Antitumor Activity of a Novel Bispecific Antibody That Targets the ErbB2/ErbB3 Oncogenic Unit and Inhibits Heregulin-Induced Activation of ErbB3 Free

The family of cell surface ErbB receptors, ErbB1–4, play an essential role in development and physiology, mediating cellular growth and differentiation in multiple tissues, both in the developing embryo and in the adult (1). Escape of ErbB-activated pathways from normal control mechanisms results in tumorigenesis and growth (1). A prominent example is the amplification and overexpression of ErbB2 that occurs in 20% to 30% of breast and gastric carcinomas. Although lacking a high affinity ligand ErbB2 efficiently propagates prosurvival signals by forming heterodimers with other ErbB family members such as ErbB3. Therapeutic agents targeting ErbB2 (2), have greatly improved outcomes for many patients with ErbB2-overexpressing tumors but a significant proportion do not benefit and initial responders often develop resistance (3). ErbB3 activation following engagement of its ligand, heregulin, can mediate resistance to ErbB-targeted therapies (4–8). ErbB3 expression is upregulated in trastuzumab resistant tumor cells (9) and associated with poor prognosis in breast cancer (10, 11). Despite lacking an active kinase domain ErbB3 forms heterodimers with other members of the ErbB family to generate robust cellular signals through activation of the phosphoinositide 3-kinase (PI3K) pathway. ErbB3 is a preferred dimerization partner of ErbB2 (12, 13) and the ErbB2/ErbB3 heterodimer has been identified as the most potent activator of AKT compared to other ErbB receptor heterodimers with a strong dependence on ErbB3 for maintenance of oncogenic signaling (14, 15). However, as ErbB3 lacks an active kinase domain and does not become amplified or significantly overexpressed in tumors, ErbB2 has largely remained the major therapeutic target in this heterodimeric oncogenic unit.

Recently, a computational analysis of the ErbB signaling network identified ErbB3 as the principal perpetrator of PI3K activation following ligand engagement (16). These data support direct therapeutic targeting of ErbB3 and subsequently an anti-ErbB3 antibody, MM-121, was developed that shows potent AKT inhibition and attenuation of tumor growth (16, 17). Traditionally, ErbB-mediated tumor growth was thought to arise through dependence of tumor cells on a mutated or amplified receptor, as exemplified by ErbB1 and ErbB2, respectively. These data generated by Schoeberl and colleagues (16) support an additional emerging paradigm of ErbB-dependent tumor growth through a combinatorial ligand-induced mechanism without the requirement for receptor overexpression or mutation. In the case of ErbB2-amplified tumors it seems that both oncogenic addiction and ligand activation may combine to drive potent aberrant signaling through the ErbB2/ErbB3 heterodimer and combinations of targeted therapies may be required for optimal antitumor activity.

In this study, we show that the approved ErbB2-targeted therapies, trastuzumab and lapatinib, are poor inhibitors of ligand-induced ErbB3 activation in ErbB2-overexpressing tumor cells. Likewise, pertuzumab, a monoclonal antibody in clinical development that binds to the extracellular ErbB2 dimerization domain, cannot effectively suppress ligand-induced ErbB3 phosphorylation. Thus, we have extended the mathematical ErbB network modeling approach described by Schoeberl and colleagues (16) to guide the design of an optimal inhibitor of the ligand activated ErbB2/ErbB3 oncogenic unit in ErbB2-overexpressing tumor cells. These simulations predicted that a bispecific molecule that targets the ErbB2/ErbB3 heterodimer and promotes the formation of inactive trimeric complexes is more effective at inhibiting ErbB3 activation than either an ErbB2 or ErbB3 monoclonal antibody. Consequently, we have developed MM-111, a bispecific antibody fusion protein consisting of fully human anti-ErbB2 and anti-ErbB3 single chain antibody moieties linked by modified human serum albumin (HSA). MM-111 binds with both avidity and specificity to tumor cells expressing ErbB2 and ErbB3 and blocks ligand-induced signaling and tumor growth in several preclinical models.

Trastuzumab was recently shown to effectively inhibit basal ErbB3 signaling in the absence of ligand stimulation (18). However, in agreement with our findings, trastuzumab did not effectively block ligand-induced activation of ErbB2/ErbB3 signaling thus providing tumor cells with an escape mechanism from trastuzumab therapy (18). ErbB3 may also provide escape from lapatinib therapy, which causes ErbB3 upregulation in both preclinical models and patient samples following treatment (19). We hypothesized that the combined inhibition of ErbB2 and ErbB3 provided by addition of MM-111 to either trastuzumab or lapatinib would synergistically inhibit tumor cell growth. Indeed, we report here that such combinations are effective and concurrent treatment of MM-111 and an ErbB2-targeted agent may deter resistance through ErbB3 activation.

Cell lines were obtained from American Type Culture Collection (ATCC) with the exception of BT-474-M3, provided by Dr. Daryl Drummond (Hermes Biosciences) and NCI/ADRr, obtained from the NCI. All cell lines were passaged for fewer than 6 months after resuscitation, and purchased cell lines were cultured using the protocol provided. To obtain the BT-474-M3 cell line, BT-474 cells, obtained from ATCC, were passaged twice through mice with the fastest growing 2 tumors out of 10 selected for ex vivo propagation during each round of selection. Tumors were excised and cultured ex vivo to obtain the M3 subline that was verified by single-nucleotide polymorphism analysis. MM-111 was subcloned downstream of the human glyceraldehydes-3-phosphate dehydrogenase promoter between 2 Matrix Attachment Region elements (20). HSA containing C34S and N503Q mutations was obtained by gene synthesis (Codon Devices). MM-111–binding variant MM-111ΔErbB2 was constructed by mutating amino acids in the CDR3 of the B1D2 V_H domain, and variant MM-111ΔErbB3 was constructed by replacing the H3 scFv with the mutated B1D2 scFv. MM-111, MM-111ΔErbB2, and MM-111ΔErbB3 were stably expressed in CHO-K1 cells in shake flasks or 10 L WAVE bags and purified from conditioned media by Blue Sepharose chromatography. The extracellular domain of human ErbB2 was expressed in CHO-K1 cells as a histidine-tagged fusion protein and purified by nickel affinity chromatography. Pertuzumab was produced as described previously (16). Trastuzumab was obtained from pharmacy. Lapatinib was obtained by custom synthesis (Biomol International). ErbB3 extracellular domain Fc fusion protein (ErbB3ecd-Fc) and heregulin 1-β were obtained from R&D Systems. The H3 scFv was cloned into the pCYN bacterial expression vector and expressed in Escherichia coli as a histidine-tagged fusion protein.

A computational model of heregulin-induced p-ErbB3 signaling, as well as models of the inhibitors pertuzumab and lapatinib, were used as previously described (16). For format simulations, computational models were constructed according to the following reactions:

E2 + BsAb ↔ E2:BsAb (A)

BsAb + E3 ↔ BsAb:E3 (B)

E2 + BsAb:E3 ↔ E2:BsAb:E3 (C)

E2:BsAb + E3 ↔ E2:BsAb:E3 (D)

E2 + IgG ↔ E2:IgG (E)

E2:IgG + E2 ↔ E2:IgG:E2 (F)

E3 + IgG ↔ E3:IgG (G)

E3:IgG + E3 ↔ E3:IgG:E3 (H)

Interaction parameters between each inhibitor arm and its target were k_on = 10⁵ (M s)⁻¹ and k_off = 10⁻³ (s⁻¹). Reactions (C), (D), (F), and (H) represent avid inhibitor binding leading to receptor cross-linking; for these reactions, all inhibitors were assumed to have equal cross-linking potency. For format simulations, assumed receptor levels were ErbB1 = 0, ErbB2 = 10⁶/cell, and ErbB3 = 4 × 10⁴/cell. Simulated conditions were 24-hour preincubation of inhibitor followed by 10 minutes of 5 nmol/L heregulin stimulation. All computations were conducted in MATLAB 7.7 (The MathWorks).

In vitro signaling experiments were carried out as described previously (16). Briefly, serum-starved cells were preincubated with MM-111, pertuzumab, trastuzumab, lapatinib, or combinations followed by stimulation with 5 nmol/L heregulin 1-β (R&D Systems) for 10 minutes. p-ErbB3 and p-AKT were measured by ELISA as described previously (16). Inhibitor IC₅₀ values were calculated by fitting dose–response data to a 4-parameter sigmoidal curve (GraphPad Prism). As appropriate, computational and experimental data for ligand-induced signaling were compared by subtraction of the unstimulated control and normalization to maximum observed signal.

ErbB1, ErbB2, and ErbB3 receptor levels were determined by quantitative FACS as described previously (16). ErbB3 scFv H3 blocking of heregulin/ErbB3 binding was assessed in a BIAcore. Heregulin was coupled to a CM5 sensor chip, and 50 nmol/L ErbB3ecd-Fc in HBS alone or mixed with 500 nmol/L H3 scFv or 500 nmol/L control IgG (C225) was flowed over the chip. To measure cell-binding affinities, MALME-3M cells (1,00,000 per well) were seeded into a 96-well plate and MM-111, MM-111ΔErbB2, and MM-111ΔErbB3 was added at concentrations indicated. Cells were then incubated with a 1:500 dilution of Alexa 647–labeled (Invitrogen) goat anti-HSA antibody (Bethyl Laboratories) and stained with 0.5 μg/mL propidium iodide followed by FACS analysis.

Cells were incubated overnight following seeding, and inhibitor was added for 24 hours (lapatinib) or 6 days (MM-111, trastuzumab, pertuzumab, or combinations). For experiments with ligand stimulation, cells were serum-starved overnight before addition of inhibitor and 2 nmol/L heregulin 1-β was added 1 hour postinhibitor treatment in media containing 5% FBS. Cell viability was measured by the CellTiter-Glo Luminescent Cell Viability Assay (Promega).

BT-474-M3 cells (1,00,000 per well) were plated into a 6-well plate and the following day 1 μmol/L MM-111 was added for 72 hours. Cells were treated with propidium iodide and RNAse A and cell-cycle distribution was measured by FACS.

BT-474-M3 cells were treated with MM-111 for 72 hours and then lysed with RIPA buffer (Sigma-Aldrich) supplemented with protease and phosphatase inhibitor cocktails (Roche). Cell lysate proteins were resolved by SDS-PAGE, and immunoblots were probed with rabbit anti-cyclin D1 antibody (Thermo Fisher).

For p27 staining, BT-474-M3 cells (40,000) were seeded on glass coverslips and incubated with 1 μmol/L MM-111 for 6 hours. Fixed and permeabilized cells were blocked and then incubated with mouse anti-p27 antibody (BD Biosciences).

MM-111 was incubated in mouse serum (Sigma-Aldrich) at 37°C for 120 hours. Samples were removed and MM-111 was detected by a bispecificity ELISA. Briefly, a 96-well plate was coated with ErbB2ecd domain overnight followed by blocking (Pierce) and incubation with MM-111. Plates were then incubated with Fc-ErbB3 followed by goat anti-human-Fc-HRP conjugate (Jackson ImmunoResearch). To determine the stability of MM-111 in circulation 5- to 6-week-old female CD-1 nude mice (Charles River Labs) were dosed with 30 mg/kg MM-111 by bolus intravenous injection. Serum was harvested 0.5, 4, 8, 24, 48, 72, and 120 hours postinjection and MM-111 was detected by an HSA ELISA (Bethyl Laboratories) and the bispecificity ELISA.

Tumor xenografts were established by subcutaneous injection of tumor cells into the flanks of 5- to 6-week-old female athymic nude mice (nu/nu; Charles River Laboratories) except for MDA-MB-361 cells that were injected into 5- to 6-week-old female nonobese diabetic/severe combined immunodeficient (NOD/SCID) mice (Charles River Laboratories). For the BT-474-M3 and ZR75-1 models, mice received a subcutaneous 60-day, slow-release estrogen implant in the opposite flank (0.72 mg pellet; Innovation Research of America), 24 hours before the injection of cells. Once tumors reached a mean volume of 150 to 500 mm³, mice were randomized into groups of 8 or 10 and dosed by intraperitoneal injection.

We conducted inhibitor dose–response assays to investigate the ability of the ErbB2-directed therapies, lapatinib, trastuzumab, and pertuzumab to inhibit p-ErbB3 in heregulin-stimulated BT-474-M3 cells that overexpress ErbB2 (Table 1). We found that all 3 molecules weakly inhibited ErbB3 phosphorylation with IC₅₀ values of 96 and 260 nmol/L for pertuzumab and lapatinib, respectively (Fig. 1A), whereas trastuzumab was unable to inhibit heregulin-induced ErbB3 activation. We then applied a previously developed computational model of heregulin-induced signaling of the ErbB receptor signaling network (16) to explore optimal inhibitor formats for specifically disrupting signaling through the ErbB2/3 heterodimer in ErbB2-overexpressing cells. The protein–protein interactions, biochemical reactions, and kinetic parameters incorporated into the model are described by Schoeberl and colleagues (16). To validate the model, we generated in silico representations of lapatinib and pertuzumab ErbB3 inhibition, which compared accurately with experimental data (Fig. 1A).

We next created in silico models of 3 paradigms for inhibiting signaling from the ErbB2/ErbB3 heterodimer: an ErbB2 monoclonal antibody, an ErbB3 monoclonal, and an ErbB2/3-bispecific antibody. The inhibition of ErbB3-mediated signaling by the in silico ErbB2 antibody occurs through sequestration of ErbB2 receptors from ErbB3, thereby preventing the formation of ErbB2/3 heterodimers. In contrast, the ErbB3 antibody and ErbB2/3 bispecific antibody function by blocking heregulin binding to ErbB3. To isolate the role of inhibitor format in driving the efficacy of ErbB3 inhibition, these generic inhibitor models used identical kinetic binding parameters [k_on = 10⁵ (M s)⁻¹ and k_off = 10⁻³ (s⁻¹)] and ability to bivalently cross-link their targets. The relative ability to inhibit ligand-induced ErbB3 phosphorylation was simulated in a model cell expressing 1 × 10⁶ ErbB2 receptors per cell and 4 × 10⁴ ErbB3 receptors per cell under 5 nmol/L heregulin stimulation. In this model system our simulations suggest that an ErbB2/3-bispecific antibody offers superior p-ErbB3 IC₅₀ potency (0.2 nmol/L) compared with either an ErbB2 (IC₅₀ = 58 nmol/L) or ErbB3 (IC₅₀ = 3.1 nmol/L) monoclonal antibody. Moreover, the bispecific antibody is more potent than a combination of both ErbB2 and ErbB3 antibodies (IC₅₀ = 1.1 nmol/L; Fig. 1B).

Using simulations of on-cell binding, we further explored the relative ability of a bispecific antibody to bind to ErbB3 receptors in cells with different levels of ErbB2. In model cells with equal expression of ErbB2 and ErbB3 (4 × 10⁴ receptors per cell; “HER2 1+”), 50% receptor occupancy of ErbB2 results in 50% occupancy of ErbB3 by the bispecific antibody. However, simulated overexpression of ErbB2 to levels of 2 × 10⁵ (corresponding to an immunohistochemistry score of approximately “HER2 2+”) and 1 × 10⁶ (corresponding to an immunohistochemistry score of approximately “HER2 3+”) receptors per cell led to increasingly strong occupancy of ErbB3 receptor to 95% and 99%, respectively. These results predicted that an ErbB2/3-bispecific antibody would potently target ErbB3 only in cells overexpressing ErbB2 (Fig. 1C).

The ErbB2 and ErbB3 scFv binding arms, B1D2 and H3, respectively, were chosen for building MM-111. The ErbB2 scFv component of MM-111, B1D2 (21), is an affinity matured version of the C6.5 scFv (22) that binds receptor with an affinity of 0.3 nmol/L (Supplementary Fig. S1A) providing ErbB2 targeting, whereas the ErbB3 scFv component of MM-111, H3 (23), binds to ErbB3 with an affinity of 16 nmol/L (Supplementary Fig. S1B). Both the B1D2 scFv and H3 scFv bind specifically to ErbB2 and ErbB3, respectively, and do not interact with other ErbB family members (Supplementary Fig. S2). We investigated the ability of the H3 scFv to block heregulin binding to ErbB3. Preincubation of ErbB3ecd-Fc with H3 scFv prevented binding of ErbB3ecd-Fc to heregulin immobilized on a CM5 chip (Fig. 2A and B). A mutated variant of HSA, mHSA, was inserted between the H3 and B1D2 scFvs of MM-111 with short connector peptide linkers, AAS and AAAL, inserted at the amino and carboxyl terminus of mHSA, respectively. The long serum half-life of HSA of approximately 21 days has been reported to be due to its recycling by the FcRn receptor by a similar mechanism to IgG recycling (24) and incorporating HSA into therapeutic biologics is an established strategy for enhancing serum half-life. To achieve greater homogeneity of the HSA linker, we made 2 point mutations. A cysteine residue at position 34 of native HSA was mutated to serine to reduce potential protein heterogeneity due to oxidation at this site. An asparagine residue at amino acid 503 of native HSA, which in native HSA is sensitive to deamidation, was mutated to glutamine. Analysis of purified MM-111 and its variants MM-111ΔErbB2 and MM-111ΔErbB3 by size exclusion chromatography showed that greater than 95% of each purified protein eluted in the monomeric fraction (Supplementary Fig. S3).

The ability of MM-111 to bind cells avidly by engaging both ErbB2 and ErbB3 was tested on the melanoma tumor cell line Malme-3M, which expresses approximately equivalent levels of the 2 receptors as determined using quantitative FACS methods (Table 1), thus allowing assessment of binding avidity. Although the ErbB3 scFv component of MM-111, H3, specifically binds ErbB3 and blocks heregulin (Fig. 2A and B) incubation of MM-111ΔErbB2, which lacks ErbB2 binding activity, with Malme-3M cells resulted in no measurable cell binding (Fig. 2C), likely due to its monovalent affinity of 16 nmol/L. MM-111ΔErbB3, which retains a functional, high affinity ErbB2 binding scFv but lacks ErbB3 binding activity had an apparent K_D of 10 nmol/L (Fig. 2C). MM-111 bound cells with an apparent K_D of 0.3 nmol/L showing avidity binding of greater than 30-fold than the single binding arms (Fig. 2C) and indicating that MM-111 interacts with ErbB2 and ErbB3 simultaneously. A dose–response experiment on BT-474-M3 cells showed that following 24 hours incubation MM-111 potently inhibits ErbB3 phosphorylation, with an IC₅₀ value of 3 nmol/L, whereas a combination of MM-111ΔErbB2 and MM-111ΔErbB3 providing an equivalent dose of ErbB2 and ErbB3 binding moieties is ineffective (Fig. 2D). Together these data show that MM-111 is a potent inhibitor of p-ErbB3 in ErbB2-overexpressing cells and inhibition requires simultaneous binding of both ErbB2 and ErbB3 receptors by MM-111 to form a trimeric inhibitory complex.

We subsequently assessed MM-111 potency on a panel of tumor cell lines expressing a range of ErbB2 levels. We observed that IC₅₀ values for p-ErbB3 inhibition were consistently in the low nanomolar range, regardless of ErbB2 expression level, but the ability of MM-111 to inhibit ligand-activated p-ErbB3 to basal levels was strongly positively correlated with ErbB2 expression levels (Table 1). To further investigate the dependency of MM-111 activity on ErbB2 levels we examined the potency of MM-111 in ovarian ADRr cells, expressing 4 × 10⁴ ErbB2 receptors per cell and an ErbB2-overexpressing transfectant of ADRr, ADRr-E2, which express 7 × 10⁵ ErbB2 receptors per cell. We observed a 3-fold increase in heregulin-stimulated p-ErbB3 levels in the ADRr-E2 cells compared with the parental ADRr line (Fig. 2E). Despite this, elevation in ErbB3 activation MM-111 displayed far greater potency and percentage of inhibition of p-ErbB3 in the ADRr-E2 cell line relative to the parent ADRr cells showing the specificity of MM-111 for the ErbB2/ErbB3 oncogenic unit in tumor cells overexpressing ErbB2 receptors (Fig. 2F).

Potent inhibition of ErbB3 phosphorylation was achieved in BT-474-M3 (IC₅₀ = 3 nmol/L) cells and an additional ErbB2-overexpressing breast tumor cell line, ZR75-30 (IC₅₀ = 5 nmol/L; Fig. 2G) following 1 hour incubation with MM-111. In addition, we observed inhibition by MM-111 of p-AKT in BT-474-M3 (IC₅₀ = 10 nmol/L) and in ZR75-30 (IC₅₀ = 4 nmol/L; Fig. 2G). We found that the ability of MM-111 to inhibit heregulin-induced ErbB3 activation was superior to lapatinib and pertuzumab (Fig. 2G) and the relative IC₅₀ value for each inhibitor was consistent following up to 24 hours incubation with inhibitors, indicating treatment times had little effect on the potency of the inhibitors (Fig. 2G). Despite moderately inhibiting ErbB3 phosphorylation, pertuzumab did not effectively inhibit AKT whereas lapatinib was a poor inhibitor (Fig. 2G). Consistent with our earlier observations (Fig. 1A), trastuzumab was found to be an ineffective inhibitor of ligand-induced ErbB3 and AKT phosphorylation following 1- and 24-hour treatment in both cell lines tested (data not shown).

MM-111 is a fusion protein with multiple components, including unnatural peptide linkers. Although proteolytic resistance was a criterion for selecting connector peptides for MM-111, we wished to confirm the stability and proteolytic resistance of MM-111 both in vitro and in circulation. First, we incubated MM-111 in serum at a predicted therapeutic dose of 100 nmol/L and investigated its stability over a 5-day time course. MM-111 retained its ability to bind both recombinant ErbB2 and ErbB3 when incubated in mouse (Fig. 3A) and human (Supplementary Fig. S4) serum at 37°C with similar activity for all time points compared with the 0 hour control. MM-111 also remained stable in circulation in mice with comparable serum levels of MM-111 measured by an HSA assay and an assay that measured active circulating levels of MM-111 that retain simultaneous interaction with both ErbB2 and ErbB3 (Fig. 3B). We also measured the serum levels of MM-111 in mice administered 5, 15, and 45 mg/kg of bispecific antibody. Pharmacokinetic data were subjected to noncompartmental analysis to estimate the terminal half-life. Nude mice dosed with 5, 15, 30, or 45 mg/kg had similar terminal half lives of 16.6, 16.2, 22.6, and 17.5 hours, respectively (Supplementary Table S1).

MM-111 efficacy in vivo was initially investigated in the BT-474-M3 breast tumor xenograft model. HSA was administered as a control at an equimolar concentration to MM-111. Statistical significance was observed between HSA and 30 and 3 mg/kg MM-111 treatment groups from days 8 and 14, respectively (Fig. 3C). The 0.3 mg/kg MM-111 treatment group was not significantly different from HSA treatment (Fig. 3C). To thoroughly investigate the relationship between MM-111 antitumor activity and ErbB2 expression levels, MM-111 was studied in a panel of 9 models (ADRr, ACHN, IGROV1, ZR-75-1, MDA-MB-361, ADRr-E2, Calu-3, NCI-N87, and SK-OV-3) expressing a range of ErbB2 from 4.0 × 10⁴ to 1.4 × 10⁶ receptors per cell (Table 1) that showed relative MM-111 activity was dependent on ErbB2 overexpression (Fig. 3D). The ADRr-E2 xenograft model of the ErbB2-overexpressing engineered cell line derived from ADRr cells responded well to MM-111 treatment whereas the parental ADRr xenografts did not respond to MM-111 (Fig. 3D). This observation in xenografts of ADRr-E2 transfectants is consistent with the inhibition of ErbB3 phosphorylation we observe in vitro (Fig. 2F).

The effect of MM-111 on the accumulation of BT-474 cells in G₁ phase and the concomitant decrease in S-phase of the cell cycle was examined. MM-111 modestly decreased the percentage of cells in S-phase by 9.5% and the population of cells in G₁ phase increased by 11% (Fig. 4A). We subsequently examined the ability of MM-111 to inhibit signaling molecules downstream of ErbB3 that regulate cell-cycle progression or cell death. MM-111 downregulated cell-cycle modulator cyclin D1 (Fig. 4B) and induced nuclear translocation of cell-cycle inhibitor p27 in BT-474-M3 cells following 72 hours of treatment (Fig. 4C). Annexin V staining of BT-474-M3 cells treated with MM-111 did not show an apoptotic effect (data not shown).

As MM-111 is a potent inhibitor of ligand-induced ErbB3 activation in ErbB2-overexpressing cancer cells, we hypothesized that its combination with ErbB2 inhibitors, trastuzumab or lapatinib, would have additive or synergistic effects on growth in tumors sensitive to both agents. First, we investigated the differential ability of MM-111, lapatinib, and trastuzumab to inhibit cell proliferation in the presence of heregulin. Under basal conditions, we found that lapatinib, trastuzumab, and MM-111 maximally inhibit BT-474-M3 cell proliferation by 50%, 32%, and 24%, respectively (Fig. 5A). When cells are cultured in the presence of 5 nmol/L heregulin the effect of both lapatinib and trastuzumab is compromised, decreasing inhibition of cell proliferation to 23% and 9%, respectively (Fig. 5A). The inhibition of tumor cell growth by MM-111 is enhanced when heregulin is present, with 33% growth inhibition observed (Fig. 5A). This observation suggested that both lapatinib and trastuzumab may work additively in combination with an inhibitor of heregulin-driven signaling such as MM-111.

We next investigated the ability of the combination of MM-111 and lapatinib or MM-111 and trastuzumab to inhibit AKT phosphorylation. While we found that lapatinib alone inhibited p-AKT in the presence of heregulin the combination of MM-111 and lapatinib was extremely effective, inhibiting p-AKT well below basal levels at therapeutically relevant concentrations (Fig. 5B). Trastuzumab does not inhibit heregulin-activated ErbB2/3 signaling (Fig. 1A; ref. 18). However, as we increased combination doses of MM-111 and trastuzumab we observed improved p-AKT inhibition to basal levels suggesting an additive effect of the combination (Fig. 5B).

The combination of MM-111 with trastuzumab or lapatinib was further investigated in vivo using the BT-474-M3 breast cancer xenograft model. Suboptimal monotherapy doses of MM-111 (3 mg/kg dosed every 3 days; Fig. 3B) and trastuzumab (1 mg/kg dosed weekly) were selected for combination experiments to allow observation of any differences in activity between monotherapy and combination groups. Tumor growth inhibition in groups dosed with the combination of 3 mg/kg MM-111 and 1 mg/kg trastuzumab was more potent than the monotherapy-treated groups and on day 17 posttumor implantation reached statistical significance compared with MM-111 alone (P = 1.4 × 10⁻⁵) and trastuzumab alone (P = 1.4 × 10⁻⁷). MM-111 and lapatinib were each dosed at an optimal efficacious dose weekly and everyday, respectively. The combination of MM-111 and lapatinib provided more potency than either drug alone reaching statistical significance to MM-111 (P = 3.9 × 10⁻⁴) and lapatinib (P = 5.1 × 10⁻³) on day 13 (Fig. 5C).

The combination of trastuzumab and pertuzumab has been shown to be synergistic (25, 26). We compared the combination of MM-111 and trastuzumab to pertuzumab plus trastuzumab and found that the MM-111 combination was superior at inhibiting proliferation of both BT-474-M3 and NCI-N87 tumor cells (Fig. 5D). In addition, activity of MM-111 as a single agent was similar to trastuzumab and superior to pertuzumab in these models (Fig. 5D).

ErbB3 has been identified as a key node in the ligand-activated ErbB receptor PI3K signaling axis (16). Unlike ErbB1 and ErbB2, ErbB3 does not require adaptor proteins to interact with the regulatory p85 subunit of PI3K, instead interacting directly through up to 6 potential phosphotyrosines. Consequently, ErbB3 activation proficiently triggers PI3K signaling and activation of this pathway is frequently observed in tumors that escape from currently marketed ErbB-directed inhibitors (4, 6, 8, 27), as well as hormonal therapy (28, 29) and chemotherapy (30, 31). Thus, inhibition of ErbB3 activation by heregulin is a worthy therapeutic strategy that is being actively pursued (16, 17). However, in ErbB2-overexpressing tumors ErbB2/3 heterodimers preform in the absence of ligand to create a primed, picomolar-affinity binding site for heregulin that is several 100-fold higher in affinity than the ErbB3/heregulin interaction (32) and challenging for a subnanomolar affinity inhibitor to impede. MM-111 is designed to deliver ErbB3 inhibition through docking to the overexpressed ErbB2 receptor and overcomes the high affinity ErbB2/3/heregulin complex through enhanced apparent affinity for ErbB3 as a result of binding avidity. Computational modeling predicts that the bispecific design provides close to 100% occupancy of ErbB3 by MM-111 at therapeutically relevant doses in cells that overexpress ErbB2.

At nonamplified levels of ErbB2, ErbB3 occupancy by MM-111 is greatly reduced indicating that the bispecifc design also promotes tumor selectivity. Robinson and colleagues (33) have shown the superior tumor selectivity of a bispecific antibody with ALM, an ErbB2/ErbB3-bispecific antibody generated by linking an anti-ErbB2 scFv and anti-ErbB3 scFv with a synthetic peptide linker. Tumor-specific inhibition of ErbB2-targeted agents is desirable as ErbB2 plays a role in ensuring survival and growth of cardiomyocytes (34). Trastuzumab treatment has been shown to result in cardiac dysfunction due to left ventricular ejection fraction decline that becomes more prevalent with coadministered anthracyclines (35). Similar though less severe toxicities have occurred with lapatinib (36). We anticipate that the high specificity of MM-111 for tumor cells coexpressing ErbB2 and ErbB3 may prevent target-related toxicity in the clinic. To further mitigate against cardiotoxicity, we selected B1D2 as the anti-ErbB2 scFv moiety of MM-111 because it does not inhibit ErbB2 signaling (37).

The ErbB2/ErbB3-bispecific antibody, ALM, has shown antiproliferative activity on tumor cells in vitro (33) but with a molecular weight of approximately 55 kDa this protein is expected to have a short circulating half-life due to rapid renal clearance and thus is unlikely to achieve serum levels required for activity in patients. Others have found that using HSA as a linker between 2 scFvs can extend serum half-life significantly (38). We found that incorporating HSA between the H3 and B1D2 scFvs extended the serum half-life of MM-111 to 16 to 20 hours in mice compared with approximately 5 hours typically observed for tandem scFvs reported in the literature (39). In cynomolgus monkeys a circulating half-life of MM-111 of up to 99 hours was achieved (Supplementary Table S1), and we anticipate that MM-111 half lives in patients may exceed those observed in monkey as species compatibility is achieved.

The partition of signaling and kinase function to ErbB3 and ErbB2, respectively, presents significant challenges to inhibiting the ErbB2/ErbB3 heterodimer. Trastuzumab blocks ligand independent ErbB2/ErbB3 activation (18), but we show here that trastuzumab is an ineffective inhibitor of heregulin-activated ErbB2/3 signaling. Thus, when autocrine or paracrine heregulin is available tumor cells may continue to thrive under trastuzumab therapy. Indeed, heregulin expression is observed in the majority of primary breast tumors (40, 41) and is upregulated in preclinical ErbB2-overexpressing models following trastuzumab treatment (8, 42). Pertuzumab binds to the dimerization domain of ErbB2, blocking the formation of ligand-induced ErbB2/ErbB3 heterodimers, and a combination of pertuzumab and trastuzumab synergistically inhibits the survival of breast cancer cells (15, 25, 26). Although some of the activity of both trastuzumab and pertuzumab is mediated through antibody-dependent cell-mediated cytotoxicity, the synergistic activity of the combination is attributed to the complementary pathway inhibitory mechanisms of each antibody rather than their ability to recruit effector cells (26). The combination of trastuzumab and pertuzumab has achieved success in the clinic. A phase II trial in patients with Her2-positive metastatic breast cancer who had previously progressed on trastuzumab therapy achieved a 50% clinical benefit rate and 24.2% objective response rate (43). We show here that MM-111 inhibition of ligand-activated ErbB3 phosphorylation is superior to pertuzumab and the combination of MM-111 and trastuzumab is more effective at inhibiting tumor cell growth than pertuzumab plus trastuzumab. Pertuzumab indirectly inhibits ErbB3 activation by precluding ErbB2 dimerization whereas MM-111 binds directly to ErbB3 and blocks heregulin. As ErbB2 is expressed at an order of magnitude or greater than ErbB3, even a high percentage of ErbB2 inhibition still leaves enough free receptors to form active ErbB2/3 heterodimers. Ironically, this overexpression of ErbB2 in tumors makes it more difficult to inhibit with ErbB2-directed therapies in tumors relative to normal tissues. MM-111 on the other hand is potentiated by ErbB2 overexpression making this an attractive approach to tumor specific inhibition of receptor tyrosine kinases.

We found the addition of MM-111 to trastuzumab was very effective at inhibiting tumor growth, observing a significant combination effect in the BT-474 breast tumor model in vitro and in vivo, and it is possible that the combination of MM-111 and trastuzumab could offer benefit to Her2-positive patients whose tumors escape trastuzumab therapy through heregulin activation of ErbB3. Lapatinib reversibly inhibits the kinase activity of ErbB1 and ErbB2 and while lapatinib, unlike trastuzumab, can perturb heregulin-activated ErbB3 phosphorylation, MM-111 is far more effective and the activity of lapatinib is attenuated by heregulin. ErbB3 upregulation in lapatinib-treated patients (19) indicates that concurrent ErbB3 inhibition is desirable and we found that addition of MM-111 to lapatinib increased p-AKT suppression and weekly MM-111 added to daily lapatinib administration improved tumor growth inhibition in vivo. Interestingly, we did not observe an improvement in antitumor activity in xenografts when daily lapatinib dosing was combined with MM-111 dosed every 3 days suggesting that the dosing schedule or sequence of administration of these 2 agents may be critical for activity and warrants further investigation.

In conclusion, we have used computational modeling and cell signaling insights to develop an engineered antibody fusion molecule, MM-111, that potently and specifically inhibits ErbB3 signaling in ErbB2-positive tumors by exploiting the overexpression of ErbB2 through incorporation of a bispecific design. As the critical role of ErbB3/heregulin activation in ErbB2-positive refractory disease unravels the stage is set for the clinical development of MM-111, a specialized ErbB3 inhibitor that can act in concert with ErbB2 therapies to deter resistance or restore sensitivity.

All authors were employees of Merrimack Pharmaceuticals with ownership interests in the company at the time this work was undertaken.

The authors thank Gege Tan and Maja Razlog for excellent technical assistance and Drs. James Marks, Louis Weiner, and Clet Niyikiza for helpful discussions.

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.