H and V_L domains, introduction of charged amino acids within the V_H and V_L domains, or grafting the complementarity-determining regions (CDR) into different framework regions (FWR; refs. 8–10). In the contest of scFvs assembled into a CAR format, the tonic signaling caused by the scFv derived from the murine 14g2a mAb is specifically attributed to the Ab's FWRs (6). Remarkably, the engraftment of the CDRs of the scFv derived from the FMC63 mAb into the FWRs of the 14g2a Ab is sufficient to cause tonic signaling of the FMC63 scFv when used to generate a CAR (6), indicating that the FWRs play a critical role in causing tonic signaling of scFvs in the CAR format.

Here, we showed that amino acid substitutions into the FWRs can stabilize the scFv and correct the tonic signaling of the CAR. We additionally showed that substitution of murine FWRs with stable human FWRs also prevents antigen-independent activation of CAR-T cells. Thus, CAR tonic signaling is caused by unstable scFvs, and we provide structure-based strategies that can be used to correct tonic signaling.

The tumor cell lines WM115 and SK-MEL-2 were obtained from the ATCC between 2016 and 2018, whereas M14 cells were provided by Dr. Ferrone (Massachusetts General Hospital). The tumor cell lines MDA-MB-468 and MDA-MB-231 were obtained from German Collection of Microorganism and Cell Cultures GmbH (ACC 738 and ACC 732). The 293T cells used for the production of retroviral vectors were obtained from the ATCC in early 2000. All cells were maintained in culture with the appropriate media, either RPMI-1640 (Gibco), DMEM (Gibco), or MEM (Gibco) supplemented with 10% FBS (Sigma), 1% l-glutamine (Gibco), and 1% penicillin/streptomycin (Gibco) in a humidified atmosphere containing 5% CO₂ at 37°C. WM115 cells were modified to express the fusion protein firefly luciferase and enhanced GFP (eGFP-FFluc; ref. 11). Cells were kept in culture for less than 6 consecutive months, after which, aliquots from the original expanded vial (cultured for four to six passages after being received) were used. All tumor cell lines were routinely tested to exclude contamination with Mycoplasma and assessed for the expression of transgenes and tumor markers by flow cytometry to confirm identity. Glioblastoma-derived neurospheres (GBM-NS) were generated as previously described (12). The variable regions of the immunoglobulin genes expressed by the 763.74 hybridoma were amplified with a set of proprietary primers from cDNA generated from the hybridoma cells using an RT-PCR protocol and sequenced using a standard dye-terminator capillary sequencing method (Synbuild).

Humanization of the murine scFv 763.74(A) was performed by grafting its CDRs into the stable human framework rFW1.4 (13). To design the first humanized variant, six CDRs of the murine 763.74(A) were combined with the unmodified FWRs of the rFW1.4. The sequence of the framework, referred here as rFW1.4, is as follows: EIVMTQSPSTLSASVGDRVIITC *CDRL1* WYQQKPGKAPKLLIY *CDRL2* GVPSRFSGSGSGTEFTLTISSLQPDDFATYYC *CDRL3* FGQGTKLTVLG (GGGGSGGGGSGGGGSGGGGS) EVQLVESGGGLVQPGGSLRLSCTASG *CDRH1* WVRQAPGKGLEWVG *CDRH2* RFTISRDTSKNTVYLQMNSLRAEDTAVYYCAR *CDRH3* WGQGTLVTVSS. Asterisks separate the amino acid sequence of the FWRs from the CDR sequences. A polypeptide linker consisting of (Gly₄Ser)₄ was used to join the V_L and V_H chains and shown in round brackets. The scFv 763.74(A) sequence and the human framework rFW1.4 were aligned, and 24 critical murine amino acids were identified as described (14). Twenty-six humanized variants were designed by introducing the identified murine amino acid residues back into the human FWRs. Some variants were generated by shuffling humanized heavy and light chains. The numbers of “back to mouse” mutations for all humanized heavy- and light-chain variants are shown in Supplementary Table S1. The positions of murine amino acids are shown for the four best selected humanized scFvs (h763.74 #2, h763.74 #3, h763.74 #4, and h763.74 #5). DNA sequences of humanized scFvs were optimized for Escherichia coli codon usage, synthesized by DNA2.0, and cloned into the expression vector with the isopropyl b-d-1-thiogalactopyranoside (IPTG)–inducible T7 promoter by DNA2.0. The expression vector had a His(6) fusion tag on the carboxyl terminus. A His(6) fusion tag allowed detection of humanized scFvs in binding studies. Each DNA synthesis at DNA2.0 included a stringent quality control process with 100% accuracy. In addition, the genes encoding for scFvs were confirmed by sequencing at Microsynth AG.

E. coli BL21-DE3 (100 μL, Invitrogen, C600003) transformed with the corresponding expression plasmids (5 ng) containing His-tag (scFv h763.74 #2, h763.74 #3, h763.74 #4, or h763.74 #5) were grown at 37°C in Luria Broth (LB) medium (VWR Chemicals) containing kanamycin (30 μg/mL; AppliChem GmbH). Protein expression was initiated by addition of 1 mmol/L isopropyl 1-thio-β-d-galactopyranoside (AppliChem) at an absorbance (A600) of 1. Four hours after induction, E. coli cells were harvested and disrupted by sonication. Inclusion bodies were isolated by repeated washing with 10 mmol/L Tris-HCl (AppliChem) and 150 mmol/L NaCl (AppliChem, pH 7.3) containing 0.5% N,N-dimethyldodecylamine N-oxide (Sigma-Aldrich) and centrifugated and solubilized at a concentration of 10 mg/mL in the presence of 6 mol/L guanidine HCl (AppliChem), 100 mmol/L Tris-HCl, and 1 mmol/L EDTA (Sigma-Aldrich). Solubilized inclusion bodies were reduced by adding 20 mmol/L dithiothreitol (Sigma-Aldrich). Refolding was performed in refolding buffer (4 mol/L urea, 50 mmol/L glycine, 2 mmol/L cysteine, 2 mmol/L cysteine, pH 10.0, all chemicals were purchased from AppliChem) overnight at room temperature. After up-concentration and buffer exchange using tangential flow filtration with a 10 kDa cutoff (Centramate T-Series Cassette OSO010T12), scFvs were purified using hydrophobic interaction chromatography followed by size-exclusion chromatography (SEC), as described below. ScFv 763.74(B) and h763.74 #5 were expressed in EXPI293 transient expression system (Thermo Fisher) following the manufacturer's protocol. The Expi293 cells were transfected and incubated for 4 days in Expi293 medium (Thermo Fisher) at 37°C, 125 rpm, and 8% CO₂ atmosphere (5% was also acceptable) with 80% humidity. Cells were harvested by centrifugation at 2,000 × g for 15 minutes. Supernatant was collected and then filtered through a 0.22-μm filter, followed by purification on a Ni-Penta affinity column (Marvelgent Biosciences).

Refolded scFvs were purified by hydrophobic interaction chromatography on an AEKTA Purifier system (GE Healthcare), which was operated with Unicorn software, with a Phenyl Sepharose 6 FF (low sub) XK26/60 column (GE Healthcare). The column was equilibrated with the column volumes of 20 mmol/L Na-Phosphate, 50 mmol/L NaCl (AppliChem), and 1 mol/L (NH₄)₂SO₄ (Merck), pH 7.0, at 5 mL/min flow rate. Humanized scFv solution was mixed with (NH₄)₂SO₄ to a final concentration of 1.0 mol/L. This solution was loaded on a Phenyl Sepharose 6 FF column at 0.2 mL/min flow rate. After washing with buffer 20 mmol/L Na-Phosphate, 50 mmol/L NaCl, and 1 mol/L (NH₄)₂SO₄, pH 7.0, the bound scFv was eluted with 100% buffer 20 mmol/L sodium phosphate and 50 mmol/L NaCl, pH 7.0, at 5 mL/min flow rate. Fractions containing scFvs were pooled and concentrated with a Vivaspin-20 concentrator (Sartorius) using the Multifuge X3R (Thermo Fisher) at 3,800 × g.

The polishing step was conducted by SEC on an AEKTA Pure system (GE Healthcare) with a HiLoad 26/60 Superdex 75 prep grade column (GE Healthcare). The column was equilibrated with three column volumes of PBS, pH 7.4 (Gibco), at a flow rate of 2.6 mL/min. The scFv sample of 10 mL was fractionated with a HiLoad 26/60 Superdex 75 column at 2.6 mL/min flow rate using a mobile phase buffer PBS (pH 7.4) to separate dimers from monomers. Fractions containing monomeric scFvs were analyzed by SDS-PAGE electrophoresis using NuPAGE 12% Bis-Tris Gels (Novex/Invitrogen), and absorbance of samples was measured at 280 and 320 nm using a BioSpectrometer basic. ScFv 763.74(B) and h763.74 #5, labeled with His tag in C terminus, were purified by Ni-Penta affinity column (Marvelgent Biosciences) and eluted with an imidazole gradient. Protein identity and purity were determined by SDS-PAGE gel.

For binding studies, soluble scFvs were used at 5 μg/mL. For the titration studies, soluble scFvs were used at the specific concentrations reported in the figures. ScFvs were detected either by protein L-biotin (5 mg/mL used at 1:3000; PI-29997, Pierce, Thermo Scientific) and streptavidin-phycoerythrin (PE; used at 1:2,000; BD Pharmingen, cat. no. 55406) or by anti-Myc tag FITC conjugate (Abcam, ab1394). In some experiments, scFvs were detected by anti–His-biotin (QIAGEN, 34440, diluted 1:1,000) and SAV-PE. Tumor cell lines (MDA MB231, MDA-MB-468, M14, and SK-MEL-2; 0.2 × 10⁶ cells/sample) were incubated in the presence of 763.74 and control scFvs in PBS with 2% FCS and 5 mmol/L EDTA. Cells were acquired on a FACSAria III or BD LSRFortessa (BD Biosciences) instrument using the FACSDiva software. The EC₅₀ was calculated with GraphPad Prism.

Humanized scFvs were formulated in PBS (pH 7.2; Gibco) at 1 mg/mL. After 48 hours of storage at 4°C or 37°C, samples were inspected visually, and protein concentration was measured at 280 and 320 nm using a BioSpectrometer basic (Eppendorf). The following equation was used to calculate protein concentration: (A280 – A320) x Dilution factor x Molecular weight/Extinction coefficient equal to scFv concentration in mg/mL. Samples were analyzed by SEC)-high-performance liquid chromatography (HPLC) to determine the percentage of monomers, dimers, and high molecular weight oligomers in relation to total peak area. A Tosoh TSKgel G2000 SWXL column, phase diol, L × I.D. 30 cm × 7.8 mm, 5 μm particle size (Sigma-Aldrich, 08540) was used for size-exclusion chromatography. Five microliters of scFvs at 1 mg/mL were loaded. The mobile phase was PBS (pH 7.2).

The monomer purity of the samples was analyzed using SEC-HPLC. The analysis was performed using an Ultimate 3000 HPLC system (Dionex) and Chromeleon software and UV detection at 280 nm absorbance, TOSOH TSKgel G2000 SWXL column (Sigma-Aldrich), and corresponding Guard column. The column was washed with two column volumes of degassed MilliQ water and equilibrated with two column volumes of degassed PBS (pH 7.2) at a flow rate of 0.3 mL/min. Five microliters of scFvs were injected and separated using a mobile phase containing PBS (pH 7.2) at a flow of 0.5 mL/min.

Nanoscale differential scanning fluorometry (nanoDSF) was performed with the scFvs 763.74(B) and h763.74 (20 μmol/L in PBS) using a NanoTemper Prometheus NT.48. Fluorescence emission at 350 and 330 nm was collected from 25°C to 95°C at a ramp rate of 1°C/min. Analysis of melting temperature (T_m) was performed using the PR. ThermControl Software (NanoTemper, https://nanotempertech.com/prometheus-software).

Retroviral supernatants were prepared by transient transfection of 293T cells and used to transduce T cells (15). Buffy coats from healthy volunteer blood donors were purchased from the Gulf Coast Regional Blood Center. Peripheral blood mononuclear cells (PBMC) were isolated by Lymphoprep (Accurate Chemical and Scientific Corporation) density-gradient centrifugation. T cells isolated from PBMCs were cultured in complete T-cell medium, consisting of 45% Click's medium (Irvine Scientific), 45% RPMI-1640 (Hyclone), 10% FBS (Hyclone), 1% l-glutamine (Gibco), and 1% penicillin/streptomycin (Gibco). T cells were activated, transduced, and expanded in complete medium with IL7 (10 ng/mL, PeproTech) and IL15 (5 ng/mL, PeproTech) as previously reported (15).

Mouse experiments using a melanoma model were performed in accordance with University of North Carolina (UNC) Animal Husbandry and Institutional Animal Care and Use Committee (IACUC) guidelines and were approved by UNC IACUC (ID: 17029). Mouse experiments using the glioblastoma model were performed following directives of Fondazione IRCCS Istituto Neurologico Carlo Besta in Milan in accordance with the Italian Principle of Laboratory Animal Care (D. Lgs. 26/2014) and European Communities Council Directives (86/609/EEC and 2010/63/UE). For the melanoma model, female and male NSG mice (7–9 weeks of age, obtained from the UNC Animal Core) were injected s.c. with 0.5 × 10⁶ eGFP-FFluc–labeled WM115 tumor cells. Seven days after tumor cell injection (day 0), mice were infused i.v. with 5 × 10⁶ CAR-T cells. Melanoma tumor cell growth was monitored weekly with caliper measurement for s.c. tumors and by bioluminescence (BLI; total flux, photons/second) using the IVIS kinetic in vivo imaging system (PerkinElmer). Mice were sacrificed according to UNC guidelines for tumor growth or occurrence of sign of discomfort. At sacrifice, peripheral blood was collected from heart in 1.7 mL tubes (GeneMate) with 20 μL of 0.5 mol/L EDTA (Corning). Spleen and liver were harvested and smashed on cell strainers and washed with 2 mL of PBS. Peripheral blood, spleen, and liver were analyzed to detect the presence of human T cells using the following antibodies: CD3 (APC-H7, clone SK7), CD45 (APC, clone 2D1), and PD-1 (PE/Cy7, clone EH12.1) from BD Biosciences, and CAR-specific anti-idiotype by flow cytometry, described below in “Immunophenotyping,” using CountBright absolute counting beads (Invitrogen). For the glioblastoma model, antitumor activity of CAR-T cells was evaluated in CD1 nude mice (CD1-Foxn1^nu purchased from Charles River Laboratories) engrafted with GBM-NS. Five- to 6-week-old mice were injected intracaudate nucleus (i.c.) with 0.1 × 10⁶ GBM-NS in 2 μL 1X PBS. The coordinates, with respect to the bregma, were 0.7 mm post, 3 mm left lateral, 3.5 mm deep, and within the nucleus caudatum. On day 15 after tumor cell injection, CAR-T cells were injected i.c. in 5 μL PBS 1X using the same tumor coordinates. Glioblastoma xenograft mice underwent MRI. For survival studies, mice were monitored 3 times a week and euthanized when signs of discomfort appeared in accordance with the institutional guidelines.

Glioblastoma xenografts were explanted at different time points after CAR-T cell injection. Tumor masses were cut into small pieces using a scalpel. The tissue pieces were enzymatically and mechanically digested, using the Human Tumor Dissociation Kit (Miltenyi Biotec) in combination with the gentleMACS Dissociator (Miltenyi Biotec). First, tumor tissues were transferred in C-Tubes (Miltenyi Biotec) containing the digestion enzyme mix of the kit, and the gentleMACS Dissociator was used for the mechanical dissociation using h_tumor_1 program and h_tumor_2 program. The dissociated tissue was filtered using Pre-Separation Filters (30 μm, Miltenyi Biotec), and a cell suspension was ready for flow cytometry analysis.

T cells were stained with antibodies against CD3 (APC-H7, clone SK7), CD45RA (PE, clone HI100), CCR7 (FITC, clone 150503), CTLA-4 (BV421, clone BNI3), PD-1 (PE-Cy7, clone EH12.1), LAG3 (PE, clone T47-530), TIM3 (BV711, clone 7D3), and CD45 (APC, clone 2D1) from BD Biosciences. Anti-CD45 (PerCP, clone REA747) and anti-CD69 (APC, clone REA824) were obtained from REAffinity by Miltenyi Biotec. Tumor cells were stained with antibodies against CD276 (BV421, clone 7-517) from BD Biosciences and with anti-763.74 [anti-CSPG4 (chondroitin sulfate proteoglycan 4); kindly provided by Dr. Ferrone (Massachusetts General Hospital)], followed by the staining with a secondary rat anti-Mouse (PE, clone X56) from BD Biosciences. The expression of the 763.74(A) and 763.74(B) CARs and the anti-CD19 CAR was assessed using specific anti-idiotypic antibodies [MK2-23 mAb for scFv 763.74(A) and 763.74(B) and 233-4A mAb for anti-CD19 CAR; kindly provided by Dr. Ferrone], followed by the staining with a secondary rat anti-Mouse antibody (PE, clone X56) from BD Biosciences. The expression of the humanized 763.74 CARs was assessed using a biotinylated monoclonal anti-rFW1.4 antibody (generated and biotinylated internally at Cell Medica) followed by the staining with streptavidin protein (PE-conjugated) from Invitrogen (catalog number 12-4317-87). Data acquisition was performed on BD LSRFortessa or Canto II flow cytometer using the BD FACSDiva software or on a MACSQuant (Miltenyi Biotec). Data analysis was performed with the FlowJo software (Version 9 or 10) or FlowLogic software (Version 7.2, Miltenyi Biotec).

T cells expressing the CAR fused with GFP were fixed with Cytofix buffer (BD Biosciences), stained with 4′,6-diamidino-2-phenylindole (DAPI; Invitrogen) according to manufacturer's protocol, washed with PBS, and mounted on glass-bottom microwell dishes (MatTek corporation). Data acquisition was performed on LSM700 Zeiss laser scanning confocal microscope (objective lens 63X/1.4 Plan Apo Oil, pixel size 0.07 μm, pinhole size 1 AU) using ZEN software (ZEISS Microscopy). All groups of images were acquired using the same settings. Data analysis was performed with Fiji software. The samples were scored by drawing region of interests and calculating the pixel mean fluorescence intensity (MFI). The coefficient of variation was calculated as SD/average pixel MFI.

For spontaneous IFNγ release assays, 1 × 10⁶ CAR-T cells were plated in 24-well plates in 2 mL of complete T-cell media without cytokines. T cells (2 × 10⁴ cells/well) were cocultured with tumor cell lines [M14-WT (wild-type) or WM115, 10⁵ cells/well in 24-well plates], in complete medium, in the absence of cytokines [effector (E):target (T) = 1:5]. After 5 days of culture, cells were harvested and stained with CD3 (APC-H7, clone SK7 from BD Biosciences) and CD276 (BV421, clone 7-517 from BD Biosciences) mAbs to detect T cells and tumor cells, respectively (16). Surface staining was performed in 200 μL of PBS for 20 minutes at 4°C. Residual tumor cells in culture were enumerated by flow cytometry. Data acquisition was performed on BD LSRFortessa or Canto II flow cytometer using the BD FACSDiva software. Data analysis was performed with the FlowJo software (Version 9 or 10). Culture supernatants were harvested after 24 hours of culture, and IFNγ and IL2 concentration was measured using the DuoSet Human IFNγ and IL2 ELISA Kit (R&D Systems). For the detection of IFNγ, 10 μL of culture supernatants was diluted with 90 μL of reagent diluent prepared according to R&D Systems protocol. For the detection of IL2, 100 μL of culture supernatants was used without any dilution. Data acquisition was performed on a Synergy2 microplate reader (BioTek) using Gen5 software (BioTek). IFNγ and IL2 concentration was calculated using Microsoft Excel. GBM-NS were plated at 5 × 10⁵ cells in 24-well plates with T cells at E:T ratio of 1:5 in GBM-NS medium without serum and in the presence of 1X B27 supplement (Thermo Fisher). T cells were maintained in GBM-NS medium for 3 days before plating the cocultures (12). GBM-NS and T cells were collected at different time points following 2, 4, 6, and 24 hours, and residual tumor cells and T cells were measured by flow cytometry based on CSPG4 and CD45 expression, respectively. The activation of CAR-T cells was measured by evaluating the expression of CD69. Data acquisition was performed on a MACSQuant (Miltenyi Biotec). Data analysis was performed with the FlowLogic software (Version 7.2, Miltenyi Biotec).

To generate the three-dimensional (3-D) conformation of scFvs, we initially considered the primary sequence of scFv and performed BLAST search against the Research Collaboratory for Structural Bioinformatics (RCSB) database (https://www.rcsb.org/pages/about-us/index) to identify homologous template structures with high sequence similarity. BLASTp (protein BLAST) analysis identified scFv 1696 (ref. 17; resolution: 2.70 Å; sequence identify: 70.51%; PDB ID: 1JP5), anthrax-neutralizing single-chain antibody 14b7 (resolution: 1.30 Å; sequence identify: 69.92%; PDB ID: 3ESU), the omalizumab scFv (resolution: 2.30 Å; sequence identify: 72.08%; PDB ID: 6TCS), and the single-chain disulfide-stabilized antibody variable fragments derived from the G6-Fab (resolution: 2.40 Å; sequence identify: 73.50%; PDB ID: 3AUV) as potential templates of scFv 763.74(B), FMC63, h763.74 #2, and h763.74 #5 for homology modeling. We generated 40 models using Modeller-9v19 (https://salilab.org/modeller; ref. 18) and selected the structure with least molecular objective function score as the representative conformation of the scFv. Because steric clashes are common in modeled and low-resolution structures, we employed Chiron (https://dokhlab.med.psu.edu/chiron; ref. 19) to optimize the structure of the scFv. Chiron resolves atomic clashes by performing short-DMD (Discrete Molecular Dynamics) simulations (20–22) on protein structure with minimal or no perturbation to the backbone. After the high energy caused by atomic clashes is relaxed by Chiron, the scFv structure was subsequently considered for in silico mutagenesis studies using Eris molecular suite (https://dokhlab.med.psu.edu/eris; ref. 23). The Eris protocol induces mutations in protein and estimates free energies of mutant (ΔG_mut) and WT (ΔG_WT) conformations. Eris performs rapid side-chain repacking and backbone relaxation around the mutated site using Monte-Carlo algorithm and subsequently evaluates ΔG_WT and ΔG_mut using Medusa force field (23, 24). The Eris algorithm then computes change in free energy of protein upon mutation by employing the following formula: ΔΔG_mut = ΔG_mut − ΔG_WT. We evaluated ΔΔG_mut values to estimate the stabilizing (ΔΔG_mut < 0) or destabilizing (ΔΔG_mut > 0) mutations. Eris has been extensively validated and used in designing novel proteins (25–27).

MRI was performed using a horizontal-bore preclinical scanner (BioSpec 70/20 USR, Bruker). The system has a magnetic field strength of 7 T (1H frequency 300 MHz) and a 20 cm bore diameter. The scanner was equipped with an actively shielded gradient system with integrated shims set up to second order. The maximum gradient amplitude was 440 mT/m. All acquisitions were carried out using a cross-coil configuration: a 72 mm linear birdcage coil was used for radiofrequency excitation, and a mouse brain surface coil received signal. Mice injected with GBM-NS and treated with CAR-T cells underwent high-resolution MRI at the following time points:+0, +14, +21, +28, +35, +44, +54, and +61 days after treatment. Animals were anesthetized with 1.5% to 2% isoflurane (AErrane, Baxter) vaporized in a mixture of medical air and oxygen (60: 40 vol:vol, flow rate 0.8 L/min) via a nose cone. Exhaled gas was actively vacuumed away via a built-in vacuum line. To detect the depth of anesthesia and the animal health condition during the study, the respiratory rate was monitored by a pneumatic sensor. The animals' temperature was kept at 36.5 ± 0.5°C by means of a warm-water circuit integrated into the animal holder. Mice were positioned on an animal bed equipped with a nose cone for gas anesthesia and a three-point fixation system (tooth bar and ear plugs). Mice underwent high-resolution MRI with the following protocol: a T2-weighted Rapid Acquisition with Reduced Echoes (RARE) sequence (TR = 3,360 ms, TE = 35 ms, in-plane resolution = 100 × 100 μm², slice thickness = 400 μm, 4 averages, total acquisition time of 5 minutes 36 seconds) and two T1-weighted RARE sequences (TR = 510 ms, TE = 8 ms, in-plane resolution = 78 × 78 μm², slice thickness = 400 μm, 6 averages, total acquisition time of 9 minutes 47 seconds) acquired before and after i.p. administration of 100 μL of Gadolinium-based contrast medium (Gadovist 1.0 mmol/mL, Bayer). All imaging sequences were acquired along the same coronal geometry (400-μm-thick continuous slices), with slice package posterior to olfactory bulb and anterior to cerebellum. T1-weighted images were visually inspected, and every slice was evaluated in terms of contrast agent–induced signal enhancement, which was interpreted as being due to a blood–brain barrier lesion.

Data were summarized as the mean ± SD. The Student t test or two-way ANOVA was used to determine statistically significant differences between treatment groups, with Bonferroni's correction for multiple comparisons when appropriate (Prism 6: GraphPad Software). Survival analysis was performed using the Kaplan–Meier method, and the Mantel–Cox log-rank test was applied (Prism 6: GraphPad Software). All P values less than 0.05 were considered statistically significant.

We have previously reported on the generation of a CAR targeting the CSPG4 (12, 28). This CAR was obtained using the scFv 763.74(A) from the 763.74 murine mAb. T cells expressing CSPG4-specific CAR show antitumor effects in multiple tumors expressing CSPG4 (12, 28). However, T cells expressing the scFv 763.74(A) CAR encoding either CD28 or 4-1BB costimulatory endodomains showed release of IFNγ in the absence of antigen stimulation (Fig. 1A and B), a phenomenon defined as CAR tonic signaling (6). Spontaneous IFNγ release by T cells expressing the scFv 763.74(A) CARs was strictly dependent on CAR signaling because mutations of the tyrosine of the immunoreceptor tyrosine-based activation motifs (ITAM) of the CAR-CD3ζ chain (Fig. 1C), which prevent tyrosine phosphorylation, completely abrogated the spontaneous IFNγ release (Fig. 1D and E). To study the distribution of CAR molecules on the cell surface of T cells, we generated scFv 763.74(A) CARs in which the CD3ζ chain of the CAR was fused at COOH terminal with GFP. Using confocal microscopy imaging, we found that scFv 763.74(A) CARs form membrane clusters in the absence of CAR cross-linking, likely indicating self-aggregation of CAR molecules (Fig. 1F). The sequence of the scFv 763.74(A) was obtained from an early passage of the hybridoma 763.74 secreting the murine IgG1 mAb that recognizes a peptide epitope of human CSPG4 (29). It is well-established that culture passages of hybridomas affect their growth rate and the yield of the secreted antibody (30). It has also been described that upon culture passage, amino acid substitutions may occur in both CDRs and FWRs in subclones derived from the hybridoma (31). In light of this possibility, we sequenced the V_L and V_H domains of a late passage of the 763.74 hybridoma. We obtained two V_L and V_H sequences in which amino acid substitutions were identified in the FWRs (FWR1 and FWR3) of both V_L and V_H (Supplementary Fig. S1A). We assembled a new scFv called scFv 763.74(B), generated new scFv 763.74(B) CARs, and compared them side by side with scFv 763.74(A) CARs for evidence of tonic signaling. All CARs were equally expressed in T cells (Fig. 2A; Supplementary Fig. S1B), and CAR-T cells equally expanded in vitro (Supplementary Fig. S1C). However, T cells expressing the scFv 763.74(B) CAR encoding either CD28 or 4-1BB costimulatory endodomains did not show spontaneous release of IFNγ (Fig. 2B). We also observed a more homogeneous distribution of the scFv 763.74(B) CARs on the membrane of T cells using confocal microscopy of GFP-tagged CARs (Fig. 2C). We quantified the relative variability of GFP-tagged CAR expression by calculating the coefficient of variation. The distribution of scFv 763.74(A) was significantly more variable than scFv 763.74(B) (P < 0.0001), regardless of the type of costimulatory molecule (Fig. 2D). Cross-linking of CARs expressed in T cells mediated by an anti-idiotypic mAb (MK2-23 mAb) caused significant cluster formation of CAR molecules, regardless of the type of scFv expressed, further indicating that clusters identified by confocal microscopy reflected the formation of CAR aggregates (Supplementary Fig. S1D). Phenotypic analysis of T cells expressing scFv 763.74(A) CARs or scFv 763.74(B) CARs did not show differences in the expression of memory and exhaustion markers (Supplementary Fig. S1E and S1F), suggesting that tonic signaling may not induce an exhaustion phenotype identified by phenotypic markers during the 10 to 14 days of culture usually required to manufacture CAR-T cells for clinical use (32, 33). Overall, these data indicated that amino acid substitutions within the FWRs of a scFv are sufficient in causing self-aggregation of the scFv in the CAR format and tonic signaling in T cells.

To study if differences in amino acids between the scFv 763.74(A) and scFv 763.74(B) cause destabilization of the scFv, we generated the 3-D conformation of the scFv 763.74(B) through homology modeling and optimized the structure for in silico mutagenesis (Fig. 3A). We employed the Eris tool to delineate the effect of FWR mutations, L3K, T5S, A9S, E83Q, I123V, Q124K, V126K, Q127E, and L230V on the structure of the scFv 763.74(B). Eris estimated the ΔΔG_mut for the aforementioned mutations as 2.41, 1.26, 2.29, 0.47, 0.79, 1.87, 4.58, 2.70, and 0.67 kcal/mol, respectively (Fig. 3B). Specifically, the mutations destabilized the scFv 763.74(A) structure (ΔΔG_mut > 0) and subsequently affected the spontaneous CAR aggregation. To cross-validate the structural conformation of scFv 763.74(B), we performed Eris analysis to identify stabilizing mutations (ΔΔG_mut < 0) at the FWR-mutated sites. Our analysis indicated that mutations, such as E83L, E83I, T5M, and Q124M, resulted in negative ΔΔG_mut, notifying their potential stabilizing capability (Supplementary Fig. S2). We also identified that the residues, such as I123 and Q127, were critical for the stability of the scFv 763.74(B) structure. The substitution of any of the other 19 amino acids at I123 or Q127 positions destabilized the scFv 763.74(B) structure (ΔΔG > 0) and thereby affected the protein aggregation (Supplementary Fig. S2). Overall, these data indicated that amino acid substitutions within the FRWs of a scFv destabilize the protein and cause self-aggregation of the scFv.

To validate the computational model, we selected the murine scFv FMC.63 (recognizes the human CD19), which is known to lack tonic signaling in the CD19-specific CAR. We generated the 3-D conformation of the scFv FMC.63 through homology modeling and optimized the structure for in silico mutagenesis (Fig. 3C). We then employed the Eris tool to predict destabilizing mutations of this scFv and identified three sets of mutations (Q8W|C25R|S28P, Q136W|P139L|Q146P, and L148Y|V150W|C152W) within the FWRs that destabilized the FMC.63 scFv (Fig. 3D; Supplementary Fig. S3A). Eris estimated the ΔΔG_mut for the three mutations as 95.3, 70.6, and 118.3 kcal/mol, respectively. We generated CARs containing these three predicted destabilized scFvs encoding either CD28 or 4-1BB [FMC.63(m#1), FMC.63(m#2), and FMC.63(m#3)]. Upon transduction of T cells, only the FMC.63(m#2) CAR was detectable by flow cytometry (Fig. 3E; Supplementary Fig. S3B and S3C). We compared T cells expressing the FMC.63(m#2) CARs side by side with T cells expressing the FMC.63 CARs for evidence of tonic signaling. The destabilized FMC.63(m#2) CAR showed spontaneous production of IFNγ, especially when encoding 4-1BB (Fig. 3F). Overall, these data validated the computational model's ability to predict scFv stability, supporting our finding that destabilizing mutations of the FWRs are sufficient in causing self-aggregation of the scFv in the CAR format and tonic signaling in T cells.

To determine if the amino acid substitutions within the FWRs of the scFv affected the antitumor activity of CAR-T cells, we used the melanoma cell lines WM115 (CSPG4⁺) and M14 (CSPG4⁻; Supplementary Fig. S4A). Paralleling the spontaneous release of IFNγ, T cells expressing the 763.74(A) CAR with 4-1BB showed compromised capacity to eliminate tumor cells in vitro (residual tumor cells 43.6% ± 26.0%) after a 4-day coculture of CAR-T cells and tumor cells at the ratio 1:5 (Fig. 4A and B). In contrast, CD28 costimulation seemed to allow complete tumor elimination for both 763.74(A) and 763.74(B) CARs (residual tumor cells 2.2% ± 2.7% and 2.9% ± 2.6%, respectively). T cells expressing the 763.74(B) CAR with 4-1BB costimulation showed improved antitumor effects compared with T cells expressing the 763.74(A) CAR with 4-1BB, but did not completely eliminate the tumor cells (residual tumor cells 14.1% ± 8.0% and 43.6% ± 26.0%, respectively; Fig. 4A and B). CAR-T cells did not eliminate the melanoma cell line M14 (lacks CSPG4 expression), indicating that antigen specificity was not affected by amino acid substitutions within the FWRs. Only T cells expressing the 763.74(B) CARs consistently released detectable amounts of IFNγ in the culture supernatant with WM115 tumor cells (Supplementary Fig. S4B). The superior antitumor effects of 763.74(B) CAR-T cells were more evident in vivo using the eGFP-FFluc WM115 xenogeneic NSG mouse model (Fig. 4C). T cells expressing the 763.74(B) CAR with CD28 exhibited the most prominent antitumor effects measured as both tumor bioluminescence (Fig. 4D; Supplementary Fig. S4C) and tumor size (Fig. 4E). Enhanced functions of CAR-T cells expressing the 763.74(B) CAR were confirmed in our previously described glioblastoma tumor model in which nude mice were engrafted in the brain with primary GBM-NS cells and treated via intratumoral inoculation of CAR-T cells (Fig. 5A; ref. 12). Tumor engraftment and progression were monitored by MRI. As previously described (12), we observed rapid tumor progression in mice treated with control T cells or T cells expressing the 763.74(A) CAR encoding CD28, and in these animals, tumor masses occupied whole hemispheres and infiltrated the contralateral one (Fig. 5B and C). Mice treated with T cells expressing the 763.74(B) CAR encoding CD28 showed the most evident antitumor effects, indicated by smaller and more circumscribed lesions (Fig. 5D), but tumor control was also observed in mice treated with T cells expressing either 763.74(A) or 763.74(B) CAR encoding 4-1BB, although antitumor effects were less dramatic (Fig. 5E and F). T cells expressing either 763.74(A) or 763.74(B) CARs encoding 4-1BB prolonged survival compared with mice treated with control T cells (P < 0.0001). However, T cells expressing the 763.74(B) CAR encoding CD28 were the most effective in prolonging survival [P < 0.0001 vs. CTR; P = 0.04 vs. 763.74(A) with 4-1BB, P = 0.01 vs. 763.74(B) with 4-1BB; Fig. 5G; ref. 12]. To further characterize the antitumor effects of T cells expressing the 763.74(B) CAR encoding CD28 in the glioblastoma model, we investigated the activation status of T cells immediately after intracranial infusion. We explanted the tumor masses 2, 4, 6, 12, 24, and 48 hours after CAR-T cell infusion and observed that T cells expressing the 763.74(B) CAR with CD28 upregulated CD69 within 2 hours after inoculation (42.5% ± 2.1%) and maintained high CD69 expression for 24 hours (22.5% ± 1.5% CD69⁺ T cells), which is consistent with previous reports indicating the fast activity of CAR-T cells encoding the CD28 endodomain (Fig. 5H; refs. 34, 35). Additional experiments in vitro further demonstrated the rapid antitumor effects of T cells expressing the 763.74(B) CAR with CD28 (Supplementary Fig. S5). Overall, these data indicated that the correction of the tonic signaling caused by the self-dimerization of the scFv improves the antitumor effects of CAR-T cells.

Humanization of murine-derived scFvs is a strategy proposed to prevent humoral and T-cell responses to the CAR (36, 37). We asked the question whether substituting the murine FWRs of the scFv with human FWRs could be used to abrogate the CAR tonic signaling. The sequence of 763.74(A) and the stable human framework rFW1.4 were aligned, and the critical amino acids were identified (14). One CDR graft with no mutations in the rFW1.4 sequence and 7 variants with up to 24 mutations in the critical regions were designed in the first round of engineering (Supplementary Fig. S6A). Humanized scFv variants alone (i.e., only extracellular domain of CARs) and the wild-type murine scFv were expressed in E. coli, refolded, and purified by SEC. The wild-type murine 763.74(A) scFv variant was not refoldable due to aggregation. Therefore, we were unable to purify it in a soluble form. This result further suggested that the murine 763.74(A) scFv was unstable. The humanized variant with no mutations in the rFW1.4 was expressed but did not bind CSPG4. Nearly all other humanized scFv variants were successfully expressed and able to bind CSPG4⁺ cells. In the next engineering rounds, humanized variants with the minimal number of 763.74(A) murine FW residues were further subjected to chain shuffling of the V_H and V_L. A total of 26 humanized scFvs were produced (Supplementary Table S1). Four humanized scFvs (h763.74 #2, h763.74 #3, h763.74 #4, and h763.74 #5) with the minimal number of murine FWR residues and retained CSPG4-binding activity were selected for further studies (Supplementary Fig. S6B). The scFvs were characterized for binding affinity (EC₅₀) to CSPG4-expressing cells by flow cytometry. The humanized scFvs h763.74 #2, h763.74 #3, h763.74 #4, and h763.74 #5 bound CSPG4 with comparable affinities (Supplementary Fig. S6C). In a separate experiment, we compared the scFvs h763.74 #5 and 763.74(B). The latter bound CSPG4 with a higher binding affinity than the scFv h763.74 #5 (P < 0.0001 and P = 0.0007 at 500 and 1,000 ng/mL, respectively; Supplementary Fig. S6D). We generated CARs with all four humanized scFv 763.74 with CD28 endodomain and performed in vitro coculture experiments with tumor cells. T cells expressing h763.74.CAR #2 and h763.74.CAR #5 showed a trend for better antitumor activity and higher production of IFNγ and IL2 in vitro and were selected for further studies (Supplementary Fig. S6E–S6G). To further characterize the two selected humanized scFvs h763.74.CAR #2 and h763.74.CAR #5 (Supplementary Fig. S7A), we performed storage stability studies with purified soluble scFvs. Proteins were prepared at 1 mg/mL and stored for 48 hours at 4°C and 37°C. After incubation, samples were analyzed by SEC to estimate the percentage of monomeric proteins. Under tested conditions, no detectable protein loss was observed, and the percentage of monomers remained above 91% to 97% (Supplementary Fig. S7B). This method demonstrated that humanized scFvs were monomeric under the tested conditions and did not dimerize or aggregate upon storage at 4°C and 37°C. We also measured the thermal unfolding temperature of h763.74 #5 scFv and the murine 763.74(B) scFv with nanoDSF and observed that both proteins remain folded up to 60°C (Supplementary Fig. S7C). We also employed DMD (20–22) to perform a computational molecular dynamics simulation to compare the stability. The free energy [763.74(B) < h763.74 #2 < h763.74 #5] and the average root mean square fluctuation [RMSF; 763.74(B) < h763.74 #2 < h763.74 #5] both suggested that 763.74(B) was more stable than h763.74 #2 and h763.74 #5 (Supplementary Fig. S7D). The h763.74.CAR #2 and h763.74.CAR #5 were equally expressed in T cells (Fig. 6A; Supplementary Fig. S8A), and T cells did not show spontaneous release of IFNγ (Fig. 6B). T cells expressing the h763.74.CAR #2 and h763.74.CAR #5 successfully controlled the CSPG4⁺ WM115 melanoma cell growth in vitro (residual tumor cells 11% ± 15% and 10% ± 17%, respectively), whereas they did not target the CSPG4⁻ M14 melanoma cell line, indicating that antigen specificity was maintained (Fig. 6C; Supplementary Fig. S8B). The antitumor activity of T cells expressing h763.74.CAR #2 and h763.74.CAR #5 was corroborated by specific production of IFNγ and IL2 (Supplementary Fig. S8C). Finally, we compared T cells expressing the h763.74.CAR #2 and h763.74.CAR #5 with T cells expressing the 763.74(B) CAR encoding the CD28 endodomain in the xenogeneic WM115 melanoma mouse model (Fig. 6D). T cells expressing the h763.74.CAR #2 and h763.74.CAR #5 showed potent antitumor activity (Fig. 6E; Supplementary Fig. S8D). Furthermore, T cells were detectable in the peripheral blood of treated mice at different time points (Supplementary Fig. S8E) and in the liver and spleen at the time of euthanasia (Fig. 6F), and T cells retained CAR expression (Fig. 6G). T cells expressing h763.74.CAR #2 and h763.74.CAR #5 did not show increased expression of PD-1 compared with T cells with the 763.74(B) CAR with CD28 (Supplementary Fig. S8F and S8G). Overall, these data showed that the humanization of an scFv can be used to eliminate tonic signaling of CAR molecules, maintaining specific antitumor effects.

Unstable murine scFvs cause self-aggregation, and murine sequences can theoretically induce immune responses in human subjects. Here, we demonstrated that the instability of the scFv was critical in causing tonic signaling when the scFv was assembled into the CAR format, and the CAR was expressed in T cells. We demonstrated that tonic signaling can be corrected either by the substitution of the amino acids, causing instability within the murine FWRs of the scFv, or by humanization of the FWRs. Correction of the tonic signaling enhanced the antitumor effects of the CAR-T cells.

TCR-mediated tonic signaling is a well-characterized homeostatic property of naïve T cells and plays a critical role in promoting their long-term persistence (38). However, TCR-mediated tonic signaling requires TCR engagement with self-peptides presented either in Class I or II, and is strictly confined to T cells located in lymphoid organs (39). In sharp contrast, CAR-mediated tonic signaling in T cells, in its strictest sense, refers to CAR signaling that is independent from any specific CAR engagement and defined as spontaneous release of cytokines, such as IFNγ (6). The event triggering CAR-mediated tonic signaling has been identified as the spontaneous aggregation of a sufficient number of CAR molecules, which leads to initiation of signaling (6). Here, we confirmed that tonic signaling was due to self-aggregation of CAR molecules and further demonstrated that the CAR-CD3ζ chain was exclusively responsible of the spontaneous cytokine release because loss of function of the CAR-CD3ζ chain completely abrogated the spontaneous release of IFNγ. Spontaneous cytokine release, rather than detection of cell surface markers associated with T-cell exhaustion, was the most robust and reliable assay to define the presence of CAR tonic signaling and reduced functionality.

The instability of synthetic scFvs has long been recognized to cause spontaneous aggregation, hindering the generation of soluble reagents (7, 9). Here, we demonstrated that an unstable scFv caused self-aggregation of CAR molecules, leading to CAR tonic signaling in T cells. The CSPG4-specific CAR we generated used a murine scFv from the 763.74 mAb. This scFv expressed in E. coli showed a tendency to aggregate, and the scFv was not producible in a soluble form. We have previously demonstrated that structural modeling and mutagenesis driven by computational protein design can be used to restore specificity of scFvs distorted by fusing V_L and V_H domains (40). Here, we showed that the instability of the scFv was the exclusive cause of self-aggregation and tonic signaling of the CAR. In general, the thermodynamic stability of a protein (ΔG) was crucial for its biological functionality. Amino acid mutations in proteins can disrupt important residue interactions and alter protein active sites and protein stability. These unstable mutant protein conformations are the underlying sources of numerous human disorders (41). Hence, quantifying the effect of mutations on protein structure is important to estimate the protein stability and their functionality. Here, we employed computational mutagenesis, facilitated by the Eris tool, to delineate the effect of FWR mutations on the structure of the scFv 763.74(A). Using a stable scFv like the scFv FMC.63, we validated the computational modeling, demonstrating how specific amino acid substitutions within the FWRs affected the stability of the scFvs which caused tonic signaling. Finally, we demonstrated that substituting the murine FWRs with a stable human FWRs, such as the framework rFW1.4, also corrected CAR tonic signaling without modifying the antigen specificity (13). The stable human framework rFW1.4 can accommodate CDRs of different origin and allows analyses of soluble scFv proteins to assess their physicochemical stability. This approach thus can be used to select scFvs with optimal physical and binding properties to design CARs. Taken together, the data suggested that stable and monomeric scFvs are applicable to generate CARs that avoid antigen-independent tonic signaling. ScFvs engineered with the human framework rFW1.4 may also have a low likelihood for immunogenicity and could potentially be repeatedly infused to the patients with full efficacy.

Conflicting data have been reported on the role of the costimulatory CD28 and 4-1BB endodomains in exacerbating or attenuating the CAR tonic signaling in T cells using CARs containing different hinge/spacer regions or different vectors/promoters (6, 42–44). Despite these conflicting data, it is possible to conclude that an intrinsic instability of the scFv cannot be corrected by the type of costimulation used or by modifying other components of the CAR structure. Indeed, CAR tonic signaling is frequently considered a good reason to abandon a specific scFv from further CAR development, despite the original antibody having excellent antigen specificity and affinity. Our data indicated that using CARs with the intracytoplasmic tail of the costimulatory endodomain CD28 and 4-1BB can cause tonic signaling. Stabilization of the scFv abrogated tonic signaling regardless of the costimulation used, indicating that an unstable scFv can be rescued by either mutations or humanization of the FWRs and that both CD28 and 4-1BB costimulation can be used.

CAR tonic signaling attributes negative effects in T cells and in particular causes poor antitumor effects due to rapid exhaustion (6). Our data in vitro and in vivo support the notion that CAR tonic signaling causes impaired antitumor activity. In our xenograft models, correction of the stability of the scFv and abrogation of the CAR tonic signaling significantly enhanced the antitumor effects of CAR-T cells encoding the CD28 endodomain. This observation is in line with previous work suggesting that 4-1BB slows antitumor effects and can be mechanistically related to the 4-1BB–mediated recruitment of phosphatases within the CAR synapse (34, 35). We propose that tonic signaling, defined as the spontaneous release of IFNγ, by CAR-T cells due to self-aggregation of the scFv should be considered as a distinct phenomenon from the spontaneous proliferation of CAR-T cells (43). The latter may be a positive attribute of the CAR-T cells that persist in the absence of an immediate antigen stimulation, especially in patients with solid tumors. Spontaneous proliferation and enhanced survival of CAR-T cells encoding 4-1BB may be due to sustained NF-κB signaling, rather than proximal CAR signaling, and this phenomenon requires further studies to be fully mechanistically defined (42, 45, 46).

Overall, we demonstrated that tonic signaling of a CAR was due to self-aggregation of an unstable scFv and that it could be abrogated by stabilization of the FWRs of scFv obtained by amino acid substitutions or humanization. Correction of the scFv stability and elimination of CAR tonic signaling enhanced the antitumor effects of CAR-T cells.

E. Landoni reports a patent for CSPG4.CAR pending. B. Savoldo reports grants from Cell Medica during the conduct of the study, as well as personal fees and other from Tessa Therapeutics (consultant) and grants from bluebird bio and NIH outside the submitted work. B. Kuhlman reports grants from NIH during the conduct of the study. G. Dotti reports grants from Cell Medica during the conduct of the study; grants from Bellicum Pharmaceuticals and bluebird bio and personal fees from Bellicum Pharmaceuticals, Catamaran, and Tessa Therapeutics outside the submitted work; as well as a patent for CSPG4.CAR pending. No disclosures were reported by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

E. Landoni: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. G. Fucá: Investigation, methodology. J. Wang: Software, formal analysis, investigation, visualization, methodology. V.R. Chirasani: Software, investigation. Z. Yao: Investigation, visualization, methodology. E. Dukhovlinova: Investigation. S. Ferrone: Conceptualization. B. Savoldo: Supervision. L.K. Hong: Investigation. P. Shou: Resources. S. Musio: Formal analysis, investigation, visualization, methodology. F. Padelli: Formal analysis, investigation, visualization. G. Finocchiaro: Supervision. M. Droste: Formal analysis, investigation, visualization. B. Kuhlman: Supervision, funding acquisition. A. Shamshiev: Conceptualization, formal analysis, methodology, writing–review and editing. S. Pellegatta: Conceptualization, resources, supervision, funding acquisition, writing–review and editing. N.V. Dokholyan: Conceptualization, supervision, writing–review and editing. G. Dotti: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing.

The Microscopy Services Laboratory, Department of Pathology and Laboratory Medicine, is supported in part by P30 CA016086 Cancer Center Core Support Grant to the UNC Lineberger Comprehensive Cancer Center. This work was supported by R01 CA193140 (G. Dotti), 1R35 GM134864 (N.V. Dokholyan), UL1 TR002014 (N.V. Dokholyan), R35 GM131923 (B. Kuhlman), RO1DE028172 (S. Ferrone), RO3CA239193 (S. Ferrone), and RO3CA216114 (S. Ferrone) from NIH; Passan Foundation (N.V. Dokholyan); U.S. Department of Defense W81XWH-16-1-0500 (S. Ferrone); and Il Fondo di Gio Onlus (S. Pellegatta). G. Dotti also received support from Cell Medica. Sequences of the 763.74 mAb and its humanized form have been submitted as provisional patent.

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