CD137 (TNFRSF9, 4-1BB) agonist antibodies (mAb) have demonstrated potent antitumor activity with memory response while causing hepatotoxicity in mouse models. In clinical trials, the degrees of liver toxicity of anti-CD137 vary from grade 4 transaminitis (urelumab) to nonexistent (utomilumab). To exploit the antitumor potential of CD137 signaling, we identified a new class of CD137 agonist mAbs with strong antitumor potency without significant transaminitis in vivo compared with CD137 agonists previously reported. These mAbs are cross-reactive to mouse and cynomolgus monkey and showed cross-linking–dependent T-cell costimulation activity in vitro. Antitumor efficacy was maintained in Fc gamma receptor (FcγR) III–deficient mice but diminished in FcγRIIB-deficient mice, suggesting the critical role for FcγRIIB to provide cross-linking in vivo. Interestingly, a single dose of an affinity-reduced variant was sufficient to control tumor growth, but a higher affinity variant did not improve efficacy. These observations suggest that binding epitope and FcγR interaction, but not necessarily high affinity, are important for antitumor efficacy and reduced liver toxicity of CD137 mAb. Our study suggests the possibility of CD137 agonist therapy with improved safety profile in humans.
CD137 (TNFRSF9, 4-1BB) is a type I transmembrane receptor, a member of the tumor necrosis factor receptor superfamily. It is a T-cell costimulatory molecule which is highly expressed on activated T cells and natural killer (NK) cells, while constitutively expressed at a lower level on monocytes, neutrophils, and dendritic cells (1, 2). CD137 is also expressed on lung and liver endothelial cells (3), tumor capillaries (4), and activated articular chondrocytes (5). In comparison, CD137 ligand (CD137L, TNFSF9) is expressed on activated B cells, dendritic cells, and macrophages (6). Binding of CD137 with its ligand or with an agonistic monoclonal antibody (mAb) induces clustering of the receptors and recruitment of specific TNF receptor–associated factors (TRAF1, TRAF2, and TRAF3; ref. 7). Association of these adaptor proteins in the cytoplasmic domain of CD137 induces activation of different T-cell signaling pathways (NF-κB, ERK, JNK, and p38 MAPK), upregulation of Bcl-xL, and inhibition of proapoptotic protein BIM (8–10). CD137 signaling results in T-cell proliferation, cytokine release, enhancement of CD8+ T-cell cytotoxicity, and protection from activation-induced cell death. CD137 signaling also enhances antigen-specific memory CD8+ T-cell accumulation (11, 12) and rescues nonresponsive, anergic CD8+ T cells (13, 14). In addition, it has been shown that a cytoplasmic domain of CD137 engineered into chimeric antigen receptor T cells (CAR-T) enhances the survival and induces development of a central-memory phenotype (15).
Agonistic CD137 antibodies (mAb) have demonstrated strong antitumor activity in many mouse tumor syngeneic models. For examples, 1D8 regressed tumor growth in sarcoma (16), glioma (17), and lymphoma models (18). 2A demonstrated efficacy in colon carcinoma, lymphoma (19), and myeloma models (20). 3H3 and MAB9371 showed efficacy in both CT26 and MC-38 models (21). Most of these mAbs have been used in combination studies and demonstrated synergistic effect with checkpoint inhibitors such as anti-CTLA4 (22) and anti–PD-1 (21, 23), chemotherapeutics (24, 25), radiotherapy (26), and vaccines (27). However, all these CD137 mAbs are known to increase transaminases in mice when dosed repeatedly (28, 29).
Interestingly, 2 anti-CD137 in clinical trials, urelumab (BMS-663513), a fully human IgG4 antibody, and utomilumab (PF-05082566), a fully human IgG2 antibody, have shown distinct liver toxicity profiles. The maximum tolerable dose of urelumab is low (0.1 mg/kg every 3 weeks) due to the tendency to cause transaminitis (30), which may result in a suboptimal therapeutic window. In contrast, utomilumab can be dosed up to 10 mg/kg every 4 weeks without hepatotoxicity, but it had limited clinical activity in solid tumors [3.8% objective response rate (ORR)] and Merkel cell carcinoma (13.3% ORR) as a single agent (31).
Given the limitations of the current CD137 mAbs in clinical trials, we have pursued opportunities to identify CD137 mAbs with potent antitumor efficacy and reduced liver toxicity. Under the assumption that the mechanism of liver toxicity is conserved between human and mouse, CD137 agonist mAbs with human and mouse cross-reactivity were selected based on their antitumor efficacy and reduced liver toxicity in mouse syngeneic models. In this article, we report the identification of a novel class of CD137 mAbs and characterization of their functions in vitro and in vivo.
Materials and Methods
Female, 5- to 6-week-old BALB/c and C57BL/6 mice were purchased from Charles River Laboratories. FcγRIIB knockout mice (C.129S4(B6)-Fcgr2btm1TtK/cAnNTac N12) were purchased from Taconic, and FcγRIII knockout mice (B6.129P2-Fcgr3tm1Sjv/J) were purchased from The Jackson Laboratory. Mice were housed under specific pathogen-free conditions in an AAALAC accredited facility. All mouse procedures were performed in accordance with protocols approved by the internal Institutional Animal Care and Use Committee (IACUC).
Hybridoma generation and screening
Rat monoclonal mAbs were generated according to standard hybridoma technology. Recombinant mouse and human CD137-human Fc fusion (Creative Biomart) and mouse CD137-His proteins were generated in-house and used as immunogens. Hybridoma were screened for antigen-specific binding as well as for the ability to activate NFκB by a reporter assay using HEK-293 cells transduced with pLenti-NFκB-Luciferase vector and a plasmid expressing human CD137.
Cell lines and antibodies
CT26, B16-F10, CHO-K1, and HEK-293 were purchased from the ATCC, and MC-38 cell line was obtained from the University of Chicago under a license agreement with the NIH (Rockville, MD). Cell lines overexpressing FcγRIIB and human or murine CD137 were generated by stably transfecting CHO-K1 and HEK-293, respectively, and maintained in the media containing 0.5 mg/mL G418. All cell lines were tested negative for mycoplasmas and banked after purchasing or genetic modifications. They were maintained in culture for no more than 10 passages from master cell banking.
CD137 mAbs 106, 107, and their isotype variants, murine IgG1 isotype control, and human IgG1 isotype control were produced in-house as described in PCT/US2017/034687 (106 and 107 were stated as TABBY106 and 107, respectively). Sequence of the 1D8 was from US patent#7,754,209 B2 (SEQ ID Nos.: 101 and 103). 3H3 and rat IgG1 were purchased from Bio X Cell. MAB9371 was purchased from R&D Systems. Anti-human CD137 control reagents, Mab A (20H4.9) and Mab B (MOR.7480.1), were expressed using published sequences in US patent#7,288,638 and PCT/US2012/032704, respectively. Rat 106 mAb was humanized using VH1-69 and Vκ6-57 frameworks as acceptors with rodent complementarity-determining regions (CDR) along with back mutations (106M). 106L is a low-affinity variant that contains an amino acid substitution in a CDR, whereas 106H was isolated from yeast display library and contains several amino acid substitutions in CDRs. All constructs were subcloned into in-house expression vectors and purified by protein A from the supernatant of HEK293-6E cells that had been transiently transfected with heavy and light chain vectors as previously described (32).
Syngeneic tumor models
Log-phase cells of CT26 (2.5 × 105), MC-38 (5 × 105), and B16-F10 (2.5 × 105) were implanted by subcutaneous (s.c.) injection in the right flank of BALB/c or C57BL/6 mice. Tumor-bearing mice were randomized into treatment groups based on mean tumor volume of ∼100 mm3 and were treated with CD137 mAb or isotype control (murine or rat) by intraperitoneal (i.p.) injections. Tumor volume was monitored using electronic calipers and calculated according to the formula: V = ½ × length × width × height. Mice were euthanized when tumor volumes reached a maximum of 2,500 mm3 or if animal health was compromised, per IACUC-approved protocol.
Liver immunohistochemistry (IHC)
Mouse livers were perfused with PBS prior to harvesting and processed as formalin-fixed paraffin-embedded blocks. The sections were stained using rat anti-mouse CD45 (BD), rabbit anti-CD8 (Cell Signaling Technology), and rabbit anti-Iba-1 (Wako) after the antigen retrieval process. The sections were then stained using a standard IHC processing protocol with secondary mAbs; Envision-HRP (Dako) for CD8 and Iba-1 and a rat anti-mouse-HRP for CD45. Images were captured by Aperio AT2 digital scanner (Leica) and analyzed by HALO imaging analysis system.
Human T-cell costimulation assay
A 96-well plate was coated with 0.5 μg/mL anti-CD3 (OKT3, eBioscience) in PBS. The plate was washed with assay medium (AIM-V, Life Technologies) and blocked dry before adding 50 μL of CD137 mAb at 30 μg/mL and 50 μL of medium with or without Fc cross-linker (F(ab')2 fragment of human Fcγ-specific goat polyclonal mAbs, Jackson ImmunoResearch Labs) at 120 μg/mL. CD8+ T cells isolated by negative selection (Stem Cell Technologies) from peripheral blood mononuclear cells of healthy donors were then added (105 cells/well). In some experiments, 4 × 104 CHO-K1 cells expressing FcγRs were seeded overnight as cross-linkers and 105 CD8+ T cells, CD137 mAb, and OKT3 (0.5 μg/mL) were added for a final volume of 200 μL. Plates were incubated for 3 days, and supernatants were collected for cytokine analysis. Cells were pulsed with 0.5 μCi/well methyl-[3H] thymidine for 6 hours and proliferation was assessed by thymidine incorporation (Wallac 1450 MicroBeta Liquid Scintillation Counter).
Mouse T-cell costimulation assay
A 96-well plate was coated with 1 μg/mL of anti-CD3 (145-2C11, BD) in PBS. CD8+ T cells were isolated from the splenocytes using a mouse CD8+ T-cell isolation kit (STEMCELL Technologies) from spleens of naïve BALB/c mice and resuspended in media to 5 × 105 cells/mL. CD8+ T cells (100 μL) with 1 μg/mL of anti-CD28 (37.51, BD) were added to each well. CD137 mAbs were mixed with F(ab')2 fragment of rat or mouse Fcγ-specific goat polyclonal mAbs used as cross-linkers (Jackson ImmunoResearch Labs) at a molar ratio of 1:4 and incubated at room temperature for 1 hour. The mixture was serially diluted in 100 μL and then added to the cells and incubated for 66 hours before supernatants were collected for cytokine measurement. The methyl-[3H] thymidine incorporation measurement was carried out as previously described.
Single-cell suspensions from tumors were prepared by digesting tumors with a mouse Tumor Dissociation Kit in combination with the gentle MACS Octo Dissociator system (Miltenyi Biotec). Fluorescent-conjugated mAbs used in flow cytometry analysis were purchased from BD: CD45 (30–11); CD3 (145-2C11); CD4 (cRM4-5); CD8 (53-6.7); CD11b (M1/70), CD25 (PC61); CD49b (DX5), CD62L (MEL-14); B220 (RA3-6B2), Ly6G (1A8), Ly6C (AL-21), F4/80 (T45-2342), and FoxP3 (MF23). Dead cells were excluded from analysis using the Fixable Viability Dye eFluor 780 (Thermo Fisher Scientific). Binding of mouse anti-human CD137 mAbs were detected by PE-conjugated goat F(ab')2 anti-mouse IgG (Jackson ImmunoResearch). H-2Ld MuLV gp70 Tetramer-SPSYVYHQF was purchased from MBL. Flow data were analyzed by FlowJo V10.
Epitope grouping of anti-CD137 was determined by competition assays using Biacore T200 instrument (GE Healthcare) at 12°C to slow the off-rate of mAb: antigen interactions. Recombinant human and mouse CD137 and CD137 ligand proteins (His-tags) were produced in-house (HEK-293). Chip preparation and binding kinetic measurements were made in the assay buffer HBS-EP+ (10 mmol/L HEPES, pH 7.4, 150 mmol/L NaCl, 3 mmol/L EDTA, 0.05% Tween 20). Approximately 2,000 RU of goat anti-mouse or anti-rat IgG Fc polyclonal mAb (Thermo Fisher Scientific), diluted to 25 μg/mL in 10 mmol/L sodium acetate (pH 4.5), was directly immobilized across a CM5 biosensor chip using a standard amine coupling kit according to the manufacturer's instructions. Each assay cycle consisted of the following steps: (i) capture of first test mAb at 10 mg/mL on test surface only; (ii) blocking injection of isotype control cocktail at 100 mg/mL over both reference and test surface; (iii) analyte injection; (iv) second test mAb injection at 10 mg/mL; (v) regeneration of capture surface by 10 mmol/L Glycine-HCL, pH1.5 injections.
Measurement of liver enzymes and cytokines/chemokines in serum
Blood chemistry was measured by VetScan VS2 analyzer (ABAXIS) using Mammalian Liver Profile rotors. Human or mouse cytokines/chemokines in culture or serum samples were analyzed using the MILLIPLEX MAP Human or Mouse Cytokine/Chemokine kit (Millipore) and measured by FLEXMAP 3D (Luminex).
Statistical significance between multiple groups was determined by one-way ANOVA (single variable) or two-way ANOVA (two variables). Student t test (unpaired, two-tailed, 95% confidence interval) was used to compare 2 groups with one variable. For tumor survival analysis, Kaplan–Meier test was used. Error bars indicate SEM. All graphs and statistical analysis were generated using Prism 7 software (GraphPad Software).
Novel CD137 mAbs 106 and 107 demonstrated strong antitumor activity with low alanine aminotransferase (ALT) level
Because CD137 mAbs are known to induce transaminitis in both human and mouse, we hypothesized that the biological pathway causing the liver cell damage may be conserved between human and mouse. Therefore, it is critical to identify mAbs that cross-react to mouse in order to screen in vivo for potential clinical CD137 agonists with strong antitumor efficacy and low liver toxicity using mouse syngeneic tumor models. Three selected anti-CD137 were tested in the CT26 syngeneic model for their antitumor efficacy and liver function measured by the level of liver enzyme ALT in serum (Fig. 1A and B). Even though all 3 clones were similarly effective in controlling the tumor growth, mAb 105 treatment showed increased ALT level but not mAbs 106 or 107 treatment under the condition tested (10 mg/kg, q7d × 3). Both 106 and 107 mAbs showed cross-reactivity to human, cynomolgus monkey, and mouse CD137 (Supplementary Table S1). Furthermore, the epitopes of 106 and 107 clones were characterized by competition to other mAbs and CD137 ligand (Fig. 1C and Supplementary Fig. S1). CD137 mAbs 106 and 107 compete with each other but not with either human or mouse CD137 ligand, suggesting these 2 mAbs share similar epitopes that do not overlap with the ligand binding site. 106 and 107 mAbs also did not compete with other anti-mouse CD137 (1D8, 3H3, and MAB9371) or an anti-human CD137, Mab A, suggesting they bind to a unique epitope. Interestingly, 106 and 107 mAbs compete with a ligand blocking anti-human CD137, Mab B. These results suggest the epitopes of 106 and 107 mAbs are different from that of Mab B even though they partially overlap.
To further investigate the efficacy and liver toxicity of 106 mAb, we performed a dose titration study in CT26 tumor model using 3H3 as a control CD137 mAb. Compared with 3H3, 106 mAb showed a similar dose-dependent antitumor activity (Fig. 2A) while induced lower serum ALT levels (Fig. 2B). It has been previously suggested that CD137 mAbs caused liver damage by increasing the infiltration of CD8+ T cells and macrophages (28, 33), so we examined liver tissues from the treated animals to evaluate the level of immune cell infiltration. Livers from 3H3 treated mice showed higher number of infiltrated leukocytes including CD8+ T cells and Iba1+ macrophages in comparison with 106 mAb (Fig. 2C). Plasma concentrations of 106 and 3H3 mAbs 24 hours after last dose were comparable (Supplementary Fig. S2), suggesting that the drug exposure is not a factor for the differences in liver toxicity. These data demonstrated the antitumor activity and liver toxicity with anti-CD137 agonist treatment can be segregated in mice.
In vivo antitumor activity is Fc and Fc gamma receptor (FcγR) IIB dependent
Fc-mediated biology is known to have an important role in driving antitumor efficacy by immunomodulatory mAbs. To assess if the in vivo activity of our CD137 mAbs is Fc dependent, we first compared the antitumor potency induced by a murine 107 IgG2a and 107 IgG2a DANA (D265A-N297A, mutant that eliminates FcγR binding) in CT26 and MC-38 syngeneic tumor models. As shown in Fig. 3A and B, an IgG2a has stronger antitumor activity than the IgG2a DANA variant of 107, which indicates that FcγR-mediated activity is important for in vivo antitumor activity. Murine IgG1 consistently showed strong efficacy in both models; however, no statistical difference was observed between IgG1 and IgG2a under the condition tested. A similar result was observed in 106 mAb (Supplementary Fig. S3C). Interestingly, 107 with a murine IgG1 resulted in more tumor-free mice than IgG2a-treated mice (5/8 vs. 0/8 in CT26 and 2/8 vs. 0/8 in MC-38 for IgG1 and IgG2a, respectively). It is also important to note that the serum ALT remained low in both BALB/c or C57BL/6 mice treated with 106 and 107 (Supplementary Fig. S3A, S3B, and S3D), indicating that they induce less liver toxicity. We observed a slight increase of ALT levels when MC-38 tumors in C57BL/6 were treated with either 106 or 107 in IgG1 but not with IgG2a (Supplementary Fig. S3B and S3D). Similarly, a small increase in ALT levels compared with control group was observed when CT26 and MC-38 tumors were treated with 106 mouse G1 (Fig. 2B and Supplementary Fig. S3D). It is important to note that these increases in liver enzyme levels are ∼10-fold lower than those observed in serum treated with 105 (mean 529 U/L) or 3H3 mAbs (mean 388 U/L), Fig. 1B and Fig. 2B, respectively. Thus, we conclude that these CD137 mAbs (106 and 107) induce significantly reduced liver toxicity compared with typical CD137 agonist mAbs reported elsewhere, regardless of mouse genetic background.
Because both murine IgG1 and IgG2a engage FcγRIIB and III, we evaluated whether FcγR interactions are critical for the bioactivity of CD137 mAbs using FcγR knockout mice. Efficacy was completely lost for both IgG1 and IgG2a when CT26 tumor was grown and treated in FcγRIIB−/− background (Fig. 3C). In MC-38, the difference between 107 IgG1 and IgG2a was insignificant (Fig. 3B) while they remained active in FcγRIII−/− mice (Fig. 3D). Because murine IgG1 engages only FcγRIIB and FcγRIII, FcγRIIB that remained in the FcγRIII−/− mice can still provide sufficient activity to control MC-38 tumor growth. These results demonstrated that FcγRIIB is crucial for inducing antitumor activity in mice in vivo.
mAbs 106 and 107 require Fc cross-linking for T-cell activation in vitro
To investigate the intrinsic costimulation activity of 106 and 107 mAbs, a mouse CD8+ T-cell costimulation assay was performed with 3H3 mAb as a positive control. In the absence of Fc cross-linking, only 3H3 mAb has significant bioactivity, indicating 106 and 107 mAbs are not strong agonists intrinsically; however, in the presence of anti-rat IgG as cross-linkers, 106 and 107 mAbs induced GM-CSF cytokine release (Fig. 4A). Similarly, in a human CD8+ T-cell proliferation assay, Mab B-IgG2, 106-hIgG1AA (rat mAb chimeric with human IgG1 Fc with L234A/L235A mutations for reduced FcγR binding), and 107-hIgG1AA induced T-cell proliferation following the effect of cross-linking with a secondary mAb, except for Mab A (IgG4 or IgG1AA), which is active even in the absence of cross-linking (Fig. 4B). These data suggested that, similar to Mab B, 106 and 107 mAbs require Fc cross-linking to function as T-cell costimulators, whereas Mab A is a strong agonist that is capable of inducing signal by itself. In order to exclude the possibility that IgG4 isotype induces cross-linking–independent activity, T-cell costimulation of Mab A, Mab B, 106, and 107 was also compared with IgG4 (Supplementary Fig. S4A). Only Mab A showed activity without cross-linking, indicating the activity is due to its intrinsic epitope instead of Fc isotype being G4. To further confirm the effect of Fc-mediated cross-linking, different IgG variants of 106 and 107 mAbs were compared in a T-cell costimulation assay in the presence of FcγRIIB expressed on CHO-K1. The human IgG4 isotype induced higher IFNγ secretion than the IgG1 and IgG2 isotypes for both mAbs (Fig. 4C). In the absence of FcγRIIB-expressing CHO-K1 cells, no activity was observed for all variants. These data indicate that 106 and 107 mAbs require Fc cross-linking to induce T-cell costimulation activity, and human IgG4 is much more efficient than IgG2 in mediating the activity in the presence of FcγIIB.
In vivo pharmacodynamics analysis of anti-CD137 agonists in mouse syngeneic models
To understand the mechanism of action of 106 and 107 mAbs in vivo, we used an MC-38 tumor model and included 1D8 mAb as the positive control to allow comparison as mouse IgG1. 1D8 mAb is also known to induce liver toxicity similar to 3H3 mAb (29). Consistent with previous results (Fig. 3B), i.p. injection of IgG1 isotypes of 106, 107, and 1D8 mAbs at 10 mg/kg on days 7, 14, and 21 resulted in significant regression of established MC-38 tumors (Fig. 5A). Systemic administration of 106 and 107 mAbs resulted in significant decrease of tumor-infiltrating CD4+ T cells at early time point (day 15; Fig. 5B). All 3 CD137 mAbs induced a significant increase of tumor-infiltrating CD8+ T cells at later time point (day 22). There was a slight decrease in T regulatory cells (Treg) (106 mAb at day 15) and an increase of CD8+/Treg ratio, although they are not statistically significant at day 22. We also observed that total T cells, B cells, and NK cells were decreased in the circulation 24 hours after the second and third doses (days 15 and day 22, respectively); while in the tumor-draining lymph nodes, changes in immune cell populations were usually small and transient between the control group– and CD137 mAb–treated groups (Supplementary Fig. S5A).
Serum cytokine and chemokine levels were measured 24 hours after the second (day 15) and third doses (day 21). No significant changes were observed in the levels of IL2, TNFα, and IL6 cytokines in sera after treatment (Supplementary Fig. S5B). Detectable levels of IFNγ, IL10, MIG, MCP-1, MIP-1β, and RANTES were identified in groups treated with CD137 mAbs (Fig. 5C). The level of these cytokines/chemokines seems to increase from day 15 to day 22.
An antigen-specific memory T-cell response has been demonstrated in CD137 mAb–treated CT26 mice based on tumor rechallenge studies following treatment with mIgG1 isoforms of 106 and 1D8 mAbs at 10 mg/kg on days 7, 14, and 21 (Supplementary Fig. S5C). Accumulation of antigen-specific CD8+ T cells in the spleen has also been observed by gp70–peptide–MHC tetramer staining (34) following treatment with the 106 mAb (Fig. 5D).
CD137 mAb binding affinity and in vitro activity may not correlate with in vivo efficacy and hepatotoxicity
Although binding epitope is critical to its liver toxicity, it is possible that affinity difference plays a role for the activity of CD137 agonist mAbs. To investigate whether the affinity of a CD137 mAb would also affect liver toxicity and affect its functional activity, two 106 variant mAbs with different CD137 binding affinity were created: a high-affinity mutant 106H with a KD (dissociation constant) of 0.54 nmol/L and a low-affinity variant 106L with a KD of 220 nmol/L (KD values were determined by Biacore; see Supplementary Table S1) generated by amino acid substitutions in CDRs. In comparison, the parental 106 mAb (106M) has an intermediate KD of 17 nmol/L. Binding affinities were also confirmed by flow cytometry using CD137 stable transfectant (Fig. 6A). T-cell proliferation assay showed high-affinity 106H induces a strong T-cell activation, whereas low-affinity 106L has very limited bioactivity, which indicates a direct correlation between affinity and in vitro activity (Fig. 6B). In vivo activity of these variants was tested using the CT26 tumor syngeneic model. Interestingly, the efficacy of 106L was comparable with parental and high-affinity mAbs following a single-dose treatment at high concentration (3 mg/kg; Fig. 6C). Surprisingly, at low dose (0.3 mg/kg), parental (106M) and low-affinity mAb (106L) controlled tumor growth better than the high-affinity 106H without causing significant increase in ALT. All of these mAbs slightly increased the level of serum ALT at 3 mg/kg, less than 2-fold of the control (Fig. 6D), which is consistent with the results in Fig. 2B at weekly dose. These results suggest the affinity of CD137 mAbs may not play a major role in hepatotoxicity, and a low-affinity mAb might be more efficacious in vivo in a suboptimal dose. Finally, these results clearly indicate that binding affinity and bioactivity in vitro do not predict in vivo antitumor efficacy with CD137 mAbs in mice.
Based on the observation that anti-CD137 induced transaminitis in both mouse and human, we used mouse syngeneic models to screen for lead candidates with low liver toxicity and potent antitumor efficacy. In this study, we successfully generated and identified 2 rat mAbs, 106 and 107, which cross-react to human, cynomolgus monkey, and mouse CD137. These mAbs have unique properties that are important for improved safety while maintaining strong antitumor potency in vivo.
TNFR superfamily signaling depends on receptor clustering and conformation change induced by ligand binding. In this study, we found that cross-linking dependency of an agonistic mAb to induce CD137 signaling could be independent from the ability to compete with ligand binding. Although 106/107 mAb and Mab B compete in binding to CD137, they likely bind to different epitopes because 106/107 do not compete with ligand binding while Mab B does. This could be potentially explained by the differences in conformational change induced by mAb binding; further study such as X-ray cocrystallography will be necessary to understand the molecular mechanism.
Bioactivity of CD137 mAb described in this study is dependent on FcγRIIB in vivo, which is in agreement with mAbs against CD40 (35) and DR5 (36). In the literature, 2 other CD137 mAbs have been tested in FcγRIIB−/− mice. First, clone 2A (rat IgG2a) was shown to be efficacious in FcγRIIB−/− mice in an EL4E7 lymphoma model (37). Second, clone LOB12.0 (mIgG1) showed enhanced efficacy in FcγRIIB−/− mice in a CT26 syngeneic tumor model (38). The antitumor activity of LOB12.0 was reported to dependent primarily on depletion of intratumoral Tregs, which explains why FcγRIIB-mediated cross-linking was not essential for its in vivo efficacy. It is likely that 2A mAb also has some Treg-depleting activity because rat G2a binds to FcγRIII; however, these observations may simply reflect differences in tumor microenvironment in the models used and/or size of tumor when treatment was initiated. Interestingly, in mice lacking all activating FcγRs (Fcer1g−/−), yet still retaining FcγRIIB, both 2A and LOB12.0 (mIgG1) achieved 100% long-term survival in EL4E7 and EG7 models, respectively (37, 38). These results indicate that activating FcγRs interfered with FcγRIIB-mediated CD137 signaling and negatively affected the in vivo efficacy under certain circumstances. Hence, an anti-CD137 with Fc engineered to enhance FcγRIIB binding with minimal interaction with activating FcγRs would likely produce an optimal antitumor response. We have compared the in vivo efficacy of murine IgG1 and IgG2a isoforms of 106, 107, and 1D8 in CT26 or MC-38 models and found that the efficacy of murine IgG1 is more consistent than the murine IgG2a isoform, resulting in higher numbers of tumor-free mice at the end of the study. Similar to rat IgG2a, murine IgG1 binds FcγRIIB and FcγRIII, whereas murine IgG2a binds strongly to FcγRI and FcγRIV, yet weakly to FcγRIIB. Higher efficacy of IgG1 over IgG2a is likely due to its increased binding affinity to FcγRIIB (39). It is known that Fc-engineered CD40 mAbs with enhanced binding specifically to FcγRIIB results in a dramatic enhancement of their antitumor efficacy (40). With the common mechanism in receptor clustering of TNFR superfamily members, similar Fc mutations is likely applicable to 106/107 to further enhance activity in vivo.
Because the functions of FcγRIIB are conserved in both mouse and human, the choice of isotype for a clinical CD137 agonistic mAb that can engage FcγRIIB becomes critical to its therapeutic potential and liver toxicity. Our results show that human IgG4 is the most potent isotype, whereas IgG2 is the weakest in human T-cell costimulation. The observations were not limited to 106 and 107, but also applied to Mab A and Mab B (Supplementary Fig. S4B). These results are in agreement with the binding affinity of isotypes to FcγRIIB, IgG4 is about 2-fold higher than IgG1, and 10-fold higher than IgG2 (41). The clinical outcomes of 2 current agonistic CD137 mAbs also agree with our hypothesis. Urelumab is a strong CD137 agonist like Mab A, and the human IgG4 isotype will likely boost its in vivo activity, which explains why it has severe hepatotoxicity and must be dosed at 0.1 mg/kg or lower. On the other hand, utomilumab is human IgG2, the weakest isotype for FcγRIIB cross-linking, which may result in a weak agonist unless G2 hinge causes the activity to be Fc-independent as described in the example of a CD40 agonist mAb (42). Assuming the activity of utomilumab is independent from this hinge effect, it may explain why utomilumab can be dosed at 10 mg/kg without any liver toxicity issues. In addition, utomilumab blocks CD137 ligand whereas urelumab does not, which may also explain the difference in activity (43). Urelumab and utomilumab represent two extremes of CD137 agonist and their choice of isotype and subsequent clinical outcomes provide valuable lessons for choosing the optimal isotype for future CD137 mAbs.
Treatment with CD137 agonist induced a variety of cytokines and chemokines that are detectable in serum, which might be important for CD137-mediated antitumor efficacy. For example, IFNγ has been shown to be essential for antitumor activity of CD137 mAb because it is required for the infiltration of tumor-specific CD8+ cytotoxic T cells into the tumors (44). MIG is induced by IFNγ as a chemoattractant for activated T cells, which inhibits angiogenesis and promotes cytotoxic T-cell activity (45). The presence of IL10 may be in part due to a Treg-mediated anti-inflammatory response or may also play a critical role in the proliferation of antigen-specific CD8+ T cells (46, 47). MCP-1 may be particularly important for initiating liver toxicity because its receptor CCR2 has been shown to be involved in CD137-mediated hepatotoxicity (33). However, we did not detect TNFα, which has been shown to be essential for CD137-induced liver toxicity (28).
An inverse correlation between affinity and potency has been observed for agonist mAbs against Fas in vitro (48) and erythropoietin receptor (EPOR) in vivo (49). The authors suggested that these low-affinity mAbs have fast off-rates that favored the formation of receptor clusters with active conformation and hence enhanced signaling. What surprised us is the strong in vivo antitumor activity of the low-affinity 106L mAb, which showed the weakest T-cell costimulation activity in vitro. It is known that CD137 can be expressed in a variety of immune cells, including myeloid cells in liver that have been reported to play a role in toxicity mediated by CD137 agonist mAbs (33). It is possible that a low-affinity agonist provides selective binding and activation of immune cells with higher CD137 expression such as cytotoxic CD8+ T cells, which are directly involved in antitumor activity in vivo.
The major concern of targeting CD137 for immunotherapy is liver toxicity. We addressed this problem by using mice as a screening platform to select for mAbs with high potency and low liver toxicity. These mAbs demonstrated minimum liver enzyme elevation while retaining antitumor efficacy, accompanied by increased CD8+ T cells in the tumor which resulted in an antigen-specific memory T-cell response. This characteristic of inducing efficacy without significant liver toxicity is likely attributed by the specific epitope. It is also noteworthy that clone 105 showing high ALT in Fig. 1B belonged to the same epitope group as 1D8. Because hepatic myeloid cells express both CD137 and FcγR are thought to drive liver toxicity (33), one could speculate that certain mAbs in an epitope group may activate these liver myeloid cells more than the others by coengagement of FcγR on the same cell. Further investigation will be required to fully understand the mechanism(s) utilized by different epitopes of CD137 mAbs resulting in differential liver toxicity.
In conclusion, this study demonstrated the importance of epitope and FcγRIIB-mediated activity to develop a new class of CD137 agonistic mAbs that have an enhanced therapeutic window for effective cancer immunotherapy. Synergistic antitumor activity by combination of CD137 mAb with anti–PD-1 or anti–PD-L1 in B16-F10 tumor model was confirmed as previously reported (Supplementary Fig. S6; refs. 21, 50); however, it is important to address the toxicity effect because the checkpoint inhibitors may alter lymphocytes infiltration not only to tumors but also to other organs. Thus, enhanced safety monitoring will be critical for any CD137-related human therapy development.
Disclosure of Potential Conflicts of Interest
E. DiGiammarino is a principal scientist at AbbVie, Inc., reports receiving a commercial research grant from the same. L. Zhou is a senior scientist at AbbVie. C.M. Forsyth is a principal scientist at AbbVie. D.B. Powers is a research fellow at AbbVie. D.T. Chao is a director at AbbVie. H.M. Alvarez is a principal research scientist at AbbVie. Y. Akamatsu is a senior principal research scientist at, reports receiving commercial research grant from, and has ownership interest (including patents) in AbbVie. No potential conflicts of interest were disclosed by the other authors.
Conception and design: D.B. Powers, D.T. Chao, D. Hollenbaugh, Y. Akamatsu
Development of methodology: S.K. Ho, Z. Xu, A. Thakur, E. DiGiammarino, L. Zhou, V. Zhao, M. Xiong, J. Samayoa
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.K. Ho, Z. Xu, A. Thakur, M. Fox, S.S. Tan, E. DiGiammarino, L. Zhou, M. Sho, M. Xiong, C.M. Forsyth
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.K. Ho, Z. Xu, A. Thakur, S.S. Tan, E. DiGiammarino, L. Zhou, B. Cairns, V. Zhao, J. Samayoa, H.M. Alvarez
Writing, review, and/or revision of the manuscript: S.K. Ho, Z. Xu, A. Thakur, M. Fox, E. DiGiammarino, L. Zhou, B. Cairns, V. Zhao, J. Samayoa, C.M. Forsyth, D.B. Powers, D.T. Chao, D. Hollenbaugh, H.M. Alvarez, Y. Akamatsu
Study supervision: H.M. Alvarez, Y. Akamatsu
We thank Jieyi Wang (Lyvgen BioPharma) for his earlier works to the study when he was an employee of AbbVie. We also thank the Protein Science Core group in AbbVie Redwood City for producing reagents.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.