A human CD137×PD-L1 bispecific antibody promotes anti-tumor immunity via context-dependent T cell co

Immune checkpoint inhibitors demonstrate clinical activity in many tumor types, however, only a fraction of patients benefit. Combining CD137 agonists with these inhibitors increases anti-tumor activity preclinically, but attempts to translate these observations to the clinic have been hampered by systemic toxicity. Here we describe a human CD137xPD-L1 bispecific antibody, MCLA-145, identified through functional screening of agonist- and immune checkpoint inhibitor arm combinations. MCLA-145 potently activates T cells at sub-nanomolar concentrations, even under suppressive conditions, and enhances T cell priming, differentiation and memory recall responses. In vivo, MCLA-145 anti-tumor activity is superior to immune checkpoint inhibitor comparators and linked to recruitment and intra-tumor expansion of CD8 + T cells. No graft-versus-host-disease is observed in contrast to other antibodies inhibiting the PD-1 and PD-L1 pathway. Non-human primates treated with 100 mg/kg/week of MCLA-145 show no adverse effects. The conditional activation of CD137 signaling by MCLA-145, triggered by neighboring cells expressing >5000 copies of PD-L1, may provide both safety and potency advantages.

Cancer immunotherapy has made impressive strides in the last decade. A key factor behind this success is the progress made in effectively directing cytotoxic T cell responses against tumors via adoptive T cell therapy (with engineered T cells or expanded tumor-infiltrating lymphocytes) or immune checkpoint inhibitor (ICI) therapies1. Cytotoxic CD8+ T lymphocytes (CTLs) provide long-term protection against pathogens2 and carry out immune surveillance to eliminate tumor cells3. Tumor-specific CTLs are generated by the presentation of tumor-associated antigens (TAAs), derived from mutations and epigenetic changes on antigen-presenting cells (APC), driving tumor-specific T cell expansion, effector cell differentiation, and memory cell generation4. The extent and nature of CD8+ T cell activation is strongly influenced by the quality of T cell receptor (TCR)/peptide MHC interaction (signal 1) and the presence, type and strength of signaling through costimulatory receptors (signal 2); lack of appropriate costimulation has been shown to impact effective anti-tumor immune responses5.

Tumors exploit various mechanisms to evade immune destruction. One key barrier to effective anti-tumor immunity is immune checkpoints as exemplified by the PD-1/PD-L1 axis6. PD-1 is an inhibitory receptor expressed on activated T cells as well as other immune cells, and its expression is associated with T cell activation and subsequent exhaustion7. PD-1 binds to two ligands, PD-L1 (B7-H1) and PD-L2 (B7-DC), that are primarily expressed on dendritic cells (DCs) and myeloid lineage cells but can also be expressed on tumors, particularly in the case of PD-L18. Furthermore, PD-L1 expression can be increased in the tumor microenvironment by IFNγ, a cytokine released by activated T cells9. Such PD-1/PD-L1 interactions deliver a negative signal to T cells dampening anti-tumor responses. This immune suppression can be blocked by ICIs such as antagonistic antibodies against PD-1 or PD-L18. ICI treatment has demonstrated remarkably durable responses in a subset of cancer patients that are correlated with activated CD8+ T cell infiltration and proliferation10,11Combinations of ICIs (e.g., anti-PD-1 and anti-CTLA-4) have been shown to further enhance efficacy, however at the cost of toxicity, as the majority of patients experience grade 3 or 4 treatment-related adverse events12. Strategies that add to the benefit of ICI therapies without further increasing toxicity are needed.

CD137 (4-1BB, tumor necrosis factor receptor superfamily 9) is an inducible costimulatory receptor expressed on activated T and natural killer (NK) cells. CD137 signaling is triggered by receptor clustering through interaction with its trimeric ligand (tumor necrosis factor superfamily 9), expressed on professional APCs, activating the NF-κB and MAPKs pathways13. On T cells, CD137 is transiently expressed after TCR engagement and once activated provides CD28-independent costimulation resulting in enhanced cytokine production, proliferation, survival, effector function, and immunological memory formation14. In the tumor microenvironment, CD137 is a marker for tumor-specific CTLs and is co-expressed with PD-115,16. A body of orthogonal evidence suggests that the activity of tumor-specific CTLs can be enhanced by CD137 activation. Ectopic expression of CD137 ligand or an agonist antibody fragment on the surface of tumors has been shown to result in tumor elimination17,18. CD137 agonist antibodies drive CD137 clustering to induce signaling, analogous to the trimeric CD137 ligand, and demonstrate potent anti-tumor responses in preclinical models19,20,21,22,23,24. CD137 activation has been shown to upregulate PD-1 expression on effector T cells and PD-L1 on tumors via release of IFNγ and is linked to therapeutic resistance22,23. The combination of PD-1/PD-L1 pathway blocking antibodies with CD137 agonist antibodies in these models synergistically enhances anti-tumor responses19,20,21,22,23. Given the preclinical data, along with the high frequency of PD-1 and CD137 co-expression on tumor-specific CD8+ tumor-infiltrating lymphocytes (TILs) found in humans, there is a clear mechanistic rationale for dual targeting of the PD-1/PD-L1 axis and CD137 to optimally engage anti-tumor immunity.

The two most advanced therapeutic CD137 agonist antibodies in clinical testing are urelumab (IgG4) and utomilumab (IgG2). Utomilumab, in contrast to urelumab, requires Fc-mediated crosslinking to activate CD137 and shows weaker agonist activity25,26. Development of urelumab has been halted as a consequence of dose-dependent hepatitis resulting from systemic activation of the CD137 pathway in patients27. Safe administration of urelumab required a reduced dose of 0.1 mg/kg, which was chosen for combination studies with PD-1 inhibitors27. Utomilumab is better tolerated by patients but has modest anti-tumor activity as a monotherapy28 and no clear synergy with PD-1 blockade in combination therapy29. To address the shortcomings of current therapeutic strategies targeting CD137, we undertook an unbiased functional screening approach that identified a full-length bispecific antibody (bAb) that potently activates CD137 via engagement of PD-L1 to enhance tumor-specific T cell responses.

Identification of MCLA-145 using unbiased functional screening

To produce dual T cell agonist and checkpoint blocking bispecific IgG antibodies, diverse common light chain (cLC) Fab panels against CD137, PD-1 and PD-L1 were generated. The Fab panels were reformatted into human immunoglobulin G1 (IgG1) and extensively characterized (Supplementary Table 1). A total of ~440 bAbs were produced using 47 different CD137 Fab arms (representing 11 epitope bins) combined with high affinity and functionally active PD-L1 [8×] or PD-1 [3×] Fab arms (Supplementary Table 1). The bAbs were expressed as IgG1 molecules with Fc CH3 modifications to force heavy chain heterodimerization30 and CH2 modifications to abrogate binding to Fc receptors31. Initial screening assays showed the CD137xPD-L1 and CD137xPD-1 bAbs weakly induced luciferase expression in anti-CD3-activated Jurkat NFκB−luc CD137 reporter cells (Fig. 1a), despite the fact that the CD137 reporter cells upon anti-CD3 treatment co-expressed CD137 with PD-1 and PD-L1 (Supplementary Fig. 1a). The addition of an anti-human IgG to crosslink CD137 antibodies did not meaningfully increase activation of CD137 reporter cells (Fig. 1a). Our data demonstrate that dual engagement of CD137 and PD-1 or PD-L1 by bAbs either on the same T cell (in cis) or possibly on adjacent T cells (in trans) was not able to effectively induce CD137 receptor downstream signaling.

a Left panel activity of CD137xPD-L1 and CD137xPD-1 bispecifics in a CD137 NFκB-luc reporter assay (fold induction of luciferase signal) Right panel experiment repeated with the addition of an Fc binding antibody; b Left panel levels of IL-2 released by activated T cells after 3 days of incubation with eight candidate CD137xPD-L1 bAb combinations in a dose titration in the presence of CHO-PD-L1 cells, Right panel, experiment repeated in the presence wildtype CHO cells; c levels of IL-2, IFNγ and TNFα released by activated T cells after 3 days of incubation with the three best performing CD137xPD-L1 bAbs and control antibodies in a dose titration in the presence of CHO-PD-L1 cells (bAb001 = MCLA-145).

Full size image

When CHO cells expressing human PD-L1 were co-cultured with CD137 reporter cells under the same assay conditions, we observed several CD137xPD-L1 bAbs able to induce activation of CD137 reporter cells. We then performed serial titrations of a panel of CD137xPD-L1 bAbs in a human primary T cell assay containing WT or PD-L1 expressing CHO cells (Fig. 1b). Urelumab* potently activated T cells as measured by IL-2 secretion with or without PD-L1 expression on the co-cultured CHO cells. The CD137xPD-L1 bAb panel demonstrated a range of potency, some comparable to urelumab*, but only in the presence of PD-L1 antigen on the co-cultured CHO cells. The T cell agonist activity of the CD137xPD-L1 bAb panel was confirmed in a transactivation assay using purified T cells and staphylococcal enterotoxin B (SEB) activated PBMCs (Supplementary Fig. 1b). The most potent CD137xPD-L1 bAbs shown in (Fig. 1c) led to greater cytokine release than urelumab* or the combination of urelumab* and atezolizumab*. In the SEB assay, the polyclonal stimulation of T cells is also associated with the induction of high levels of PD-L1 expression on a subset of CD14+ cells and CD137 induction on CD8+ T cells (Supplementary Fig. 1c) demonstrating that effective costimulation can be achieved under inflammatory conditions. MCLA-145 was selected from the CD137xPD-L1 bAbs panel based on its potent function and optimal biophysical characteristics to facilitate IgG manufacturing.

MCLA-145 exhibits high affinity for its targets and binds a distinct epitope on CD137

MCLA-145 binds with high affinity to CHO cells expressing either human CD137 or PD-L1 and to activated human T cells as determined by flow cytometry (Supplementary Fig. 2a). The affinity of MCLA-145 for its targets was assessed by surface plasmon resonance (SPR) with recombinant proteins and confirmed in a cell-based system where the targets are in their native conformation (Fig. 2a, b). By SPR, MCLA-145 was shown to bind PD-L1 (0.51 ± 0.05 nM) and CD137 (1.9 ± 0.2 nM) with high affinity (Supplementary Table 2). MCLA-145 was capable of binding to both targets simultaneously on an SPR sensor (Fig. 2c). The cell-based affinity of MCLA-145 for human PD-L1 and human CD137 in native conformation was confirmed by Scatchard analysis (0.49 ± 0.03 nM and 2.02 ± 0.02 nM respectively) using radiolabeled antibody (Supplementary Table 3). The affinity and binding kinetics of MCLA-145 are in the same range as comparator antibodies atezolizumab* and avelumab* for PD-L1 (Fig. 2a) and urelumab* and utomilumab* for CD137 (Fig. 2b).

a Summary of SPR and cell-based affinity measurements for binding to PD-L1; b Summary of SPR and cell-based affinity measurements for binding to CD137; c SPR affinity analysis showing dual binding of human PD-L1 and CD137 to immobilized MCLA-145. Green curves: human PD-L1 injected first followed by injection of human CD137 or running buffer. Red curves: human CD137 injected followed by human PD-L1 or running buffer; d surface homology model of the CD137 ectodomain in complex with CD137L (PDB: 6MGP), the CRD domains 1–4 are shaded in yellow, green, purple and cyan respectively, CD137L is depicted as a gray ribbon, the backbone and residues of urelumab* (dark red) and utomilumab* (orange) epitopes are highlighted, critical binding residues for MCLA-145 as determined by alanine scanning are shown as solid blue spheres; e binding of labeled CD137L to CHO-CD137+ cells was measured by flow analysis in a titration series of MCLA-145 and control antibodies f NFAT activity of PD-1 reporter cells cocultured with CHO-PD-L1+ and incubated with a titration of MCLA-145 and control antibodies.

Full size image

The binding location of CD137 mAbs strongly influences their functional properties26. The epitope of MCLA-145 on CD137 was first explored using a domain swap approach, exploiting the observation that MCLA-145 binds with low affinity to murine CD137. MCLA-145 was only capable of binding with high affinity to HEK293T cells transiently expressing constructs that retained the human cysteine-rich domain (CRD) 2 (Supplementary Fig. 2b). Using a separate mapping approach, critical interaction sites of MCLA-145 on CD137 were mapped through an alanine scanning experiment to residues R66, G70, V71, and F72 located in CRD2 (Fig. 2d, Supplementary Fig. 2c). Alanine replacement at C133 reduced MCLA-145 binding but this effect was considered due conformational alterations rather than directly interfering with MCLA-145 binding. Collectively the mapping data show that the MCLA-145 epitope on CD137 resides in CRD2, a region critical for CD137L interaction32 but distinct from the binding epitopes of utomilumab and urelumab (Fig. 2d). The CD137 epitope mapping is consistent with the functional characteristics of the CD137 binding arm of MCLA-145 which is not agonistic without crosslinking and is able to block CD137L binding to CD137 (Fig. 2e). It is noteworthy that these characteristics were shared by the most potent CD137xPD-L1 bAbs in the panel (indicated in bold in Supplementary Table 1) suggesting this binding region to be optimal for trans-activation of CD137 by bAbs, at least in the IgG1 format.

The PD-L1 binding arm of MCLA-145 plays a dual role; enabling context-dependent activation of CD137 while also blocking interaction with PD-1 on T-cells. The monovalent anti-PD-L1 activity of MCLA-145 was evaluated in a PD-1 reporter assay and compared to the bivalent monoclonal antibody atezolizumab* (Fig. 2f). Although we measured the monovalent affinity of MCLA-145 for PD-L1 to be equivalent to atezolizumab* (Fig. 2a) the ED50 of atezolizumab* was significantly lower than MCLA-145 in the reporter assay. The maximal PD-1 inhibition reached by MCLA-145 was similar to atezolizumab* and at concentrations comparable to those maximally activating CD137 on T cells in vitro. Collectively, these data suggest that MCLA-145 could intensify antigen-specific T cell responses in the tumor micro-environment by simultaneously blocking the inhibition of TCR signaling by PD-1 in addition to providing costimulation through CD137 signaling.

MCLA-145 potently activates T cells, promotes immunological memory, and overcomes immunosuppression

Antigen naïve T cells that encounter their cognate antigen will transiently express CD13714. We assessed the effects mediated by MCLA-145 on the priming of human naïve CD8+ T cells in an antigen-specific priming assay with peptide-pulsed human DCs expressing PD-L1. MCLA-145 significantly expanded the absolute number of antigen-specific CD8+ T cells (MCLA-145 compared to Neg. Ctrl Ab, p < 0.01, as determined by one-way ANOVA and Tukey’s multiple comparisons test) and preferentially enhanced the proportion of antigen-specific T cells with an effector and memory phenotype (Fig. 3a and Supplementary Fig. 3a). The absolute number of antigen-specific T cells was greater after MCLA-145 treatment compared to urelumab* while the proportion of T cell populations was more donor than treatment related. These results are consistent with studies in mouse models that show the engagement of CD137 during T cell priming is a critical determinant of secondary immune responses and clearly linked to the expansion of the antigen-specific memory CD8+ T cell pool33. The effect on T cell expansion or the proportion of differentiated T cells was unaltered when atezolizumab* was combined with urelumab* nor did atezolizumab* alone have activity (Supplementary Fig. 3b). Experiments with PD-1 and PD-L1 blocking antibodies confirmed that T cell expansion and differentiation was independent of the PD-1 signaling axis (Supplementary Fig. 3c). Taken together these observations indicate the potent enhancement of T cell priming by MCLA-145 is mediated by CD137 activation and not PD-1 axis blockade.

a Effect of MCLA-145 on naïve CD8+ T cell priming. Graphs show the absolute number of dextramer+ antigen-specific CD8+ T cells from four donors after 10 days of co-culture with colored segments indicating the number of T naïve/memory stem cells (CD45RA + CCR7+), central memory T cells (CD45RA-CCR7+), effector memory T cells (CD45RA-CCR7-) and terminally differentiated effector T cells (CD45RA + CCR7−) as determined by flow cytometry (Dotted line represents control co-cultures without peptide priming). b Transactivation assay. EC50 of human T cell cytokine production (IL-2, IFNγ, and TNFα) in the presence of CHO-PD-L1 cells (n = 7); c mixed lymphocyte reaction (MLR) measuring IFNγ production by human donor leukocytes (n = 3 independent donors) stimulated with allogeneic DCs (n = 1) in the presence of 10 µg/ml of test compounds, (error bars are SEM), IFNγ concentrations were significantly greater for MCLA-145 compared to urelumab*, atezolizumab* and Neg. Ctrl Ab (p < 0.001) and significantly less compared to ure* + atezo* (p < 0001) as determined by two-way ANOVA and Holm-Sidak’s multiple comparisons test; d SEB assays. Fold change in IL-2 or IFNγ production by PBMC (left panel, n = 3 independent donors, error bars are mean ± SD) or Whole blood (right panel, n = 7 independent donors, error bars are SEM) respectively after stimulation with SEB in the presence of increasing concentrations of antibodies; e Recall assay measuring fold induction of IFNγ and TNFα production by human CD4+ memory T cells re-stimulated with a viral peptide pool (CEFT) in the presence of 10 µg/ml antibodies, (n = 4 independent donors, error bars are SEM), *p < 0.0001 for IFNγ and p = 0.0012 for TNFα production (Neg. Ctrl Ab vs. MCLA-145) and p = 0.0026 for IFNγ and p = 0.0025 for TNFα production (MCLA-145 vs. ure* + atezo*) as determined by two-way ANOVA and Tukey’s test; f Treg assay measuring fold induction of TNFα and IFNγ production by activated leukocytes in the presence of autologous regulatory T cells (1:1 ratio) and in the presence of 10 µg/ml antibodies (n = 4 independent donors error bars are SD of the mean). *p = 0.0005 for IFNγ and p = 0.012 for TNFα production (Vehicle vs MCLA-145), as determined by one-way ANOVA and Tukey’s test; g PBMC:M2 suppression assay measuring IFNγ production by human leukocytes co-cultured with activated M2-polarized macrophages together with 10 µg/mL antibodies (n = 5 independent donors, error bars are SEM, *p < 0.0001 by one-way ANOVA and Tukey’s test).

Full size image

To assess the potency of MCLA-145 on pre-activated T cells, we used a battery of orthogonal in vitro assays. In a transactivation assay using purified T cells collected from 7 different donors, EC50 values for MCLA-145 ranged from 10 to 117 ng/mL and for urelumab* from 22 to 1752 ng/mL for the various cytokines (Fig. 3b). MCLA-145 also demonstrated potent activity in a mixed lymphocyte reaction (MLR) using allogenic human DCs (Fig. 3c) and a SEB assay using PBMCs or whole blood (Fig. 3d). In contrast to CD8+ T cells, there is no consensus on the role CD137 signaling has on CD4+ T cells34. MCLA-145 was incubated with human CD4+ memory T cells re-stimulated with their cognate antigen to investigate its activity in this population. MCLA-145 treatment enhanced IFNγ and TNFα release, whereas urelumab* had no effect. Atezolizumab* treatment induced an increase in IFNγ and TNFα levels that could not be increased further by the addition of urelumab* (Fig. 3e). This suggests the effect of MCLA-145 in the CD4+ T cell recall response was mainly due to inhibition of the PD-1 pathway.

Tumors in patients that do not respond to checkpoint blockade are often infiltrated by large numbers of suppressive immune cells, in particular T regulatory cells (Treg) and tumor-associated macrophages (TAM) that may dampen immune responses35,36. To test whether MCLA-145 could overcome the suppressive effect of these cell types, T cell activation assays were conducted in the presence of Tregs or suppressive M2 macrophages (Fig. 3f, g). The addition of Tregs was able to suppress T cell-dependent IFNγ and TNFα cytokine secretion by 79% and 69%, respectively. The addition of MCLA-145 was able to fully overcome the inhibition in cytokine secretion (Fig. 3f). In contrast, the combination of atezolizumab* and urelumab* partly overcame the Treg-induced inhibition of cytokine secretion. Similarly, the addition of MCLA-145 to PBMC:M2 macrophage co-cultures was able to overcome the immune suppression induced by M2 macrophages, as measured by IFNγ production, while the negative control antibody or the urelumab* and atezolizumab* combination could not (Fig. 3g). Together these results point to the multifunctional role MCLA-145 can play in an immune-suppressive environment.

MCLA-145 recruits and activates T cells ex vivo and in vivo to enhance antitumor immunity

To investigate the activity of MCLA-145 under therapeutically relevant conditions, fresh tumor explants containing tumor-specific effector T cells and Tregs were evaluated. Five primary endometrial tumors were dissociated into a single-cell suspension and analyzed by flow cytometry for cell counts or incubated with MCLA-145 or reference antibodies in the presence of soluble anti-CD3 antibody to measure IFNγ production. The percentage of CD3+T cells, the proportion of CD4+, CD8+ and Treg subsets and the level of IFNγ production was heterogeneous between tumor samples (Fig. 4a). Treatment with MCLA-145 increased IFNγ production in all tumor samples relative to control antibody and, to a lesser extent, comparator antibodies. This effect was observed even in the context of high Treg numbers (Fig. 4a), consistent with our in vitro results (Fig. 3f). Like our observations in the SEB assay (Supplementary Fig. 1c), small numbers of PD-L1hi myeloid and/or tumor cells (Supplementary Fig. 4a) appear to be sufficient to drive MCLA-145 activation of T cells and cytokine release.

a Top panel, donut chart of relative proportion of T cell subsets including percentage of CD3+ cells in single-cell suspensions derived from five human endometrial tumors; Bottom panel, IFNγ levels in supernatant of each tumor explant 6 days after antibody incubation. Error bars: mean ± SD of three replicates, ND = not done, **p < 0.0001 in Tumor #1 and Tumor #5, *p = 0.0018 (Neg. Ctrl Ab vs. MCLA-145) and p = 0.0042 (MCLA-145 vs. atezolizumab*); Tumor #2, **p = 0.0007 (Neg. Ctrl Ab vs. MCLA-145); Tumor #4 *p = 0.0074 (MCLA-145 vs. atezolizumab*), determined by one-way ANOVA and Tukey’s test; b Ly95 T cells expressing an NY-ESO-specific TCR were transferred to NSG mice bearing A549-PD-L1hi tumors and treated with indicated antibodies, end point data are shown, (n = 7 mice); c Stacked histogram, percentage of human CD3+ lymphocytes in total live cell population in blood after RBC lysis (open bar) and tumor (solid colored bar) in each of the treatment groups in (b); d Proportion of NY-ESO-1-specific T cells in tumors per treatment group (b), expressed as a percentage of CD3 + TILs, (n = 7, error bars are SEM), *p < 0.0216 (MCLA-145 vs. Ly95), determined by one-way-ANOVA and Tukey’s test; e Ly95 T cells expressing an NY-ESO-specific TCR were transferred to NSG mice bearing A549-PD-L1hi or A549-PD-L1null tumors and treated with indicated antibodies, end point data are shown. (n = 9 mice), **p < 0.0018 MCLA-145 treatment (A549-PD-L1hi vs A549-PD-L1null groups), determined by one-way-ANOVA and Tukey’s test. f Tumor volume in individual human CD34+ engrafted NSG mice following grafting with MDA-MB-231 cells and treatment with indicated concentrations MCLA-145 or reference antibody (n c= 9 mice per group). Tumor growth and moment of death are indicated for the individual mice per group, a representative experiment of two independent experiments is shown, *p < 0.05, **p < 0.005, ***p < 0.001 of treatment group vs negative control antibody, determined by a two-sided mixed-effects model and Dunnett’s test. gProportion of CD8+ (upper panel) and CD4+ (middle panel) TILs in tumors from each treatment group, expressed as a percentage of the total population of tumor cells, bottom panel, proportion of PD-L1+ monocytes (bottom panel) in tumors per treatment group, expressed as a percentage of all monocytes (CD11b+ cells), (n = 7 mice per group, error bars are mean ± SD, *p = 0.0492 for CD8+ T cells and p = 0.0014 for PD-L1+ monocytes, determined by one-way ANOVA and Tukey’s test.

Full size image

Efficacy evaluation of MCLA-145 in vivo was carried out in murine xenograft models. The binding arms of MCLA-145 do not cross react with the murine homologues of PD-L1 and CD137, so two different strategies were used to humanize the murine xenograft models. In the first model, a human T cell clone expressing an NY-ESO-specific TCR (Ly95) was adoptively transferred into immunodeficient mice bearing human A549 tumors. The NSCLC cell line A549 expresses NY-ESO antigen in the appropriate HLA context37 and was modified for the study to stably express high levels of PD-L1 (Supplementary Fig. 4b). Ly95 cells express PD-1 and CD137 and a greater percentage of double-positive Ly95 cells were observed in tumors at the end of the experimental period (Supplementary Fig. 4c). In mice engrafted with A549 PD-L1hi cells, treatment with urelumab* and atezolizumab alone or in combination along with Ly95 cells did not significantly alter tumor growth compared to control (Fig. 4b). All mice treated with MCLA-145 were able to control tumor growth (Fig. 4b). Importantly, MCLA-145 dependent tumor growth inhibition was clearly associated with a skewed distribution of adoptively transferred huCD3+ Ly95 cells to tumor vs blood (Fig. 4c) and increased NY-ESO antigen-specific T cells within the tumor compared to mice treated with controls (Fig. 4d).

To further explore the context-dependent nature of MCLA-145 activity, a second A549 line was prepared in which PD-L1 was knocked out (Supplementary Fig. 4b). MCLA-145 was unable to activate T cells when co-cultured with the A549 PD-L1null line in contrast to A549 PD-L1hi cells (Supplementary Fig. 4d). Similarly, MCLA-145 treatment in vivo had no effect on A549 PD-L1null tumor size compared to control and in contrast to A549 PD-L1hi tumors (Fig. 4e).

To evaluate the impact of MCLA-145 treatment in the context of a more heterogenous T cell population, we engrafted NSG mice with human CD34+ hematopoietic stem cells and then implanted the PD-L1 expressing human MDA-MB-231 breast cancer line (Fig. 4f). In animals receiving control IgG, tumors grew progressively over the first 40 days after which more rapid growth kinetics were observed in most animals (Fig. 4f). The group treated with CD137 agonist, urelumab* showed a very similar tumor growth pattern to control. In the groups treated with pembrolizumab, atezolizumab or the combination of atezolizumab and urelumab* there was evidence of tumor growth control in some mice but, by the end of the observation period, tumors in all surviving mice were growing rapidly (Fig. 4f). In contrast, all animals treated with MCLA-145 experienced tumor control including two complete responses and showed significantly slower growth kinetics at the end of the treatment period (Fig. 4f). A cohort of mice treated with a 10-fold lower dose of MCLA-145 (0.5 mg/kg) also showed similar inhibition of tumor growth (Fig. 4f). Consistent with the NY-ESO model, analysis of TILs in this model showed that MCLA-145 treatment resulted in an increased frequency of CD8+ T cells in the tumor (Fig. 4g). Strikingly, and consistent with the NY-ESO model, urelumab* treatment alone significantly reduced the number of CD8+ cells in the tumor compared to MCLA-145 and this was associated with an increase in the frequency of CD4+ T cells and PD-L1+ monocytes (Fig. 4g). These results demonstrate that the potent T cell agonism observed with MCLA-145 in vitro also translated into PD-L1 dependent tumor control in vivo.

We also noted that some mice in the pembrolizumab, atezolizumab or urelumab* groups and all mice in the combination treatment group exhibited ruffled fur and skin and a hunched posture, in some cases associated with excessive weight loss (Supplementary Fig. 4e), that required euthanasia (3–5 mice in each group). No animals in the control or MCLA-145 groups showed signs of stress (although one animal in the MCLA-145 group was euthanized for unrelated reasons). In particular, we observed that the combination treatment of urelumab* and atezolizumab was significantly more toxic than the control or MCLA-145 treatment (Supplementary Fig. 4f). One possible explanation for the observed toxicity is an acceleration of graft vs host disease (GvHD) which has been reported for ICI agents38. If true, the apparent lack of GvHD in MCLA-145 treated mice may be attributed to its tumor-restricted T cell agonism.

MCLA-145 is well tolerated in non-human primates at high doses

To further investigate safety, MCLA-145 was evaluated in non-human primates. MCLA-145 is fully cross-reactive with the cynomolgus homologues of PD-L1 and CD137 (Supplementary Fig. 5a, b) and showed similar functionality in a cynomolgus whole blood SEB assay (Supplementary Fig. 5c). In a repeat dose study, male and female cynomolgus monkeys received five weekly doses of MCLA-145 with a subset of animals being observed for a further 31 (males) or 32 (females) days (Fig. 5a). All dosed animals were shown to be exposed to MCLA-145. Pharmacokinetic analysis of MCLA-145 showed dose-proportional systemic exposure with a PK profile and half-life consistent with standard monoclonal IgG antibodies (Fig. 5b). Exposure increased with escalating dose in a manner that was dose-proportional on Day 1, and greater than proportional on Day 22, reflecting the loss of exposure in the lower dose levels after repeat dosing (Fig. 5c). In the 100 mg/kg/week dose group 70% of animals showed continuous exposure throughout the treatment period. The systemic exposure data are consistent with the expected anti-drug antibodies (ADA) titers observed in animals receiving lower doses. MCLA-145 was shown to be well tolerated at all doses up to the maximal dose of 100 mg/kg. No change in organ weight or any gross or microscopic pathological findings were noted. Blood chemistry was within normal ranges at all doses and time points including the liver enzymes AST and ALT (Fig. 5d). There was no effect on hematological parameters including neutrophil and platelets numbers (Fig. 5e). An integrated review of safety data from clinical trials of the agonist mAb urelumab noted elevated transaminases, neutropenia, thrombocytopenia as the most common serious adverse events along with fatigue27. None of these symptoms were observed even at the highest dose tested which is 1000 times higher than the maximum tolerated dose (MTD) of urelumab in man (0.1 mg/kg). Liver toxicity has been associated with the appearance of immune infiltrates in the liver39,40. Pathological examination of the liver and other tissues did not show any evidence of increased immune infiltration. Minor fluctuations in the lymphocyte numbers and T cell subpopulations were noted in both the vehicle and 100 mg/kg/week dose group (Fig. 5f). Collectively these data demonstrate that MCLA-145 is capable of potent T cell agonism resulting in tumor growth reduction in vivo but does not appear to exhibit the safety liabilities that have been observed with other potent CD137 agonists.

a Schematic showing the design for the repeat dose safety study; b mean serum concentrations of MCLA-145 measured over one cycle (n = 10 animals for vehicle and 100 mg/kg, n = 6 animals for other treatment groups); c mean maximum serum concentration (Cmax) and area under the curve (AUC) at 10 mg/kg (blue), 30 mg/kg (green) and 100 mg/kg (orange) of MCLA-145 measured 24 h after the first dose and 24 h after the fifth dose combined for males and females; n = 10 animals for vehicle and 100 mg/kg, n = 6 animals for other treatment groups, error bars are mean ± SD); d Mean serum concentrations of AST and ALT and e mean number of circulating neutrophils and platelets measured over the study period at the indicated doses (n = 10 animals for vehicle and 100 mg/kg, n = 6 animals for other treatment groups, error bars are SD). Reference mean (male and female bold blue and pink lines respectively) and mean +1 or −1 SD (male and female bold blue and pink lines respectively) as previously reported;69 f Number of circulating B (CD20+), NK (CD16+) and T cells (CD3+) with CD4+ and CD8+ T cell numbers and proportion of different subpopulations shown for the 100 mg/kg/week dose group or vehicle, mean ± SD of male and female groups combined (n = 10 animals).

Full size image

MCLA-145 transactivates CD137 in a PD-L1 dependent manner via receptor clustering

Under no experimental setting could we show that MCLA-145 was able to activate CD137 without the presence of PD-L1 expressed on an accessory cell. To investigate the effect of PD-L1 expression level on MCLA-145 agonistic activity, experiments were conducted with the reporter assay Jurkat cells co-cultured with CHO cell lines overexpressing human PD-L1 at different protein levels (Fig. 6a) or human tumor cell lines with different levels of endogenous PD-L1 expression (Fig. 6b). In both settings, there was a direct relationship between PD-L1 expression on accessory cells and MCLA-145 mediated CD137 receptor activation. Based on this set of experiments, we estimate a threshold of >5000 molecules of PD-L1 per cell is required to drive CD137 receptor clustering and activation.

Fold induction of CD3-stimulated Jurkat NFκB-luc-CD137 reporter activity after 24 hours of co-incubation with either a CHO-PD-L1 cells or b human tumor cells endogenously expressing PD-L1 at increasing concentrations of MCLA-145; numbers in brackets indicate PD-L1 binding sites per cell as determined by QIFIKIT analysis; c Schematic illustrating the VeraTag assay to measure CD137 clustering; d CD137-expressing activated Jurkat T cells and PD-L1-expressing CHO cells cocultured with 10 µg/mL test antibody, fixed and measured for CD137 proximity using two different CD137 detection antibodies; e Schematic illustrating the VeraTag assay to measure CD137 and PD-L1 proximity; f Cells cultured and prepared as in (d) to measure PD-L1 and CD137 in proximity, RPA relative peak area; confocal images of CD137 receptor internalization in cocultures of activated CD8+ T cells and CHO-PD-L1+ incubated with (g) MCLA-145, (h) urelumab* or (i) negative control antibody for 15 min, nuclear (DAPI) staining (blue), CD8 (green), CD137 (magenta). Single plane view, magnification ×20 (insets ×63). Data are representative of two repeat experiments for Fig. 6g–i.

Full size image

The natural ligand for CD137 is a trimer and engages three copies of CD137 at the cell surface resulting in the clustering of CD137 to trigger downstream signaling. Agonistic CD137 mAbs are similarly capable of activating CD137 via clustering of receptor dimers resulting in rapid internalization and continued signaling of the receptor complex41. We performed receptor proximity assays (RPA) to address experimentally whether MCLA-145 induces CD137 clustering. RPA is capable of quantifying protein expression and dimerization in formalin-fixed, paraffin-embedded (FFPE) tissues or cells42. Expression of cell surface targets can be detected directly using this method and the signal for both CD137 and PD-L1 was equivalent in the presence of MCLA-145 and controls (Supplementary Fig. 6a–c). In the co-culture assay system, CD137 expression was specifically detected on activated Jurkat cells, and detection of PD-L1 was restricted to CHO-PD-L1 cells (Supplementary Fig. 6d–e). We first probed the cocultures of T cells and CHO-PD-L1 cells for the presence of CD137 clusters (schematically illustrated in Fig. 6c). MCLA-145 was shown to enhance clustering of CD137 (Fig. 6d). The activity of MCLA-145 was comparable to the agonist antibody urelumab* that is active without crosslinking or accessory cells (Fig. 1a). We next focused on the measurement of PD-L1 and CD137 in proximity (experimental set up schematically illustrated in Fig. 6e). Based on published structures, the distance between the CD137 epitope of MCLA-145 and the cell membrane is estimated to be 8–9 nm and for the PD-L1 epitope 7–8 nm; the opposing Fab arms of an IgG1 can reach 12 nm apart. Thus MCLA-145 could be expected to bring the two surface molecules within range of the proximity assay detection (30–100 nm)42. Indeed PD-L1 and CD137 proximity could only be measured in co-cultures of Jurkat and CHO-PD-L1 cells treated with MCLA-145; addition of urelumab* and negative control antibody induced no signal above background (Fig. 6f). These data indicate that binding of MCLA-145 translocates CD137 and PD-L1 towards a contact zone between neighboring cells resulting in CD137 clustering.

To investigate whether MCLA-145 triggers CD137 clustering and internalization in the presence of PD-L1 expressing cells, we performed confocal imaging of co-cultures of activated human primary CD8+ T cells with CHO-PD-L1 or CHO-WT cells (Fig. 6g-i). In co-cultures with CHO-PD-L1 cells, an intense punctate staining pattern of CD137, indicative of receptor clustering and internalization, was visible on CD8+ T cells 15 min after treatment with MCLA-145 (Fig. 6g) or urelumab* (Fig. 6h) but not a negative control antibody (Fig. 6i). In co-cultures with CHO-WT cells, treatment with MCLA-145 failed to induce CD137 clustering or internalization (Supplementary Fig. 6f). A 3D rendering of confocal images shows MCLA-145 activated CD8+ T cells with punctate CD137 staining at the interface of a monolayer of CHO-PD-L1+ cells (Supplementary Fig. 6g). Clustering and internalization results in brighter CD137 staining, which is significantly enhanced in MCLA-145 treated co-cultures compared with urelumab* (Supplementary Fig. 6h), consistent with our in vitro functional data. Collectively these data show direct evidence of MCLA-145 induced CD137 clustering resulting in receptor internalization that is dependent on adjacent cells expressing PD-L1 above a threshold of >5000 molecules per cell.

ICI therapies are now standard of care in many cancer indications and result in remarkably durable responses. However, most patients receiving ICI therapies do not respond to treatment and many whom do respond eventually relapse43. New approaches to enhance anti-tumor immunity are urgently needed. Pharmaceutical intervention to activate co-stimulation pathways has long been hypothesized as a logical companion to ICI therapies analogous to “hitting the gas and releasing the brakes”44.

A number of T cell costimulatory molecules including CD137, OX40, CD40, ICOS, GITR, and CD27 are being targeted pharmacologically with agonists to improve outcomes in cancer patients24,45. They are transiently expressed upon ligation of the TCR with cognate antigen (e.g., tumor neo-antigens). Clustering of these receptors results in downstream signaling cascades that enhance T cell activity. Targeting CD137 to improve therapeutic response is clinically validated based on its application in CAR-T and TIL therapies46,47. Activation of the CD137 pathway appears to favor extending persistence and activity of pre-existing tumor-specific CD8+ T cell responses14. Relative to CD28 (also sometimes used in CAR-T constructs) CD137 co-stimulation expands memory CD8+ T cells and results in longer retention of effector T-cell functions48. Unfortunately, it has been difficult to translate the success in cell-based therapies to antibody-based therapeutic approaches targeting CD137. Strong agonists such as urelumab are hampered by major dose-limiting liver toxicity27 while other antibody approaches that depend on FcγR cross-linking (e.g., utomilumab) are weakly active preclinically and have not shown clinical activity even in combination with checkpoint inhibitors29. Here we describe an unbiased functional screen evaluating a large panel of bAbs targeting the PD-1/PD-L1 axis and the costimulatory receptor CD137. Applying this approach, MCLA-145 was identified as a potent T cell agonist with several advantageous mechanistic attributes. Our screen included antagonistic Fab arms to PD-1 and PD-L1 in combination with a panel of Fab arms binding across the ectodomain of CD137, and we evaluated T cell activity alone or in co-culture. We observed that only specific combinations of PD-L1 and CD137 binding arms in the bAb format were capable of potent T cell activation and only in the context of accessory cells expressing PD-L1. These bAbs all bind to the CRD2 of CD137, a region that is also associated with CD137L binding. Indeed, fine epitope mapping carried out with MCLA-145 identified critical binding residues that overlap with the CD137L epitope, and cross-blocking assays showed that MCLA-145 inhibits CD137L binding. We hypothesize that the CRD2 epitope bound by both MCLA-145 and CD137L may be the optimal for “trans” based engagement and activation of CD137. Furthermore, MCLA-145 engagement could drive the formation of an ‘immunological synapse’, as demonstrated for other bispecific constructs49. In the MCLA-145 driven synapse, T cells would benefit from effective TCR signaling relieved of PD-1 inhibition and potent CD137 activation (see Fig. 7 for a schematic representation). In this way, MCLA-145 may effectively and specifically promote antigen presentation and target cell lysis50,51.

MCLA-145 is shown binding to PD-L1 expressed on a tumor cell or APC (depicted in pink) and CD137 expressed on a T effector cell (depicted in green). Engagement of MCLA-145 results in clustering of CD137 molecules on the surface of the effector T cell and downstream activation of the NF-kB signaling pathway. In addition, engagement of PD-L1 by MCLA-145 blocks interaction with its receptor PD-1 reversing PD-1 dependent suppression of downstream signaling from the T cell receptor complex. The net effect of MCLA-145 binding in trans to tumor cell/APC and T effector cells is shown: upregulation of CD137 (on T cells) and PD-L1 (on tumor cells and APCs), the release of pro-inflammatory cytokines from effector T cells resulting in T cell differentiation and expansion and activation of effector T cells to promote cytotoxic attack of tumor cells.

Full size image

MCLA-145 is a fully human IgG1 based bAb with substitutions in the Fc region to prevent FcR binding and thus higher-level crosslinking via FcR bearing cells. A prevailing assumption has been that trivalent52,53 or at least bivalent engagement54,55 is required to optimally cluster and activate members of the TNFRSF. We demonstrate that monovalent engagement of CD137 with a bAb anchored in trans is capable of clustering CD137 followed by rapid internalization and potent activation of T cells. The clustering and internalization activity induced by MCLA-145 resembled urelumab* in our experiments and CD137 agonist mAbs described in other studies41. The monovalent engagement of CD137 in this context has several potential advantages. Monovalent constructs often demonstrate a greater level of cell binding compared to bivalent constructs because in the latter case the two binding arms may sterically compete at higher concentrations. This property is likely to be an advantage as T cells and PD-L1-bearing cells are brought in proximity. Even under saturating conditions MCLA-145 is not able to activate CD137 without the presence of a neighboring cell expressing PD-L1. In addition, T cell activation required signal 1 (i.e., TCR stimulation via CD3 mAb or polyclonal activation). Furthermore, we show that the monovalent binding to CD137 places strict criteria on the number/density of PD-L1 molecules required to trigger the CD137 clustering and signaling. We estimate that expression of >5000 PD-L1 molecules per cell is required to promote effective engagement and activation of CD137 by MCLA-145.

Using a battery of orthogonal functional assays, we demonstrate MCLA-145 engages CD137 and PD-L1 to enhance T cell responses relevant for anti-tumor immunity. In a T cell priming assay where PD-1/PD-L1 axis inhibition has no functional impact, MCLA-145 increased antigen-specific T cell numbers in comparison to negative control antibodies and urelumab*. In some donors we observed a clear expansion of the effector memory compartment consistent with the unique role CD137 has among the TNFRSF costimulatory receptors in promoting durable antigen-specific CD8+ T cell responses13,33,56. Conversely, in a CD4+ T cell recall assay, where urelumab* treatment was ineffective, we showed MCLA-145 to be more active than atezolizumab*. These experiments demonstrate that both targeting arms of MCLA-145 are functionally active. In more heterogenous cell-based assays, such as the SEB and MLR assay, the synergy of targeting both CD137 and PD-L1 pathways was even more apparent and we show MCLA-145 is capable of T cell activation comparable to the combination of urelumab* and atezolizumab*.

Tumors in patients unresponsive to ICI therapy are often characterized by a lack of T cell infiltration and/or a highly immunosuppressive microenvironment4,10. Several pieces of evidence in this study suggest MCLA-145 may benefit patients with suboptimal responses to ICIs. First, we demonstrate in vitro that enhancement of T cell responses by MCLA-145, but not urelumab* + atezolizumab*, was evident in co-cultures of Tregs or M2 macrophages reversing their suppressive activity on T cell cytokine release. Secondly, in fresh tumor explants MCLA-145 increased T cell cytokine release in the presence of a significant proportion of Treg cells and this effect was greater than atezolizumab* in the majority of donors. We hypothesize that T cell hyperactivation induced by MCLA-145 in co-cultures containing PD-L1 bearing cells counteract the suppressive signals that are released by Tregs and TAMs. We note that CD137 expression is upregulated on Tregs57 and that CD137 agonism can modulate Treg activity58 thus MCLA-145 could be acting directly on the Treg population in addition to increasing effector T cell activity. Finally, in two different humanized xenograft models treated with MCLA-145, we demonstrated an expansion of CD8+ T cell numbers in tumors that were associated with greater tumor growth control than negative control or comparator antibodies. In the A549 model where human PD-L1 is only present on the tumor, we observed recruitment of adoptively transferred NY-ESO specific T cells from the periphery to the tumor and this was confirmed in the MDA-MB-231 model where PD-L1 expressing myeloid cells are also present. Unexpectedly, treatment with urelumab* alone in both models resulted in fewer tumor CD8+ T cells while increasing the number of CD4+ T cells and myeloid cells. While both models suffer the limitation of not providing a complete ‘human’ tumor microenvironment (e.g., with Tregs and TAMs), the differential effect of MCLA-145 on tumor growth and immune cell infiltration may point to a potential benefit of localized activation of CD137.

Systemic activation of CD137 by agonistic antibodies results in toxicity in animal models and humans27,44. We observed that immunocompromised mice engrafted with CD34+ human cord blood exhibited signs of accelerated GvHD when treated with urelumab*, atezolizumab or pembrolizumab. In some cases, the GvHD was severe enough to require euthanasia on the study. In contrast, MCLA-145 treated mice remained healthy throughout the treatment period, resembling negative control-treated animals. This suggests the context-dependent nature of MCLA-145 T cell activation did not extend to graft reactive T cells. In a GLP repeat dose non-human primate (NHP) study, MCLA-145 was well tolerated at suprapharmacological doses with no evidence of elevated liver enzymes, neutropenia or thrombocytopenia, these being the most common serious adverse events reported in clinical studies with urelumab27. Based on an integrated safety analysis, the MTD of urelumab was determined to be 0.1 mg/kg, a dose three orders of magnitude lower than the highest MCLA-145 dose tested in NHPs. To our knowledge, this is the first report of a fully human and cynomolgus cross-reactive CD137 agonist tested in NHPs at this dose level. While NHP studies have failed to pick up safety issues with immune agonists previously59, MCLA-145 was evaluated as safe at supraphysiological doses. Taken together with the rodent experiments our data suggest MCLA-145 will have a favorable safety profile in humans.

MCLA-145 is a targeted drug capable of both T cell costimulation and checkpoint blockade which may have broad clinical utility. The preclinical data presented suggest MCLA-145 has the potential to be active as a monotherapy in settings where ICI treatments fail to improve patient outcomes. MCLA-145 could also be utilized as an adjuvant to boost responses to cancer vaccines and other treatment modalities (e.g., chemotherapy) that promote neoantigen presentation in tumors or draining lymph nodes. Professional APCs express the highest levels of PD-L160,61. Based on our data, MCLA-145 could be expected to enhance naïve T cell priming and promote long-term T cell immunity. Importantly, MCLA-145 is unlikely to have the same safety liabilities that have limited the clinical development of CD137 agonists. MCLA-145 is currently being evaluated in an open label, dose-escalation single-agent study with expansion cohorts for dose confirmation/safety and preliminary efficacy in advanced or metastatic malignancies (NCT03922204).