Research ArticleCancer

Tumor-targeted CD28 bispecific antibodies enhance the antitumor efficacy of PD-1 immunotherapy

See allHide authors and affiliations

Science Translational Medicine  24 Jun 2020:
Vol. 12, Issue 549, eaba2325
DOI: 10.1126/scitranslmed.aba2325

Immunotherapies combining forces

Although immunotherapy has been achieving increasing prominence in cancer treatment, these methods are not perfect, and many tumors still do not respond to treatment. One approach used for cancer immunotherapy is bispecific antibodies, which recognize a receptor on the surface of T cells and a tumor antigen, helping to bring the two types of cells together and activate the T cells. Building on recent advances in this field, Waite et al. used bispecific antibodies targeting CD28 on T cells and one of two different tumor antigens and then combined them with immune checkpoint therapy, showing that the two treatments reinforce each other.

Abstract

Monoclonal antibodies that block the programmed cell death 1 (PD-1) checkpoint have revolutionized cancer immunotherapy. However, many major tumor types remain unresponsive to anti–PD-1 therapy, and even among responsive tumor types, most of the patients do not develop durable antitumor immunity. It has been shown that bispecific antibodies activate T cells by cross-linking the TCR/CD3 complex with a tumor-specific antigen (TSA). The class of TSAxCD3 bispecific antibodies have generated exciting results in early clinical trials. We have recently described another class of “costimulatory bispecifics” that cross-link a TSA to CD28 (TSAxCD28) and cooperate with TSAxCD3 bispecifics. Here, we demonstrate that these TSAxCD28 bispecifics (one specific for prostate cancer and the other for epithelial tumors) can also synergize with the broader anti–PD-1 approach and endow responsiveness—as well as long-term immune memory—against tumors that otherwise do not respond to anti–PD-1 alone. Unlike CD28 superagonists, which broadly activate T cells and induce cytokine storm, TSAxCD28 bispecifics display little or no toxicity when used alone or in combination with a PD-1 blocker in genetically humanized immunocompetent mouse models or in primates and thus may provide a well-tolerated and “off the shelf” combination approach with PD-1 immunotherapy that can markedly enhance antitumor efficacy.

INTRODUCTION

Monoclonal antibodies (mAbs) that block the programmed cell death 1 (PD-1)/programmed cell death ligand 1 (PD-L1) checkpoint have been clinically approved in a number of cancer types, including melanoma, renal cell carcinoma, non–small cell lung cancer, and advanced metastatic cutaneous squamous cell carcinoma (15). However, even for these responsive cancer types, most patients do not respond; moreover, many important tumor classes—such as prostate, pancreatic, colorectal, and breast—respond poorly or not at all. Considerable efforts are ongoing to identify patients who are more likely to respond to checkpoint inhibition through identification of biomarkers (6); expression of PD-L1 and tumor mutational burden (especially when associated with microsatellite instability) are imperfect markers of responsiveness (7). Extensive combination approaches are being evaluated for their ability to improve the efficacy of PD-1 blockade and the durability of the antitumor response; many of these come with increased toxicities (8). Clearly, it would be desirable to have a targeted and well-tolerated immune therapy that could extend the benefit of PD-1 therapy to more patients.

T cell activation occurs when the T cell receptor/CD3 complex (TCR/CD3) binds to specific peptides in the context of the major histocompatibility complex (MHC) expressed by professional antigen-presenting cells (APCs), virally infected cells, or tumor cells; this initial trigger for T cell activation has been termed “signal 1.” TCR/CD3 complexes cluster together at the interface of the T cells and their target cells at the so-called “immune synapse,” concentrating intracellular signaling molecules, cytokines, and granzymes (9). Costimulatory and coinhibitory receptors also accumulate at the immune synapse, to either enhance (referred to as “signal 2”) or attenuate (termed “checkpoint” regulation) T cell activation, respectively. Cross-talk between the CD28 costimulatory pathway and the PD-1 checkpoint has been demonstrated in vitro and in vivo (10, 11), with PD-1 attenuating T cell activation through dephosphorylation of the TCR/CD3 complex and/or CD28.

An emerging class of immunotherapies termed TSAxCD3 bispecifics work by cross-linking a tumor-specific antigen (TSA) with the TCR/CD3 complex, thereby substituting for—and mimicking—signal 1. An early version of TSAxCD3 bispecifics, requiring administration via constant infusion, recently received regulatory approval for B cell acute lymphoblastic leukemia, with one arm binding to CD19 on the leukemia cells and the other arm binding to CD3 (1214). More advanced versions of CD3 bispecifics, which pharmacokinetically behave more like conventional antibodies and can be administered weekly or even less frequently, have demonstrated impressive clinical activity against non-Hodgkin’s lymphomas, by linking CD20 to CD3 (1519). Despite their promising clinical efficacy in some settings, CD3 bispecifics still do not work for many patients. Moreover, very few tumor antigens have been identified that have the requisite high abundance and the restricted tissue specificity that allows the use of CD3 bispecifics without risking damage to essential normal tissues. We have recently described a class of “costimulatory bispecifics” that cross-link a TSA to CD28 on T cells (TSAxCD28 bispecifics), thereby mimicking signal 2 and markedly synergizing with TSAxCD3 bispecifics. TSAxCD28 bispecifics can allow for increased responsiveness to TSAxCD3 bispecifics and even allow for targeting a TSA that is poorly expressed and thus cannot be addressed by a TSAxCD3 bispecific alone (20). However, combining these two classes of bispecifics still depends on a target for the TSAxCD3 bispecific that is highly restricted; otherwise, there is substantial risk of damage to the normal tissues expressing the TSA.

Here, we describe an immunotherapeutic approach combining TSAxCD28 bispecifics with PD-1 inhibition. We provide two examples of TSAxCD28 bispecifics—one for a tissue-specific antigen [prostate-specific membrane antigen (PSMA)] and one for a broadly expressed antigen [epidermal growth factor receptor (EGFR)] and show that this approach does not depend on highly restricted expression of the TSA. Each of these TSAxCD28 bispecifics enhanced responsiveness to anti–PD-1—even endowing responsiveness to tumors that do not respond to anti–PD-1 alone—and induced long-lived antitumor immunity and promoted robust intratumoral T cell activation and T cell memory without systemic cytokine release in animal models. Studies in genetically humanized immunocompetent mice and in cynomolgus monkeys demonstrated that TSAxCD28 bispecifics were well tolerated when administered alone or in combination with anti–PD-1 antibody. Collectively, these data suggest that combining this class of CD28-based bispecifics (TSAxCD28) with PD-1 inhibition may provide an “off the shelf” biological combination therapy with markedly enhanced, specific, and synergistic antitumor activity and without the need for patient customization.

RESULTS

Antitumor efficacy of anti–PD-1 can be potentiated by enforced overexpression of “natural” CD28 ligands on tumor cells

Previous studies have shown that engineered expression of CD28 ligands on tumor cells can stimulate T cell–mediated tumor immunity against immunogenic tumor lines (2124). To determine whether the antitumor efficacy of anti–PD-1 can be enhanced when CD28 binds its natural ligand(s), we used the well-established C57BL6 syngeneic MC38 mouse tumor model; MC38 cells were engineered to overexpress murine CD86, one of the costimulatory ligands for CD28 (MC38/CD86, fig. S1A). MC38/CD86 tumor growth was reduced compared to control MC38 cells transfected with an empty vector control (MC38/EV) in immunocompetent wild-type (WT) mice but not immunodeficient Rag2−/− mice, demonstrating tumor growth inhibition due to a specific immune response and not tumor cell intrinsic difference in growth rate (fig. S1B). The antitumor efficacy of anti–PD-1 treatment, as assessed by tumor growth and survival (Fig. 1, A and B), was more impressive in mice implanted with MC38/CD86 tumors as compared with mice implanted with control MC38/EV; 8 of 10 mice implanted with MC38/CD86 tumors became tumor-free after anti–PD-1 treatment, whereas 1 of 10 mice implanted with control MC38/EV tumors became tumor-free after anti–PD-1 treatment. Depletion of CD8+ T cells during the course of treatment completely abrogated the antitumor benefit of combining anti–PD-1 therapy with MC38/CD86 cells (Fig. 1C), demonstrating that antitumor immunity was dependent on CD8+ T cells. Mice that had initially been implanted with MC38/CD86 cells and became tumor-free after treatment with anti–PD-1 developed long-term antitumor memory and immunity after “tumor rechallenge.” That is, more than 60 days after the implantation of the primary tumor, these mice were then able to reject a second tumor challenge of implanted MC38 parental cells without requiring any additional treatment (Fig. 1D). Consequently, these data demonstrate that the antitumor efficacy of anti–PD-1 can be enhanced when CD28 engages its natural ligand(s), resulting in CD8-dependent antitumor memory and long-term immunity.

Fig. 1 Expression of a CD28 ligand (CD86) on tumor cells potentiates PD-1 mAb treatment to induce CD8-dependent antitumor immunity.

MC38 tumor cells were transduced with the ligand for CD28, CD86 (MC38/CD86), or empty vector control (MC38/EV). WT C57BL6 mice were initially implanted with 1 × 106 tumor cells per mouse and treated with PD-1 mAb or rIgG2a isotype control at 5 mg/kg on days 0, 3, 7, 10, and 14 after tumor implants. (A) Average tumor volume over time. Data are represented as means ± SEM. Statistical significance was determined with two-way analysis of variance (ANOVA) and Tukey’s multiple comparisons tests. (B) Survival over time (percentage of mice with tumors <2000 mm3). Data are represented as means. Statistical significance at day 60 after implantation was determined with the log-rank (Mantel-Cox) test. Data shown in (A) and (B) are representative of two independent experiments with n = 10 mice per group. (C) Mice were treated with CD8-depleting antibody (CD8 depleted, dashed lines) or isotype control (no depletion, solid lines). Average tumor volume over time. Data are represented as means ± SEM. Statistical significance was determined with two-way ANOVA and Tukey’s multiple comparisons tests. Data represent two independent experiments with n = 5 mice per group. (D) Tumor-free mice previously implanted with MC38/CD86 tumor and treated with PD-1 mAb were rechallenged with a secondary tumor implant with 0.5 × 106 parental MC38 cells. Average tumor volume over time. Data are represented as means ± SEM. Data represent two independent experiments with n = 5 to 8 mice per group. Statistical significance is indicated (*P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001).

Previous studies have shown that MC38 tumors respond weakly to anti–PD-1 monotherapy treatment that is delayed until tumors are established (25, 26). Therefore, we tested whether delayed treatment with anti–PD-1 starting on day 7 after tumor implantation would induce rejection of established MC38/CD86 tumors. Antitumor efficacy with delayed anti–PD-1 treatment was increased in mice implanted with MC38/CD86 tumors as compared with mice implanted with control MC38/EV; 4 of 10 mice implanted with MC38/CD86 tumors became tumor-free after anti–PD-1 treatment, whereas 0 of 10 mice implanted with control MC38/EV tumors became tumor-free after anti–PD-1 treatment (fig. S1, C and D). As expected, delaying anti–PD-1 treatment reduced efficacy compared to immediate treatment described above (Fig. 1, A and B) although the enhanced effect of combination with MC38/CD86 was maintained.

PD-1 can exclude CD28 from the immune synapse

Efficient T cell activation depends on coclustering of TCR/CD3 and CD28 complexes at the immune synapse. PD-1 can directly inhibit signaling from both TCR/CD3 and CD28, in part because of its recruitment of the SHP-2 phosphatase to the synapse (10, 27). To determine the relative localization of CD28 and PD-1 at the immune synapse, we developed an in vitro model system using Jurkat T cells overexpressing PD-1 and Raji cells engineered to overexpress PD-L1. A fluorescently labeled CD20xCD3 bispecific antibody (15) was used to simulate peptide MHC/TCR binding and to visualize T cell interactions with the target cells forming an immune synapse. We used two different fluorescently labeled monoclonal PD-1 antibodies: a PD-1 mAb that does not block the interaction with PD-L1 (PD-1 nonblocker; table S1 and fig. S2, A and B) that was used to visualize the localization of PD-1 and a blocking PD-1 mAb (PD-1 mAb and REGN2810 cemiplimab) (28). As might be expected, PD-L1 expression on target cells promoted accumulation of PD-1 on T cells at the immune synapse, while excluding CD28 from the synapse (Fig. 2A and fig. S2, C to E). In contrast, in the presence of the blocking PD-1 mAb, PD-1 was no longer localized at the immune synapse, whereas CD28 localization into the immune synapse was present (Fig. 2B). The distribution of PD-1 and CD28 was quantified by calculating the ratio of antibody staining inside versus outside of the immune synapse (Fig. 2C). The quantitative data analysis confirmed that expression of PD-L1 on target cells enhanced PD-1 localization at the immune synapse while simultaneously decreasing CD28 localization at the synapse, and in contrast, PD-1 blocking mAb reduced the relative amount of PD-1 and restored CD28 expression, thereby increasing the relative ratio of CD28 to PD-1 at the synapse. In the same experimental setup, PD-L1 blocking or nonblocking mAbs resulted in similar redistribution of PD-L1 and CD28, suggesting that the change in CD28 localization is due to the presence or absence of PD-1/PD-L1 in the synapse and not caused by anti–PD-1 binding on T cells (table S1 and fig. S2, H to L). These experiments support an “immune synapse localization model” explaining how PD-1 blockade can potentiate CD28 activation: PD-L1:PD-1 engagement can act to exclude CD28 from the immune synapse and thus dampen T cell activation, whereas blocking PD-1 can instead allow CD28 to accumulate in the synapse, thus allowing it to more efficiently promote T cell activation when encountering its ligand on the target cell.

Fig. 2 TSAxCD28 bispecific and PD-1 blockade synergistically promote T cell activation in vitro.

(A and B) Images of T cell (Jurkat/PD-1) and target cell (Raji/PD-L1) conjugates in the presence of CD20xCD3 bispecific and PD-1 mAb nonblocker (A) or PD-1 blocker (B). PD-1 mAbs, CD20xCD3, and CD28 mAb were directly labeled with Alexa647 (shown in red), Alexa488 (shown in green), and PE (shown in blue). Nuclei were stained with Hoechst 33342 (shown in gray). Dashed lines are outlines of cells, drawn based on the bright-field image. Scale bar, 7 μm. (C) PD-1 and CD28 localization at the immunological synapse. Data are represented as means ± SEM. Statistical significance was calculated with an unpaired t test (****P < 0.0001). PD-1 nonblocker, n = 150; and PD-1 blocker, n = 175. Data in (A) to (C) are representative of three independent experiments. (D and E) Human T cell cytokine release assay with DU-145/hPSMA cells cocultured with the indicated antibodies. (D) Schematic of cytokine release assay setup. (E) IL-2 release at 96 hours. Data are represented as mean ± SEM. Data shown from one of three human T cell donors are tested and are representative of two independent experiments.

TSAxCD28 bispecifics enhance the ability of PD-1 blockade to induce T cell activation in vitro

Because we had demonstrated that antitumor efficacy of PD-1 blockade could be enhanced by allowing CD28 to engage its natural ligand(s) on tumor cells, perhaps in part explained by the above immune synapse localization model, we next wanted to test whether we could use a TSAxCD28 bispecific to similarly potentiate the efficacy of PD-1 blockade, first in vitro and then in vivo. For this purpose, we generated a PSMAxCD28 bispecific, specific for the prostate-specific antigen known as PSMA, using previously described approaches (15, 20).

We first assessed the effect of combining our PD-1 blocking mAb with the PSMAxCD28 bispecific on primary human T cell activation in vitro. We used a modified mixed lymphocyte reaction (MLR) to simulate physiological PD-L1 expression and TCR/CD3 stimulation. In a one-way MLR, incompatibility of allogeneic determinants causes T cell activation, which can be quantified by cytokine production. To generate a one-way MLR assay, T cells from healthy donors were incubated with allogenic DU145/PSMA cells, an engineered prostate cancer cell line that endogenously expresses PD-L1 and is engineered to overexpress PSMA (Fig. 2D). In this MLR assay, addition of 20 nM PD-1 mAb modestly increased interleukin-2 (IL-2) release [~3- to 4-fold over immunoglobulin G4 (IgG4) isotype control, white triangles, Fig. 2E and fig. S3A]. Similarly, the PSMAxCD28 bispecific caused a modest dose-dependent increase in IL-2 release (~3- to 4-fold over the IgG4 isotype control, orange circles, Fig. 2E and fig. S3A). However, in combination, PSMAxCD28 and 20 nM PD-1 mAb markedly potentiated T cell activation, such that IL-2 concentrations were increased by ~20-fold (Fig. 2E, blue squares). At a concentration of 20 nM, the IL-2 release induced by either PSMAxCD28 or PD-1 mAb alone was ~200 pg/ml while the combination resulted in over 1200 pg/ml, which is three times the additive effect of each monotherapy. These results demonstrate that the PSMAxCD28 bispecific can potently and synergistically combine with PD-1 blockade to promote T cell activation in the presence of tumor cells.

To further show that the enhanced T cell activation by PSMAxCD28 bispecific and PD-1 blockade combination was due to blocking PD-1 interaction with PD-L1, we tested the same blocking PD-1 mAb and compared it with the nonblocking PD-1 mAb and a blocking PD-L1 mAb as described above (table S1, Fig. 2, and fig. S2) (28). A blocking PD-L1 mAb does not engage the T cells and avoids any potential effects on agonizing PD-1 receptors on T cells other than preventing PD-L1 binding. Here, we used an MLR with T cells from healthy donors incubated with allogenic Raji cells that are engineered to express PSMA (tumor target for PSMAxCD28) and PD-L1 (the ligand for PD-1) and knocked out for CD80 and CD86 (the ligand for CD28) (fig. S3B). Increasing amounts of PSMAxCD28, in the presence of a blocking anti–PD-1 or anti–PD-L1 mAb but not the nonblocking PD-1 mAb, enhanced T cell activation, resulting in IL-2, interferon-γ (IFN-γ), and tumor necrosis factor–α (TNFα) cytokine release (fig. S3C). These results extend the above findings from the allogeneic DU145/PSMA MLR showing that in combination with PSMAxCD28 bispecific activation of CD28, blocking PD-1 and PD-L1 interaction, rather than simply engaging PD-1 on T cells, can promote T cell activation and production of multiple cytokines, which can contribute to a robust immune response.

We next asked whether the increased T cell activation was due to direct signaling through CD28 or simply due to increased T cell and target cell cross-linking and bringing T cells in close proximity with tumor cells. To test this, we compared the costimulation effect on T cells using bispecifics that bind to CD28 or CD27. CD27 is a member of the TNF receptor super family constitutively expressed on T cells that costimulates activation when bound to its ligand CD70 expressed on APCs (29). Using an in vitro allogeneic T cell response assay, T cells from healthy donors were incubated with allogeneic human embryonic kidney–293 (HEK293) cells expressing human CD20 and human MUC16. Addition of MUC16xCD28 (20) caused a dose-dependent increase in proliferation (fig. S4A) and IL-2 (fig. S4B), whereas the addition of MUC16xCD27 or isotype control antibody did not increase T cell proliferation or IL-2 release. The binding of MUC16xCD28 and MUC16xCD27 antibodies to primary T cells and target cells was tested by flow cytometry, showing that both of these antibodies bind to T cells and to MUC16-expressing target cells (HEK293/hCD20/hMUC16) (fig. S4C). These results demonstrate that the MUC16xCD28 bispecific, but not the MUC16xCD27 bispecific, can potently promote T cell activation in the presence of MUC16-expressing cells despite similar binding ability to T cells and MUC16-expressing cells. Although this does not exclude the possibility that CD27 agonists in other formats as well as other costimulatory molecules may have a role in mounting an antitumor response in similar settings (20, 30), this result demonstrates that TSAxCD28 specific agonism of CD28 receptors on T cells has potent antitumor efficacy.

PSMAxCD28 bispecific enhances the antitumor efficacy of PD-1 mAb treatment in a syngeneic tumor model

Because we had demonstrated that TSAxCD28 bispecifics could promote T cell activation by PD-1 blockade in vitro and that CD28 engagement by its natural ligand(s) could potentiate the antitumor efficacy of PD-1 inhibition in vivo, we next wanted to determine whether a TSAxCD28 bispecific could similarly potentiate antitumor efficacy in combination with a PD-1 mAb in vivo. Using the syngeneic MC38 tumor model, we expressed the human PSMA (hPSMA) gene in the MC38 cells, creating MC38/hPSMA cells as previously described (20). To avoid the possibility that the mice would spontaneously reject these otherwise syngeneic tumors simply because they were expressing an introduced human tumor antigen, we created mice that normally expressed (and would thus be tolerant to) the hPSMA protein by humanizing the PSMA gene in these mice using VelociGene technology (20, 31, 32). In addition, we also humanized the CD3γ-δ-ε and CD28 genes using VelociGene technology as previously described (20), so that the bispecific antibodies would recognize the host T cells (via hCD3 or hCD28), as well as the human tumor antigen (hPSMA) in both normal tissues and tumors (33). We confirmed that humanized CD3 and CD28 function was intact and did not disrupt normal T cell development (20). We confirmed that human CD28 interacts with mouse CD80 and CD86 similarly to human CD80 and CD86 (table S2 and fig. S5), validating previous studies (34). In this model, the PSMAxCD28 bispecific markedly synergized with the PD-1 mAb to block tumor growth (Fig. 3A) and resulted in markedly improved long-term survival (Fig. 3B). Monotherapy PSMAxCD28 or PD-1 mAb resulted in 15.6 and 12.1% survival, respectively, whereas the combination resulted in 70.8% survival, which is a more than 2.5-fold increase over the additive effect of the monotherapies. Once again, combining CD28 activation with PD-1 inhibition generated long-term antitumor memory and immunity: that is, after the initial tumor response due to treatment with the PSMAxCD28 bispecific and the PD-1 mAb, these mice were able to reject a second tumor challenge with parental MC38 tumor cells without any additional therapy, demonstrating again the generation of endogenous immune memory just as observed above with enforced expression of CD28 ligand (compare Fig. 1D with Fig. 3C). These results suggest that an endogenous antigen-specific TCR signal (signal 1) is being generated against a peptide/MHC complex on the MC38/PSMA implanted tumor cells, which is then being amplified by the combination of anti–PD-1 with the PSMAxCD28 bispecific (providing signal 2). Consistent with this, we and others have shown that MC38 tumor cells express reactivated endogenous retroviral proteins such as p15E and that C57BL6 mice implanted with these MC38 cells generate intratumoral T cells responsive to this p15E antigen (KSPWFTTL) (20, 35, 36). Furthermore, we observed that such mice only induced high numbers of peripheral T cells (outside of the tumor) reactive to the p15E peptide when they were treated with the PD-1 mAb together with the PSMAxCD28 bispecific, explaining the generation of antitumor immune memory by this combination treatment (fig. S6A). In addition, depletion of CD8+ but not CD4+ T cells abrogated the memory response, demonstrating that rejection of a second MC38 tumor rechallenge is CD8+ T cell–dependent (Fig. 3D). Furthermore, MC38 immune mice did not reject an irrelevant B16F10.9 tumor cell line, showing the specificity of the response (Fig. 3E). In experiments where treatment with PSMAxCD28 in combination with PSMAxCD3 induced MC38/hPSMA tumor rejection, those tumor-free mice failed to reject secondary tumor rechallenge (fig. S6, B and C), indicating failure to induce antitumor immune memory with this combination. Together, these results show that CD28 bispecifics can enhance the antitumor efficacy of PD-1 mAb in this syngeneic tumor model by boosting endogenous TCR/CD3-dependent T cell responses, which also result in long-term immune memory, and supporting the notion that this combination can improve the durability of antitumor responses.

Fig. 3 PSMAxCD28 synergizes with PD-1 mAb treatment to induce antitumor immunity.

MC38/hPSMA tumor cells were implanted in hCD3/hCD28/hPSMA mice subcutaneously (1 × 106 tumor cells per mouse). PSMAxCD28 bispecific antibody, PD-1 mAb, or rat IgG2a isotype control (Iso Ctrl) was administered as monotherapy or in combination by intraperitoneal injection at 5 mg/kg each. (A to E) Immediate treatment regimen (dosing indicated by arrows on days 0, 7, and 14). (A) Average tumor volume on day 21 after implant. Symbols represent the average from five independent experiments. Bars are means ± SEM. Statistical significance was determined with two-way ANOVA and Tukey’s multiple comparisons test. (B) Survival over time (mice with tumors < 2000 mm3). Data are represented as means from five independent experiments combined with n = 32 to 39 mice per group. Statistical significance at day 60 after implantation was determined with the log-rank (Mantel-Cox) test. Comparison with isotype control (*), PD-1 mAb (#), and PSMAxCD28 (^). (C) Tumor-free mice previously treated with PSMAxCD28 and PD-1 mAb combination were rechallenged with a secondary tumor implant with 1 × 106 MC38 parental tumor cells per mouse. Data are represented as means ± SEM. n = 5 to 7 mice per group. Data represent three independent experiments. (D) Mice that rejected the primary tumor implant or naïve mice were treated with CD4 or CD8 depletion antibodies or isotype control 1 day before implant with 1 × 106 MC38/PSMA tumor cells. Tumor volumes on day 16 after implant are shown for individual mice (n = 4 to 5 mice per group). Bars represent the means ± SEM. (E) Mice that rejected the primary tumor implant or naïve mice were implanted with 0.5 × 106 MC38 tumor cells and 0.2 × 106 B16F10.9 tumor cells on opposite flanks. Tumor volumes from individual mice are shown (n = 9 mice). Bars represent the means ± SEM. Data in (D) and (E) are from one experiment. (F to J) Delayed/therapeutic treatment regimen. Dosing indicated by arrows on days 9, 16, and 22 (F to G) and days 11 and 14 (H to J). (F) Average tumor volume over time. Data are represented as means ± SEM. Statistical significance was determined with two-way ANOVA and Tukey’s multiple comparisons tests. n = 7 mice per group. Data represent three independent experiments. (G) Ex vivo splenic and intratumoral cytokines from tissues harvested on day 29. Symbols represent data from individual mice. Bars are the means ± SEM. n = 4 to 6 mice per group. Data represent two independent experiments. (H to J) Tumors were harvested on day 17 for tumor CD8+ T cell immunophenotyping. (H) Tumor CD8+ T cell CITRUS analysis. Color gradient indicates normalized expression of each marker (red-blue, high-low expression). (I) MFI heatmap of clusters 1 and 2 (C1 and C2). (J) Relative frequency of cells in C1 and C2 from the indicated treatment groups normalized to the isotype control. Data are represented as means ± SEM. Data in (H) to (J) are representative of two independent experiments with n = 8 to 10 mice per group in each experiment. Data in (J) are combined from both experiments.

The above combination study was done “prophylactically,” by treating simultaneously with tumor implantation. In general, it is much harder to show treatment benefit when treatment is delayed until the tumors are well established. To examine the combination potential for treating established tumors, we used a delayed treatment protocol and found that the combination of PSMAxCD28 bispecific with PD-1 mAb (given 9 days after implantation) also inhibited growth of these established MC38/hPSMA tumors (Fig. 3F). This tumor-targeted combination therapy selectively increased intratumoral cytokines, including IFN-γ (Fig. 3G) and other proinflammatory cytokines (fig. S7A). No splenic or systemic cytokine induction was observed in these mice (Fig. 3G and fig. S7, A and B). To further characterize the responding T cell subsets upon combination treatment, we profiled tumor-infiltrating CD8+ T cells on day 17 after tumor challenge by high-dimensional flow cytometry using CITRUS (cluster identification, characterization, and regression) (37) analyses to independently stratify significantly different T cell clusters (P value cutoff is 0.05) (Fig. 3H). Mean fluorescence intensity (MFI) for each cluster is shown in Fig. 3I and fig. S8A. We found that PD-1 blockade expanded effector (CD44highCD62Llow) CD8+ T cells (cluster C1), which express high amounts of activation/exhaustion markers [PD-1, T cell immunoglobulin and mucin domain-containing protein 3 (TIM3), lymphocyte-activation protein 3 (LAG3), and Ki67] (Fig. 3J). However, only combination treatment was able to drive an expansion of intratumoral CD8+ T cells (cluster C2) with a memory-like phenotype [high transcription factor T cell factor 1 (Tcf1), Eomes, and CD62L] and less exhausted phenotype (low PD-1, LAG3, TIM3, CD38, and KLRG1, and higher CD5 and CD28) (Fig. 3J and fig. S8B) (38, 39). CD28 expression on responding CD8+ T cell populations is consistent with CD28 targeting by PSMAxCD28 bispecific antibody. We further separated out MC38 tumor antigen-specific T cells by p15E pentamer staining (fig. S9A). Although the frequency of p15E-specific T cells was unchanged with treatment (fig. S9B), we found that PSMAxCD28 and PD-1 combination treatment caused a decrease in frequency of cells with higher expression of exhaustion markers (LAG3 and PD-1) and increase in frequency of cells with higher expression of memory-like markers (Tcf1, Eomes, and CD62L), which express CD28 (clusters 6 and 13, respectively, fig. S9, C to F), similar to what we observed in total CD8 T cells. These results demonstrate that PSMAxCD28 bispecific and PD-1 mAb combination therapy drives robust antitumor immunity associated with tumor antigen-specific intratumoral T cell activation and a memory-like phenotype.

EGFRxCD28 bispecific in combination with PD-1 mAb controls and eradicates human tumor xenografts in humanized mice

To confirm and generalize the above findings that CD28 bispecifics can synergize with PD-1 blockade, we generated a second CD28 bispecific, EGFRxCD28, using the technology referenced above for the PSMAxCD28 bispecific (15). In brief, we generated conventional fully human antibodies specific for the human EGFR and human CD28 using VelocImmune mice (40, 41). We then combined these antibodies to make an EGFRxCD28 bispecific antibody using the aforementioned technology and a hinge-stabilized, effector function–minimized IgG4P isotype as previously described (15). Surface plasmon resonance (SPR) analysis showed that EGFRxCD28 bound hEGFR monomer with a dissociation constant (KD) of 9.6 nM and the hCD28 dimer with a KD of 52 nM (table S1).

We first verified that the EGFRxCD28 bispecific bound specifically to CD28-positive Jurkat cells and the EGFR expressed endogenously on the ovarian cancer cell line PEO-1 (42) (fig. S10, A and B). Then, using cocultures of human T cells containing peripheral blood mononuclear cells and PEO-1, we tested the ability of EGFRxCD28 to enhance cellular cytotoxicity and T cell activation (fig. S10C). In agreement with previous results using TSAxCD28 bispecifics (20), EGFRxCD28 increased the ability of MUC16xCD3 bispecific (43) to induce T cell killing of tumor cells, from 18.8 to 71.5% (fig. S10D). Consistent with this enhancement of T cell cytotoxicity, EGFRxCD28 also boosted T cell activation as assessed by CD25 expression and IFN-γ cytokine release (fig. S10, E and F). EGFRxCD28 in the presence of nontargeted CD3 binding control (absence of signal 1) had no effect on T cell cytotoxicity or activation (fig. S10, D to F).

We next tested the effectiveness of EGFRxCD28 bispecific antibody in combination with the PD-1 blocking Ab [REGN2810, (28)] using a human tumor xenograft model. A431 epidermoid carcinoma human tumor cells were implanted subcutaneously into an immunodeficient genetically humanized mouse model SIRPAh/h TPOh/m Rag2−/− Il2rg−/− (STRG) mice, a variant of M-CSFh/h IL-3/GM-CSFh/h SIRPah/h TPOh/h RAG2−/− IL2Rg−/− (MISTRG) mice (44, 45) optimized for engraftment and function of introduced human hematopoietic stem cells. To generate STRG mice with functional human T cells, human fetal liver CD34+ cells were implanted subcutaneously into these mice (44, 45). The successful generation of human immune cell populations in the STRG mice was validated by flow cytometry (fig. S11A). Before tumor implantation, the mice were randomized into treatment groups based on fetal liver donors, human immune cell engraftment, and gender. The expression of EGFR and PD-L1 on A431 tumor cells was demonstrated by flow cytometry (fig. S11B). Consistent with the above findings with the PSMAxCD28 bispecific in the syngeneic MC38/hPSMA model, the EGFRxCD28 bispecific in this xenograft A431/STRG model also enhanced responsiveness to PD-1 blockade (Fig. 4A). The combined efficacy with the EGFRxCD28 was entirely dependent on its recognition of tumor-expressed EGFR antigen, as a different CD28 bispecific that did not recognize the A431 tumor (PSMAxCD28) had no benefit in combination with PD-1 blockade (Fig. 4B). The lack of PSMA expression on A431 tumor cells was confirmed by flow cytometry (fig. S11C).

Fig. 4 EGFRxCD28 cooperates with PD-1 mAb treatment to induce antitumor immunity in VelociHum mice engrafted with human fetal liver CD34+ cells.

A431 epidermoid carcinoma tumor cells were implanted subcutaneously in SIRPAh/h TPOh/m Rag2−/− Il2rg−/− mice that were engrafted with fetal liver CD34+ cells. Mice were segregated into indicated treatment groups to have similar distribution of fetal liver donors, human immune cell engraftment frequency, and sex. Mice were treated intraperitoneally starting on the day of tumor challenge and treated every 2 to 3 days through the experiment. Dose for anti–PD-1 was 10 mg/kg and dose for EGFRxCD28/PSMAxCD28 was 5 mg/kg. (A and B) A431 tumor volumes for each treatment group plotted against days after tumor challenge. (A) EGFRxCD28. n = 7 to 11 mice per group. (B) PSMAxCD28. n = 5 to 6 mice per group. Data shown are means ± SEM (left) and individual mice (right). Two-way ANOVA, Tukey comparison: **P < 0.01, ***P < 0.001, and ****P < 0.0001. Data in (A) and (B) are representative of two independent experiments. (C and D) Tumor CD8+ T cell CITRUS analysis. (C) Color gradient indicates normalized expression of each marker (red-blue, high-low expression). (D) Frequency of cells in each cluster from the indicated treatment groups. Data are represented as means ± SEM. (E and F) Tumor CD4+ T cell CITRUS analysis. (E) Color gradient indicates normalized expression of each marker (red-blue, high-low expression). (F) Frequency of cells in each cluster from the indicated treatment groups. Data are represented as the means ± SEM. Data in (C) to (F) are representative of two independent experiments with n = 7 to 11 mice per group.

To investigate whether the combination of EGFRxCD28 bispecific and PD-1 Ab treatment had an impact on human T cell activation in A431 tumor xenograft model, we profiled the human intratumoral T cells as described above (Fig. 4, C to F) using CITRUS analyses to identify responding human CD8+ and CD4+ T cell clusters (fig. S12A). In addition, the strategy used to gate out non–T cells, which excludes CD8a+ dendritic cells, is shown in fig. S12A. We found that combination therapy robustly increased a cluster of CD8+ T cells with a highly activated phenotype (hCD8 C2, high expression of CD2, CD86, GITR, TIGIT, CD38, and ICOS; Fig. 4D, bottom) and repressed a cluster of CD8+ T cells with high expression of CD45RA, CCR7, PD-L1, Ki67, and ICOS (hCD8 C1; Fig. 4, C and D, and fig. S12B). EGFRxCD28 and PD-1 mAb single-agent treatments also reduced the CD8+ T cell cluster hCD8 C1, whereas PD-1 blockade alone drove the expansion of a highly activated CD8+ T cell cluster hCD8 C2 (Fig. 4D, bottom). We also observed more diverse CD4+ T cell clusters that responded to single or combination treatments (Fig. 4, E and F, and fig. S12C). Only combination treatment significantly (P < 0.05) expanded CD4+ cells with a proliferating effector memory-like phenotype (high Ki67 and low CD45RA and CCR7; hCD4 cluster 2) and less exhausted effector memory-like cluster hCD4 C3 (low CD38, CD45RA, and CCR7) (Fig. 4F). Similar to the CD8+ T cells, combination treatment significantly (P < 0.05) reduced the CD4+ T cell cluster hCD4 C1 that contained cells with higher PD-L1 expression. PD-1 blockade monotherapy promoted the expansion of effector memory CD4+ T cells with higher CD38 expression (hCD4 C4), resembling a more activated phenotype (46, 47). EGFRxCD28 monotherapy also maintained a small pool of naïve-like CD4+ T cells in the tumor (hCD4 C5), which could be further primed/activated upon being combined with PD-1 Ab treatment. CD28 expression was detected on all responding CD4+ and CD8+ T cell populations, consistent with CD28 targeting by EGFRxCD28 bispecific antibody.

The results of these high-dimensional flow cytometry analyses reveal immune response mechanisms in human immune system reconstituted mice. Although PD-1 blockade monotherapy was able to drive expansion of CD8+ and CD4+ T cells with an activated phenotype in this model, it was not sufficient to induce detectable tumor growth inhibition. Only PD-1 mAb combined with EGFRxCD28 bispecific achieved a balanced T cell activation state together with potent antitumor responses.

TSAxCD28 bispecifics alone, or in combination with PD-1 mAb, do not induce systemic T cell activation and cytokine release in vivo

A clinical trial in a group of healthy volunteers showed that a bivalent CD28-activating antibody, termed “CD28 superagonist,” broadly activated T cells and resulted in profound toxicity associated with cytokine release syndrome (48). To compare the effects of the CD28 superagonist with the tolerability of TSAxCD28 bispecifics alone or in combination with PD-1 mAb, we performed exploratory studies in cynomolgus monkeys and in genetically engineered triple-humanized mice. Three monkeys per group received a single intravenous dose of CD28 superagonist (10 mg/kg) (20) and compared to three monkeys per treatment group receiving a single intravenous dose of either the PSMAxCD28 or EGFRxCD28 bispecific alone (each at 10 mg/kg) or in combination with PD-1 mAb (REGN2810, cemiplimab 10 mg/kg; combination groups received sequential intravenous infusions) (Fig. 5). These doses of EGFRxCD28, PSMAxCD28 bispecifics, and PD-1 mAb in monkey studies were selected to provide systemic exposure above that projected to saturate the target. Assessment of tolerability and pharmacologic activity in monkeys was based on clinical observations, qualitative food consumption, body weight, vital signs (body temperature, heart rate, pulse oximetry, and respiration rate), and clinical and anatomic pathology upon completion of the experiment. In addition, blood samples were collected for cytokine profiling and immunophenotyping analysis by flow cytometry.

Fig. 5 TSAxCD28 bispecific alone or in combination with PD-1 mAb does not induce systemic T cell activation in cynomolgus monkeys or humanized mice.

(A to G) Cynomolgus monkeys were treated with a single dose of bispecifics or antibodies as indicated at 10 mg/kg. Time is indicated after dose (hours). (A to C) PSMAxCD28 or PSMAxCD28 + PD-1 mAb. (D to F) EGFRxCD28 or EGFRxCD28 + PD-1 mAb. (A and D) Plasma cytokines. (B and E) Relative peripheral blood T cell counts. (C and F) Frequency of Ki67+ T cells (% of CD3). Data in (A) to (F) are represented as the means ± SEM, n = 3 animals per group. (G) Hematoxylin and eosin staining of the indicated tissues. Scale bar, 60 μm. Arrows indicate areas of infiltration. Data in (A) to (G) are representative of two independent experiments. (H) CD3/CD28/PSMA humanized mice were treated with a single dose of antibody at 2.5 mg/kg. Blood was collected from the submandibular vein at 4 hours after dosing for plasma cytokine analysis. Symbols represent individual mice. n = 4 to 5 mice per group. Lines represent the means ± SEM. Statistical significance was calculated with one-way ANOVA and Tukey’s multiple comparisons test. *P < 0.05, **P < 0.01, and ***P < 0.001. Data in (H) are representative of at least three experiments.

Consistent with the expected pharmacologic immune system activation by a CD28 superagonist, cytokine release, lymphocyte margination, and T cell activation were observed in monkeys administered the CD28 superagonist alone, with maximum cytokine release and lymphocyte margination seen at about 5 hours after CD28 superagonist administration (20). Changes in clinical chemistry parameters associated with CD28 superagonist also included increased C-reactive protein (CRP) and fibrinogen (on day 2) as well as increase in globulins (on day 15) and decreased albumin (on day 15), consistent with an acute phase response (table S3). Other changes noted in animals administered CD28 superagonist included small but detectable increases in the following: activated partial thromboplastin time, large unstained cells, basophils, and red cell distribution width (table S3). In addition, mononuclear or mixed cell infiltrates were observed in the kidneys, brain, and seminal vesicles of animals administered the CD28 superagonist (Fig. 5G). It should be noted, however, that these changes in animals administered CD28 superagonist were not associated with observed changes in clinical signs, vital signs, or body weight parameters.

In contrast to the CD28 superagonist, neither PSMAxCD28 nor EGFRxCD28 bispecifics alone or in combination with PD-1 mAb produced any significant cytokine release, T cell margination, or T cell activation (Fig. 5, A to F). In addition, these CD28 bispecifics were well tolerated when administered alone or in combination with PD-1 mAb, with no test article–related changes in clinical observations, vital signs, body weights, or organ weights. With the exception of one of the animals administered PSMAxCD28 and PD-1 mAb, which had a slight increase in fibrinogen, there were no changes in clinical pathology parameters (hematology, serum chemistry, coagulation, and urinalysis) and no macroscopic or microscopic changes noted at the terminal necropsy in animals administered PSMAxCD28 and EGFRxCD28 bispecifics alone or in combination with PD-1 mA (Fig. 5G and table S3). These results demonstrated that in contrast to CD28 superagonists, which directly activate the immune system, these costimulatory CD28 bispecific antibodies do not independently stimulate the immune system in the absence of signal 1, consistent with the in vitro data.

Consistent with the above studies in cynomolgus monkeys, no cytokine elevation was observed in tumor-bearing or non–tumor-bearing naïve triple humanized mice (hCD3/hCD28/hPSMA) dosed with PSMAxCD28 bispecific alone or in combination with PD-1 mAb (Fig. 5H and fig. S13, A and B). In contrast, dosing these mice with CD28 superagonist induced significant increases in IFN-γ (P < 0.05), TNFα (P < 0.01), IL-2 (P < 0.001), IL-4 (P < 0.05), IL-5 (P < 0.05), and IL-6 (P < 0.05) detected in the serum at 4 hours after dose (Fig. 5H and fig. S13, A and B). Similarly, in the human immune cell reconstituted STRG mice implanted with A431 tumor as described above (Fig. 4), we did not observe any increase in cytokines with EGFRxCD28 bispecific alone or in combination with PD-1 mAb (fig. S14A), whereas the CD28 superagonist induced significant increases in IFN-γ (P < 0.05), TNFα (P < 0.01), IL-2 (P < 0.01), and IL-4 (P < 0.05) (fig. S14B). Consistent with the results described above, we have previously shown that a TSAxCD28 bispecific (as well as the parental bivalent nonsuperagonist CD28 antibodies used to make these bispecifics) failed to induce human T cell proliferation in the Food and Drug Administration–recommended in vitro dry- and wet-coated assay (49), again in contrast to the strong proliferation induced by the CD28 superagonist (20). Overall, these results suggest that TSAxCD28 bispecifics are qualitatively different from CD28 superagonists and are likely to be well tolerated.

DISCUSSION

Here, we introduce and validate a tumor-targeted immunotherapy using TSAxCD28 bispecifics in combination with PD-1 blocking mAb, which induces long-lived antitumor immunity and promotes robust intratumoral T cell activation in animal tumor models. Studies in genetically humanized immunocompetent mice and in cynomolgus monkeys demonstrated that these bispecifics were well tolerated as monotherapy or in combination with PD-1 mAb, suggesting that this combination approach may provide immunotherapies with markedly enhanced and specific antitumor activity with favorable safety profile.

Checkpoint inhibition with PD-1 blocking mAbs is able to release the brake on T cell activation; however, for most tumors, the efficacy of PD-1/PD-L1 mAbs as single agents can be insufficient to achieve tumor clearance and durable antitumor control. A number of approaches to improve responses to PD-1 inhibition are currently being evaluated. However, these strategies have, in some cases, resulted in worse outcomes for the patients (50). It has previously been shown that tumor-targeted activation of CD28 through engineered expression of CD28 ligands on tumor cells stimulates T cell–mediated tumor immunity without systemic toxicity (2124). Here, to improve the antitumor efficacy of PD-1 mAb, we introduced the concept of using a TSAxCD28 costimulatory bispecific to enhance T cell signaling and activation at the tumor site. This combination immunotherapy was validated using two different tumor targets (PSMA and EGFR) and demonstrated that CD28 costimulatory bispecific antibodies combine with PD-1 mAb not only to generate robust T cell activation but also to provide durable antitumor responses without off-target toxicity. By virtue of being tumor-targeted, this combination therapy may provide more tumor-focused T cell activation and less toxicity to normal organs compared to other nontargeted combination approaches described previously. In particular, using CD28 bispecific antibodies, which do not directly activate CD28 unless clustered on tumor cell surfaces, offers the possibility of promoting T cell costimulation only at the tumor site, avoiding the systemic toxicity of CD28-activating antibodies (48), as well as reducing the potential toxicity observed with the combination of CTLA-4 and PD-1 blockade (5153) or other costimulatory bivalent agonist antibodies (54).

Our investigation of human-specific clinical candidates in xenograft and genetically humanized immunocompetent mouse models and in cynomolgus monkeys shows potent antitumor efficacy and favorable safety profile. However, these models have limitations. In the in vivo studies in monkeys, we demonstrated that unlike CD28 superagonist, TSAxCD28 alone or in combination with PD-1 mAb did not induce cytokine release or lymphocyte margination. Previous reports suggest a lack of clinical predictivity of cytokine release with CD28 superagonist in monkeys due to the absence of CD28 expression on the CD4+ effector memory T cells in monkeys (55). However, we have previously shown that PSMAxCD28 enhanced activation of monkey T cells in the presence of PSMAxCD3 and PSMA expressing tumor cells in vitro (20). In our previous study, the peak increase in cytokines with CD28 superagonist was observed at 5 hours after dose (20), a time point considered optimal to detect changes in key cytokines such as IL-6 (48) but not previously interrogated with CD28 superagonist in this species. Although it has been suggested that PSMA is not expressed in the prostate of monkeys (56), a PSMAxCD3 bispecific antibody has been shown to induce cytokine release in cynomolgus monkeys (57). This is likely driven by T cell activation after engagement with PSMA expressed in other tissues such as the kidney (58). Although our studies show that PSMAxCD28 does not induce cytokines, this model does not replicate the human setting with metastatic tumor burden and thus increased tumor target throughout the body, which could result in a higher cytokine response.

Previous studies have shown that CD28 costimulation with a tumor-targeted bispecific together with CD3-based bispecifics enhances antitumor immunity and can be localized to the tumor site (20, 59). In a phase 1 clinical trial, EGFRxCD3 and a different EGFRxCD28 combination was directly coinjected with activated T cells into the glioma tumor site, resulting in complete response for 2 of 10 patients, but infusions were associated with adverse effects (60). This study highlights not only the potential antitumor efficacy but also the toxicities associated with using CD3-based bispecifics, which may broadly activate T cells and exaggerate toxicity. In contrast, our approach to combine EGFRxCD28 with anti–PD-1 may reduce the toxicity as we have shown for PSMAxCD28. In addition, our approach does not require coinjection of activated T cells. Others have used CD28 costimulation to enhance T cell response to endogenous tumor antigens, but using a vaccine approach that requires ex vivo pretreatment of T cells and tumor cells (61). In our study, we have described a mechanistically different approach using tumor-targeted CD28 agonist and PD-1 blockade to safely amplify T cell response to endogenous TSAs in situ without the need for ex vivo manipulation of tumor cells or T cells. PD-1 targeting has been shown to cooperate with both CD3 and CD28-based bispecifics to enhance T cell activation and antitumor activity in vitro (62), with one study testing the combination of a CD3 bispecific and anti–PD-1 in xenograft tumor models (63). However, this study did not evaluate combination with CD28 costimulation and did not examine the durability of immune responses, T cell phenotype, or tolerability/safety as we have done in our study.

We recently demonstrated that combining CD28 bispecifics can enhance the antitumor efficacy of CD3 bispecifics, although such combinations do not generate strong antitumor memory (20). This is presumably because CD3 bispecifics activate T cells independently of their TCR specificity and may not activate T cells reactive to endogenous tumor antigens and thus do not induce long-lived tumor-specific immunity (64); although CD28 bispecifics enhance the effect of CD3 bispecifics, they would not be expected to change the character of the response. In contrast, it has long been known that PD-1 blockade relies on the endogenous antigen-specific T cell response to tumor peptides (6568), although PD-1 therapies often do not yield a permanent memory response. CD28 bispecifics seem to enhance the antitumor activity of PD-1 blockade—primarily by enhancing the activation of endogenous T cell responses to the tumor—and this enhancement seemingly also results in long-term antitumor memory by generating an enhanced endogenous T cell memory component. In the syngeneic MC38 tumor model we used here, tumor cells have high expression of reactivated endogenous retroviral peptides such as p15E and that syngeneic C57BL6 mice generate endogenous T cells that recognize and respond to this neoepitope (35, 36). In our study, we demonstrated that PSMAxCD28 and PD-1 mAb combination therapy markedly increases the number of T cells responsive to this p15E neoantigen found in the spleen, consistent with the ability of this combination to generate long-term antitumor memory. In addition, we found that PSMAxCD28 combination with PD-1 mAb results in less dysfunctional CD8 tumor-infiltrating lymphocytes and promotes a strong intratumoral memory T cell phenotype expressing memory cell markers (Tcf1, Eomes, and CD62L) with low expression of exhaustion markers (PD-1 and LAG3) (39). These cells express intermediate amounts of memory cell markers CD127 and CD122 possibly due to transient down-regulation associated with increased level of IL-2 upon combination therapy (69). Similarly, memory-like T cells in the human immune cell-engrafted tumor xenograft model were increased. In this model, we detected CD86 expression on the intratumoral T cells. Human T cells may have an allogeneic response to the engrafted tumor cell line due to HLA mismatch. Allogeneic T cell response induced in the tumor would mediate higher intratumoral T cell activation and IL-2 production, which may explain the unusual expression of CD86, which has been previously described (70). Together, these data show that CD28 bispecifics in combination with PD-1 blockade can boost endogenous TCR/CD3-dependent T cell responses, thereby driving durable antitumor responses.

Previously, two studies demonstrated that the antitumor activity of PD-1 mAb is CD28 dependent and that PD-1 inhibition of T cell activation reduces signaling through TCR/CD3 and/or CD28, which may be affecting the spatial localization of those molecules (10, 11). Here, we validated and expanded these findings. We showed that PD-1 is accumulated at the immune synapse when PD-L1 is expressed by target cells, and its accumulation is associated with a reduction of CD28 at the synapse, suggesting that PD-1 could exercise T cell inhibition by preventing CD28 localization to the synapse. The PD-1 nonblocker mAb used in the imaging studies does not prevent PD-1 accumulation at the synapse and does not rescue PD-1–mediated inhibition of human T cell activation in vitro. This contrasts with a recently described PD-1 nonblocker mAb, which has potent PD-1 antagonist activity (71). Although both PD-1 mAbs do not block PD-1 interaction with PD-L1, they may bind to PD-1 at distinct epitopes, which could differentially affect PD-1 conformation, localization within the synapse, and proximity to associated signaling molecules. On the other hand, we found that PD-1 blockade prevented PD-1 synaptic localization, whereas CD28 accumulation at the synapse was increased, allowing a TSAxCD28 bispecific to markedly enhance the ability of PD-1 mAb to promote T cell activation. In addition, a PD-L1 blocking mAb also reduced PD-1 and increased CD28 localization in the synapse but without engaging receptors on T cells. These data suggest that the effects of PD-1 and PD-L1 blocking mAbs are through changes in PD-1 and CD28 localization and not agonistic or conformational effects on PD-1 directly. Overall, the visualization of PD-1 and CD28 localization in the immunological synapse resulting from PD-1:PD-L1 interaction in the presence of PD-1/PD-L1 mAbs enables better understanding of the effect of PD-1 blockade on T cell activation, as well as the cooperation between TSAxCD28 and PD-1 mAb at the level of the immune synapse.

Similar to chimeric antigen receptor (CAR) T cell approaches that use chimeric receptors that artificially activate both signal 1 and signal 2 to improve their antitumor activity (72, 73), we now show the potential benefit of combining PD-1 inhibition (which promotes signal 1) with CD28 bispecifics (which provide signal 2) to enhance antitumor activity. This approach has several practical benefits over CAR T cell therapies in that it does not require a laborious cell therapy preparation that must be individually customized for each patient, nor does it require that patients be preemptively lymphodepleted via toxic chemotherapy that is often associated with adverse effects (74).

Collectively, these results show that combining CD28-based bispecifics with clinically validated PD-1 mAbs may provide well-tolerated, off-the-shelf biologic solutions with markedly enhanced antitumor activity. The safety profile together with the similar enhancement of antitumor efficacy by two different TSAxCD28 (PSMA and EGFR) with PD-1 mAb across syngeneic and xenograft models suggests that this therapeutic modality is robust, not limited to a specific tumor model, and could have broader utility as a combination immunotherapy.

MATERIALS AND METHODS

Study design

The objective of this study was to develop TSAxCD28 bispecific antibodies and demonstrate that TSAxCD28 bispecifics potentiate T cell activation enhanced by PD-1 targeting in vitro and safely enhances antitumor efficacy in vivo. The dynamics of PD-1 and CD28 interaction in vitro were demonstrated by showing images of bispecific antibodies, PD-1, and CD28 localized at the immunological synapse of T cell and target cell conjugates and enhancement of T cell cytokine release when both PD-1 was inhibited and CD28 was activated. In vivo antitumor efficacy was evaluated in two different mouse tumor models (xenograft and syngeneic). Tumor growth was monitored over time, phenotypic changes in tumor-infiltrating T cells were profiled by high-dimensional computational flow cytometry, and serum cytokines were monitored to show response to TSAxCD28 and PD-1 combination treatment. Cynomolgus studies were performed to determine the safety and tolerability (pharmacologic and toxicologic profile) of TSAxCD28 as monotherapy or in combination with anti–PD-1 in nonhuman primate. Animals were examined for toxicity by clinical observations and blood sample collections to analyze serum cytokines and T cell phenotype. The numbers of animals and experimental replicates are indicated in the figure legends. Sample sizes were chosen empirically to ensure adequate statistical power and were in the line with field standards for the techniques used in the study. In the xenograft model, human immune cell mice were randomized before tumor implantation into treatment groups based on fetal liver donors, frequency of human immune cells found in the blood, and gender. In therapeutic experiments, mice were randomized into treatment groups based on the size of established tumors. Blinding was not used in this study.

Animal studies

All procedures were carried out in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocols were approved by the Regeneron Pharmaceuticals Institutional Animal Care and Use Committee (IACUC).

Syngeneic tumor studies

MC38/EV and MC38/CD86 were cultured according to American Type Culture Collection (ATCC) guidelines. MC38/EV or MC38/CD86 (1 × 106) were implanted subcutaneously on the flank of C57BL/6 mice. Mice were treated with PD-1 mAb (RPM1-14, BioXcell) or rat IgG2a isotype control (BioXcell) at 5 mg/kg by intraperitoneal injection on days 0, 3, 7, 10, and 14 after tumor implant. Tumor-free mice previously implanted with MC38/CD86 tumor and treated with PD-1 mAb were rechallenged with a secondary subcutaneous tumor implant with 0.5 × 106 parental MC38 cells on the opposite flank.

Mice expressing human CD28, human CD3, and hPSMA in place of the corresponding mouse genes have been described and were generated using VelociGene technology at Regeneron Pharmaceuticals (20, 31, 32) (referred to as hCD3/hCD28/hPSMA humanized mice). hCD3/hCD28/hPSMA mice (8 to 16 weeks old) were injected with 1 × 106 MC38/hPSMA tumor cells subcutaneously on the flank. Indicated antibodies or bispecifics were administered as a monotherapy or in combination by intraperitoneal injection on days 0, 7, and 14 (prophylactic treatment) or days 9, 13, and 22 (delayed treatment) at 5 mg/kg. Tumor-free mice previously treated with PSMAxCD28 and PD-1 mAb combination were rechallenged with a secondary subcutaneous tumor implant with 1 × 106 MC38/hPSMA, 1 × 106 parental MC38, or 0.2 × 106 B16F10.9 (a subline from B16F10, ATCC-CRL6475) cells per mouse.

To deplete CD4 or CD8 T cells, mice were treated with CD4 (GK1.4 clone, BioXcell), CD8 (2.43 clone, BioXcell), or rat IgG2b isotype control (LTF-2 clone, BioXcell) at 15 mg/kg twice per week starting 1 day before tumor rechallenge.

For all tumor experiments, tumor sizes were measured twice per week using calipers (Roboz RS-6466). Tumor volume was calculated using the formula X × Y × (X/2), where Y is the longest dimension and X is the perpendicular dimension. Mice with tumors larger than 2000 mm3 or with ulcerated tumors were euthanized by CO2 asphyxiation.

Ex vivo tissue cytokine analysis

On day 29 after implant, MC38/hPSMA tumor-bearing CD3/CD28/PSMA mice were euthanized by CO2 asphyxiation. Spleens and tumors were collected and stored in medium on ice. All subsequent steps were performed on ice or at 4°C unless noted differently. Tumors were cut into small pieces and fragments were processed into single-cell suspension using a Miltenyi mouse tumor dissociation kit following the manufacturer’s protocol (Miltenyi 130-096-730). Spleens were processed into single-cell suspension using gentle MACS (magnetic-activated cell sorting) mechanical dissociation (spleen 4 program) and mashing through a 70-μm filter using the rubber end of a 3-ml syringe. Cells were pelleted by centrifugation at 1200 rpm for 5 min. Red blood cells were lysed by resuspending the cell pellet in 1-ml ACK (ammonium-chloride-potassium) lysis buffer and incubating on ice for 5 min. ACK lysis buffer was quenched with flow cytometry buffer 3% fetal bovine serum (FBS) and 2 mM EDTA in Dulbecco’s phosphate-buffered saline solution (D-PBS, Irvine Scientific). Cells were pelleted by centrifugation at 1200 rpm for 5 min. Cell suspension was resuspended in 1 ml of medium, and 0.2 ml was plated in 96-well plates (20,000 to 400,000 tumor cells or 50,000 to 70,000 spleen cells). Cells were incubated overnight at 37°C, and culture supernatant was collected. Cytokines in tissue culture supernatant were measured using a V-Plex Proinflammatory Meso Scale Diagnostics (MSD) kit following manufacturer protocol (MSD K15048D-4). The number of cells plated per well was determined by flow cytometry analysis. The concentrations of cytokines were normalized to the number of cells plated. Calibration beads were run together with cells to accurately measure the number of cells using the following calculation:

Cell # = (# input beads × # cells counted by flow cytometry)/# of beads counted by flow cytometry.

Measurement of p15E specific IFN-γ+ T cells in the spleen

MC38/PSMA tumor cells were implanted in CD3/CD28/PSMA humanized mice and treated with isotype control, PSMAxCD28, PD-1 mAb, or combination all dosed at 5 mg/kg on days 10 and 14 after implant. Spleens were harvested on day 17. Splenocytes were cultured overnight in T cell medium [RPMI 1640 (Irvine Scientific) supplemented with 10% FBS (Seradigm), penicillin-streptomycin-glutamine (Thermo Fisher Scientific), nonessential amino acids (Irvine Scientific), sodium pyruvate (Invitrogen), and beta-mercaptoethanol (Sigma-Aldrich)] with peptide (10 μg/ml; p15E or ovalbumin) and anti-CD28 (2 μg/ml). After overnight incubation, intracellular cytokine staining was performed following the manufacturer protocol (Thermo Fisher Scientific).

Measurement of serum cytokine concentrations in mice

At the indicated time points, blood was collected from the submandibular vein into microtainer serum tubes (Becton Dickinson, BD 365967) and processed following the manufacturer’s protocol. Cytokine concentrations were analyzed using a V-plex Human ProInflammatory-10 Plex kit following the manufacturer’s instructions (Meso Scale Diagnostics).

Flow cytometry analysis

For flow cytometry analysis of in vivo experiments, tumors were harvested, single-cell suspensions were prepared, and red blood cells were lysed using ACK lysis buffer (Thermo Fisher Scientific). Live/dead cell discrimination was performed using a live/dead fixable blue dead cell staining kit (Thermo Fisher Scientific). Samples were acquired on Symphony (BD Bioscience) and analyzed using Cytobank software (Cytobank). Analysis was performed with equal numbers of events per sample. The range in events was determined by the sample with the fewest events acquired. For unbiased clustering analysis, t-distributed stochastic neighbor embedding method was used. To identify T cell clusters automatically on the basis of selected markers, CITRUS analysis from Cytobank and FlowSOM from OMIQ were used as indicated in the text.

Xenograft tumor studies

A431 epidermoid carcinoma tumor cells (CRL-1555, ATCC) were implanted subcutaneously in SIRPAh/h TPOh/m Rag2−/− Il2rg−/− mice that were engrafted with fetal liver CD34+ cells. Mice were segregated into four groups to have similar distribution of fetal liver donors, human immune cell engraftment frequency, and sex. Mice were dosed by intraperitoneal injection twice per week starting on the day of implant (day 0) with isotype control, EGFRxCD28 (5 mg/kg), anti–PD-1 (10 mg/kg), or combination. Antibody injections were then administered every 2 to 3 days through the experiment. Tumors were measured two-dimensionally (length by width), and tumor volume was calculated (length × width2 × 0.5). Mice were euthanized when the tumor reached a designated tumor end point (tumor volume >2000 mm3 or tumor ulceration).

Studies in cynomolgus monkey

The cynomolgus monkey study was conducted in accordance with IACUC guidelines. Studies with EGFRxCD28 were carried out at Shin Nippon Biomedical Laboratories (USA, accredited by the Association for Assessment and Accreditation of Laboratory Animal Care, Animal Welfare Assurance issued by Office of Laboratory Animal Welfare, registered with the United States Department of Agriculture, and IACUC) using female cynomolgus monkeys (Macaca fascicularis). Studies with PSMAxCD28 were carried out at Charles River Laboratories using male cynomolgus monkeys (M. fascicularis). Monkeys (three animals per group) received a single dose of each test article via intravenous infusion for about 30 min (combination treatment was administered as separate infusion for a total of 1 hour). For flow cytometry, blood was collected from peripheral vein of restrained, conscious animals into potassium EDTA tubes. Whole blood was stained with the indicated antibodies. Flow cytometric data acquisition was conducted using FACSCanto II. The absolute counts of the individual cell populations were calculated from their relative percentages as derived from the parent/grandparent population gate and the total parent/grandparent population counts from a validated hematology analyzer ADVIA 120 Hematology System (Siemens) according to the following formula: absolute population count (×103/μl) = (population relative % × total parent/grandparent population count)/100. For plasma cytokines, blood was collected into plasma separator tubes with anticoagulant and separated via centrifugation at 1000 to 2000g at 4°C for 10 to 15 min. Cytokines were analyzed using the MSD U-Plex platform or Luminex multiplex assay. CRP concentrations were analyzed on a Roche Modular P800 system. Studies in cynomolgus monkeys also included evaluation of clinical observations, qualitative food consumption, body weights, vital signs (body temperature, heart rate, pulse oximetry, and respiration rate), and clinical and anatomic pathology.

Statistical analysis

Data are presented as means or medians ± SD or ± SEM as stated in the figure legends. Statistical significance was determined as indicated in figure legends, with P < 0.05 considered statistically significant. Original data are shown in data file S1.

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/12/549/eaba2325/DC1

Materials and Methods

Fig. S1. MC38/CD86 tumor growth inhibition is immune cell dependent and potentiated by therapeutic anti–PD-1.

Fig. S2. Localization of PD-1/PD-L1 and CD28 at the immunological synapse in the presence of PD-1/PD-L1 mAbs.

Fig. S3. PSMAxCD28 bispecific and PD-1 or PD-L1 blockade promote T cell activation in vitro.

Fig. S4. MUC16xCD28, but not a MUC16xCD27 T cell binding control, promotes T cell activation.

Fig. S5. SPR-Biacore sensorgrams for binding of human and murine CD28 to human and murine CD80 and CD86.

Fig. S6. PSMAxCD28 and PD-1 mAb combination increases the frequency of tumor-specific T cells.

Fig. S7. PSMAxCD28 cooperates with anti–PD-1 to induce intratumoral but not splenic or systemic cytokines.

Fig. S8. Expression of T cell activation markers used to determine CITRUS clusters shown in Fig. 3.

Fig. S9. Tumor antigen-specific (p15E+) CD8 T cell CITRUS analysis.

Fig. S10. EGFRxCD28 bispecific potentiates T cell activation only in the presence of TCR stimulation.

Fig. S11. A431 human xenograft tumor model.

Fig. S12. MFI data for CITRUS clusters shown in Fig. 4.

Fig. S13. PSMAxCD28 alone or in combination with PD-1 mAb does not induce cytokine in non–tumor-bearing mice, in contrast to CD28 superagonist.

Fig. S14. EGFRxCD28 alone or in combination with PD-1 mAb does not induce cytokines in human immune cell-engrafted STRG mice, in contrast to CD28 superagonist.

Table S1. SPR-Biacore kinetics.

Table S2. Measuring binding kinetics of human and mouse CD28/CD80/CD86 interactions using SPR.

Table S3. Histopathology, organ weights, body weight, and clinical pathology findings in cynomolgus monkeys.

Data file S1. Primary data.

References (7577)

REFERENCES AND NOTES

Acknowledgments: We acknowledge A. Skinner, T. Gorenc, and S. Chandwani for project management. Funding: Regeneron Pharmaceuticals Inc. This work was funded, in part, by Sanofi. Author contributions: Concept was conceived by D.S. and G.D.Y. The strategy and the overall study were designed by D.S. and J.C.W. Additional experiments were designed by J.C.W., B.W., L.H., A.H., E.U., X.Y., D.D., R.S., D.K.A., S.J.G., R.B., G.C., and D.S. Experiments were performed by J.C.W., E.U., X.Y., D.D., I.R., Q.W., E.O., P.P., J.G., D.G., J.S., A.P., J.F., E.H., H.A., V.K., A.D., E.N., J.X., J.K., S.N.Y., J.W., C.L., L.C., T.S., and R.F. Data were analyzed by J.C.W., B.W., L.H., A.H., E.U., X.Y., D.D., R.S., D.K.A., S.J.G., V.K., A.D., P.K., E.G., and A.R. Genetically modified mice were generated and provided by C.-J.S., N.S.O., and W.P. Manuscript was written by D.S., J.C.W., and G.D.Y. Technical support and conceptual advice were given by C.D., C.G., J.M., L.M., D.M., W.P., E.S., I.L., G.T., W.O., J.C.L., M.A.S., G.D.Y., A.J.M., and D.S. Competing interests: G.D.Y., D.S., J.C.W., A.J.M., E.S., E.U., A.H., and L.H. are inventors on patent applications US16/448,462 and PCT/US2019/038460, and US16/717,273 and PCT/US2019/067109 held/submitted by Regeneron Pharmaceuticals that cover PSMAxCD28 and MUC16xCD28, respectively. G.D.Y., A.J.M., D.S., L.H., and J.C.L. are inventors on patent applications US62/822,124 held/submitted by Regeneron Pharmaceuticals that cover EGFRxCD28. G.D.Y., E.S., and L.H. are inventors on patent applications US2018/0112001, PCT/US2017/053113, US2018/0118848, and PCT/US2017/053114 held/submitted by Regeneron Pharmaceuticals that cover MUC16xCD3. A.J.M. is an inventor on US patent no. 9,987,500 and on patent applications US2018/0185668 and WO2015/112800 held/submitted by Regeneron Pharmaceuticals that cover PD-1. A.J.M. is an inventor on US patent no. 9,938,345 and on patent applications US2018/0186883 and WO2015/112805 held/submitted by Regeneron Pharmaceuticals that cover PD-L1. All authors are employees of Regeneron Pharmaceuticals Inc. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials.

Stay Connected to Science Translational Medicine

Navigate This Article