Research ArticleCancer

A class of costimulatory CD28-bispecific antibodies that enhance the antitumor activity of CD3-bispecific antibodies

See allHide authors and affiliations

Science Translational Medicine  08 Jan 2020:
Vol. 12, Issue 525, eaaw7888
DOI: 10.1126/scitranslmed.aaw7888

Double trouble for tumors

Bispecific antibodies, which have one arm that binds to a cancer cell antigen and one arm that binds to CD3 on T cells, can help T cells recognize and eliminate cancer cells. However, they are not always sufficient to activate the T cells. To provide a boost to their activity, Skokos et al. designed a second type of bispecific antibodies that bind tumor antigens on one side and a T cell costimulatory receptor on the other side. These costimulatory bispecific antibodies had little activity alone but were effective in combination with CD3-targeting bispecific antibodies in multiple mouse models while showing no toxicity in mice or primates.


T cell activation is initiated upon binding of the T cell receptor (TCR)/CD3 complex to peptide–major histocompatibility complexes (“signal 1”); activation is enhanced by engagement of a second “costimulatory” receptor, such as the CD28 receptor on T cells binding to its cognate ligand(s) on the target cell (“signal 2”). CD3-based bispecific antibodies act by replacing conventional signal 1, linking T cells to tumor cells by binding a tumor-specific antigen (TSA) with one arm of the bispecific and bridging to TCR/CD3 with the other. Although some of these so-called TSAxCD3 bispecifics have demonstrated promising antitumor efficacy in patients with cancer, their activity remains to be optimized. Here, we introduce a class of bispecific antibodies that mimic signal 2 by bridging TSA to the costimulatory CD28 receptor on T cells. We term these TSAxCD28 bispecifics and describe two such bispecific antibodies: one specific for ovarian and the other for prostate cancer antigens. Unlike CD28 superagonists, which broadly activate T cells and resulted in profound toxicity in early clinical trials, these TSAxCD28 bispecifics show limited activity and no toxicity when used alone in genetically humanized immunocompetent mouse models or in primates. However, when combined with TSAxCD3 bispecifics, they enhance the artificial synapse between a T cell and its target cell, potentiate T cell activation, and markedly improve antitumor activity of CD3 bispecifics in a variety of xenogeneic and syngeneic tumor models. Combining this class of CD28-costimulatory bispecific antibodies with the emerging class of TSAxCD3 bispecifics may provide well-tolerated, off-the-shelf antibody therapies with robust antitumor efficacy.


The ability of T cells to recognize and kill their cellular targets—such as virally infected cells or tumor cells—depends on a coordinated set of interactions. Foremost among these is the recognition and binding of the target cell by the T cell receptor (TCR) complex (which includes the associated CD3 γ, δ, ε, and ζ chains); this interaction has been referred to as “signal 1” for T cell activation (fig. S1A). The TCR can recognize viral or tumor peptide presented in the groove of a major histocompatibility complex (MHC) protein expressed on the surface of the target cells. This binding is typically of low affinity; therefore, successful triggering of signal 1 requires clustering of many TCR complexes along the interface between a T cell and its target cell (fig. S1B), and this interface has been referred to as the immune synapse (1). T cell activation and proliferation are then further promoted by additional interactions with costimulatory receptors such as CD28 (“signal 2”) (2). When a T cell recognizes a target cell via the TCR complex and engages signal 2 via CD28 binding to its cognate ligand(s) (CD80/B7.1 and/or CD86/B7.2) on a professional antigen-presenting cell or the target cell, T cell activation is enhanced (fig. S1C). As with signal 1, CD28-mediated signal 2 is thought to occur via coclustering at the immune synapse (fig. S1D) (3).

Conventional monoclonal antibodies targeted against tumor-specific antigens (TSAs) have been used as antitumor therapeutics over the past two decades (48). However, this class of antibodies had limited ability to induce T cell–mediated cytotoxicity and instead acted by promoting antibody-dependent cellular cytotoxicity (ADCC) and/or complement-dependent cytotoxicity or by delivering a toxin to the tumor cells. Recently, a class of bispecific antibodies (TSAxCD3) with a native immunoglobulin format (9) has emerged that can efficiently trigger T cell–mediated killing of tumor cells (fig. S2A) by linking a T cell to a tumor cell and activating the CD3/TCR complex (usually via the ε chain of CD3) via a surrogate mechanism, thus mimicking signal 1 (fig. S2B). Furthermore, a single-chain variable fragment (ScFv) bispecific T cell engager with no Fc region (one arm binding to CD19 on leukemia cells, whereas the other binds to CD3), blinatumomab, received regulatory approval for B cell acute lymphoblastic leukemia (10, 11). Recently, alternative versions of bispecifics have been shown to have good activity against non-Hodgkin’s lymphomas, targeting CD20 (9, 1215). However, although TSAxCD3 bispecifics are emerging as an important class of immunotherapy in hematologic malignancies, cross-study comparisons (16) suggest that in some cases, they may not be achieving the efficacy seen with chimeric antigen receptor T cell (CAR-T) therapies.

One of the reasons for the strong efficacy of CAR-T therapies is that the chimeric antigen receptor (CAR) is engineered to provide both signal 1 (via a portion of the CD3ζ cytodomain) and signal 2 (via a portion of the CD28 cytodomain) upon binding to its target on a tumor cell. Two CAR-T therapies have recently received U.S. Food and Drug Administration (FDA) approval for B cell malignancies, and they both act by binding and targeting the antigen CD19 (17, 18). CAR-T approaches can be associated with severe adverse effects such as cytokine release syndrome (CRS) and neurotoxicity (1921), and because of the highly personalized manufacturing processes and requirement for preconditioning chemotherapeutic regimens (17, 18, 22), many patients are not deemed suitable candidates.

The advantages of TSAxCD3 bispecifics as relatively well-tolerated and off-the-shelf therapeutic solutions for broader patient populations would be enhanced if their antitumor activity could be further optimized, especially if this could be done without sacrificing tolerability, or perhaps even increase specificity for tumor cells as opposed to normal cells. Toward this end, we hypothesized that pairing TSAxCD3 bispecifics with a class of bispecifics that independently activates signal 2 could provide potential increased efficacy as well as an opportunity for enhanced specificity. Therefore, we designed a second class of bispecifics that could engage either a second epitope on the same TSA or a second separate tumor antigen (for the same indication), with the costimulatory receptor CD28 (TSAxCD28 bispecifics; see fig. S3, A and B) expressed on T cells. We reasoned that combining TSA1xCD3 with a TSA2xCD28 should allow directed and enhanced surrogate activation of T cells by triggering both signal 1 and signal 2, with specificity targeted only against tumor cells expressing both epitopes or both antigens, allowing for greater antitumor activity together with an opportunity for increased specificity. However, in 2006, a clinical trial testing a superagonist anti-CD28 bivalent antibody (TGN1412) resulted in life-threatening complications in six human volunteers (23) due to massive CRS. This catastrophe led to cessation of any further testing of CD28-activating antibodies in humans. Consequently, in this study, we gave great consideration to the selection of an appropriate anti-CD28 arm of the bispecific to minimize the chance of CRS.

Here, we describe the generation and testing of TSAxCD28 costimulatory bispecific antibodies targeted against two different TSAs: one for ovarian cancer [MUC16xCD28, which binds MUC16, a large integral membrane glycoprotein highly expressed in certain cancers (24), and which is cleaved to release the ovarian tumor biomarker CA-125 (25)], and a second TSAxCD28 bispecific targeted to prostate cancer (PSMAxCD28, which binds prostate-specific membrane tumor antigen). Toxicology studies in genetically humanized immunocompetent mice as well as in cynomolgus monkeys demonstrate that these bispecifics exhibit limited activity and no toxicity as single agents. However, these costimulatory bispecifics can be effectively combined with the emerging class of TSAxCD3 bispecifics to potentiate antitumor responses in both xenogeneic and syngeneic tumor models. Collectively, these data suggest that combining this class of CD28-based bispecifics (TSAxCD28) with the CD3-based bispecifics (TSAxCD3) is a safe and effective immunotherapeutic solution that could replace the need for laborious and expensive personalized therapeutic approaches.


TSAxCD3 and TSAxCD28 bispecifics form artificial immune synapses between T cells and their target cells

Multiple pathways have been shown to play costimulatory roles for T cell activation (26). We screened various costimulatory pathways by forced overexpression of costimulatory ligands on a panel of syngeneic tumor cell lines (MC38, B16F10.9, and EL4; fig. S4), confirming that the CD80/CD86-CD28 and 41BBL-41BB pathways can play a predominant role in promoting T cell activation and antitumor activity. Here, we describe the generation of CD28 bispecifics that attempt to mimic the normal function of this pathway. We have previously described the generation and characterization of a CD20xCD3 bispecific antibody (9). Using this same technology, we engineered PSMAxCD28 as our first example of a TSA2xCD28 bispecific. We generated conventional fully human antibodies specific for the human orthologs of prostate-specific membrane antigen (PSMA) and CD28 using VelocImmune mice (table S1) (27, 28) and then combined these antibodies to make a PSMAxCD28 bispecific antibody using the aforementioned technology; the constant region (Fc) of this PSMAxCD28 bispecific is based on an effector function–minimized immunoglobulin G4P (IgG4P) isotype to reduce or eliminate the binding to Fcγ receptors (FcγRs) by replacing four amino acids (E233FLG236) in the lower hinge region of IgG4 with three amino acids (P233VA235) derived from the corresponding region of human IgG2 as previously described for CD3-based bispecifics (referred to here as hIgG4s) (9). In addition, our bispecifics are “hinge stabilized” and contain a serine-to-proline amino acid substitution (S228P, Eu numbering) in the hinge region that reconstructs the human IgG1 hinge sequence (CPPC) to promote stabilization of disulfide bonds between the two heavy chains (29). Biacore analysis showed that PSMAxCD28 bound human PSMA (hPSMA) with a KD (dissociation constant) of ~2.2 × 10−10 and hCD28 with a KD of 2.7 × 10−7 (table S1). Additional Biacore data show that the modifications to the Fc region of the bispecifics described above abolished binding to mouse FcγRs (table S2 and fig. S5).

To understand whether TSA2xCD28 bispecifics could enhance the activity of TSA1xCD3 and validate this proof of concept in vitro, we first engineered an experimental target cell line [human embryonic kidney (HEK) 293/hCD20/hPSMA] expressing two well-characterized human tumor antigens, hCD20 (TSA1) specific for B cell–related malignancies and hPSMA (TSA2) specific for prostate cancer. We then used this HEK293/hCD20/hPSMA cell line together with human Jurkat T cells to evaluate combinations of our CD20xCD3 and our PSMAxCD28 bispecific antibodies (hereinafter referred to as CD20xCD3 and PSMAxCD28). Because efficient T cell activation depends on coclustering of TCR/CD3 and CD28 complexes at the immune synapse (fig. S1D), we determined whether our CD20xCD3 and PSMAxCD28 bispecific antibodies could promote formation of and become clustered to form an immune synapse between T cells and the experimental target cells expressing both CD20 and PSMA. When Jurkat T cells alone were bound to a fluorescently labeled CD20xCD3 bispecific (CD20xCD3Alexa 488) or to a fluorescently labeled bivalent CD28 antibody, immunofluorescence analyses demonstrated that CD3 and CD28 remain uniformly distributed around the surface of the T cell (Fig. 1A). Similarly, when isolated target cells were bound to the fluorescently labeled CD20 bispecific (CD20xCD3Alexa 488) or to the fluorescently labeled PSMA bispecific (PSMAxCD28CF647), both CD20 and PSMA tumor antigens remained uniformly distributed around the surface of the target cell (Fig. 1B). However, when T cells and target cells were brought together by the CD20xCD3 and PSMAxCD28 bispecifics, both bispecifics clustered at an artificial synapse that formed between the two cells, mimicking a natural immune synapse (Fig. 1C). Quantification of the CD28 fluorescent signal inside versus outside of the synapse confirmed that the combination of CD20xCD3 with PSMAxCD28 significantly (P < 0.0001) increased the accumulation of CD28 within the artificial synapse (Fig. 1D).

Fig. 1 TSAxCD28 bispecifics potentiate T cell activation only in the presence of TCR stimulation by TSAxCD3.

(A to D) TSA-engineered target cells (HEK293) and human Jurkat T cells were cocultured with the indicated fluorescently labeled bispecifics. Fluorescence signal from each marker is shown as indicated at the top of the image panel. Cells are outlined with white dotted lines. Representative images of (A) T cells, (B) HEK293/hCD20/hPSMA, and (C) T cell–HEK293/hCD20/hPSMA doublets. Scale bar, 10 μM. (D) Quantification of the percentage of CD28 in the immunological synapse. n = 172 (isotype), n = 288 (CD20xCD3), n = 143 (PSMAxCD28), and n = 94 (combo). Statistical significance was calculated using one-way analysis of variance (ANOVA) and Tukey’s multiple comparisons test (****P < 0.0001). mAb, monoclonal antibody. (E and F) Proliferation of human T cells cultured with TSA-engineered target cells and bispecifics as described in the schematics on the left. Data are shown as means ± SEM. (E) Dose titration of CD20xCD3 in the presence of 0.5 nM hIgG4s isotype control or PSMAxCD28. (F) Dose titration of PSMAxCD28, CD28-SA, or NonTargetxCD28 in the presence of 5 pM hIgG4s isotype control or CD20xCD3. Data are representative of at least two experiments. CPM, counts per minute.

TSAxCD28 bispecifics have limited activity on their own but potentiate the ability of TSAxCD3 bispecifics to activate T cell proliferation and cytokine release

To test whether CD28 bispecifics could potentiate the effects of CD3 bispecifics on T cell activation, we performed assays using these bispecifics in cocultures of primary human T cells and engineered target cells overexpressing hCD20 and hPSMA (HEK293/hCD20/hPSMA). We first determined the dose response of the CD20xCD3 bispecific driving T cell activation as assessed by T cell proliferation (Fig. 1E) and repeated this dose response in the presence of a fixed amount of the PSMAxCD28 bispecific (Fig. 1E). These experiments showed that the PSMAxCD28 markedly potentiated the effect of the CD20xCD3 bispecific, shifting the dose response by 34-fold. We then performed the reciprocal experiment: A dose-response assay using the PSMAxCD28 bispecific demonstrated limited activity on its own (Fig. 1F), but robust activity was detected when using a minimal dose of CD20xCD3 (Fig. 1F) (5 pM minimal dose was determined using the CD20xCD3 dose-response assay and indicated by the arrow in Fig. 1E). In agreement with previous studies (30), in the soluble format assay, an in-house generated bivalent anti–CD28 superagonist (CD28-SA) showed only minimal T cell proliferation and cytokine production (Fig. 1F and fig. S6). Furthermore, control studies showed that the activity observed after combination of the two bispecific antibodies was dependent on target cell expression of the appropriate TSAs (fig. S6), as well as on the specificity of the CD28 bispecific for the correct tumor antigen (an irrelevant CD28 bispecific that did not bind to the target cells had no activity even in combination with the CD3 bispecific). In addition, the results seen in the T cell proliferation assays were consistent with those seen in T cell activation assays based on interleukin-2 (IL-2) and interferon-γ (IFN-γ) cytokine release (fig. S6). Together, these experiments demonstrate that the TSAxCD28 has limited activity on its own but markedly enhances activity of the TSAxCD3 bispecific in a tumor antigen–specific fashion.

TSAxCD28 bispecifics enhance the ability of CD3 bispecifics to induce in vitro T cell killing of ovarian and prostate cancer cell lines endogenously expressing tumor antigens

We next explored whether the ability of CD28 bispecifics to potentiate the activity of CD3 bispecifics occurred when using tumor cell lines that endogenously express the target antigens of interest and whether this potentiation extended to tumor cell killing in vitro. Toward this end, we used the prostate cancer line C4-2 that endogenously expresses PSMA and the ovarian cancer cell line PEO-1 that endogenously expresses MUC16. We used CD3 and CD28 bispecifics targeting two different epitopes on PSMA and CD3 and CD28 bispecifics targeting two different epitopes on the “nub” of MUC16 (the cell surface remnant after cleavage and release of CA-125).

Using cocultures of human peripheral blood mononuclear cells (PBMCs; containing human T cells) and C4-2 cells, we tested the ability of PSMAxCD28 in combination with PSMAxCD3 to induce tumor cytotoxicity and T cell activation (Fig. 2, A to E). We first verified that both PSMAxCD28 and PSMAxCD3 bispecifics bound specifically to the C4-2 prostate cancer cell line (fig. S7A) and that these bispecifics did not compete for binding to PSMA-expressing cells and therefore bind to different epitopes (fig. S7B). We found that the PSMAxCD28 bispecific markedly increased the ability of PSMAxCD3 to induce T cell killing of tumor cells, shifting the EC50 (median effective concentration) by almost 50-fold (Fig. 2B). Consistent with this enhancement of T cell cytotoxicity, PSMAxCD28 also shifted the EC50 and boosted the maximal induction of IFN-γ release (Fig. 2C), as well as the expansion and activation of CD4+ and CD8+ T cells as assessed by CD25 expression (Fig. 2, D and E). As a control, a nontargeting TSAxCD3 (EGFRvIIIxCD3, which does not bind to C4-2 cells) had no activity on its own, nor did adding PSMAxCD28 to this CD3 bispecific generate any activity (Fig. 2, B to E). The combination of PSMAxCD28 and PSMAxCD3 did not induce cytotoxicity or T cell activation when ovarian tumor cell line (OVCAR-3), which does not express the tumor target of interest (PSMA), was used as a negative control (fig. S8, A and B), in contrast to a prostate tumor cell line (22RV1) that endogenously expresses PSMA (fig. S8, C and D). In addition, we tested these PSMAxCD28 and PSMAxCD3 bispecifics in assays using cynomolgus monkey PBMCs and C4-2 cells and obtained similar results (fig. S9). These results confirm and extend the previous results seen with engineered HEK293/hCD20/hPSMA cells, demonstrating that the PSMAxCD28 bispecific has limited activity as a monotherapy but can potently enhance the ability of a CD3 bispecific to promote T cell activation, cytokine release, and killing of tumor cells endogenously expressing PSMA in vitro.

Fig. 2 TSAxCD28 bispecifics potentiate T cell cytotoxicity and activation only in the presence of TCR stimulation by TSAxCD3.

Human T cells were cultured with cancer target cells with endogenous PSMA expression (prostate cancer line C4-2) (A to E) or endogenous MUC16 expression (ovarian cancer line PEO-1) (F to J) and the indicated bispecifics for 96 hours. (A and F) Schematic of assay setup. (B and G) Tumor cell kill, percent viable cells. (C and H) IFN-γ release. (D and I) CD4 T cell counts and percent CD25+ in CD4. (E and J) CD8 T cell counts and percent CD25+ in CD8. Data are representative of at least two experiments.

To expand these findings, we generated a second TSAxCD28 bispecific (MUC16xCD28) using the same technologies that were used to generate the PSMAxCD28 bispecific (from VelocImmune-derived conventional MUC16 and CD28 antibodies). The constant region of MUC16xCD28 was subcloned into hIgG4s isotype as described (9). Biacore analysis showed that MUC16xCD28 bound MUC16 with a KD of ~9.1 × 10−10 and CD28 with a KD of ~1.7 × 10−7 (table S1). Both MUC16xCD28 and MUC16xCD3 (31) bound to PEO-1 ovarian cancer cells endogenously expressing MUC16 (fig. S10A) and did not compete for binding to MUC16 (fig. S10B). As described above for the PSMA bispecific, we examined the potential of the MUC16xCD28 bispecific to enhance cellular cytotoxicity using human PBMCs cocultured with PEO-1 cells (Fig. 2, F to J). We observed that the MUC16xCD3 bispecific by itself induced less than 40% killing of PEO-1 cells by T cells, even at maximal concentrations (Fig. 2G). Notably, addition of the MUC16xCD28 bispecific increased both the potency and depth of cytotoxicity induced by MUC16xCD3, resulting in greater than 80% killing of PEO-1 cells (Fig. 2G). The addition of MUC16xCD28 led to increased IFN-γ release by more than 10-fold over release with MUC16xCD3 alone (Fig. 2H). The combination of bispecific antibodies also induced expansion of CD4+ and CD8+ T cells and increased the expression of the activation marker CD25 (Fig. 2, I and J). As a control, a nontargeting TSAxCD3 (EGFRvIIIxCD3, which does not bind to PEO-1 cells) had no activity on its own, nor did adding MUC16xCD28 to this CD3 bispecific generate any activity (Fig. 2, G to J). MUC16xCD28, in combination with MUC16xCD3, did not activate T cells and had no cytotoxic effect on the prostate tumor cell line, 22RV1, which does not express the tumor target (MUC16), in comparison to the ovarian tumor cell line OVCAR-3 that endogenously expresses MUC16 (fig. S8). We also tested these MUC16xCD28 and MUC16xCD3 bispecifics in assays using cynomolgus monkey PBMCs and PEO-1 cells and obtained similar results to those described above (fig. S11).

To determine whether the enhanced cytotoxicity effect observed with the combination of MUC16xCD3 and MUC16xCD28 bispecific antibodies is T cell dependent, we performed additional experiments using purified human T cells (fig. S12A). Similar to the results obtained using human PBMCs (Fig. 2), MUC16xCD3 and MUC16xCD28 bispecific combination resulted in strong cytotoxicity against PEO-1 cells and induced T cell activation (fig. S12, B and C) that was not observed with the negative control cell line (fig. S12, D and E). This result demonstrated that this effect was not due to ADCC, because monocytes were absent from the coculture. To further validate that the cytotoxicity observed was not due to ADCC, one of our CD28-based bispecific antibodies, PSMAxCD28, was tested in an in vitro ADCC cell-based assay (fig. S13). In this assay, ADCC occurs via antibody binding to target cells and engagement of natural killer (NK) effector cells through FcR. Cell surface binding of PSMAxCD28 to target cells was confirmed by flow cytometry (fig. S13A). PSMAxCD28 bispecific antibody did not exhibit ADCC against any of the cell lines tested in the presence or absence of NK effector cells (fig. S13, B to E). Positive control anti-CD20 antibody (human IgG1 isotype) did mediate ADCC in an NK cell–dependent manner, confirming the capacity of the NK effector cells to induce ADCC of the target cells used in this assay (fig. S13, F to I). The lack of effector function activity of our PSMAxCD28 bispecific antibody is consistent with previously published results for MUC16xCD3 (31) and the minimal effector function activity of our stealth modified IgG4-based antibodies (IgG4s) with reduced binding to FcRs as previously described (9). Because human IgG4 is known to induce ADCC by murine macrophages (32), we sought to further test our human IgG4s isotype in a mouse ADCC assay using anti-CD20 antibodies with hIgG1, hIgG4, or hIgG4s. As expected, strong ADCC activity was induced with human IgG1 but not our IgG4s isotype (fig. S13J), confirming our Biacore data (table S2 and fig. S5).

Together, these data establish the general principle that CD28 bispecifics show limited single-agent activity, but they can potently enhance the ability of CD3 bispecifics to promote T cell killing of tumor cells in vitro through enhanced T cell activation, T cell proliferation, and proinflammatory cytokine release.

MUC16xCD28 bispecific enhances the in vivo antitumor activity of the MUC16xCD3 bispecific in a xenogenic ovarian tumor model

To test the ability of CD28 and CD3 bispecifics to promote tumor killing in vivo, we used the well-established xenogenic intraperitoneal ovarian OVCAR-3 tumor model, in which tumor cells are introduced into immunodeficient mice that are reconstituted with human PBMCs (31). Like other ovarian cancer cell lines, the OVCAR-3 cells express MUC16. Before implantation, the OVCAR-3 cells were engineered with a luciferase reporter to allow in vivo tracking of tumor growth over time using bioluminescence imaging (BLI) (Fig. 3A). Implanted OVCAR-3 tumors grew unabated in mice treated with EGFRvIIIxCD3 bispecific, a control CD3 bispecific that did not bind to these cells, and in mice treated only with the MUC16xCD28 bispecific (Fig. 3, A and B). Although the MUC16xCD3 bispecific alone demonstrated significant antitumor activity (P < 0.05 on days 8, 12, and 25; P < 0.01 on days 15 and 20), it did not completely clear the OVCAR-3 tumors (Fig. 3, A and B), whereas the addition of the MUC16xCD28 bispecific to the MUC16xCD3 bispecific enhanced the in vivo antitumor effect, resulting in further reduction of tumor burden (Fig. 3, A and B) over MUC16xCD3 alone. Consistent with enhanced antitumor activity, the combination of both bispecifics also increased the secretion of circulating cytokines (fig. S14).

Fig. 3 MUC16xCD28 bispecific enhances the in vivo antitumor activity of the MUC16xCD3 bispecific in a xenogenic ovarian tumor model.

NSG mice were pre-engrafted with human PBMCs and implanted with OVCAR-3/Luc. Mice were treated with EGFRvIIIxCD3, MUC16xCD3, and/or MUC16xCD28 by intraperitoneal injection on days 5 and 8 after tumor implantation (indicated by arrows). Tumor burden was measured by monitoring bioluminescence over time. (A) Representative bioluminescence images of mice from the indicated treatment groups. (B) Tumor burden as measured by median radiance (p/s/cm2/sr) over time. Values represent the group median plus range. P values were calculated with Mann-Whitney test for each time point. MUC16xCD3 versus EGFRvIIIxCD3: *P < 0.05 or **P < 0.01; MUC16xCD3 + MUC16xCD28 versus EGFRvIIIxCD3: ##P < 0.01. (C) Tumor burden and correlation to CA-125 concentration in serum on day 26. n = 5 mice per group. Data are representative of at least two experiments.

As mentioned previously, the MUC16 bispecifics bind to the remaining nub of MUC16 on the ovarian cancer cell surface after proteolytic cleavage has released the prognostic ovarian cancer biomarker CA-125 (25), but does not bind to soluble CA-125 (fig. S15, A and B). To determine whether the MUC16xCD28 bispecific perturbed the ability to use CA-125 as a biomarker for ovarian tumor burden, we measured the concentration of CA-125 in the serum of tumor-bearing mice treated with MUC16xCD28. We found that CA-125 concentration correlated with tumor burden regardless of treatment. The lowest CA-125 concentrations were seen in the mice treated with the combination of bispecifics (Fig. 3C), as previously demonstrated for MUC16xCD3 bispecific (31).

CD28 bispecifics enhance the in vivo antitumor activity of CD3 bispecifics in syngeneic tumor models using genetically humanized mice

To further our mechanistic understanding, we evaluated both CD28 and CD3 bispecifics in mouse syngeneic tumor models with intact immune systems. Using the established C57BL6 syngeneic MC38 tumor model, we genetically introduced the hPSMA gene (pLVX.EF1a.hPSMA) or hMUC16 gene (pLVX.EF1a.hMUC16) into the MC38 cells, creating tumor antigen–specific MC38/hPSMA or MC38/hMUC16 tumor cells. To avoid the possibility that the mice would spontaneously reject these otherwise syngeneic tumors because they were expressing an introduced human tumor antigen, we genetically humanized the PSMA and MUC16 genes in these mice using VelociGene technology (33, 34). In addition, we humanized the CD3γ-δ-ε and CD28 genes as described below, so that the bispecifics would recognize the host T cells (via hCD3 or hCD28), as well as the human tumor antigens in both normal tissues and the tumors (hPSMA or hMUC16), mimicking the actual clinical situation. Using this approach, we generated the following double- and triple-humanized mice, respectively: hMUC16/hCD3 (31) and hPSMA/hCD3/hCD28 mice (fig. S16). For the PSMA experiments, we used bispecifics targeting all the human components, whereas in the MUC16 experiments, we used a bispecific antibody with specificity for mouse CD28. The human CD3 and human CD28 expression on T cell subsets in the hPSMA/hCD3/hCD28 mice was validated by flow cytometry (fig. S17A), and the ratio of CD4+/CD8+ and the frequency of regulatory T cells (Tregs) were found to be similar between the humanized mice and their wild-type counterparts, indicating normal development of T cells (fig. S17, B and C). The T cells from the triple-humanized PSMA mice responded to anti-CD3/anti-CD28 stimulation similarly to T cells from wild-type mice in an in vitro T cell IL-2 release assay, thus functionally validating the signaling through hCD3 and hCD28 receptors (fig. S17D). Human PSMA expression was evaluated by quantitative polymerase chain reaction and immunohistochemistry (IHC), showing a range of tissue expression as expected, thus validating previous findings (fig. S18, A and B) (35). Similar validations were previously performed on the double-humanized hMUC16/hCD3 mice (31).

Appropriate humanized mice received MC38/hMUC16 or MC38/hPSMA tumor cells implanted subcutaneously on the right flank and were treated with isotype control, the individual CD3 or CD28 bispecifics, or the combinations by intraperitoneal injection (Fig. 4, A to F). Antibodies were dosed twice per week starting on the day of implantation for a total of three doses (PSMA) or six doses (MUC16). Pharmacodynamic analysis confirmed antibody concentrations in the serum (~60 to 100 μg/ml) and, as expected with human antibodies in a xenogenic mouse host, mouse anti-human antibodies (table S3). In both the MUC16 and the PSMA tumor models, the combination of CD3 and CD28 bispecifics provided the best antitumor responses (Fig. 4, A and D), as was also noted in assays of cytokine production (Fig. 4, B C, E, and F). Histological analysis of normal tissues that express PSMA by immunohistochemistry revealed minimal mononuclear infiltration in the kidneys and salivary glands of mice treated with PSMAxCD3 monotherapy or PSMAxCD3 and PSMAxCD28 combination (fig. S19, A and B). These histological findings are consistent with cytokine responses observed in these treatment groups (Fig. 4, E and F). Tissue infiltration appears to be PSMAxCD3 driven, because there were no findings in normal tissue from mice treated with PSMAxCD28 monotherapy (fig. S19).

Fig. 4 TSAxCD28 bispecific enhances TSAxCD3 antitumor efficacy and T cell activation in mouse syngeneic tumor models.

(A to C) hCD3/hMUC16 mice (seven mice per group) were implanted with 1 × 106 MC38/hMUC16 cells resuspended in PBS by subcutaneous injection on the right flank. MUC16xCD3 (0.01 mg/kg), MUC16xCD28 (0.05 mg/kg), or isotype controls were administered by intraperitoneal injection starting on day 0 (day of tumor implant) twice per week for six treatments. (D to H) hPSMA/hCD3/hCD28 mice (four to eight mice per group) were implanted with 1 × 106 MC38/hPSMA tumor cells resuspended in PBS by subcutaneous injection on the right flank. PSMAxCD28, PSMAxCD3, or isotype controls (all 5 mg/kg) were administered by intraperitoneal injection on days 0, 3, and 7 (D to F) or days 11 and 14 after tumor implant (G and H). (A and D) Tumor volume was monitored by caliper measurement over time. Values shown are means ± SEM. Statistical significance was calculated with two-way ANOVA. Combination versus isotype control: **P < 0.01 and ****P < 0.0001; TSAxCD3 versus isotype control: #P < 0.05; TSAxCD28 versus isotype control: $P < 0.05. (B, C, E, and F) Analysis of serum cytokines at 4 hours after dose on day 7 (B and C) or day 0 (E and F). Statistical significance was calculated with one-way ANOVA in comparison to isotype **P < 0.01 and ****P < 0.0001. Data are representative of three experiments. (G and H) Fluorescence-activated cell sorting analysis of T cell phenotype in spleen and tumor harvested on day 17. Data are represented as means + SEM. Statistical analysis: two-way ANOVA with Tukey’s multiple comparisons; isotype (n = 8), PSMAxCD3 (n = 8), PSMAxCD28 (n = 8 for spleen, n = 6 for tumor), and combo (n = 8 for spleen, n = 6 for tumor). (G) Percentage of cells in each cluster from each treatment group (top); overlay of indicated cluster on viSNE plot. (H) Expression (mean fluorescence intensity) of the indicated markers is overlaid on the viSNE plot. Color scale indicates expression: red, high; blue, low.

Unlike the previous in vitro and in vivo analyses in which the CD28 bispecifics had very limited single-agent activity (see above), the CD28 bispecifics in these syngeneic MC38/hPSMA and MC38/hMUC16 models had more notable activity as single agents. This suggested that signal 1 was already being activated to some degree in these MC38 models. Consistent with this, it has been previously shown that MC38 tumor cells express reactivated endogenous retroviral proteins such as p15E and that C57BL6 mice can generate endogenous T cells that recognize and respond to this neoepitope (36, 37). We confirmed that in our MC38 models, intratrumoral T cells responsive to this p15E neoantigen could easily be detected (fig. S20). Thus, CD28 bispecifics in these MUC16 and PSMA syngeneic tumor models can boost endogenous TCR/CD3-dependent T cell responses, which can then further be enhanced by providing additional signal 1 activation via a CD3 bispecific.

To determine the cellular mechanism underlying the combination therapy, we profiled tumor-infiltrating and spleen CD8+ T cells from these experiments by high-dimensional flow cytometry and used unsupervised clustering approaches (Fig. 4, G and H, and fig. S21). Antibody treatment did not change the absolute counts or frequency of T cell subsets (fig. S21, A and B). However, we found increased expression of activation marker PD-1 on CD8+ T cells in the tumor (fig. S21C). Furthermore, we found that each treatment drove unique CD8+ T cell clusters in spleen and tumor (Fig. 4, G and H, and fig. S21, D and E). Single treatment regimens reduced intratumoral CD8+ T cells with less activated phenotype (lower ICOS, KLRG1, Ki67, PD1, CD38, and LAG3; Fig. 4, G and H, and fig. S21E), as shown in cluster C35. However, combination therapy significantly drove the expansion of a more activated/memory T cell phenotype (P < 0.05 compared with PSMAxCD28 or PSMAxCD3 monotherapy, P < 0.01 compared with isotype control) (expressing TCF1, CD122, CD127, PD1, ICOS, KLRG1, CD38, and CD5; Fig. 4, G and H, and fig. S21F), as shown in cluster C4. These data suggest a phenotypic change in T cells that corresponds with antitumor efficacy induced by the combination of CD3- and CD28-based bispecific antibodies.

TSAxCD28 bispecifics do not induce cytokine release and systemic T cell activation in vitro or in cynomolgus monkeys in vivo

Bivalent CD28-activating antibodies, termed “CD28 superagonists” (CD28-SA), have caused universal and profound toxicity via CRS in human trials (23); this history raises obvious concerns for any agents that involve activating the CD28 pathway. Because CD28-SA tested in vitro in a soluble format induces only a minimal T cell activation and/or cytokine production (Fig. 1A and fig. S6) (30), we first used the FDA-recommended in vitro dry- and wet-coated human T cell proliferation assays (38) to determine whether either of our costimulatory bispecific antibodies could promote T cell activation and cytokine release in a manner equivalent to in-house generated CD28-SA (table S1). In contrast to the strong proliferation induced by CD28-SA, both MUC16xCD28 and PSMAxCD28 bispecifics (as well as the parent bivalent CD28 antibodies used to make these bispecifics, which were specifically selected so as not to be CD28-activating antibodies) failed to induce human T cell proliferation in these assays (Fig. 5, A and B). In addition, CD28-SA did induce systemic cytokine production in the genetically CD28 humanized mice (fig. S22), consistent with its superagonistic activity in vivo. Furthermore, we have shown that TSAxCD28 bispecifics could potentiate in vitro the TSAxCD3-induced activation of cynomolgus monkey T cells (figs. S9 and S11). To evaluate in vivo the tolerability of TSAxCD28 bispecifics alone, as compared with CD28-SA, we conducted exploratory studies in cynomolgus monkeys. In these studies, female monkeys were dosed with MUC16xCD28 alone, whereas male monkeys received PSMAxCD28 or CD28-SA (table S4). Blood samples were collected for cytokine and flow cytometry immunophenotyping analysis. Although CD28-SA administered to monkeys induced significant cytokine release (P < 0.01 for IFN-γ, IL-2, and IL-8; P < 0.001 for IL-6 and IL-10), lymphocyte margination (P < 0.001), and T cell activation (P < 0.0001), it was notable that no cytokine release, T cell margination, or T cell activation was observed after the administration of MUC16xCD28 or PSMAxCD28 (Fig. 5, C to G, and table S4). Overall, these preliminary observations suggest that TSAxCD28 bispecifics are well tolerated in primates and do not induce cytokine release or T cell activation as seen with CD28-SA (Fig. 5, F to H).

Fig. 5 TSAxCD28 alone or in combination therapy does not induce systemic T cell activation in comparison to CD28-SA.

MUC16xCD28 or control antibodies (A) and PSMAxCD28 or control antibodies (B) were anchored to assay plates using the dry- or wet-coating method as described. Human PBMCs were incubated in assay plates, and proliferation was measured after 72 hours. Data shown are from individual donors, with a line representing the means ± SD. n = 4 donors. (C to E) Cynomolgus monkeys received a single dose of MUC16xCD28 at either 1 or 10 mg/kg (indicated in parentheses). Blood was collected at the indicated times after dose (hours). (C) Serum cytokines, (D) relative T cell counts, and (E) frequency of Ki67+ and ICOS+ T cells (% of CD3) are shown. (F to H) Cynomolgus monkeys received a single dose of PSMAxCD28, CD28-SA, or isotype control (EGFRVIII) at either 1 or 10 mg/kg (indicated in parentheses). Blood was collected at the indicated times after dose (hours). (F) Serum cytokines, (G) relative T cell counts, and (H) frequency of Ki67+ and ICOS+ T cells (% of CD3) are shown. Data represent means ± SEM. n = 3 animals per group. P values were calculated with two-way ANOVA with comparison to isotype control. (**P < 0.01, ***P < 0.001, and ****P < 0.0001).


It has long been appreciated that T cell activation via the TCR complex (signal 1) can be markedly enhanced by costimulatory signals such as those mediated when the CD28 receptor on T cells engages its ligands (CD80/B7.1 and CD86/B7.2) on target cells (signal 2) (2). In agreement with our data, the potential for CD28 costimulation to enhance the antitumor activity of T cells was first demonstrated by studies in which B7 ligands were overexpressed on tumor cells (39, 40), which showed improved T cell rejection of such B7-expressing tumors. This potential inspired efforts to evaluate CD28-activating antibodies in human trials. Tragically, the 2006 trial of such an antibody (TGN1412) resulted in life-threatening complications in all six human volunteers (23) due to multiorgan failure resulting from massive CRS. This catastrophe led to cessation of any further testing of CD28-activating antibodies in humans.

CD28 bispecific antibodies, which would not directly activate CD28 unless clustered on tumor cell surfaces, offered the possibility of promoting costimulation only at the tumor site, without the systemic toxicity of conventional CD28-activating antibodies. Initial versions of such CD28 bispecifics were proposed and evaluated in the 1990s (4143); however, the early technology available at that time required chemical cross-linking or hybrid/hybridoma fusions to create the proposed biotherapeutics and resulted in suboptimal reagents, which had profound activity on their own, independent of their clustering on tumor cells (reminiscent of conventional CD28 antibodies, presumably due to nonspecific aggregation of these bispecifics). Moreover, these early approaches also required preactivation of T cells in vitro to observe any antitumor activity in vivo. Together, the catastrophic clinical results with the TGN1412 CD28-activating antibody, as well as the limitations of these early CD28 bispecific approaches, dissuaded further exploration of these approaches.

We have resurrected the notion of CD28 bispecifics, and we describe a class of CD28 costimulatory bispecific antibodies that can markedly and safely promote antitumor activity by providing a costimulatory signal 2. These CD28 bispecifics have limited activity on their own (in the absence of signal 1) but can markedly enhance antitumor activity in the setting of signal 1, as can be provided by pairing these CD28 bispecifics with the emerging class of CD3 bispecifics (or if these CD28 bispecifics are used in settings where there are already endogenous populations of tumor-specific T cells). The generation and preclinical testing of this CD28 bispecific approach were dependent on (i) the use of a bispecific platform that was initially developed to produce CD3 bispecifics and was recently validated both technologically (9) and clinically ( NCT02290951; NCT03564340) for these CD3 bispecifics, and which we then adapted to efficiently produce CD28 bispecifics that display minimal activity in the absence of a specific signal 1; (ii) the development of multiple xenogenic and syngeneic genetically humanized (33, 34) animal tumor models to assess these CD28 bispecifics on their own and in combination with CD3 bispecifics; and (iii) together with a much deeper knowledge of the CRS and its clinical development (21, 44, 45), the validation of a monkey model in which any potential toxicity of these CD28 bispecifics could be compared to that of conventional CD28-activating antibodies.

Here, we described the generation and testing of TSAxCD28 costimulatory bispecific antibodies targeted against two different TSAs: one for ovarian cancer (MUC16xCD28) and one for prostate cancer (PSMAxCD28). We showed that, in the absence of signal 1, these CD28 bispecifics have minimal activity in vitro or in vivo. However, these CD28 bispecifics can be paired with CD3 bispecifics to form artificial “immune synapses” containing the tumor antigens as well as the TCR and CD28 complexes. Because the organization of adhesion molecules and cytoskeletal components at the immune synapse directly affects T cell activation (46), it is worth investigating the molecular components of immune synapse induced by TSAxCD3 and TSAxCD28 in future studies. Moreover, when paired with appropriate CD3 bispecifics in vitro, these CD28 bispecifics can efficiently and specifically promote T cell activation and tumor cell killing in an antigen-dependent manner. Furthermore, these CD28 bispecifics also enhanced the antitumor activity of CD3 bispecifics in vivo in a tumor antigen–specific manner in xenogenic and syngeneic tumor models; in such models, the CD28 bispecifics have minimal single-agent activity unless tumor-specific T cells are already present, and in such settings, they appear to enhance this specific activity in a tumor antigen–dependent manner. In addition, TSAxCD28 and TSAxCD3 combination therapy drove substantial expansion of an intratumoral activated/memory T cell phenotype in vivo. Last, toxicology studies in genetically humanized immunocompetent mice, as well as in cynomolgus monkeys, demonstrated that these bispecifics exhibit limited activity and no toxicity as single agents, as directly compared with conventional CD28-activating antibodies. It should be noted that previous studies with CD28-SA in monkeys failed to predict the profound cytokine release and T cell activation seen in humans (TeGenaro AG;, and this was attributed to lower CD28 expression in monkeys (47). Although tolerability studies in cynomolgus monkeys might not be predictive of CRS in humans, the strong signals we noted with CD28-SA in monkeys suggest that this was missed by TeGenaro et al. because they did not examine the early time points when these responses can be robustly observed.

Often, the characterization of human-specific clinical candidates in the field of immuno-oncology is limited to testing in xenogeneic tumor models with engrafted human immune cells. Although these xenogenic models (such as the OVCAR-3 model we used here) can be very useful, they have limitations. The mice used in such xenogenic models do not express the human tumor target in their normal tissues, thereby precluding assessment of the test agent in the setting of normal tissue expression of the target. If a target is normally also highly expressed in normal tissues, this could limit antitumor efficacy by diverting the test agent from the tumor and could result in toxicity on these normal tissues, none of which could be assessed in a xenogenic model. An additional limitation could involve the activity of engrafted human PBMCs transferred to an immunodeficient mouse, which could differ from that of normal host T cells found in an immunocompetent system. To overcome these limitations and provide better models for testing human-specific clinical candidates, we created double and triple genetically humanized mice. In these models, the tumor antigens were genetically humanized to allow for their normal expression in appropriate host tissues (for both PSMA and MUC16), and the CD3 and/or CD28 components were genetically humanized to allow immunocompetent host cells to respond to the human-specific clinical candidates. In these genetically humanized immunocompetent syngeneic animal models, we found that—just as in the xenogenic animal models—the CD28 bispecifics for both the PSMA and MUC16 tumor targets enhanced the antitumor activity of their appropriate CD3 bispecifics. The similar enhancement of antitumor efficacy by two different TSAxCD28 bispecifics (MUC16 and PSMA) across multiple preclinical models suggests that this therapeutic modality is robust and not limited to a specific tumor model and could have broader use as a combination target class for immunotherapy. One of the constraints of testing human antibodies in mouse models with competent immune systems is the generation of neutralizing mouse anti-human antibodies, which restricts the effective dosing over time. Overall, our findings highlight that TSAxCD28 bispecifics can be combined with TSAxCD3 bispecifics and may provide a biologic solution that could markedly enhance the efficacy of the well-studied TSAxCD3 bispecifics in a reasonably safe and well-tolerated manner, justifying testing in human trials.

TSAxCD3 bispecifics represent a promising emerging class of immunotherapy, but further optimization of antitumor activity will surely be necessary in many cases. Just as CAR-T approaches have used chimeric receptors that artificially activate both signal 1 and signal 2 so as to improve their antitumor activity (16, 48), we now show the potential benefit of combining CD3 bispecifics (which provide signal 1) with CD28 bispecifics (which provide signal 2) to enhance antitumor activity. In addition to the practical benefits that such an approach might have over CAR-T therapies—in that it does not require laborious cell therapy preparation that must be individually customized for each patient, nor does it require that patients be preemptively “lymphodepleted” via toxic chemotherapy so that they can accept this cell therapy often associated with adverse effects (21, 49)—our bispecific approach offers the potential for increased efficacy as well as increased safety and specificity of action. That is, it is possible to take advantage of combinatorial targeting by pairing a CD3 bispecific for one antigen with a CD28 bispecific for a second antigen (for the same indication), such that increased efficacy will only occur on tumor cells expressing both antigens, thus focusing T cell killing only to tumor cells expressing both antigens while limiting off-target toxicity in normal tissues expressing only one of the antigens. Collectively, our data suggest that combining CD28-based bispecifics with CD3-based bispecifics may provide well-tolerated off-the-shelf biologic solutions with markedly enhanced antitumor activity.


Study design

The objective of this study was to develop TSAxCD28 bispecific antibodies and demonstrate that TSAxCD28 potentiates TSAxCD3-induced T cell activation in vitro and safely enhances antitumor efficacy in vivo. Activation of T cells in vitro was demonstrated by showing images of bispecific antibodies localized at the immunological synapse of T cell and target cell conjugates and enhancement of TSAxCD3-induced human and cynomolgus T cell proliferation, cytokine release, and cytotoxicity when cocultured with various tumor target-expressing cancer cells. In vivo antitumor efficacy was evaluated in two different mouse tumor models (xenogenic and syngeneic). In xenogenic tumor models, mice were assigned to groups matched for equal T cell engraftment and tumor burden at the start of treatment. In syngeneic models, control and experimental treatments were randomly administered to age- and sex-matched mice. Tumor burden and serum cytokines were monitored over time to assay response to bispecific antibody treatment. Single-cell suspensions from tumor and spleen were profiled by high-dimensional computational flow cytometry to identify phenotypic changes in cell populations. The purpose of the cynomolgus studies was to determine the safety and tolerability (pharmacologic and toxicologic profile) of TSAxCD28 in nonhuman primates. Animals were examined for toxicity by clinical observations and blood sample collections to analyze serum cytokines and T cell phenotype. The number of experimental replicates is 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. 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).

Generation of bispecifics

VelocImmune mice (Regeneron Pharmaceuticals) (27, 28) were immunized to generate CD28-, PSMA-, or MUC16-specific antibodies. For CD28, mice were immunized with the human CD28 protein fused to the Fc portion of mouse IgG2a. For PSMA, mice were immunized with LnCAP cells (American Type Culture Collection). Antibodies were screened for binding to their respective proteins by enzyme-linked immunosorbent assay, as well as for binding to LnCAP, Jurkat T cells, and human and monkey T cells. Generation of MUC16-specific antibodies has been described (31). The antibodies for CD28, PSMA, or MUC16 were assembled with a human IgG4 constant region containing a S228P (serine-to-proline exchange) substitution in the hinge region to minimize half-antibody formation, substitutions to minimize antibody-mediated effector functions, and a mutation on the CD28 Fc to limit protein A binding (9). Antibodies were produced in Chinese hamster ovary cells and purified using protein A chromatography as previously described (50).

Xenogenic tumor model and BLI

Experiments were performed as described (31). Briefly, 20- to 24-week-old NSG (NOD SCID gamma chain knockout) mice (the Jackson laboratory) were engrafted with 5 × 106 human PBMCs by intraperitoneal injection (ReachBio) before 2 × 106 ascites cells from the OVCAR-3/Luc cell line, previously passaged in vivo, were administered by intraperitoneal injection (day 0). Mice were treated with EGFRvIIIxCD3 at ~0.5 mg/kg (12.5 μg per mouse), MUC16xCD3 at ~0.1 mg/kg (2.5 μg per mouse), or MUC16xCD28 at ~5 mg/kg (100 μg per mouse), as indicated, by intraperitoneal injection on days 5 and 8 after tumor implant. To measure tumor burden, mice were injected intraperitoneally with 150 mg/kg of the luciferase substrate d-luciferin (PerkinElmer), suspended in phosphate-buffered saline (PBS). Ten minutes later, BLI of the mice was performed under isoflurane anesthesia using the Xenogen IVIS system (PerkinElmer). Image acquisition was carried out with the field of view at D, subject height of 1.5 cm, and medium binning level for 0.5-min exposure time. BLI signals were extracted using the Living Image Software (Xenogen). Regions of interest were drawn around each tumor mass, and photon intensities were recorded as p/s/cm2/sr. Mice that did not receive OVCAR-3/Luc cells served as a baseline reading for BLI activity. These baseline mice (n = 3) with no tumors were imaged on each day, and the lower limit of detection was calculated as the mean BLI reading across all imaged tumor-free mice.

Syngeneic tumor studies

Syngeneic tumor studies were carried out in mice genetically modified to express human CD3 and a portion of human MUC16 using VelociGene technology, as described previously (31). hCD3/hMUC16 humanized mice (seven mice per group, 8 to 16 weeks old) were implanted with 1 × 106 MC38/hMUC16 cells resuspended in PBS by subcutaneous injection on the right flank. Antibodies were administered by intraperitoneal injection starting on day 0 (day of tumor implant) with MUC16xCD3 (0.01 mg/kg), MUC16xCD28 (0.05 mg/kg), or isotype twice per week for six treatments. Tumor growth was monitored over time using caliper measurements of X and Y diameters. Tumor volume was calculated [X*Y*(X/2)]. Mice were euthanized when tumor size was greater than 2000 mm3.

Mice expressing human CD28, human CD3, and human PSMA in place of the corresponding mouse genes were generated using VelociGene technology (33, 34). hPSMA/hCD3/hCD28 humanized mice were created by breeding together three “single-humanized” mouse strains. The single-humanized CD3 mouse line was generated as follows: Nucleotide sequences encoding the ectodomains of Cd3e, Cd3d, and Cd3g were deleted and replaced with orthologous regions of human CD3E, CD3D, and CD3G, respectively, retaining mouse sequence encoding signal peptides and cytoplasmic domains. In addition, the splice acceptor regions of mouse introns directly preceding ectodomain-encoding exons were also humanized, as were any intervening introns. All other introns and intergenic sequences remained mouse. For selection purposes, a fLoxed self-deleting neomycin resistance cassette was inserted in the intron preceding the humanized region in Cd3e. The single-humanized CD28 mouse line was created such that the mouse intron 1 splice acceptor and exon 2 to part of exon 3, encoding the CD28 ectodomain, were replaced with orthologous human sequence. Mouse intron 2 was also humanized. The sequence encoding the signal peptide, a short stalk motif just N-terminal to the transmembrane domain, transmembrane, and cytoplasmic region, remained mouse. A fLoxed self-deleting neomycin resistance cassette was inserted in intron 1. The single-humanized FOLH1 (PSMA) mouse line was generated by replacing the mouse ectodomain-coding region (exons 2 to 19) with orthologous human sequence. All intervening introns were also humanized. The human 3′ untranslated region and an additional 384 base pairs of downstream intergenic sequence were also included. A fLoxed neomycin resistance cassette was inserted downstream of the human sequence. These single-humanized CD3EDG, CD28, and FOLH1 lines are referred to as Cd3edghu/hu, Cd28hu/hu, and Folh1hu/hu, respectively. VGF1 embryonic cells (50% C57BL/6NTac, 50% 129S6/SvEvTac) were targeted for partial humanization of each of the three loci. Separately, targeted cells were microinjected into eight-cell embryos from Charles River Laboratories Swiss Webster albino mice, yielding F0 VelociMice that were 100% derived from the targeted cells (34). These mice were bred once to C57BL/6NTac to leverage deletion of the antibiotic resistance cassette in the F0 germ line by Prm1-expressed cre recombinase, generating targeted, cassette-free mice in a genetic background that was 75% C57BL/6NTac, 25% 129S6/SvEvTac. C3edghu/hu, Cd28hu/hu, and Folh1hu/hu mice were subsequently bred to generate triple-humanized, heterozygous C3edghu/+ Cd28hu/+ Folh1hu/+ animals and maintained in the Regeneron animal facility during the whole period. Wild-type mice used as controls in all experiments were 75% C57BL/6NTac, 25% 129S6/SvEvTac.

hPSMA/hCD3/hCD28 mice (four to eight mice per group, 8 to 16 weeks old) were implanted with 1 × 106 MC38/hPSMA tumor cells resuspended in PBS by subcutaneous injection on the right flank. PSMAxCD28 bispecific antibody, PSMAxCD3 bispecific antibody, or human IgG4s isotype control was administered as a monotherapy or in combination by intraperitoneal injection on days 0, 3, and 7 at 5 mg/kg. Tumor growth was monitored over time using caliper measurements of X and Y diameters. Tumor volume was calculated as [X*Y*(X/2)]. Mice were euthanized when tumor size was greater than 2000 mm3.

Flow cytometry

For immunophenotyping experiments, hPSMA/hCD3/hCD28 mice (eight mice per group) were implanted with 1 × 106 MC38/hPSMA tumor cells resuspended in PBS by subcutaneous injection on the right flank. PSMAxCD28 bispecific antibody, PSMAxCD3 bispecific antibody, or human IgG4s isotype control was administered as a monotherapy or in combination by intraperitoneal injection on days 11 and 14 at 5 mg/kg. Tumors and spleens were harvested on day 17, 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 LIVE/DEAD fixable blue dead cell staining kit (Thermo Fisher Scientific). Samples were acquired on Symphony (BD Biosciences) and analyzed using the 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. To cluster T cells automatically based on specific markers, viSNE analysis from Cytobank was used.

Cynomolgus toxicology studies

The cynomolgus monkey studies were conducted at facilities accredited by the Association for Assessment and Accreditation of Laboratory Animal Care and the Animal Welfare Assurance issued by the Office of Laboratory Animal Welfare, registered with the U.S. Department of Agriculture and IACUC. For MUC16xCD28 and MUC16xCD3, studies were carried out at SNBL USA using female cynomolgus monkeys (Macaca fascicularis) (three animals per group). For PSMAxCD28 and CD28-SA, studies were carried out at Charles River using male cynomolgus monkeys (M. fascicularis) (three animals per group). Subjects received a single dose of each test article via intravenous infusion for about 30 min (combination treatment was administered as separate infusion for total of 1 hour). Assessment of toxicity was based on clinical observations, qualitative food consumption, body weight, neurological examinations, vital signs (body temperature, heart rate, pulse oximetry, and respiration rate), and clinical and anatomic pathology. Blood and tissue samples were collected for cytokine analysis, immunophenotyping analysis, histopathology, and toxicokinetic evaluation. For peripheral blood flow cytometry, blood was collected into potassium EDTA tubes; lysed; stained with anti-CD3, anti-Ki67, and anti-ICOS (BD Biosciences); and analyzed with FACSCanto II. For cytokine analysis, blood was collected into serum separator tubes with anticoagulant. Serum was separated via centrifugation at 1000g to 2000g at 4°C for 10 to 15 min and analyzed using the MSD U-Plex platform (IL-2, IL-4, IL-6, IL-8, IL-10, tumor necrosis factor–α, and IFN-γ). C-reactive protein concentrations were analyzed on a Roche Modular P800 system.

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 the figure legends, with P < 0.05 considered statistically significant. Original data are shown in data file S1.


Materials and Methods

Fig. S1. Schematic of T cell activation at the immune synapse via signal 1 and signal 2.

Fig. S2. Schematic of TSAxCD3 bispecific and mechanism for T cell activation at the immune synapse.

Fig. S3. Schematic of TSAxCD3 and TSAxCD28 bispecifics and mechanism for T cell activation.

Fig. S4. Introduction of co-stimulatory ligand expression on tumor cell lines inhibits tumor growth in vivo.

Fig. S5. Biacore sensorgrams for binding of mIgG2a, hIgG4, and hIgG4s to mouse FcγRs.

Fig. S6. PSMAxCD28 potentiates T cell activation only in the presence of TCR stimulation by CD20xCD3.

Fig. S7. PSMAxCD28 and PSMAxCD3 bispecifics bind simultaneously to PSMA+ tumor cells.

Fig. S8. TSA bispecifics potentiate T cell activation only in the presence of TSA expression on target cells.

Fig. S9. PSMAxCD28 potentiates cynomolgus T cell activation only in the presence of TCR stimulation by PSMAxCD3.

Fig. S10. MUC16xCD28 and MUC16xCD3 bispecifics bind simultaneously to MUC16+ tumor cells.

Fig. S11. MUC16xCD28 potentiates cynomolgus T cell activation only in the presence of TCR stimulation by MUC16xCD3.

Fig. S12. MUC16xCD28 potentiates purified T cell activation and cytotoxicity dependent on MUC16xCD3.

Fig. S13. MUC16 bispecifics bind to MUC16-expressing cells in the presence of soluble.

Fig. S14. MUC16xCD28 and MUC16xCD3 combination induces cytokines in humanized xenogenic mouse models.

Fig. S15. MUC16 bispecifics bind to MUC16-expressing cells in the presence of soluble CA-125.

Fig. S16. Targeting strategy for the generation of Cd3edghu/hu, Cd28hu/hu, and Folh1hu/hu mice.

Fig. S17. Expression and functional validation of CD3 and CD28 in hPSMA/hCD3/hCD28 mice.

Fig. S18. Validation of human PSMA expression in PSMA humanized mice.

Fig. S19. Histology of PSMA+ tissue in hPSMA/hCD3/hCD28 mice treated with PSMA bispecifics.

Fig. S20. Intratumoral T cells recognize p15E peptide.

Fig. S21. PSMAxCD28 enhances anti-tumor immunity and T cell activation induced by PSMAxCD3.

Fig. S22. CD28 superagonist induces systemic cytokine response in CD28 humanized mice.

Table S1. Biacore kinetics of MUC16, PSMA, and CD28.

Table S2. Biacore kinetics of mouse FcγRs.

Table S3. Pharmacokinetic analysis of PSMA bispecifics in hPSMA/hCD3/hCD28 mice.

Table S4. Summary of cynomolgus monkey toxicity study.

Data file S1. Primary data.


Acknowledgments: We acknowledge T. Gorenc and S. Chandwani for project management, and R. Foster and P. Krueger for the pharmacokinetic analysis. Funding: Funding was provided by Regeneron Pharmaceuticals Inc. This work was funded, in part, by Sanofi. Author contributions: D.S. and G.D.Y. conceived the concept. D.S. and J.C.W. designed the study. J.C.W., L.H., A.C., E.U., X.Y., B.W., Q.W., I.R., A.P., L.C., K.V., P.R., E.H., H.A., E.O., J.G., P.P., L.H., D.C., M.L., K.P., K.Y., J.K., J.J.W., C.-J.S., D.D., and T.P. performed the experiments. J.C.W., L.H., A.C., A.H., E.U., R.S., S.G., D.A., X.Y., B.W., Q.W., I.R., A.P., L.C., K.V., P.R., E.H., H.A., E.O., J.G., P.P., L.H., D.C., N.S.O., C.-J.S., D.D., T.H., T.P., J.M., A.R., J.R.K., E.S., and D.S. analyzed the data. N.S.O. and W.P. generated and provided the genetically modified mice for the experiments. D.S., J.C.W., and G.D.Y. wrote the manuscript. A.J.M., T.H., J.M., D.M., W.P., J.R.K., E.S., D.S., J.L., G.T., M.A.S., A.J.M., G.D.Y., D.S., A.O., and W.O. gave technical support and conceptual advice. 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 as well as US62/782,142 and US62/815,861 held/submitted by Regeneron Pharmaceuticals that cover PSMAxCD28 and MUC16xCD28, respectively. G.D.Y., A.C., E.S., L.H., and J.R.K. are inventors on patent applications US2018/0112001, PCT/US2017/053113, US2018/0118848, and PCT/US2017/053114 held/submitted by Regeneron Pharmaceuticals that cover MUC16xCD3. 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