Shaping Functional Avidity of CAR T Cells: Affinity, Avidity, and Antigen Density That Regulate Response

This article is featured in Highlights of This Issue, p. 761

Cancer-specific adoptive immunotherapies, including chimeric antigen receptors (CARs), have been remarkably successful. This approach of redirecting T cells to efficiently eliminate tumor cells (1) has shown durable clinical responses (2, 3). Despite initial success, mainly in hematologic malignancies, the poor outcome or adverse effects in patients of other malignancies imply the need for better design of these methodologies to achieve optimal response. Several approaches, such as increasing affinity, were suggested; however, little is known about the mechanisms that lead to optimal response.

CARs are hybrid receptors that typically link an extracellular antigen-recognizing moiety to the intracellular activatory signaling domains. CARs can be used to target T cells to a tumor-associated antigens (TAA) independently of MHC (4), or to antigens that have poor specific T-cell response or low specific T cells abundance, such as certain tumor-associated peptide–MHC complexes (5, 6). The CAR antigen-recognizing moiety is usually generated by selection of high-affinity single-chain variable fragments (scFv); however, increased affinity does not necessarily increase efficacy (7, 8), and might lead to reduced specificity (9, 10). We previously compared a low-affinity TCR to a high-affinity CAR and described a window of affinity for optimal function, as the high-affinity CAR exhibited inferior responses (6).

Together with affinity, the receptor–antigen interaction is defined by avidity and functional avidity. Although the latter refers to the functional response of a T cell (e.g., cytotoxicity or cytokine secretion), the former refers to the general multimeric interactions between the T cell and its target cell (11). Whereas increasing affinity or avidity are generally considered to induce stronger response (12, 13), recent evidence indicates that high-affinity or high-avidity T cells have attenuated responses (14, 15), demonstrating a nonmonotonic (bell-shaped) behavior with a delimited range that induces maximal response.

Although the nonmonotonic effects of affinity, avidity, and the target antigen density have been described in TCRs (16–18), these effects remain largely understudied in CAR T cells. Here, we created an experimental system in which affinity, avidity, and antigen density are controllable, in order to investigate the significance of these properties in shaping functional avidity of CAR T cells, and to describe their precise combination that leads to improved functional outcome. We found that these properties affected the functional output both individually and combinatorically. We observed nonmonotonic behaviors of both affinity and antigen density, and noted the interrelation between avidity and antigen density, as avidity could counter the nonmonotonic behavior at high antigen densities. This novel information sheds new light on the importance of these properties when designing CARs for improved functional output.

Unless otherwise mentioned, cells were cultured in RPMI-1640 medium containing 10% fetal bovine serum (FBS), 100 units/mL penicillin, 100 μg/mL streptomycin, 0.5 mmol/L HEPES, 1 mmol/L sodium pyruvate, nonessential amino acids, and 2 mmol/L glutamine at 37°C in 5% CO₂.

Jurkat cells, peripheral blood mononuclear cells (PBMCs), and T cells were cultured in T-cell culture medium (T cells CM), which was additionally supplemented with 0.05 mmol/L 2-mercaptoethanol and the described units of IL2 at 37°C in 5% CO₂.

For peptide loading of APCs, cells were incubated with BioGro medium, which contained RPMI-1640 medium, 2% BioGro (Biological Industries), 100 units/mL penicillin, 100 μg/mL streptomycin, 0.5 mmol/L HEPES, 1 mmol/L sodium pyruvate, nonessential amino acids, and 2 mmol/L glutamine at 37°C in 5% CO₂

The HEK293 Phoenix-Ampho cells were maintained in Dulbecco's modified Eagle's medium and with 10% FBS, 100 units/mL penicillin, 100 μg/mL streptomycin, 0.5 mmol/L HEPES, 1 mmol/L sodium pyruvate, nonessential amino acids, and 2 mmol/L glutamine at 37°C in 5% CO₂.

Unless otherwise mentioned, all media and cell growth reagents were purchased from Thermo Fisher Scientific. Oligonucleotides were manufactured at HPLC quality by Sigma-Aldrich. All restriction enzymes were purchased from NEB. All peptides were ordered from LifeTein.

The following commercial antibodies were used in this work: anti-mouse Ig Kappa light chain (BioLegend), anti-human Kappa (IQP), anti-mouse Fab (Jackson), anti-HLA-A2 (clone BB7.2, Bio-Rad), anti-HLA-A,B,C (clone W6/36), anti-CD8 (BD Biosciences), anti-CD3 (clone SK7, BD Biosciences), anti-CD3 (clone OKT3, eBioscience), anti-CTLA4 (eBioscience), anti–PD-1 (eBioscience), anti-CD62L (eBioscience), anti-CD45RO (eBioscience), anti-CD45RA (eBioscience), anti-CD80 (eBioscience), anti-CD83 (eBioscience), anti-CD86 (eBioscience), anti-CD273 (eBioscience), anti-CD274 (eBioscience), anti-IFNγ (eBioscience), anti-IL2 (eBioscience), anti-TNFα (eBioscience), anti-CD107a (BioGems), anti-MIP-1β (R&D Systems), anti-IL10 (eBioscience), anti-biotin (BioLegend), and anti-CD19 (eBioscience).

The TCR-like antibodies described in the work were received from AdicetBio and are characterized within the framework of this research.

Biotinylated single-chain MHC (scMHC)–peptide complexes were produced by in vitro refolding of inclusion bodies produced in E. coli upon IPTG induction, as described previously (19).

Tetramerization of biotinylated scMHC–peptide complexes was done as follows: labeled streptavidin was slowly added at RT to biotinylated MHC, at a molar ratio of 4:1. Tetramers were kept at 4°C up to one week.

MHC monomers were generated using excess labeled streptavidin to allow binding at just one site of the streptavidin molecule.

Tyr–HLA-A2 and CMV–HLA-A2 pentamers were bought from ProImmune.

Total RNA of the harvested hybridomas was extracted with a QIAGEN miRNeasy kit, according to the manufacturer's procedure. RNA was reverse-transcribed to cDNA using the Verso Reverse Transcriptase Kit (Thermo Fisher Scientific). PCR amplification of heavy-chain and light-chain genes was performed with Kappa polymerase and primers specific to murine heavy and light genes, as described previously (20). PCR productions of amplified V_H–C_H1 and V_L–C_L from cDNA synthesis were gel purified and then sequenced.

Expression of reported TCR-like Fab constructs was done in a bacterial (BL21) IPTG-induced expression system, followed by purification and dialysis, as previously described (6).

For expression of Tyr–HLA-A2 targeting antibodies (described in this work) as Fab fragments, purified PCR products (described above) were cloned into pMAZ-IgL mammalian expression vector (21), using PCR-based insertion of Hisx6 tag and BssHII and XbaI restriction sites, as described in Supplementary Table S10. Fab constructs were expressed by transfection of the heavy-chain and light-chain constructs to Expi293 cells (Thermo Fisher Scientific) according to the manufacturer's protocol, or using the PEI transfection reagent. Fabs were produced from cell media using the Ni²⁺-NTA exchange column, as described previously (22). Aliquots were kept in −80°C.

The affinities of the Fabs to Tyr–HLA-A2 complex were measured and analyzed by a surface plasmon resonance (SPR) biosensor (ProteOn XPR36; Bio-Rad Laboratories). Biotinylated soluble Tyr–HLA-A2 monomers (ligand) were immobilized onto a streptavidin-coated NLC chip (Bio-Rad). Anti-Tyr–HLA-A2 Fab constructs were flowed over the flow cells in increasing concentrations from 30 to 500 nmol/L. Equilibrium binding was measured at a flow rate of 50 μL/min. Binding curves were fitted using Langmuir model for 1:1 binding stoichiometry or the Langmuir and mass transfer limitation model.

Peptides were ordered from LifeTein and dissolved in DMSO. For peptide loading, the EBV-transformed JY B-cell lymphoblastoid line or the T2 T-lymphoid cell line were used as APCs, which express the HLA-A2 MHC-I molecule. All cells were washed in serum-free medium and incubated overnight with the indicated peptides and concentrations in BioGro medium at 37°C. After pulsing, the cells were washed to remove free peptide.

Hybridoma supernatants and Fab constructs were analyzed by measuring their reactivity toward specific and nonspecific peptide–HLA complexes, using peptide-loaded APCs (as described above) or cancer cell lines. Hybridoma supernatants were additionally analyzed against HLA-A2 primary PBMCs. For FACS analysis, cells were collected and resuspended in PBS–BSA 0.1%. After washing, the cells were incubated for 30 minutes at 4°C with Fab antibodies (30 μg/mL) in 100 μL. After three washes, the cells were incubated for 60 minutes at 4°C with 100-μL secondary antibody anti-mouse Fab FITC or anti-HLA-A2 antibody (BB7.2). After washing, cells were analyzed at the LSRII flow cytometer.

For generating a chimeric receptor, anti-Tyr TCR-like Fab TA2 (23) was converted to scFv forms by connecting the carboxyl-terminus of the VL region and the amino-terminus of the VH region by a peptide linker. DNA constructs were ordered from GeneArt, Thermo Fisher Scientific. The scFv was connected via the carboxyl-terminus of the VH region to a CD28–CD3ζ chain construct, described previously (24). The TCR-like chimeric receptor DNA constructs were cloned into retroviral pBullet vector followed by IRES and the GFP gene, using the NcoI and HindIII restriction enzymes. For generating chimeric receptors at different affinities, the above-described Fab constructs were converted to scFv forms by connecting the carboxyl-terminus of the VL region and the amino-terminus of the VH region by a peptide linker, and the CAR scFv was replaced using either XbaI and BstEII or XbaI and BglII restriction enzymes.

Retroviral transduction was first calibrated on Jurkat cells, and then performed on primary T cells, using the Phoenix amphotropic packaging cells, as described previously (6), with minor modifications. Briefly, 2 × 10⁶ Phoenix amphotropic packaging cells were cultured in 10-cm culture plated for 24 hours at 37°C with 5% CO₂. The cells were transfected with the vector constructs along with Gag-Pol and envelope GalV constructs, using TransIT-LT1 transfection reagent (Mirus) or calcium phosphate precipitation kit (Invitrogen Thermo Fisher Scientific) according to the manufacturer's protocol. After culturing for 48 hours and replacement of medium, the viral supernatants were harvested. PBMCs were obtained from donors, after obtaining their informed consent according to the Technion IRB, using density-gradient centrifugation according to the Ficoll-Hypaque technique. All donors were HLA-A2 negative. PBMCs were activated 72 hours prior to transduction using the anti-CD3 antibody OKT3 at 30 ng/mL and human IL2 (600 U/mL). For retroviral transductions, retronectin-coated (Clontech/Takara) 24-well plates were seeded with the viral supernatant, centrifuged for 30 minutes, according to the manufacturer's protocol. PBMCs were then added at 1 × 10⁶ cells per 1 mL. After 24 hours at 37°C with 5% CO₂, half the culture medium was replaced with fresh medium. After further 48-hour culture period, flow cytometry analysis was performed by a Calibur flow cytometer (BD Biosciences) to determine transduction efficiency.

Five days after transduction, cells were collected, washed, and stained with anti-CD8 antigen-presenting cells (APCs) with PBS/0.5%BSA buffer. After wash, CD8-positive and GFP-positive PBMCs were sorted by an AriaIII flow cytometer (BD Biosciences). After 3 days, cells went a rapid expansion protocol as described previously (24). Briefly, 1.5 to 2 × 10⁵ sorted PBMCs were added to upright T-25 flasks containing 30 ng/mL OKT3 (eBioscience), 3,000 U/mL IL2, and 2.7 × 10⁷ allogeneic irradiated (5,000 Rad) PBMC feeder cells obtained by pooling PBMC from normal donors (obtained from the Blood Bank) in 10 mL of T cells CM and 10 mL of AIM V (Thermo Fisher Scientific) medium. On day 5, 10 mL of the medium was replaced by fresh AIM V medium with 3,000 U/mL IL2 (final concentration). The CAR T cells were expanded until day 12, and diluted as needed with AIM V and IL2 to keep the viable cell at density of 1 to 2 × 10⁶/mL. Retroviral transduction was confirmed by sequencing of genomic DNA purified from the sorted CAR T cells.

As tetramer dissociation assays are suitable for fast K_off of TCR:pMHC interactions, low K_off of CAR:pMHC interactions could not be detected in conventional assays. We therefore utilized a modified assay that relies upon two conventional assays (14, 47). Briefly, 10⁶ CAR-positive cells were blocked with unconjugated streptavidin, resuspended in 100-μL ice-cold FACS buffer, rested 30 minutes on ice, and subsequently incubated in FACS buffer for 30 minutes on ice in the presence of pMHC tetramer (20 μg/mL). Following incubation, cells were washed in ice-cold FACS buffer and subsequently resuspended in FACS buffer containing the blocking antibodies targeting Tyr–HLA-A2 (D11 clone) and HLA-A,B,C (W6/36 clone) at 50 μg/mL. At the indicated time points of incubation at 15°C (a temperature in which TCR internalization is not observed; ref. 47), 100-μL samples were taken and directly analyzed in the Calibur flow cytometer (BD Biosciences). Relative tetramer staining was normalized by subtracting median fluorescence intensities (MFI) of stained CAR-negative cells and dividing by the initial tetramer staining (t = 0). An exponential decay was fitted into the data and used for t_1/2 calculations. For the 16 nmol/L CAR, transfected HEK293 cells were analyzed. We note that the receptor qualities (such as K_off) should not vary between cells.

Cells were thawed two days before sorting and kept in T cells medium with 500 U/mL IL2. At the sorting day, cells were washed and incubated with PBS/0.5%BSA for 30 minutes in ice at density of 10⁷ cells/mL, stained with monomer Tyr–MHC complex bound to streptavidin–APC (Jackson) for 30 minutes in ice, and finally washed. GFP-positive and Tyr–MHC-positive cells were sorted by an AriaIII flow cytometer (BD Biosciences) to three equal populations by the strength of binding of Tyr–MHC. Sorted cells were washed and grown in T-cell CM with 600 U/mL IL2.

CAR expression quantification was done using the Quantum Simply Cellular kit (Bangs Laboratories).

For avidity measurements, cells were incubated with biotin-labeled Tyr–MHC pentamer (ProImmune), washed, and stained with streptavidin–PE (Jackson). Following washing, cells were analyzed at the Calibur flow cytometer (BD Biosciences).

Functional measurements of avidity-sorted CAR T cells were performed 2 days after sorting.

Peptide loading was analyzed using TCR-like Fabs, as described previously (16). As detection signal for antigen densities lower than 100 sites per cell were close to background signal, an extrapolation of antigen density at low peptide concentrations was calculated using a four-parameter logistic curve fitting.

Transduced CAR T cells and JY cells loaded with the mentioned peptides were added at 1 × 10⁵ cells/well at 1:1 E:T ratio (effector cell number was calculated as the number of live Tyr–MHC pentamer-positive cells). Cells were cultured in sterile 96-well plates in 100 μL of T-cell CM containing brefeldin A and monensine (eBioscience). Anti-CD107a-APC antibodies were added to designated wells. After 5.5 hours at 37°C with 5% CO₂, cells were fixed and permeabilized according to the manufacturer's protocol (Cytofix/Cytoperm Kit; BD Biosciences) and intracellular cytokines staining (ICS) was performed as described in the text. Anti-CD8 or anti-CD3 antibodies were used to gate effector cells from target JY cells. Finally, cells were washed and analyzed by an LSR II flow cytometer (BD Biosciences). The transduced T cells were gated by size and granulation (FCS/SSC), GFP-positive, and CD3/8-positive cells. Reactivity was assessed by measuring percentage of reactive cells, and the MFI of the reactive cells (reactive-MFI). For cases in which there were no reactive cells (less than 3% positive cells), reactive-MFI was defined as zero.

As the production of the different cytokines showed similar trends, a unified score was applied by scaling of the vast array of production level across the four cytokines: IFNγ, TNFα, IL2 and MIP-1β. The set of results for each cytokine was normalized to a scale between 0 and 1, and then the normalized values were averaged for all four cytokines. Formally, we scaled each cytokine's production level as follows:

where y_j is the cytokine production value at condition j (e.g., 4 nmol/L affinity with high expression level cocultured with target cells with 0 antigen density), and min(y_tot) and max(y_tot) are the minimal and maximal production values for the same cytokine across all measured conditions. We termed the average of the cytokines' scaled production values (⁠|{Z_{cy,j}}$|⁠) as cytokines production score (CPS).

CAR T-cell cytotoxicity was measured using a commercially available LDH Non-Radiactive Cytotoxicity Assay (Promega) according to the manufacturer's instructions. Target melanoma cell lines 1938, 526, and 501A (1 × 10⁴ cells/well) were coincubated with CAR-positive T cells for 6 hours, and the concentrations of LDH in supernatants were measured using the BioTek Epoch Spectrophotometer. The percent-specific cytotoxicity was calculated as follows:

Six- to 8-week-old female NSG mice (The Jackson Laboratories) were injected subcutaneously with 5 × 10⁶ 501A melanoma cells or with 2.5 × 10⁶ 526 melanoma cells in sterile 1× PBS. When the tumor volume reached between 50 and 100 mm³ and one day before T cells treatment (day 0), the animals were treated with i.p. injection of cyclophosphamide (200 mg/k). Twenty-four hours after cyclophosphamide injection (day 1), 5 × 10⁶ CAR T cells or PBMCs were injected intravenously. Tumor sizes were measured with calipers three times per week, and tumor volume (in mm³) was determined using the formula W²L/2, where W is tumor width and L is tumor length.

FACS data were analyzed using FCS Express 5 (De Novo). MFIs were calculated after subtraction of the background staining calculated from isotype-control staining (for primary conjugated antibodies) or from secondary antibody staining (for nonconjugated primary antibodies). Reactive MFI was measured in a similar fashion on positive cells, with a threshold of at least 3% positive cells. In case of less than 3%, the reactive MFI was determined as zero.

A standard principal component analysis (PCA) analysis was performed using the Perseus software (26). The reactive-MFI data matrix was arranged such that each CAR T-cell line (e.g., CAR T of 16 nmol/L affinity with high expression) was represented in rows and the averages of all markers (CD107a, IFNγ, TNFα, IL2, and MIP-1β) at all antigen densities were represented in columns. Untransduced PBMCs were not analyzed in the PCA. Following replacement of negative values by zeros, the matrices were log transformed, normalized, and then analyzed by PCA.

All statistical analyses of the described measurements (MFIs, reactive-MFIs, percent-reactive cells, CPS of reactive-MFI, CPS of percent-reactive and specific killing) were performed using The R Project for Statistical Computing (27). P values were corrected for false discovery rate using the BH method (28). According to the correction, for the CD107a and ICS analysis, P values under 0.021 were considered significant. Significant values are represented with asterisks as follows: *, 0.05–0.01; **, 0.01–0.001; ***, <0.001 adjusted values. A three-way ANOVA (CAR expression × CAR affinity × antigen density) was performed on reactive MFI (in the ICS and CD107a staining) or percent-reactive cells (in the ICS and CD107a staining) of the CAR T cells, using the aov function. Untransduced PBMCs were not included in the analysis. EC₅₀ were calculated by local polynomial fitting (29) of each set of CAR T-cell line data as a function of antigen density with second-degree polynomial and span 0.6. For two-way ANOVA calculation of the variance explained by affinity or CAR expression, we divided each effect sum of squares by the total sum of squares. Untransduced PBMCs were not included in the analysis.

Statistical analysis of cytotoxicity at 10:1 E:T ratio was performed using The R Project for Statistical Computing. A two-way ANOVA (CAR affinity × tumor cell line) was performed using the aov function, followed by TukeyHSD (Tukey Honest Significant Differences). Untransduced PBMCs were not included in the analysis.

Statistical analysis of in vivo antitumor activity was performed using The R Project for Statistical Computing. Loess smoothed lines and confidence intervals were plotted using R's ggplot2 package. For hypothesis testing that data were log transformed, mixed-model ANOVA was performed using the lmerTest package. Post hoc pairwise comparisons were calculated using the Tukey method.

To address the contribution of affinity, avidity, and antigen density to CAR T-cell functionality, we generated an experimental system in which all of these binding properties are controllable. First, we compiled a collection of B-cell hybridomas, which target the Tyrosinase_369–377-HLA-A2 TAA, in order to design CARs with varying affinities to the same antigen, based upon these antibodies. The hybridoma-derived secreted antibodies were tested for affinity, specificity, and cross-reactivity toward cancer cell lines that present the Tyr antigen (Supplementary Fig. S1B and Supplementary Table S1), or toward APCs that were loaded with either Tyr-similar peptides or other peptide targets (Supplementary Fig. S1A and Supplementary Tables S1 and S2). We selected several hybridomas, with affinities ranging between 4 and 300 nmol/L, and validated their stability, affinity, and specificity when expressed as Fab constructs (Supplementary Fig. S1C and S1D). The monovalent affinity of each Fab construct was tested by SPR assay and found to be similar to those of the whole IgG antibodies, ranging between 4 and 434 nmol/L (Supplementary Fig. S1E and Supplementary Table S3).

Next, CARs were constructed by converting the various Fab constructs into the scFv format and attaching them to CD28 transmembrane and intracellular signaling domains, and CD3ζ intracellular signaling domain (Fig. 1A). These CARs were transduced and expressed in CD8⁺ T cells using gamma retroviral vector technology, as described previously (6). As antibody-based CARs were documented to have lower specificity than their parent antibody (5) or reduced binding toward their target (6), we validated CAR-specific binding by measuring reactivity of the various CAR T cells toward different peptide–HLA-A2 complexes (Fig. 1B; Supplementary Fig. S2). Furthermore, tetramer dissociation assays confirmed that the CARs maintained the differences in binding kinetics as measured for the Fab constructs (Fig. 1C) and tetramer dissociation half time correlated with Fab K_off values (Fig. 1D).

To generate CAR T cells with distinct avidities, we sorted the transduced CAR T-cell populations by CAR expression level. CARs were sorted into low, medium, and high expression levels, according to binding of monomeric Tyr–HLA-A2 complexes (Fig. 1E). The avidity difference between the sorted populations was maintained 1 day (Fig. 1F) to 15 days after sorting (Fig. 1G). The number of CARs per cell (i.e., CAR expression, as measured by antibody-binding capacity of monomer Tyr–HLA-A2) correlated significantly with avidity (measured as the brightness of Tyr–HLA-A2 multimer binding; Fig. 1H). These two Tyr–HLA-A2 binding measurements (monomer and multimer binding) are referred hereafter as CAR expression and avidity, respectively. Average receptor numbers per cell were 70,628 ± 4496, 215,889 ± 10,703, and 584,465 ± 25,983 in the low-, medium-, and high-avidity populations, respectively. Additionally, scatter graphs of the low-, medium-, and high-avidity cells did not show any difference in cell size or granulation (Supplementary Fig. S2E); therefore, the difference in brightness between avidities is not due to changes in cell size or autofluorescence. In addition, surface staining with CD62L, CD45RO, and CD45RA showed little differences in memory phenotype between the different affinities or avidities (Supplementary Fig. S3A), and surface staining for PD1 and CTLA4 expression did not show any exhaustion phenotype (as reported for high-affinity and high-avidity CAR T cells; refs. 30–33; Supplementary Fig. S3B and S3C).

Finally, after separating CAR T cells by receptor expression level and by affinity, we used APCs as target cells, loaded them with different concentrations of the Tyr peptide, and quantified the specific antigen density per cell, using TCR-like antibodies as previously described (Fig. 1I; ref. 16). Antigen densities ranged up to several thousand antigen sites per cell, similar to previously described antigen densities of melanoma cell lines (23). To exclude any other peptide loading-dependent alternation in the target cells, we validated that different peptide concentrations did not affect any costimulatory or coinhibitory receptors (Supplementary Fig. S3D).

After establishing an experimental system in which affinity, receptor expression, and antigen density could be manipulated, we turned to investigate their role in T-cell function. We used CAR T cells with CARs at affinities of 4, 16, 35, or 434 nmol/L, sorted into low, medium, and high expression levels and cocultured them with target cells presenting different antigen densities. T-cell activation was then assessed by flow cytometry measurements of five functionality markers: CD107a for cell degranulation, and IFNγ, TNFα, IL2, and MIP-1β for cytokine production. Untransduced PBMCs or CAR T cells that were not exposed to APCs were used as controls, and T cells that were exposed to unloaded APCs or to APCs loaded with the MART1_126–135 peptide were used to measure specificity. For each examined functionality marker (i.e., CD107a or cytokines), we calculated the percentage of cells that responded (termed percent-reactive), and the magnitude of response of the responding cells (termed reactive-MFI). The different activation markers showed similar trends, evidenced by high correlations between these markers (Supplementary Fig. S4A). We defined a unified CPS to summarize the T-cell response of the four measured cytokines (see Materials and Methods). The averages of reactive MFIs of lytic activity (CD107a) and CPS are shown graphically in Fig. 2. They are ordered either by expression level in each individual CAR affinity (Fig. 2A and C) or by their avidity (Fig. 2B and D; see also Supplementary Fig. S4 for each cytokine's cartographic maps and separated graphs for each CAR affinity and hierarchical clustering; and Supplementary Fig. S5 for percent-reactive analysis; statistics are described in Supplementary Table S4). As expected, CAR T-cell response increased in an antigen dose–dependent manner; nonetheless, several interesting features were revealed. We observed nonmonotonic behaviors of both affinity and antigen density on CAR T-cell responses. At high antigen densities, T-cell responses were considerably diminished, resembling the response described for TCRs (16, 17), thus creating a window of antigen densities that induces maximal T-cell response. These results were observed for both lytic response (measured by CD107a presentation) and cytokine production. Surprisingly, the antigen densities yielding highest response are similar across CARs of different affinities or avidities. This range of antigen concentrations is approximately 10^–7–10^–5 M (peptide loading concentrations), which is similar to the range of peptide concentrations in native TCRs that lead to maximal killing in CTLs (16), suggesting a common mechanism that tunes T-cell response to a certain range of antigen concentrations. Moreover, when the different CAR T cells populations are arranged by their avidity, we observed that high-avidity cells had a smaller reduction in the high antigen density–dependent reduced response (Fig. 2B and D), thus creating a wider window of maximal response. Additionally, the highest-affinity CAR (4 nmol/L) had reduced responses compared with the intermediate-affinity CARs (35 nmol/L and 16 nmol/L; Fig. 2 and hierarchal clustering in Supplementary Fig. S4K), implying an affinity threshold for maximum activation, as suggested previously (6, 8), and as described for native TCRs (14).

In addition to measurements of individual effector functions, the functional capability of CD8⁺ cells and CAR T cells to protect against viruses or cancer cells was shown to correlate with T cells' ability to induce several effector functions simultaneously, a feature known as polyfunctionality (34). We assessed polyfunctionality by measuring IL2, IFNγ, and TNFα production simultaneously (Fig. 3; Supplementary Table S5 describing statistics). The percentages of cells that produce three, two, one, or no cytokines are ordered either by expression level in each individual CAR affinity (left panels) or by their avidity (middle panels). The percentage of cells that produced all three cytokines (triple-positive cells) shows a nonmonotonic dependence on antigen density (Fig. 3A–C). Surprisingly, affinity seems to be negatively correlated with polyfunctionality, an attribute that seems to depend on the combination of CAR expression and affinity, so that medium avidity CAR T cells displayed the highest percentage of triple-positive cells. This effect probably reflects an interference in switching from production of two cytokines to three cytokines simultaneously, as all CAR affinities and avidities do succeed to reach high percentage of double-positive cells. CAR T cells having a low percentage of triple-positive cells had a higher percentage of double-positive cells (Fig. 3D–F). Apparently, only the medium-avidity cells efficiently switch to produce three cytokines simultaneously. Collectively, both cytokine production and lytic activity showed inferior response of the high-affinity CAR, and an antigen-dependent nonmonotonous curve that was affected by the level of CAR expression.

Alongside the degree of CAR T-cell response, we analyzed CAR T-cell sensitivity and specificity, as measured by the EC₅₀ values and the response to nonspecific targets (unloaded APCs or loaded with the MART1 peptide), respectively. Surprisingly, CAR T cells of different affinities and avidities had proximal EC₅₀ values for the IL2, IFNγ, and TNFα cytokines (Supplementary Fig. S6 and Supplementary Table S6). The EC₅₀ values of reactive-MFI of CD107a and MIP-1β were more diverse between different CAR T-cell populations; however, these variations could not be explained by either affinity, expression, or their interaction (Supplementary Table S6). The observed significant effects are not congruent between all markers and show only minor changes between the different CAR T cells. Together with the observation of similar antigen densities that lead to the highest responses, we can infer that the affinity and expression level of the CAR do not shape the response differently, but affect the strength of the response. The specificity of CAR T cells, however, is affected by CAR affinity and expression level (Supplementary Fig. S6 and Supplementary Table S7). Although the effect of affinity is not congruent between different markers (e.g., the 4 nmol/L CAR had high specificity for CD107a, but low specificity for MIP-1β), CAR expression level showed a congruent and major effect on specificity (Supplementary Fig. S6). The higher the expression level, the higher the response to nonspecific targets. Concomitantly, similar results are seen when measuring polyfunctionality (Fig. 4K), where the percentage of triple-negative cells for nonspecific targets is reduced as expression level increases.

To better evaluate the contribution of CAR affinity and expression levels to T-cell responses, we applied a two-way ANOVA for each antigen density and analyzed the percentage of variance attributed to each effect (the effect's sum of squares) out of the total variance (the total sum of squares), thereby denoting the percentage of variance they could explain for each antigen density (Fig. 4A–F). Affinity could explain up to 50% of the total variance for the medium antigen densities; however, it could explain only little of the total variance in the high or low antigen densities. Thus, affinity contributes in the range of antigen densities that evoke high responses. The percentage of variance explained by CAR expression showed different patterns between the magnitude of response (reactive-MFI), which was influenced by CAR expression mainly at the high antigen densities, and the percentage of responding cells (percent-reactive), which was influenced by CAR expression mainly at the low antigen densities and in the nonspecific targets. Taken together, CAR expression contributes to the number of cells responding to nonspecific targets (as described also in Supplementary Fig. S6), and to the strength of response at high antigen densities—the region of reduced function. This latter observation corresponds to the less attenuated responses of high-avidity cells at high antigen densities (Fig. 2B and D), that is, a nonmonotonous curve as function of antigen density is evident at lower avidities, but less so at higher avidities.

Next, we applied PCA in order to provide an unbiased and a more global assessment of the properties that best explain the variation in the entire data set. We applied PCA to the reactive-MFI parameter (Fig. 4G). The two major components seemed to group CAR T cells by their affinity and expression level, describing more than 60% of the total variance; however, no principal component reflected solely one factor (affinity or expression), but a combination of the two. Notwithstanding, PC1 (describing almost 40% of the total variance) resembled closely the CAR affinity. CAR T cells are clustered by their affinity, and within each affinity cluster are arranged by expression. Additionally, the 4 nmol/L high-affinity CAR and low-affinity 434 nmol/L CAR are clustered closer, thus separating the mid-range affinities (which yielded the stronger responses) from the high and low affinities. Expression level, on the other hand, is arranged from low to high, with low expression being more separate from medium and high expression levels. Altogether, an unbiased systematical analysis revealed which antigen densities were more affected by affinity or avidity, and corroborated the findings of a nonmonotonous dependence on affinity and a monotonous dependence on avidity.

To validate that the phenomena observed on APC target cells can describe antitumor activity, we compared responses of CAR T cells of different affinities toward Tyr–HLA-A2 high and low cell lines and Tyr–HLA-A2-negative cell line (501A, 526, and 1938, respectively; Fig 5A). We used a conventional lactate dehydrogenase (LDH) release assay to measure CAR T-cell in vitro cytotoxicity. All CAR T cells showed specific killing of the Tyr–HLA-A2-positive cell lines (Fig. 5C and D), with a slightly better cytotoxicity of the medium-affinity 35 nmol/L over the high and low affinities (see also statistics in Supplementary Table S8). Between the two Tyr–HLA-A2-positive cell lines, there is a difference in T-cell responses, and whereas for the 501A cell line the high-affinity CAR performs better than the low-affinity CAR, no obvious difference is observed for the 526 cell line.

Finally, we tested these findings in vivo, by conducting a series of experiments to evaluate the antitumor response of the different CAR affinities in human melanoma xenograft models in NSG mice. Antitumor efficacy of high, medium, and low-affinity CAR T cells was compared in mel 526- or mel 501A tumor–bearing mice (Figs. 5E and F and 5G and H, respectively). Control animals were injected with untransduced PBMCs. Our data demonstrate that only the medium-affinity CAR could induce significant antitumor response against the low antigen expressing melanoma 526 cell line (see also Supplementary Table S9 and Supplementary Fig. S7) and to reduce tumor size at day 15 and after, whereas 4 and 434 nmol/L CARs are not different from PBMCs. In addition, mice bearing the high antigen expressing 501A melanoma responded to both low- and medium-affinity CAR treatments; however, the response of the medium-affinity CAR was significantly higher (see also Supplementary Table S9 and Supplementary Fig. S7). These data in vivo support our in vitro studies presented in this work, which suggest that the medium-affinity CAR T cells perform in a superior fashion compared with the high-affinity and low-affinity CARs.

The biophysical properties of CAR ligand binding are imperative for CAR T-cell function, leading to full T-cell responsiveness and tumor eradication. Here, we set to define the properties that lead to differential T-cell responses. To the best of our knowledge, this is the first report demonstrating a unified system that can compare the contribution of affinity, avidity, and antigen density. We demonstrated that CAR T-cell response is nonmonotonic, and there exist both a window of antigen densities and a window of CAR affinities that yield maximal response. Although each of these properties seems to affect functional avidity separately, avidity seems to be related to antigen density, thereby increasing the magnitude of response. Logically, receptor and antigen density are intertwined, as they together assign the number of engaged complexes.

The promise of durable tumor regression using adoptive transfer of CAR T cells has initiated several works aimed at understanding the kinetics, sensitivities, and optimal response of different CAR designs. One aspect examines the contribution of CAR affinity. In this respect, we previously reported that low-affinity TCR had superior sensitivity and efficacy compared with CAR targeting the same antigen (6). High-affinity–dependent hypofunctionality of CARs is subject to ongoing studies. To date, different outcomes have been documented for increased receptor affinity: reduced function (6, 8, 10, 33), an equivocal response (7, 35) or an increased response (9, 36). We measured CAR T-cell activity and documented that the effect of affinity is nonmonotonous, and CARs of medium affinities (K_D of 10^–8 mol/L) outperformed the high-affinity CAR. Likewise, recently, Ghorashian and colleagues (8) showed that CD19 CAR of similar medium affinity was superior to high-affinity CAR.

Our experimental system also allowed us to resolve the influence of CAR avidity on CAR T-cell functionality. Avidity is a parameter that reflects the general strength of multiple interactions, affected by the interaction affinity, receptor expression levels, clustering, and coreceptors. We controlled and sorted cells by receptor expression levels and measured the avidity of these cells. We revealed a strong correlation between avidity and receptor expression levels, reflecting the dominant contribution of receptor expression to the general avidity. Overall, we describe that increasing CAR avidity resulted in augmented response, in line with previous reports (35, 37). Additionally, using an experimental system that integrates several biophysical binding properties and allows systematical analysis of their effects on T-cell response, we detected an interplay between antigen density and avidity. Increasing avidity results in increased sensitivity and response to low antigen concentrations. Furthermore, high-avidity cells had a wider window of antigen densities that induced maximal response. These observations were corroborated by a systematical analysis aimed at quantifying the contribution of avidity for each antigen density. This analysis describes that avidity could explain up to 60% of the variance in the number of responding cells at low antigen densities, and up to 40% of the variance in the magnitude of response at high antigen densities. Altogether, accumulating evidence demonstrates the significance of receptor expression levels on CAR T-cell activity and can be considered an effective candidate for manipulating and adjusting CAR T-cell responses. Concomitantly, Eyquem and colleagues (30) suggested a method for genetically controlling CAR expression levels and showed the dependence of CAR T-cell exhaustion and efficacy on both the basal and dynamic CAR expression levels.

Recently, polyfunationality was suggested as a clinical biomarker for CAR T-cell activity and toxicity (34, 38). We evaluated the percentage of polyfunctional cells, and demonstrated a nonmonotonous behavior, where medium-avidity CAR T cells were more polyfunctional. Surprisingly, these cells do not display higher amounts of single- or double-positive cells for the lower antigen densities. These data suggest that there exists a threshold switching from polyfunctional cells that produce two cytokines to polyfunctional cells that produce three or more cytokines. The mechanism underlying such a threshold is yet to be explored, as well as their contribution to clinical outcomes. A more thorough examination of different cytokine thresholds should be conducted in order to assess the relation between affinity, avidity, and polyfunctionality. In line with this, Rafiq and colleagues (39) could improve a high-affinity WT1 CAR that was not effective in vivo by “armoring” the CAR T cells to secrete enhanced levels of IL12. Possibly, the hurdle of high-affinity hypofunctionality could be surmounted by engineering these cells to have lower threshold to become polyfunctional.

Current CAR T-cell treatments have shown low efficacy in solid cancers (40). This is attributed to many factors; however, we suggest that affinity and antigen density should also be accounted for. Surface-exposed extracellular tumor antigens can be expressed up to several millions of molecules per cell, and CARs that are used for clinical trials have high affinities, reaching picomolar affinities. In light of our findings and along with recent works that describe high-affinity and high antigen density attenuation of T-cell responses, these binding properties and their combination should be considered in relation to clinical assessment.

Controlling CAR T-cell response can also be achieved through engineering of the intracellular signaling domains. The prominent costimulation signaling domains are derived from CD28 or 4-1BB receptors, where in this work we used CD28-based CARs. Generally, CD28 costimulation results in stronger, more sensitive (41), and more effector-like CAR T-cell response, however with low persistence (42). Comparing CARs of different affinities comprising either CD28 or 41BB costimulation (43) showed that the effect of affinity was similar on CD28 and 4-1BB CARs, with higher responses for CD28 CARs. Looking at the underlying activation mechanism, Salter and colleagues compared CD28 and 4-1BB CARs and showed they act on similar signaling pathways, whereas the stronger CD28 signaling correlated with higher basal LCK signaling level (44). Hence, the activation difference between CD28 and 4-1BB CARs can be bridged by altering the number of ITAMs (45), or engineering LCK/SHP1 signaling (46), thus emphasizing the role of signal summation in governing CAR T-cell activation. In light of the results we describe here, understanding the relationships of affinity and avidity to basal signaling level and signal summation can provide more accurate tools to optimize CAR T-cell response. In future studies, we will attempt to add into this complex equation the question of optimal costimulation signaling domains in the designed CARs.

In summary, this study examined and defined the contribution of the biophysical properties of CAR and ligand binding, and described their unique and combinatorial effect on CAR T-cell functional avidity. Further understanding of the mechanisms that generate optimal CAR T-cell response is important, both from a basic science perspective and for adjusting CAR T-cell responses to tumor antigens in regard to the tumor antigen load in clinical adoptive cell therapies.

G. Denkberg reports personal fees from Adicet-Bio Israel during the conduct of the study and personal fees from Adicet-Bio Israel outside the submitted work. S. Shen-Orr reports other from CytoReason outside the submitted work. Y. Reiter reports other from Adicet Bio outside the submitted work. No disclosures were reported by the other authors.

R. Greenman: Conceptualization, investigation, visualization, writing–original draft. Y. Pizem: Conceptualization, validation, investigation, visualization, writing–original draft. M. Haus-Cohen: Resources, validation, investigation, and project administration. A. Goor: Validation, investigation, visualization, and methodology. G. Horev: Software, formal analysis, and methodology. G. Denkberg: Validation, investigation, and methodology. K. Sinik: Validation, investigation, and methodology. Y. Elbaz: Validation, investigation, and methodology. V. Bronner: Investigation and methodology. A.G. Levin: Investigation and methodology. G. Horn: Investigation and methodology. S. Shen-Orr: Resources, software, formal analysis, visualization, and methodology. Y. Reiter: Conceptualization, resources, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

This work was supported by the Israel Science Foundation grant No. 461/15 awarded to Y. Reiter. We are grateful to Prof. Zelig Eshhar (Weizmann Institute of Science) for the gift of Phoenix-Ampho cells and the pBullet vector. We thank Drs. Stewart Abbot, Daulet (Dau) Satpayev (Adicet Bio), and Arie Admon (Technion) for helpful comments on the manuscript.

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.