Preventing Lck Activation in CAR T Cells Confers Treg Resistance but Requires 4-1BB Signaling for Them to Persist and Treat Solid Tumors in Nonlymphodepleted Hosts

Adoptive immunotherapy with T cells expressing tumor-specific chimeric antigen receptors (CAR) holds tremendous promise for patients with cancer (1, 2). CARs are recombinant transmembrane receptors that link an antigen-binding extracellular domain with T-cell signaling moieties, thereby reprogramming a T cell to recognize cell surface target antigens independent of MHC presentation (3). This strategy has been remarkably successful against hematologic malignancies, where anti-CD19 CARs have generated complete remission rates of up to 90% in patients with relapsed and refractory disease who were previously considered incurable (4–6). This approach has also been leveraged against solid tumors, where CARs have targeted GD2 in neuroblastoma (2, 7), HER-2 in glioblastoma (8), CAIX in renal cell carcinoma (9), and ERBB2 in metastatic breast cancer (10), among others. Although CARs have been shown to elicit therapeutic responses against solid tumors, there remains room for improvement.

Intrinsic CAR deficiencies and tumor immunosuppression are 2 leading hypotheses that may explain the failure of this approach against solid tumors (11). It is now well accepted that antitumor efficacy necessitates robust CAR expansion and persistence in vivo, and most studies have leveraged lymphodepleting chemotherapy and/or radiotherapy regimens prior to adoptive transfer for this purpose (12). Preparative host lymphodepletion promotes CAR expansion, survival, and functionality by improving the bioavailability of homeostatic gamma chain cytokines and depleting inhibitory cell populations that otherwise impede effector T-cell responses (13). In clinical practice, however, these regimens can be exceedingly toxic and sometimes intolerable (14, 15). Moreover, although T-cell numbers ultimately return to homeostatic levels during immune reconstitution, these cytoreductive measures have a significant impact on the repertoire diversity of emerging T cells, and long-term sequelae can include autoimmunity, reduced resistance to infection, and other serious clinical consequences owing to reduced immune competence (16–18). Therefore, identifying alternative methods of improving CAR persistence and counteracting immunosuppression remain a major priority.

Modern CARs uniformly include the CD3 zeta chain (CD3z) to mimic signal 1 of T-cell activation akin to endogenous T-cell receptor (TCR) mechanisms, and usually CD28 for provision of signal 2. These secondary signals have been shown to enhance CAR expansion in mice (19) and humans (20). Despite improvements in the physical and functional attributes of CARs, however, they remain vulnerable to extrinsic tumor-induced immunosuppression, of which CD4⁺FoxP3⁺ regulatory T cells (Tregs) are frequently implicated (21–24). Tregs impede effector T-cell proliferation and responsiveness both directly and indirectly through physical contact (e.g., granzyme B; ref. 25), cytokine secretion (e.g., IL10 and TGFB; ref. 26), CTLA-4–induced expression of IDO by antigen-presenting cells (APC; ref. 27), and importantly, through IL2 consumption (28). Effector T cells invariably contribute to these mechanisms by secreting IL2 upon antigen encounter, which enhances the generation of Tregs (29). Importantly, CARs designed to incorporate costimulatory signaling (e.g., CD28) have been shown to inadvertently potentiate Tregs through this IL2 axis (21).

Here, we investigated the relevance of CAR-mediated IL2 secretion on antitumor efficacy using the B16 melanoma model. To abrogate IL2 secretion, 2 amino acid substitutions were introduced in the CAR transgene to prevent lymphocyte-specific tyrosine kinase (Lck) ligation to the CD28 cytosolic tail (30). Mutagenesis was targeted to the PYAP signaling motif of CD28, which controls net IL2 secretion through posttranscriptional regulation of IL2 mRNA (31–33), and did not interfere with the YMNM motif to retain PI3K signaling. We show that elimination of CD28-mediated IL2 induction impairs CAR engraftment in vivo. When this deficiency in autocrine IL2 signaling is compensated for by 4-1BB costimulation, however, CARs accumulate in blood circulation, resist Treg immunosuppression, and suppress tumor growth in nonlymphodepleted mice, thereby reducing the need for toxic lymphodepletion.

The EGFRvIII-specific CAR vector has been described previously (12, 34). Briefly, second-generation CARs were defined by the intracellular signaling components, CD28 and CD3z, and third-generation CARs were defined by CD28, 4-1BB, and CD3z. To produce CARs deficient in Lck signaling, 2 amino acid substitutions were introduced in the PYAP motif in the CD28 signaling domain of the CAR transgene through site-directed mutagenesis (30). CARs were generated by retroviral transduction (35). Briefly, retroviral supernatant was produced by cotransfection of HEK 293T cells using Lipofectamine 2000 Transfection Reagent (Invitrogen) with MSGV1-EGFRvIII CAR retroviral vectors and pCL-Eco helper plasmid (Imgenex). On the same day, splenocytes were freshly harvested from C57BL/6NCr mice (Charles River Laboratories) and cultured in R10 mouse T-cell media supplemented with 50 U/mL IL2 and 2.5 μg/mL Concanavalin A. After 48 hours, splenic T cells were transduced with retroviral supernatant on nontissue culture 24-well plates previously coated with 0.5 mL of RetroNectin (Clontech) at a concentration of 25 μg/mL in PBS. Cells were plated at a density of 1 × 10⁶/mL in viral supernatant supplemented with 50 U/mL IL2. Cells were split every 24 hours for 2 days.

To verify expression of CARs on the surface of T cells, transduced cells were stained for anti-CD3-PE and an EGFRvIII-specific CAR-APC multimer developed in our laboratory using standard flow cytometry staining procedures (12). Cells were gated on the lymphocyte population by forward/side scatter. Transduction efficiency was determined by paired control T cells.

CARs were coincubated with 1 × 10⁵ B16 melanoma tumor cells expressing EGFRvIII (B16vIII) previously radiolabeled with chromium⁵¹ (Cr⁵¹) at the indicated ratios. B16 melanoma tumor cells expressing OVA (B16OVA) were used as negative controls to assess antigen-specific tumor cell lysis. Briefly, B16vIII or B16OVA tumor cells were harvested, washed twice in culture medium, counted, and resuspended at a concentration of 1 × 10⁷/mL. Tumor cells were labeled with 100 μCi of radioactive Cr⁵¹ for 90 minutes, washed twice in R10 mouse T-cell media, and plated on a 96-well round bottom plate. CARs were plated as a serial dilution to generate effector-to-target (E:T) ratios as indicated. Cells were incubated for a total of 4 hours. After incubation, supernatant was harvested and radioactivity was measured to determine the release of Cr⁵¹. For intracellular cytokine staining (ICS), 96-well flat bottom plates were coated overnight with PBS alone or containing 50 μg/mL of the EGFRvIII-derived 14–amino acid peptide recognized by the EGFRvIII CAR (12). A total of 1–5 × 10⁵ CARs were plated in duplicate on vehicle or peptide-coated wells in R10 mouse T-cell media containing 1 μL/mL Golgi Plug and Golgi Stop reagents (BD Biosciences). Cells were incubated for 4 hours, harvested, and stained for cell surface expression of the CAR and CD8 using anti-CD8-PerCPCy5.5 and CAR-APC multimer reagents. Next, cells were permeabilized, stained with anti-IL-2-AlexaFluor-488 and anti-IFNγ-PE, and analyzed by flow cytometry. To perform a cytometric bead array (BD Biosciences), CARs were similarly stimulated and supernatant was collected to quantify levels of IL2 and IFNγ per manufacturer instructions.

C57BL/6NCr mice were purchased from Charles River Laboratories and C57BL/6-Tg(Foxp3-DTR/EGFP)23.2Spar/Mmjax (DEREG) mice were purchased from Jackson Laboratory. The B16vIII melanoma line has been described previously (36). B16vIII tumor cells were cultured in vitro in DMEM with 10% FBS.

For in vivo experiments, tumor cells were grown to 60%–70% confluency, harvested, washed twice in PBS, counted, and diluted to 5 × 10⁵/mL in PBS. Two hundred microliters of cell suspension was injected into the left flank of mice for a total inoculum of 1 × 10⁵ B16vIII cells. Tumors were allowed to engraft for 7 days before treatment was administered. In experiments requiring lymphodepletion, mice were subjected to 5 Gy total body irradiation (TBI) on the seventh day prior to therapy. For adoptive transfer, CARs were harvested, washed twice in PBS, counted, and diluted to the appropriate concentration in a total volume of 200-μL PBS and infused intravenously. All animal experiments were performed according to Duke University Institutional Animal Care and Use Committee–approved protocols. Tumor volume (millimeters cubed) was calculated by the formula (length × width² × 0.52) in a perpendicular fashion. Mice were euthanized when tumors reached a volume of 2,000 mm³ or upon evidence of ulceration. Tumor measurements were included in analyses up to the day of euthanasia except in cases of ulceration, where the previous recording was considered the final measurement.

Splenocytes were harvested from DEREG mice, and CD4⁺ splenocytes were isolated using CD4 T-cell isolation reagents from Miltenyi Biotec. FoxP3⁺ Tregs were cell-sorted from the CD4 compartment based on GFP expression, and sorted cells were cultured in vitro for 3–7 days in the presence of CD3/CD28 MACSiBead Particles using the murine Treg Expansion Kit reagent from Miltenyi Biotec. On the day of adoptive cotransfer, cells were re-sorted for GFP expression to ensure high purity. A total of 3–5 × 10⁴ Tregs were mixed with CARs at a 10:1 CAR–Treg ratio and infused intravenously in a total of 200-μL PBS.

Five days after adoptive transfer, mice were bled retro-orbitally to enumerate the absolute count and frequency of CARs in the CD3 T-cell compartment. Briefly, peripheral blood was collected and 50 μL were diluted with 100-μL PBS containing 2% FBS. Blood was stained with anti-CD3-PE and CAR-APC multimer for 20 minutes. Samples were then diluted with 1 mL of 1× red blood cell (RBC)-lysing solution (BD Biosciences) and incubated for an additional 20 minutes. Cells were washed twice in PBS containing 2% FBS. Fifty microliters of Flow-Count Fluorospheres (Beckman Coulter) was added to each sample to enumerate absolute counts of CARs per μL of blood. Cells were analyzed by flow cytometry. Absolute counts and frequencies were determined by gating on cells using forward/side scatter, CD3⁺, and CAR⁺ gates set based on paired negative controls. To monitor Tregs in peripheral blood circulation over time, a similar procedure was adopted, where blood was surface stained for anti-CD4-FITC. After RBC lysis, cells were washed with PBS, and permeabilized in 1 mL of 1:3 fixation/permeabilization diluent (eBioscience) for 20 minutes. Cells were washed twice in 1× permeabilization buffer (eBioscience) and stained for anti-FoxP3-APC for 20 minutes. Cells were washed and analyzed by flow cytometry.

Unpaired t tests were used to compare groups with respect to percent specific lysis in standard 4-hour Cr⁵¹ release assays. A 1-way ANOVA followed by pairwise comparisons using the Tukey post hoc test, or 2-way ANOVA with interaction was used to compare groups with respect to cytokine secretion and T-cell presence in peripheral blood circulation. For analysis of tumor growth curves, mixed effect linear models, which assume a first-degree autoregressive structure among repeated measurements within each mouse, were employed using log-transformed tumor volume. Pairwise contrasts were used to assess differences in slopes of log-transformed growth curves among treatment groups. Repeated measures for calculation of slopes incorporated time between detectable tumor and maximum tumor size.

We developed second-generation CD28-CD3z (CD28z) CARs (Fig. 1A) targeting the tumor-specific type III mutation of the EGF receptor, EGFRvIII. The CAR was designed to encode a single-chain variable fragment (scFv)-derived from the EGFRvIII-specific human mAb clone 139, a CD8 transmembrane domain, and the CD28 and CD3z signaling domains within the MSGV1 retroviral backbone as described previously (12, 34). We confirmed expression of CD28z CARs using an EGFRvIII-CAR–specific multimer, and observed a transduction efficiency of 92.6% in activated CD3⁺ T cells (Fig. 1B). To evaluate CARs against a highly aggressive and syngeneic tumor model, we utilized B16-EGFRvIII (B16vIII) melanoma, which expresses the murine homolog of EGFRvIII and has been described previously (36). We confirmed in vitro cytotoxicity and antigen-specific lysis of target tumor cells in a standard 4-hour Cr⁵¹ release assay, where CD28z CARs were coincubated in the presence of Cr⁵¹-labeled B16vIII tumor cells in various E:T ratios or with B16OVA tumor cells as a negative control (Fig. 1C).

Next, we tested whether CD28z CARs were capable of mounting antitumor responses in vivo. With rare exception, the efficacy of T-cell immunotherapies in preclinical cancer models and in humans has been dependent on the induction of nonmyeloablative host lymphopenia prior to adoptive cell transfer (12, 37–39). To confirm this observation in our model, we performed experiments to determine whether CD28z CARs required preparative host lymphodepletion to mount a response against established tumors. We implanted B16vIII tumor cells in the left flank, and after 7 days, intravenously infused CD28z CARs with or without prior sublethal TBI. Immunotherapy with CD28z CARs alone mounted minimal responses on tumor growth, as tumors reached humane endpoints in a similar timeframe to control mice (Fig. 1D). In the context of lymphodepletive preconditioning, however, CD28z CARs induced cures in 100% of mice.

Given the significant depletion of Tregs and the kinetics of their sustained contraction in irradiated mice (Fig. 1E), we considered the CAR–Treg interaction to be a plausible explanation for treatment failure in nonlymphodepleted mice. This hypothesis was compelling given previous evidence documenting Treg suppression of CAR antitumor responses (23). We leveraged an opportunity to circumvent Treg inhibition in vivo by modulating the CAR-dependent IL2 axis. The introduction of CD28 into CAR vectors has been previously shown to increase the generation of Tregs by enhancing IL2 secretion and targeted disruption of this pathway reduces Treg infiltration into solid tumors (21). However, given the central role of autocrine IL2 signaling in the survival and proliferation of activated T cells, we sought to determine whether this approach would benefit or hinder CAR immunotherapy in the setting of advanced cancer in nonlymphodepleted hosts.

The CD28 cytosolic tail of the CAR transgene contains a distal PYAP signaling motif that recruits the Src family kinase Lck, which upon CAR activation, transduces the necessary signals for IL2 secretion. Although the proximal YMNM motif has also been implicated in the regulation of IL2 transcription via PI3K-dependent signaling, the PYAP motif controls net IL2 secretion as it plays a dominant role in posttranscriptional stabilization of IL2 mRNA (31–33). Two amino acid substitutions were introduced in the PYAP motif of the CAR transgene (Fig. 2A), as described previously, to prevent Lck activation and abrogate CAR-mediated secretion of IL2 following antigen encounter (21, 30). We confirmed expression of second-generation CD28-modified CARs (ΔCD28z) on the surface of T cells and observed a transduction efficiency of 93.9% (Fig. 2B). We confirmed cytotoxicity and antigen-specific target cell lysis in vitro as described above (Fig. 2C).

Next, we conducted experiments to characterize the in vitro functionality of ΔCD28z CARs to confirm that the newly introduced modification reduced IL2 production as intended. We subjected CD28z or ΔCD28z CARs to antigen-specific stimulation using a plate-bound EGFRvIII-derived 14–amino acid peptide and evaluated IL2 positivity by ICS and IL2 secretion using a cytometric bead array (CBA). As shown, we observed a mean frequency of 16.6% of CD28z CARs, which were IL2⁺, compared to just 5.41% of ΔCD28z CARs (Fig. 2D). We confirmed that modified CARs retained the ability to secrete the prototypical proinflammatory cytokine, IFNγ, (Fig. 2E; CD28z CARs, mean: 51.65%; ΔCD28z CARs, mean: 44.8%). Cytokine quantitation supported these results, as IL2 levels measured in supernatant obtained from ΔCD28z CARs were significantly reduced (Fig. 2F; CD28z CARs, mean: 31.97 pg/mL; ΔCD28z CARs, mean: 9.3 pg/mL), whereas levels of IFNγ were unchanged (Fig. 2G; CD28z CARs, mean: 453.3 pg/mL; ΔCD28z CARs, mean: 439.67 pg/mL). Importantly, these data indicate that the ΔCD28 modification does not impair CAR expression, IFNγ secretion, cytotoxicity, or antigen specificity, but does significantly impair the secretion of IL2 following antigen-specific stimulation.

We hypothesized that ΔCD28z CARs would circumvent Treg immunosuppression, given disruption of the CAR-mediated IL2–Treg axis, enabling them to inhibit tumor growth. We tested this in tumor experiments using nonlymphodepleted mice. We intravenously injected CD28z CARs or ΔCD28z CARs into mice bearing 7-day established solid tumors. As previously observed, CD28z CARs were not efficacious, but neither were ΔCD28z CARs (Fig. 2H). To determine whether ΔCD28z CARs were efficacious in the context of lymphodepletion, we repeated the experiment, but this time subjected mice to TBI prior to therapy. We irradiated mice and intravenously injected CARs 7 days after tumor establishment; CD28z CARs were highly therapeutic as expected, but surprisingly, ΔCD28z CARs had no impact on tumor growth compared with control mice (Fig. 2I). These findings suggest that preventing Lck activation in ΔCD28z CARs had irreparably compromised CAR activity in vivo.

To both avoid Treg stimulation and to overcome the detrimental effects of reduced autocrine IL2 signaling on CAR activity, we explored whether introducing an additional costimulatory domain in CAR design would provide sufficient compensatory signaling through alternative pathways. 4-1BB is a TNF receptor (TNFR) family member that is essential for optimal CD8⁺ T-cell activation, as it enhances T-cell survival and cell-cycle progression to promote T-cell proliferation and persistence in an IL2-independent manner (40). Therefore, we designed and evaluated a novel ΔCD28z CAR, which incorporated 4-1BB (ΔCD28-4-1BBz CARs; Fig. 3A) based on the rationale that 4-1BB rescue would result in markedly enhanced in vivo activity compared with ΔCD28z CARs. We confirmed cell surface expression of ΔCD28-4-1BBz CARs on T cells (91%; Fig. 3B), and cytotoxicity and antigen-specific target cell lysis in vitro as described above (Fig. 3C).

Next, we evaluated the impact of ΔCD28-4-1BBz CARs on tumors in vivo and found that treatment significantly retarded tumor growth without a need for prior host lymphodepletion, whereas all previously designed CARs had no measurable impact in the absence of lymphodepletion (Fig. 3D). Given the well-described role of 4-1BB costimulation in T-cell proliferation, and our hypothesis that deficient Lck signaling may have crippled the ability of ΔCD28z CARs to engraft in vivo, we explored whether this observed efficacy could be explained by an increased presence of ΔCD28-4-1BBz CARs. To do so, we bled mice retro-orbitally 5 days after adoptive transfer to monitor CAR presence. Importantly, mice treated with second-generation CARs exhibited no significant difference over control mice in absolute counts or frequency of CARs in blood, whereas mice treated with ΔCD28-4-1BBz CARs recorded significantly higher counts and frequencies of the CD3 compartment (Fig. 3E). These data indicate that 4-1BB costimulation enhances the in vivo engraftment of CARs deficient in Lck signaling, providing a therapeutic advantage in nonlymphodepleted mice.

To control for the possibility that 4-1BB addition alone resulted in the observed antitumor responses, we produced and tested third-generation CARs with a wild-type CD28 (CD28-4-1BBz CARs; Fig. 4A). We confirmed cell surface expression (91.7%, Fig. 4B), cytotoxicity, and antigen-specific lysis as described above (Fig. 4C). Next, we directly compared CD28-4-1BBz CARs to ΔCD28-4-1BBz CARs in vivo, and importantly, we found that CD28-4-1BBz CARs had no therapeutic impact on tumor growth, whereas ΔCD28-4-1BBz CARs significantly inhibited tumor progression in nonlymphodepleted mice (Fig. 4D). This observation was in line with our original hypothesis, as we envisioned a need to modulate CAR–Treg interactions through the introduction of the ΔCD28 modification.

Although we expected 4-1BB addition to improve CAR accumulation in peripheral blood, we hypothesized that this effect would be enhanced in Lck-inactive CARs, providing an explanation for differences in antitumor efficacy between these 2 constructs. Therefore, we performed a direct comparison between all manufactured CARs to assess how they accumulated in vivo by measuring their presence 5 days after adoptive transfer into hosts. As previously observed, second-generation CARs were not significantly higher compared with negative controls, whereas third-generation CARs were markedly enhanced in both absolute numbers and frequencies (Fig. 4E). Importantly, however, mice treated with ΔCD28-4-1BBz CARs exhibited significantly higher counts (mean of 32.8 vs. 19.9/μL blood) and frequencies (2.5% vs. 1.3% of the CD3 compartment) compared with CD28-4-1BBz CARs, suggesting that the effects of a modified CD28 domain synergized with 4-1BB signaling to enhance CAR presence in circulation.

We next tested our central hypothesis that CARs deficient in Lck signaling would be able to circumvent Treg immunosuppression. We performed experiments in lymphodepleted mice, which allowed us to sensitively study the net in vivo effect of CAR–Treg interactions by predefining either cell population in purity and number prior to adoptive cotransfer. First, we established that both CD28-4-1BBz and ΔCD28-4-1BBz CARs perform equally in the absence of Tregs by adoptively transferring CARs into tumor-bearing mice after irradiation (Fig. 5A). Next, we explored CAR susceptibility to Treg immunosuppression by reintroducing Tregs into this lymphodepleted system. We utilized C57BL/6-Tg(Foxp3-DTR/EGFP)23.2Spar/Mmjax (DEREG) transgenic mice, which display DTR-eGFP expression in fully functional CD4⁺ Foxp3⁺ T cells and, therefore, allow specific detection and purification of Treg populations (41). Isolated Tregs were activated and expanded in vitro to ensure suppressive functionality and re-sorted before adoptive transfer to ensure high purity (Fig. 5B). Tumor-bearing mice were irradiated 7 days after tumor engraftment, and administered CARs alone or with Tregs at a 10:1 CAR:Treg ratio. We found that cotransfer completely abolished the antitumor efficacy of CD28-4-1BBz CARs in lymphodepleted hosts, demonstrating that the sole addition of Tregs was sufficient to abrogate the effect of wild-type CARs (Fig. 5C). In stark contrast, mice treated with ΔCD28-4-1BBz CARs remained completely resistant to infused Tregs, and tumors in mice receiving cotransfer displayed delayed growth kinetics identically to mice which received ΔCD28-4-1BBz CARs alone. In sum, these data demonstrate that ΔCD28-4-1BBz CARs are able to resist Treg immunosuppression and mount antitumor efficacy against established solid tumors without the need for prior host lymphodepletion.

CARs are capable of inducing complete responses in patients with hematopoietic cancers and achieving similar responses against solid tumors remains a major clinical priority. Although several mechanisms of increasing efficacy remain under study, it is now clear that CARs face several challenges when targeting solid tumors (42–47), including inhibition by Tregs (48). Importantly, the recent finding that tumor-specific T cells can themselves contribute to these counterproductive axes (e.g., by promoting Treg activity; ref. 49), has provided insight into how CAR design can unfavorably perturb tumor immune milieu and how they may be redesigned to function optimally against solid tumors.

Tregs are a frequent component of human solid cancers and tumor microenvironments (50), considered a major driver of tumor immune evasion, and have been shown to suppress antitumor responses by CARs (23, 24). Tregs depend on local T cells for production of IL2, which is requisite for Treg proliferation and their suppression of effector T cells (51–53). Importantly, modern CARs can contribute to this network through CD28 signaling (21). Kofler and colleagues previously demonstrated the relevance of CAR-mediated IL2 secretion by showing that deletion of the Lck-binding motif on CD28 in second-generation CARs was coincident with reduced intratumoral Treg infiltration compared with their wild-type counterparts (21). However, these experiments were performed using a model of intratumoral CAR delivery, which limits the interpretation of the role of Lck signaling in the persistence and antitumor efficacy of systemically infused CARs.

To investigate this, we leveraged the availability of a unique B16-EGFRvIII melanoma tumor model. This poorly immunogenic model displays an aggressive growth pattern in vivo, has been shown to become heavily infiltrated with Treg, and enables sensitive study of CAR activity within the context of a fully syngeneic and immunocompetent system. Moreover, we have previously shown these CARs to be capable of recognizing and eradicating EGFRvIII-expressing tumor cells both in vitro and in vivo, allowing us to focus our efforts on studying intracellular CAR signaling without concerns over the binding affinity and antigen-specificity of the scFv and target cell surface antigen.

Here, we build on previous findings by demonstrating that Lck inactivity is detrimental to CD28z CAR engraftment and antitumor efficacy in the absence of additional 4-1BB costimulatory signaling. Although CD28z CARs were found to be efficacious in lymphodepleted mice, ΔCD28z CARs completely failed to induce a therapeutic impact despite preparative host lymphodepletion, suggesting that IL2 signaling may be an indispensable component of second-generation CAR activity following intravenous delivery. We hypothesized that the absence of autocrine IL2 signaling, which has a central role in the growth and function of T cells (54), compromised CAR durability in vivo as related to the physical and functional persistence of CARs in the host (2, 55, 56). We therefore designed CARs to provide compensatory signaling by incorporating the 4-1BB signaling domain into the CAR transgene.

4-1BB has been shown to enhance T-cell expansion, increase Th-1 cytokine secretion (57), prevent activation-induced cell death (58, 59), and promote differentiation (60, 61) and cell survival through the upregulation of antiapoptotic genes (57, 62–64). We show that 4-1BB addition significantly improved CAR accumulation in systemic blood circulation compared with second-generation CARs. Importantly, however, these third-generation CARs accumulated to significantly higher numbers and frequency when Lck signaling was defective, and only these CARs were able to mount therapeutic efficacy against tumors in nonlymphodepleted mice. We conclude that 4-1BB–mediated CAR accumulation is insufficient on its own to induce antitumor efficacy in nonlymphodepleted hosts and that ΔCD28 is also a required modification. We show that when these modifications are present together, ΔCD28-4-1BBz CARs accumulate to higher levels in blood circulation and completely resist Treg immunosuppression.

Mechanistically, we believe these data would have been strengthened by histologic or flow cytometric evidence of reduced Treg accumulation at tumor sites as a result of the ΔCD28 modification, as published previously (21). Although the data in Fig. 5 support the concept that ΔCD28-4-1BBz CARs can confer Treg resistance under conditions of cotransfer to lymphodepleted recipients, it does not prove that ΔCD28-4-1BBz CARs prevent Treg accumulation intratumorally. Further experiments will be required to investigate this. Our study would have also benefited from an analysis of the tumor microenvironment to enumerate relevant cytokine levels and to corroborate CAR infiltration and persistence following systemic infusion. The ΔCD28-4-1BBz CAR design may also have an effect upon the cell-intrinsic properties of T cells that could be relevant to antitumor efficacy (65, 66), which were not investigated here. Signaling via CD25 in CD8⁺ T cells drives differentiation towards short-lived effector cells and thus, it is also possible that ΔCD28-4-1BBz CARs adopt a “less differentiated” polyfunctional phenotype with greater persistence in vivo over time. Further studies relating to phenotype (e.g., KLRG1, CD127, T-bet, and Eomes) and survival over time will be needed to address this possibility.

To our knowledge, this is the first report showing tumor-specific CARs can induce major antitumor responses when used as a monotherapy against established solid tumors in mice without a need for preparative lymphodepletion. Although these data represent a step in the right direction, it must be noted that this survival advantage did not recapitulate the efficacy of CARs administered in the setting of lymphodepletion. It is also important to acknowledge that despite its known role as a tumor-specific driver mutation, EGFRvIII is not naturally expressed in the tumor model employed here. Although syngeneic models of glioblastoma, which naturally expresses EGFRvIII, are available, we leveraged subcutaneous B16-EGFRvIII melanoma to avoid the added variable of CNS immune privilege and the impact it could have on T-cell trafficking into the brain. Given our observations here, we are now exploring the CAR–Treg axis in an intracranial orthoptic tumor model that more accurately recapitulates the EGFRvIII expression pattern found in human disease.

Nonetheless, the work reported herein demonstrates that CARs can be designed to augment their vulnerability to tumor immunosuppression, which underscores the need to both bombard solid tumors with high numbers of tumor-specific T cells and to continue developing approaches to favorably improve cell-to-cell interactions that take place within the tumor and its microenvironment. In sum, the work highlighted here suggests that CARs deficient in Lck signaling may be a superior immunotherapy for solid tumors infiltrated with Treg, and that any induced defects can be compensated for by inclusion of 4-1BB in third-generation CARs. Importantly, this work also substantiates the notion that it is possible to modulate counterproductive immune interactions through CAR design to enhance adoptive T-cell immunotherapy.

No potential conflicts of interest were disclosed.

Conception and design: C.M. Suryadevara, R. Desai, B.D. Choi, P.E. Fecci, J.H. Sampson, L. Sanchez-Perez

Development of methodology: C.M. Suryadevara, R. Desai, P.E. Fecci, J.H. Sampson, L. Sanchez-Perez

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.M. Suryadevara, R. Desai, S.H. Farber, A.M. Swartz, S.H. Shen, P.C. Gedeon, D.J. Snyder, J.H. Sampson, L. Sanchez-Perez

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.M. Suryadevara, R. Desai, S.H. Farber, P.C. Gedeon, J.E. Herndon, P. Healy, E.A. Reap, J.H. Sampson

Writing, review, and/or revision of the manuscript: C.M. Suryadevara, R. Desai, S.H. Farber, B.D. Choi, A.M. Swartz, S.H. Shen, P.C. Gedeon, J.E. Herndon, P. Healy, P.E. Fecci, J.H. Sampson, L. Sanchez-Perez

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.M. Suryadevara, D.J. Snyder, G.E. Archer, J.H. Sampson

Study supervision: P.E. Fecci

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

The authors thank Steven A. Rosenberg and Richard A. Morgan of the Surgery Branch at the NCI for providing us with the EGFRvIII CAR retroviral construct. This work was supported by funding from the NIH [grants 1R01-NS099463-02, 1R01CA228232-01, 5R01-NS085412-04, 5R01-CA177476-045R01-NS086943-03, 5U01-NS090284-02, 5R25-NS065731-08, 5P50-CA190991-02 (all to J.H. Sampson); 4P01-CA154291-05 (to D.D. Bigner and J.H. Sampson); and T32-CA009111 (to B.R. Cullen)]. Additional support was provided by the Pediatric Brain Tumor Foundation.

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.