Functional Tuning of CARs Reveals Signaling Threshold above Which CD8⁺ CTL Antitumor Potency Is Attenuated due to Cell Fas–FasL-Dependent AICD

Approaches to cancer immunotherapy, whereby T cells are genetically modified to express chimeric antigen receptors (CAR), are the subject of considerable early-phase clinical trials (1). Whereas dramatic antitumor potency is observed in patients treated with CD19-specific CAR T cells for B-cell malignancies, such as acute lymphoblastic leukemia and non–Hodgkin lymphomas, challenges to achieve similar responses in patients harboring solid tumors are considerable (2–4). At present, the development and clinical testing of CAR-redirected T-cell adoptive therapy in cancer patients is largely empiric and constrained by a variety of technical parameters that affect feasibility of executing clinical phase I trials. Two parameters related to cell products that can be defined with greater precision are the composition of T-lymphocyte subset(s) and the tuning of CAR signaling for functional outputs that maximize their antitumor activity. Our group has studied the therapeutic activity of CAR-expressing central memory T cells, a stable antigen-experienced component of the T-cell repertoire having stem cell–like features and capacity to repopulate long-lived functional memory niches following adoptive transfer (5–9).

Moving beyond the targeting of CD19-expressing B-cell malignancies, a significant challenge for the field is the identification and vetting of cell-surface target molecules that are amenable to CAR T-cell recognition with tolerable “on” target “off” tumor reactivity (10, 11). Once identified, however, approaches to tune new CARs for signaling outputs are presently rudimentary. Parameters that are generally perceived as central to CAR development are the affinity of the target molecule CAR antigen-binding domain and the signaling modules of the cytoplasmic domain. We and others have described the significant impact of the extracellular spacer in contributing to CAR T-cell performance and the growing appreciation that CAR spacers need to adjust the biophysical synapse distance between a T cell and a tumor cell to one that is compatible for T-cell activation (12–14). Given that each new scFv and target molecule define a unique distance from the tumor cell plasma membrane, the adjustment of CAR spacers is unique to each construct and derives via empiric testing of libraries of spacer length variants.

The present study evaluates the contribution of both extracellular spacer length and cytoplasmic signaling module selection on the performance of a CAR specific for a tumor-selective epitope on CD171 (L1-CAM) that is recognized by monoclonal antibody CE7 and was previously tested as a first-generation CAR in a clinical pilot study (15–17). Using in vitro functional assays for CAR-redirected effector potency, we observed a quantitative hierarchy of effector outputs based on spacer dimension in the context of second- and third-generation cytoplasmic signaling domains. We observed a striking discordance in CAR T-cell performance in vitro versus in vivo due to fratricidal activation-induced cell death (AICD) of the most functionally potent CAR formats. These data reveal new and potentially clinically relevant parameters for inspection in the development of CAR T-cell immunotherapy for solid tumors.

CD171-specific CARs were constructed using (G₄S)₃ peptide–linked VL and VH segments of the CE7-IgG2 monoclonal antibody (18). The scFv was codon optimized and subsequently linked to variable spacer length domains based on 12AA [short spacer (SS)/“hinge-only”], 119AA [medium spacer (MS)/“hinge-CH3”] or 229AA [long spacer (LS)/“hinge-CH2-CH3”] derived from human IgG4-Fc. All spacers were linked to the transmembrane domain of human CD28 and to signaling modules comprising either the cytoplasmic domain (i) of 4-1BB alone (2G CAR) or (ii) of CD28 (mutant) and 4-1BB (3G CAR), with each signaling module fused to the human CD3-ζ endodomain (19). The cDNA clones encoding CAR variants were linked to a downstream T2A ribosomal skip element and truncated EGF receptor (EGFRt), cloned into the epHIV7 lentiviral vector, and CD171-CAR lentiviruses were produced in 293T cells (20).

Total RNA was extracted from T cells using the RNeasy Mini Kit (Qiagen). cDNA was synthesized by reverse transcription using the First Strand Kit (Life Technologies). RNA quantification was performed using FasL primers (IDT) and the CFX96 real-time detection system (Bio-Rad). Housekeeping gene actin was used as a control. Data were analyzed using CFX Manager Software version 3.0.

T cells were harvested, washed in PBS, and lysed in protease inhibitor (Millipore). Proteins were analyzed using SDS–PAGE followed by Western blotting using anti-CD247 (CD3-ζ; BD Biosciences). Signals were detected using an Odyssey Infrared Imager, and band intensities were quantified using Odyssey v2.0 software (LI-COR).

Immunophenotyping was conducted using fluorophore-conjugated mAbs: CD4, CD8, CD27, CD28, CD45RA, CD45RO, CD62L, CCR7 (Biolegend). Cell-surface expression of L1-CAM was analyzed using a fluorophore-conjugated mAb (Clone 014; Sino Biological). EGFRt expression was analyzed using biotinylated cetuximab (Bristol-Myers Squibb) and a fluorophore-conjugated streptavidin secondary reagent. To assess activation and AICD fluorophore-conjugated mAbs for CD25, CD69, CD137, CD178 (Fas Ligand) and CD95 (Fas, all Biolegend) were used. Caspase-3 activity was measured using CaspGlow (eBioscience) following the manufacturer's protocol. Flow analyses were performed on an LSRFortessa (BD Biosciences), and data were analyzed using FlowJo software (Treestar).

Samples of heparinized blood were obtained from healthy donors after written informed consent following a research protocol approved by the Institutional Review Board of Seattle Children's Research Institute (SCRI IRB #13795). Peripheral blood mononuclear cells (PBMC) were isolated using ficoll (GE Healthcare), and CD8⁺CD45RO⁺CD62L⁺ central memory T cells (T_CM) were isolated using immunomagnetic microbeads (Miltenyi). First, CD8⁺CD45RO⁺ cells were obtained by negative selection using a CD8 T-cell isolation kit and CD45RA beads, then enriched for CD62L, activated with anti-CD3/CD28 beads at a bead-to-cell ratio of 3:1 (Life Technologies) and transduced on day 3 by centrifugation at 800× g for 30 minutes at 32°C with lentiviral supernatant (multiplicity of infection = 5) supplemented with 1 mg/mL protamine sulfate (APP Pharmaceuticals). T cells were expanded in RPMI (Cellgro) containing 10% heat-inactivated FCS (Atlas), 2 mmol/L l-glutamine (Cellgro), supplemented with a final concentration of 50 U/mL recombinant human IL2 (Chiron Corporation), and 1 ng/mL IL15 (Miltenyi). The EGFRt⁺ subset of each T-cell line was enriched by immunomagnetic selection with biotin-conjugated Erbitux (Bristol-Myers Squibb) and streptavidin microbeads (Miltenyi; 21). CD171-CAR and mock control T cells were expanded using a rapid expansion protocol (9). T cells used for in vivo assays were derived by stimulation with CD3/CD28 beads (S₁) followed by 2 rapid expansions (R₂) and cryopreserved 14 days (D₁₄) after the second rapid expansion stimulation.

The neuroblastoma cell lines Be2 and SK-N-DZ were obtained from the ATCC. Be2-GFP-ffLuc_epHIV7 and SK-N-DZ-GFP-ffLuc_epHIV7 were derived by lentiviral transduction with the firefly luciferase (ffLuc) gene and purified by sorting on GFP. Both cell lines were further transduced with CD19t-2A-IL2_pHIV7 to generate IL2-secreting cell lines purified by sorting on CD19t. All neuroblastoma cell lines were cultured in DMEM (Cellgro) supplemented with 10% heat-inactivated FCS and 2 mmol/L of l-glutamine. EBV-transformed lymphoblastoid cell lines (TMLCL) and TMLCLs that expressed membrane-tethered CD3epsilon-specific scFvFc derived from OKT3 mAb (TMLCL-OKT3; ref. 6) were cultured in RPMI 1640 supplemented with 10% heat-inactivated FCS and 2 mmol/L of l-glutamine.

After coculturing 1 × 10⁶ effector and target cells for 4 to 8 minutes, cells were processed to measure Erk/MAP kinase ½ activity according to the 7-Plex T-cell Receptor Signaling Kit (Millipore). Protein concentration was measured using the Pierce BCA Protein Assay Kit (Thermo Scientific).

Target cells were labeled with ⁵¹Cr (Perkin Elmer), washed, and incubated in triplicate at 5 × 10³ cells/well with T cells (S₁R₂D_12-14) at various effector-to-target (E:T) ratios. Supernatants were harvested after a 4-hour incubation for γ-counting using Top Count NTX (Perkin Elmer), and specific lysis was calculated (16).

Neuroblastoma cell lines containing GFP-ffLuc_epHIV7 were cocultured with effector cells at a 5:1 E:T ratio. The effector cells were on their first, second, or third round of tumor cell encounter as described above. To assess the amount of viable tumor cells left after T-cell encounter, d-luciferin was added, and after 5 minutes the biophotonic signal from the neuroblastoma cells was measured using an IVIS Spectrum Imaging System (Perkin Elmer).

A total of 5 × 10⁵ T cells (S₁R₂D_12-14) were plated with stimulator cells at a 2:1 E:T ratio for 24 hours. IFNγ, TNFα, and IL2 in the supernatant were measured using Bioplex cytokine assay and Bioplex-200 system (Bio-Rad).

To mimic recursive antigen encounters, we started a coculture of adherent target cells and freshly thawed nonadherent effector cells at a 1:1 E:T ratio. After 24 (round I) and 48 (round II) hours, T-cell viability was assessed using the Guava ViaCount Assay (Millipore), and nonadherent effector cells were moved to a new set of adherent target cells at a 1:1 E:T ratio. After rounds I, II, and III (72 hours), T cells were harvested and treated with a dead cell removal kit (Miltenyi) before further analysis.

Tumor samples were obtained from patients diagnosed with neuroblastoma and treated in the Department of Pediatric Hematology-Oncology of Seattle Children's Hospital (SCRI IRB #13740).

Human neuroblastomas and mouse brains were harvested, fixed, processed, paraffin embedded, and cut into 5-μm sections. Antigen retrieval was performed using Diva decloaker RTU (Biocare Medical). Primary antibodies were incubated with sections overnight at 4°C and diluted in blocking buffer as follows: rat monoclonal anti-human CD3 (Clone CD3-12; Bio-Rad) 1:100, mouse monoclonal anti-human Ki67 (Clone MIB-1; Dako) 1:200, rabbit polyclonal anti-human cleaved caspase-3 (Biocare Medical) 1:100, rabbit polyclonal anti-human Granzyme B (Covance) 1:200, and mouse monoclonal anti-human CD171 (clone 14.10; Covance) 1:200. Secondary antibodies (Life Technologies) were incubated with sections for 2 hours at room temperature and diluted at 1:500 in PBS with 0.2% BSA.

Slides were imaged on an Eclipse Ci upright epifluorescence microscope (Nikon) equipped with a Nuance multispectral imaging system and analyzed with InForm analysis software (Perkin Elmer).

NSG mouse tumor models were conducted under SCRI IACUC-approved protocols.

Adult male NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ [NODscid gamma (NSG)] mice were obtained from the Jackson Laboratory or bred in house. Mice were injected intracranially (i.c.) on day 0 with 2 × 10⁵ IL2-secreting, ffLuc-expressing Be2 or SK-N-DZ tumor cells 2 mm lateral, 0.5 mm anterior to the bregma, and 2.5 mm deep from the dura. Mice received a subsequent intratumoral injection of 2 × 10⁶ CAR-modified CD8⁺ T_E(CM) either 7 (therapy response model) or 14 (stress test model) days later. In the stress test model, mice were euthanized 3 days after T-cell injection, and brains were harvested for immunohistochemistry (IHC) analysis.

For bioluminescent imaging of tumor growth, mice received intraperitoneal (i.p.) injections of d-luciferin (4.29 mg/mouse). Mice were anesthetized with isoflurane and imaged using an IVIS Spectrum Imaging System 15 minutes after d-luciferin injection. Luciferase activity was analyzed using Living Image Software Version 4.3 and the photon flux analyzed within regions of interest (all Perkin Elmer).

Statistical analyses were conducted using Prism software (GraphPad). Data are presented as means ± SD or SEM, as stated in the figure legends. The Student t test was conducted as a two-sided unpaired test with a confidence interval of 95%. Statistical analyses of survival were conducted by the log-rank test. Results with a P value of less than 0.05 were considered statistically significant.

The biophysical synapse between a CAR-expressing T cell and a tumor cell is influenced by the epitope location on the tumor cell-surface target molecule relative to the distance from the tumor cell's plasma membrane. We hypothesized that CAR extracellular spacer size tuning to accommodate a functional signaling synapse is a key attribute to engineering bioactive CARs. To assess the impact of CD171-specific CAR extracellular spacer size, we assembled a set of spacers using modular domains of human IgG4 as follows: “long spacer” (LS) IgG4 hinge-CH2-CH3, “medium spacer” (MS) IgG4 hinge-CH3 fusion, and “short spacer” (SS) IgG4 hinge. Each spacer variant was fused to a CD28-transmembrane domain followed by a second-generation (2G) 4-1BB:zeta-endodomain that was linked to the cell-surface EGFRt tag (Fig. 1A; ref. 21). We generated sets of spacer variant 2G-CAR⁺/EGFRt⁺ human CD8⁺ central memory–derived effector T-cell lines (T_E(CM)) from purified CD8⁺CD45RO⁺CD62L⁺ central memory precursors by immunomagnetic selection (Supplementary Fig. S1A and S1B). Following expansion, lentivirally transduced T_E(CM) cell lines were further enriched for homogeneous levels of EGFRt expression by cetuximab immunomagnetic positive selection (21). We confirmed the similar surface expression levels of each of the CAR spacer variants by anti-murine F(ab)-specific and EGFR-specific flow-cytometric staining and protein expression quantified by Western blot for CD3ζ of each T-cell line (Fig. 1B and C).

We first sought to determine whether the magnitude of 2G-CAR–triggered in vitro activation of CD8⁺ T_E(CM) is influenced by spacer domain size. The human tumor cell lines Be2 and SK-N-DZ were selected based on L1-CAM expression levels that are similar to those of clinical tumor specimens (Supplementary Fig. S2). Following activation by CD171⁺ human neuroblastoma cells, CD171-specific 2G-CAR(LS)-CD8⁺ T_E(CM) exhibited 3.1-fold higher levels of phospho-ERK (P = 0.003), and 5.7-fold higher percentage of cells expressing the activation marker CD137 (P = 0.015), as compared with their CD171-CAR(SS) counterparts (Fig. 1D and E). 2G-CAR(MS) exhibited intermediate levels of phospho-ERK and CD137 induction as compared with LS and SS 2G-CARs. Next, we sought to determine whether spacer size also modulated the magnitude of antitumor cytolytic activity in an LS>MS>SS pattern. Using a 4-hour CRA, we observed that lysis of CD171⁺ neuroblastoma target cells followed the same potency gradient of LS>MS>SS against both CD171^high Be2 and CD171^low SK-N-DZ cell lines (Fig. 1F and Supplementary Fig. S3A). Furthermore, activation for cytokine secretion followed the same incremental output hierarchy such that 2G-CAR(LS) produced 8.4-fold higher amount of IFNγ (P = 0.003), 6.3-fold more IL2 (P < 0.0001) and 6.1-fold higher levels of TNFα (P = 0.005; Fig. 1G) as compared with those elicited by 2G-CAR(SS), with the induction by 2G-CAR(MS) falling between these two extremes. These data demonstrate that the biophysical synapse created by CARs can be tuned by spacer size such that incremental levels of activation and functional outputs are achieved. Based on standard CAR development criteria typically utilized in the field, the 2G-CAR(LS) spacer variant would be a lead candidate for further development for clinical applications.

To delineate the relationship of the observed potency of CAR signaling based on in vitro assays to therapeutic activity in vivo, we performed adoptive transfer experiments in NSG mice with established human neuroblastoma xenografts stereotactically implanted in the cerebral hemisphere (Fig. 2A). Surprisingly, Be2 tumor–engrafted mice treated with intratumoral injection of 2G-CAR(LS) exhibited no therapeutic activity, necessitating animal euthanasia approximately 3 weeks after tumor inoculation (Fig. 2B and C). In comparison, biophotonic tumor signal was reduced and survival enhanced in mice treated with 2G-CAR(SS)-CD8⁺ T_E(CM) and to an intermediate extent in mice treated with the 2G-CAR(MS) variant (P = 0.001; median survival of the different groups: LS = 20 days, mock = 21 days, MS = 59.5 days, SS = 76 days; Fig. 2D). SK-N-DZ tumor–engrafted mice exhibited an SS>MS>>LS hierarchy of tumor responses with uniform tumor clearance and 100% survival of animals treated with 2G-CAR(SS)-CD8⁺ T_E(CM) (Supplementary Fig. S3C and S3D). Failure of 2G-CAR(LS)–redirected CTLs, which we inject directly into engrafted tumors, could not be attributed to their failure to survive adoptive transfer and be activated in situ, as equivalent intratumor densities of transferred T cells that expressed Granzyme B and Ki67 were observed by IHC early after adoptive transfer (day 3; Fig. 2E–H). While not statistically significant (P = 0.34), 2G-CAR(LS)-CD8⁺ T_E(CM), in which activated caspase-3 was detected, were 12.1-fold more frequent than their 2G-CAR(SS) counterparts (Fig. 2H). These data reveal an unexpected discordance between the in vitro antitumor potency of CAR-redirected T cells dictated by extracellular spacer size and their in vivo antitumor therapeutic activity.

We hypothesized that whereas in vitro activation for cytolysis in a 4-hour CRA is the consequence of a limited duration of CAR signaling, the in vivo tumor model requires recursive rounds of activation to achieve tumor eradication. Thus, the signaling performance of a particular CAR format that dictates superior metrics in vitro may fail to reveal the consequences of the signaling amplitude in vivo. To reproduce recursive serial stimulation in vitro, we devised a CAR T-cell:tumor cell coculture “stress test” assay whereby every 24 hours, CAR T cells are harvested and recursively transferred to culture dishes seeded with tumor cells adjusting for a constant viable 1:1 T-cell:tumor cell ratio (Fig. 3A). We utilized Be2 cells modified to express firefly luciferase to concurrently track tumor-cell killing upon each of three rounds of serial transfer. The recursive activation of the 2G-CAR spacer variant lines resulted in equivalent loss of antitumor activity by round III (Fig. 3B). In addition, analysis of each spacer variant-expressing effector cells after each round by flow-cytometric measurement revealed that 2G-CAR(LS)-CD8⁺ T_E(CM) displayed higher frequencies of cells expressing activation markers CD25 and CD69, as compared with their 2G-CAR(SS) counterparts (round I, 79.4% vs. 46.8%, P = 0.007; round II, 74.0% vs. 47.6%; round III, 65.7% vs. 42.1%, P = 0.037; Fig. 3C).

In contrast with the LS>MS>SS pattern of upregulation of activation markers in round I that mimicked our earlier in vitro analysis, we observed an LS/MS>SS loss of T-cell viability that was most substantial in round III (round III percent dead cells, LS 58.7%, MS 62.6% vs. SS 21.1%, LS vs. SS P = 0.024 and MS vs SS P = 0.007; Fig. 3D). To substantiate that the asymmetric loss of T-cell viability by 2G-CAR(LS)-CTLs occurring with recursive activation was the result of exaggerated AICD, we assessed the mechanism of cell death focusing on FasL–Fas-mediated T-cell fratricide (22). We observed tumor-induced CAR activation-dependent upregulation of FasL followed an LS>MS>SS hierarchy as 2G-CAR(LS)-CD8⁺ T_E(CM) displayed 4.8-fold and 2.5-fold higher FasL surface expression, and 5.5-fold and 3.3-fold higher FasL mRNA abundance than the short or medium spacer CAR T cells, respectively (long vs. short: P < 0.0001 and P = 0.002; long vs. medium: P < 0.0001 and P = 0.016; Fig. 3E and F). To link FasL expression with increased Fas-mediated apoptosis, we analyzed caspase-3 activity and observed 13.2-fold higher levels of cleaved caspase-3 in 2G-CAR(LS)-CD8⁺ T_E(CM) as compared with their SS counterpart (P < 0.0001; Fig. 3G). Finally, we subjected 2G-CAR(LS)-CD8⁺ T_E(CM) to siRNA knockdown of Fas or FasL, then exposed them to tumor and observed a 1.4-fold (Fas; P = 0.005) and 1.6-fold (FasL; P = 0.0001) increase in T-cell viability after round III, respectively (Fig. 3H). To verify that our siRNA knockdown led to a reduction of Fas/FasL, we assessed their surface expression on the 2G-CAR(LS)-CD8⁺ T_E(CM) and observed a 91.3% reduction in Fas⁺ (P < 0.0001) and 80.1% reduction in FasL⁺ CTLs (P < 0.0001) than 2G-CAR(LS)-CD8⁺ T_E(CM) treated with scrambled siRNA (Supplementary Fig. S4A and S4B). In aggregate, these data demonstrate that tuning of CAR spacer size can modulate downstream signaling events that result in not only differential magnitudes of antitumor functional outputs but coordinated increases in susceptibility to AICD. The balance between these two processes for optimal in vivo antitumor activity may not always be achieved by spacer tuning to achieve the highest levels of CAR-signaling outputs, as exemplified by our comparison of 2G-CAR(LS) and 2G-CAR(SS) structural variants.

Third-generation CARs contain two costimulatory endodomain modules in series with the CD3-ζ activation module and have been reported to augment the magnitude of cytolysis and cytokine production levels over their second-generation counterparts (23). We assembled a CD171-specific 3G-CAR through the addition of a CD28 endodomain to the 2G 4-1BB:zeta-endodomain (Fig. 4A). CD8⁺ T_E(CM) expressing comparable levels of 2G-CAR(SS) and 3G-CAR(SS) were derived from purified T_CM precursors by immunomagnetic selection (Fig. 4B and C). 3G-CAR(SS)-CD8⁺ T_E(CM) demonstrated an 8.4-fold higher induction of CD137 expression upon tumor contact than their second-generation counterparts (P < 0.0001; Fig. 4D), a 1.3-fold increase in cytolytic activity against Be2 targets (E:T of 1:10; P = 0.0001; Fig. 4E) and 5.1-fold more IL2 and 2.5-fold more TNFα secretion (P < 0.0001 and P = 0.003; Fig. 4F).

Next, we assessed whether the 3G endodomain, in the context of an extracellular short spacer, could selectively enhance antitumor activity in vivo without exacerbation of AICD. Surprisingly, the in vivo antitumor activity of 3G-CAR(SS)-CD8⁺ T_E(CM) against both Be2 (Fig. 5A) and SK-N-DZ (Fig. 5B) was inferior, though not to a statistically significant degree to their 2G-CAR(SS) counterparts. These findings were not attributable to differences in short-term persistence of CAR T cells within tumors based on similar densities of human CD3⁺ T cells detected 3 days after adoptive transfer (Fig. 5C). Despite the finding of higher frequencies of Granzyme B⁺ 3G-CAR(SS) T cells compared with 2G-CAR(SS) intratumoral T cells, we again observed augmented numbers of third-generation T cells with activated caspase-3, suggesting that the augmented costimulation through a combined effect of CD28 and 4-1BB was capable of hyperstimulation resulting in heightened AICD, despite the context of a short spacer extracellular domain (Fig. 5D). This was confirmed by comparing their performance using the in vitro stress test assay. Following each round of tumor stimulation, we observed higher frequencies of CD25⁺CD69⁺ T cells in the 3G-CAR(SS)-T-cell population (Fig. 6A) accompanied by increased frequencies of dead T cells through successive rounds of activation (Fig. 6B). Augmented AICD was again associated with heightened levels of FasL expression by surface staining and mRNA content, which in turn coincided with increased levels of activated caspase-3 (Fig. 6C–E). These data demonstrate that overtuning of CAR-signaling outputs based on intracellular signaling domain composition negatively impacted on a tuned short spacer dimension in a combinatorial manner by enhancing FasL-mediated T-cell AICD.

CARs are capable of mediating multiplexed signaling outputs that trigger redirected antitumor T-cell effector function (24–26). It stands to reason that the tuning of CARs for effective T-cell antitumor activity will be more stringent in solid tumor applications, and that empiric designs of CARs based on limited understanding of the impact of their composition on in vivo antitumor function will only hamper progress in human clinical applications. Here, we systematically interrogate CAR structure function in human central memory–derived CD8⁺ effector CTLs focusing on the combinatorial effects of extracellular spacer dimension in the context of cytoplasmic signaling module composition. By surveying CAR-signaling strength using in vitro assays, we have identified a potency hierarchy of CAR structural variants. These analyses have revealed a range of CAR-signaling outputs permissive for in vivo antitumor activity above which in vivo potency is attenuated by heightened AICD.

The evolution of CAR design has proceeded to date via a largely empiric process, and has focused predominantly on the augmentation of signaling outputs through combinatorial modules of costimulatory receptor cytoplasmic domains fused in series to immunoreceptor tyrosine-based activation motif (ITAM)-containing activation domains (27–29). Comparisons of the function of CTLs expressing first-, second-, or third-generation CARs have typically been made in the context of a “stock” extracellular spacer domain preferred by a particular laboratory, ranging from full-length IgGs to relatively short CD8α hinges or membrane-proximal portions of CD28 (30–32). Our group and others have studied the impact of spacer dimension on CAR signaling and functional activity (14, 19). Unlike a T-cell receptor contact with peptide-loaded HLA class I or II, which defines a scripted biophysical gap between T-cell plasma membrane and target-cell plasma membrane that is permissive for assembly of a supramolecular activation complex, CARs do not conform to this dimensional relationship as a consequence of the target molecule's structural dimensions, the scFv's epitope location on the target molecule, and the CAR's spacer size (33). While the first two dimensions are unique to each selected antigen and antibody-binding domain, the CAR spacer is size tunable and can compensate to some extent in normalizing the orthogonal synapse distance between CAR T cell and target cell. This topography of the immunological synapse between a T cell and a target cell also defines a distance that cannot be functionally bridged by a CAR due to a membrane-distal epitope on a cell-surface target molecule that, even with a short spacer CAR, cannot bring the synapse distance in to an approximation for signaling (13). Likewise, membrane-proximal CAR target antigen epitopes have been described for which signaling outputs are only observed in the context of a long spacer CAR (34).

Using a CD171-specific scFv-binding domain derived from the CE7 mAb, we first assessed the impact of extracellular spacer size on signaling outputs from a 4-1BB:zeta second-generation CAR. We observed incremental gains of function in signaling outputs based on in vitro assays as spacer size increased from the short IgG4 hinge spacer, to an intermediate hinge:CH3, to the full-length IgG4-hinge:Fc spacer. Because prior studies have revealed reduced survival of LS CAR T cells due to interaction between FcγR⁺ cells in the lung and the Fc portion of the CAR after intravenous injection (14, 35), we used for in vivo testing a direct intratumoral route of CD8⁺ CTL administration to study the direct effect of spacer length on CAR T cells within a solid tumor that provides IL2 locally, as a surrogate for infusional IL2 and/or codelivery of CD4⁺ Th1 T cells. Unexpectedly, the antitumor potency of intratumorally injected CAR-CD8⁺ CTLs was inversely correlated to spacer size (i.e., SS>MS>>LS) and in vitro functional potency. Given these findings, we hypothesized that commonly employed in vitro assays that assess CAR-T-cell function upon a single limited-duration tumor cell encounter fail to detect the subsequent fate of CAR T cells upon recursive tumor exposure, as would be predicted to occur within solid tumors in vivo. To better assess this possibility, we devised an in vitro assay in which CAR T cells are recursively exposed to equal numbers of biophotonic reporter gene-expressing tumor cells. Tumor-cell killing can thereby be quantified biophotonically and retrieved CAR T cells can be interrogated for activation status, viability, and caspase activity after each round of tumor coculture. We observed, upon three recursive tumor encounters, disproportionate increases in the frequency of T cells undergoing apoptosis among 2G-CAR(LS) T cells as compared with 2G-CAR(SS)-T-cells. The exaggerated AICD correlated with heightened LS CAR-induced expression of FasL and activated caspase-3 relative to SS CAR. AICD in LS CAR T cells was reduced by siRNA knockdown of FAS or FasL before exposure to tumor cells. These in vitro findings reveal a T cell–intrinsic Fas–FasL-dependent mechanism of AICD that correlates with limited intratumoral persistence of LS CAR T cells. In aggregate, these data demonstrate that the nonsignaling extracellular spacer is a major tunable CAR design element that affects not only signaling activity but persistence of CAR T cells in solid tumors in a T cell–intrinsic manner, independent of interactions with Fc⁺ cells of the reticulo-endothelial system (14).

Given the relation between spacer dimension and in vivo survival in the context of a 4-1BB:zeta CAR, we sought to understand whether the short spacer dimension would be generically optimal in the context of the augmented signaling outputs of a third-generation CD28:4-1BB:zeta CAR endodomain format. Consistent with observations made by multiple other groups, the CD171-specific 3G-CAR(SS) stimulated heightened levels of cytolytic activity and cytokine synthesis compared with the 2G-CAR(SS) upon in vitro tumor stimulation. However, the augmented signaling outputs of the 3G-CAR in the context of its short spacer also increased FasL expression, exacerbated apoptosis as indicated by increased levels of activated caspase-3 and resulted in higher frequencies of cell death. Correspondingly, we observed impaired in vivo antitumor efficacy of the 3G-CAR(SS) T cells, as compared with the 2G-CAR(SS) due to attenuated in vivo intratumoral survival. While CD28 costimulates T cells upon initial antigen activation and enhances T-cell viability by deflecting AICD through Nuclear Factor of Activated T Cells (NFAT)-regulated increases in cFLIPshort, published studies have also revealed that recursive CD28 costimulation of previously activated T cells can reduce their subsequent survival via augmented FasL expression and, consequently, increased AICD (36). It is interesting, therefore, to speculate whether recursive CD28 signaling mediated by anti-CD19 CD28:zeta CAR-T-cells is responsible for the relatively short persistence duration in treated ALL patients, as compared with the often prolonged persistence of anti-CD19 4-1BB:zeta-treated patients in reported clinical trials (2, 37). In aggregate, these data demonstrate that in vivo potency of CAR-redirected T cells is dependent, in part, on identifying permissive combinations of size-optimized extracellular spacer domains in the context of a particular cytoplasmic signaling domain composition. Furthermore, we describe an in vitro assay for assessing the proclivity of a CAR construct to induce AICD in primary human CD8⁺ CTLs upon recursive activation events. These studies, using a solid tumor model system, reveal that “overtuning” of CARs based on in vitro functional assays can lead to the selection of constructs that exhibit suboptimal in vivo potency due to excessive AICD.

There is as yet no predictive structural model that can reliably direct a priori how CARs should be built based on target molecule epitope location relative to the plasma membrane of the tumor cell. Moreover, commonly used surrogate in vitro bioassays may instruct away from a definitive choice of CAR composition that results in the greatest differential between high-level functional antitumor CAR-T-cell outputs and low-level AICD. Our work here demonstrates that a CAR structural library screen technique using the in vitro stress test assay may be a valuable additional parameter to integrate into CAR engineering. It is conceivable that genetic strategies might limit the susceptibility of hyperactive CAR constructs to undergo AICD, such as forced overexpression of cFLIP or Toso, or, vector-directed synthesis of siRNAs that knock down FasL or Fas (38, 39). Additional secondary consequences of CAR overtuning also require interrogation, such as the predilection of hyperactive CARs to trigger expression of inhibitory receptors, such as PD-1, capable of enforcing an exhausted T-cell functional status within PD-L1⁺ solid tumors (40, 41). Our data demonstrate in a solid tumor model that (i) CAR structure function in vitro testing using commonly employed functional assays can misdirect the selection of candidate constructs as common practice is to focus on those constructs that display the highest functional activity, and (ii) potency tuning of CAR-redirected effector CTLs has an upper limit above which gains in the magnitude of effector outputs are negated by augmentation in AICD upon recursive triggering through the CAR. These results have guided our selection of a CD171-specific short spacer CAR for a phase I study in children with relapsed/refractory neuroblastoma.

M.C. Jensen reports receiving commercial research support from, has ownership interest (including patents) in, and is a consultant/advisory board member for Juno Therapeutics, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Künkele, K.S. Kelly-Spratt, M.C. Jensen

Development of methodology: A. Künkele, A.J. Johnson, L.S. Rolczynski, C.A. Chang, K.S. Kelly-Spratt, M.C. Jensen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.S. Rolczynski, C.A. Chang, V. Hoglund, K.S. Kelly-Spratt

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Künkele, A.J. Johnson, L.S. Rolczynski, C.A. Chang, K.S. Kelly-Spratt, M.C. Jensen

Writing, review, and/or revision of the manuscript: A. Künkele, A.J. Johnson, L.S. Rolczynski, C.A. Chang, K.S. Kelly-Spratt, M.C. Jensen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Künkele, L.S. Rolczynski, C.A. Chang, K.S. Kelly-Spratt, M.C. Jensen

Study supervision: A. Künkele, K.S. Kelly-Spratt, M.C. Jensen

The authors thank A. DeWispelaere, A. Ravanpay, and C. Crane for helpful advice. The Jensen laboratory at BTCCCR is the recipient of financial support from the Ben Towne Pediatric Cancer Research Foundation, Make Some Noise: Cure Kids Cancer Foundation Incorporation, Seattle Children's Guild Association, Andrew McDonough B+ Foundation, Journey4A Cure, Zulily, St. Baldrick's Foundation, and Life Sciences Discovery Fund.

A. Künkele was supported by the German Research Foundation (DFG, Deutsche Forschungsgemeinschaft). M. Jensen was supported by a Stand Up To Cancer–St. Baldrick's Pediatric Dream Team Translational Research grant (SU2C-AACR-DT1113). Stand Up To Cancer is a program of the Entertainment Industry Foundation administered by the American Association for Cancer Research.

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.