Abstract
In recent years, HER3 has increasingly been implicated in the progression of a variety of tumor types and in acquired resistance to EGFR and HER2 therapies. Whereas EGFR and HER2 primarily signal through the MAPK pathway, HER3, as a heterodimer with EGFR or HER2, potently activates the PI3K pathway. Despite its critical role, previous attempts to target HER3 with neutralizing antibodies have shown disappointing efficacy in the clinic, most likely due to suboptimal and indirect mechanisms of action that fail to completely block heterodimerization; for example, tumors can escape inhibition of ligand binding by upregulating ligand-independent mechanisms of HER3 activation. We therefore developed 10D1F, a picomolar affinity, highly specific anti-HER3 neutralizing antibody that binds the HER3 heterodimerization interface, a region that was hitherto challenging to raise antibodies against. We demonstrate that 10D1F potently inhibits both EGFR:HER3 and HER2:HER3 heterodimerization to durably suppress activation of the PI3K pathway in a broad panel of tumor models. Even as a monotherapy, 10D1F shows superior inhibition of tumor growth in the same cell lines both in vitro and in mouse xenograft experiments, when compared with other classes of anti-HER3 antibodies. This includes models demonstrating ligand-independent activation of heterodimerization as well as constitutively activating mutations in the MAPK pathway. Possessing favorable pharmacokinetic and toxicologic profiles, 10D1F uniquely represents a new class of anti-HER3 neutralizing antibodies with a novel mechanism of action that offers significant potential for broad clinical benefit.
10D1F is a novel anti-HER3 antibody that uniquely binds the receptor dimerization interface to block ligand-dependent and independent heterodimerization with EGFR/HER2 and thus more potently inhibits tumor growth than existing anti-HER3 antibodies.
This article is featured in Highlights of This Issue, p. 323
Introduction
Members of the EGFR family (EGFR/HER family) are commonly implicated in the formation and progression of many tumor types and therefore represent attractive targets for therapeutic intervention (1). Despite the relative clinical success of small-molecule tyrosine kinase inhibitors (TKI), the anti-EGFR antibody cetuximab, and the anti-HER2 antibodies pertuzumab and trastuzumab, tumors frequently develop resistance and patients relapse. HER3 has emerged as a central player in both tumor progression and acquired resistance to EGFR and HER2-targeted therapies, as aberrant expression and/or activation of HER3 and its ligand NRG1 is associated with poor responses and low survival rates in multiple indications (2, 3). Although critical in early development, HER3 is expressed at only low levels in adult tissues such as skin and colon, yet is commonly activated in many cancers, most notably breast, colorectal, and gastric cancer (4). Furthermore, knockdown of HER3 in representative cancer models has been shown to inhibit proliferation and reduce tumor growth, even in cells that are resistant to TKIs (5).
The HER family signal through the PI3K/AKT/mTOR and the MAPK/ERKK pathways to promote cell survival and proliferation (1, 6, 7). However, whereas HER2 and EGFR primarily activate the MAPK/ERKK pathway, HER3 potently activates the PI3K/AKT/mTOR pathway (8–10), as its intracellular domain contains multiple docking sites for the regulatory p85 subunit of PI3K (11). HER3 lacks kinase activity and does not form stable homodimers; therefore, HER3 must be transphosphorylated by binding to a kinase-active heterodimer partner (commonly EGFR or HER2) for signal transduction to take place (12, 13). The HER3 extracellular domain exists in a reversible equilibrium between a “closed” inactive conformation and an “open” active conformation, in which the dimerization arm within domain II is exposed to allow dimerization along the domain II dimerization interface, and in particular through the cysteine-rich CR1 region (Fig. 1; refs. 14–21). HER3 is “activated” when the equilibrium is shifted in favor of the open conformation, increasing the probability of forming active heterodimers. The conventional model for activation is ligand-dependent, that is, the equilibrium shifts when HER3 in the open conformation is stabilized by binding of the NRG1 ligand (Fig. 1A, left). However, any dimerization partner at sufficient concentration will also shift the equilibrium as it binds to and stabilizes HER3 that is transiently in the open conformation, known as ligand-independent activation (Fig. 1A, right; refs. 19–21).
Multiple mechanisms contribute to the increased HER3 activation associated with resistance to EGFR or HER2 directed therapy, including (i) transcriptional upregulation of HER3 (22, 23); (ii) increased levels of NRG1 (24, 25); and (iii) HER2 amplification (26, 27). More recently, oncogenic fusions of SLC3A2, CD74, or VAMP2 to NRG1 isoforms have been identified in patients with lung-invasive mucinous adenocarcinoma (28, 29). These fusions promote paracrine secretion of the EGF-like domain of NRG1 and increased HER3 activation. Furthermore, efforts to target constitutively activating mutations in the downstream signaling cascades, such as BRAF V600E, which confer TKI resistance in thyroid and colon carcinomas (30–32), may fail through HER3 activation. Notably, activation of HER3 by NRG1 promotes resistance to the specific BRAF V600E inhibitor vemurafenib (33, 34).
In support of the role of HER3 in drug resistance, anti-HER3 antibodies restore sensitivity to vemurafenib in BRAF-V600E–mutant colon cancer (34) and blockade of HER2:HER3 signaling with combination therapy overcomes trastuzumab resistance in HER2-positive breast cancer (35). HER3, therefore, represents a promising therapeutic target for the treatment of HER3-driven tumors and HER therapy–resistant tumors.
Although embryonic lethal in knockout mice (36), HER3 inhibition has been shown to be broadly safe in the clinic, with only low-grade toxicity observed, such as skin rash and gastric complications, consistent with HER3′s low expression in normal adult tissues (37). Unfortunately, previous anti-HER3 antibodies have exhibited only limited clinical efficacy. First-generation HER3-targeted antibodies (e.g., seribantumab), developed to block the NRG1-binding region (class 1), did not address ligand-independent heterodimer formation (Fig. 1B) (38). Furthermore, second-generation antibodies (e.g., elgemtumab), developed to lock HER3 in the closed conformation (class 2; ref. 39), were hampered by conformation-dependent binding that may have limited their potency.
To overcome these limitations, we developed 10D1F, a humanized IgG1 antibody selected for its ability to inhibit HER3 by binding an epitope on the heterodimerization interface, accessible irrespective of receptor conformation (Fig. 1B). We demonstrate that 10D1F harnesses this novel mechanism of action to inhibit ligand-dependent and independent HER3-driven tumor growth more potently than other HER3 therapies, suggesting significant potential for clinical benefit.
Materials and Methods
Antibody isolation
Six- to 8-week-old female BALB/c mice were repeatedly immunized with antigenic peptide, recombinant target protein, or cells expressing the target protein. Twenty-four hours after the final immunization, total splenocytes were isolated and fused with the myeloma cell line P3X63.Ag8.65 (ATCC) using ClonaCell-HY Hybridoma Cloning Kit, in accordance with the manufacturer's instructions (Stemcell Technologies). After 7 to 10 days, single hybridoma clones were isolated and antibody-producing hybridomas were selected by screening supernatants for antigen binding using ELISA and flow cytometry. Variable regions of selected clones were sequenced and expressed in CHO cells for testing (Supplementary Methods). One clone, 10D1P, was selected for development and subsequently humanized and further affinity matured (Supplementary Methods) into the final antibody 10D1F (Clone 10D1_c89, PCT/EP2019/058035; WO/2019/185878; ref. 40).
Antibody production
For production of all antibodies, see supplemental methods.
Antibody affinity
All proteins were from Sino Biological. Antibody affinity was calculated using Bio-Layer Interferometry on an Octet QK384 (ForteBio). First, anti-human IgG capture (AHC) sensors (ForteBio) were loaded with anti-HER3 IgG antibodies (25 nmol/L). Kinetic measurements were performed in the absence or presence of NRG1. NRG1 was used at 1:1 molar ratio with HER3 wherein the complex was allowed to form at room temperature for 2 hours. His-tagged human HER3 or HER3–NRG1 complexes were loaded to antibody coated AHC sensors at different concentrations for 120 seconds, followed by a 120-second dissociation time. All measurements were performed at 25°C with agitation at 1,000 rpm. Sensorgrams were referenced for buffer effects and then analyzed using the Octet QK384 -software (ForteBio). Kinetic responses were globally fitted using a one-site binding model to obtain values for association (Kon), dissociation (Koff) rate constants and the equilibrium dissociation constant (KD).
ELISA
All proteins were from Sino Biological. Antibodies were analyzed for binding to recombinant human HER3 ectodomain, as well as mouse, rat and cynomologus homologues of HER3 and human EGFR and human HER2. ELISAs were carried out according to standard protocols. Ninety-six–well Maxisorp plates (Nunc) were coated with 1 μg/mL of target antigen in PBS for 16 hours at 4°C. After blocking for 1 hour with 1% BSA in TBS at room temperature, anti-HER3 antibodies were serially diluted and added to the plate. After 1-hour incubation at room temperature, plates were washed three times with TBS containing 0.05% Tween 20 (TBS-T) and were then incubated with goat anti-human IgG Fc-HRP (Abcam, #ab97225) for 1 hour at room temperature. After washing, plates were developed with colorimetric detection substrate 3,3′,5,5′-tetramethylbenzidine (Turbo-TMB; Pierce) for 10 minutes. The reaction was stopped with 2 mol/L H2SO4, and OD was measured at 450 nm on a BioTek Synergy HT.
Inhibition of dimerization
All proteins were from Sino Biological. Maxisorp plates (Nunc) were coated with 1 μg/mL of HER2-Fc or EGFR-Fc antigen diluted in PBS for 16 hours at 4°C. Following overnight incubation, plates were washed three times with washing buffer and blocked with 1% BSA in TBS at room temperature for 1 hour. In a separate plate, samples were prepared by preincubating for 1 hour at room temperature serially diluted anti-HER3 antibodies with 3 μg/mL of human HER3-His antigen (EC50 of HER3-His binding to HER2-Fc). NRG1 (0.1 μg/mL) was also added to the samples (EC80 of HER3-His binding with NRG1). After blocking, plates were washed twice with washing buffer and incubated with samples for 1 hour at room temperature; plates were washed three times and further incubated with anti-his HRP antibody (Abcam, #ab97225) for 1 hour at room temperature and developed using standard ELISA protocol.
Flow cytometry
Wild-type HEK293T cells (Supplementary Methods), which do not express HER3, were transiently transfected with HER3 cDNA expression plasmid (Sino Biological #HG10201-UT) using lipofectamine 2000 (Thermo Fisher Scientific) following the manufacturer's protocol. Twenty-four hours posttransfection, cells were harvested and used for analysis. Cells were incubated with 10 μg/mL of antibodies at 4°C for 1.5 hours. Cells were washed three times with FACS buffer (PBS with 5 mmol/L EDTA and 0.5% BSA) and resuspended in FITC-conjugated anti-Fc antibody (Thermo Fisher Scientific) for 40 minutes at 2°C to 8°C. Cells were washed again and resuspended in 200 μL of FACS flow buffer (PBS with 5 mmol/L EDTA) for flow cytometric analysis using MACSQuant 10 (Miltenyi Biotec). After acquisition, all raw data were analyzed using Flowlogic software. Cells were gated using forward and side scatter, and the profile on percentage of positive cells was determined.
Epitope binning and mapping
For epitope binning, human HER3-His protein (Sino Biological) in PBS was immobilized to Anti-Penta His sensor (HIS1K, Fortebio) on an Octet QK384 (Fortebio), for 5 minutes. Baseline signals were measured for 30 seconds before loading 400 nmol/L saturating antibody in PBS for 10 minutes at a shake speed of 1,000 rpm. Subsequently, biosensors were immersed in 300 nmol/L competing antibody in PBS for 5 minutes at a shake speed of 1,000 rpm. Association of antibodies was monitored on the sensorgram at each step.
A peptide-based epitope mapping study was also conducted to identify the potential binding region of the antibody (Supplementary Methods).
In vitro tumor growth assays
Cell lines (Supplementary Methods) were treated with 10-point serially diluted concentrations of therapeutic antibodies as indicated. Cell viability was measured using CCK-8 assay (Dojindo) at 3 to 5 days posttreatment. The percentage of cell inhibition is shown relative to cells treated with buffer only (PBS). Data points represent the average of three replicates. IC50s were calculated by plotting percent inhibition as a function of antibody concentration and fitting the data points to a four-parameter logistic model.
HER3 phosphorylation and pathway activation
Cell lines (Supplementary Methods) were seeded in 6-well plates with 10% serum O/N at 37°C, 5% CO2. Cells were starved with 0.2% FBS culture medium for 16 hours, followed by treatment with different antibodies at the IC50 determined by in vitro tumor growth assays for 0.5 or 4 hours. Before harvesting, cells were stimulated with 100 ng/mL of NRG1 (Sino Biological). Protein was harvested in RIPA lysis buffer (Thermo Fisher Scientific) according to the manufacturer's protocol and quantified by Bradford assay. Protein samples (50 μg) were fractionated by SDS-PAGE and transferred to nitrocellulose membrane using Bio-Rad semidry transfer cell. Membranes were blocked with 5% BSA in TBS and immunoblotted with the indicated antibodies, with either anti-mouse IgG-HRP secondary antibody (Lifetech #A24512) or anti-rabbit IgG-HRP secondary antibody (Cell Signaling Technology #7074S). Additional details in are provided in Supplementary Methods. Blots were visualized with Bio-Rad Clarity Western ECL substrate and a Syngene Gel Doc (Thermo Fisher Scientific). Blots were quantified using densiometric analysis and normalized to β-actin.
Animal experiments
All animals were purchased from InVivos, housed under specific pathogen-free conditions, and treated in strict compliance with the Institutional Animal Care and Use Committee guidelines.
In vivo tumor growth assays
Tumor xenografts were established by subcutaneous injection in the right flank of NCr nude or NPG mice, approximately 6 weeks old with 1 × 106 tumor cells (except A549 cells, where 5 × 106 cells were used). Tumor volume was measured using calipers (Supplementary Methods). Once mean tumor volume reached 100 to 150 mm3, mice were treated intraperitoneally with 25 mg/kg of either 10D1, elgemtumab, cetuximab, trastuzumab, or vehicle control (PBS) as indicated.
Pharmacokinetics
Single-dose pharmacokinetic profiles were analyzed in 6- to 8-week-old female NCr nude mice or female Sprague–Dawley rats (SD rats). 10D1F was administered in a single dose at the indicated concentration via tail vein slow intravenous injection (2 minutes) or intraperitoneal injection in SD rats and NCr nude mice, respectively. Vehicle (PBS) was administered as a negative control. Blood was obtained at baseline (−24 hours), 6, 24, 96, 168, and 336 hours after administration. Antibody in the serum was quantified by ELISA using HER3-His (Sino Biological) coated plates and anti-human IgG Fc-HRP (Abcam #ab97225). The parameters for the pharmacokinetic analysis were derived from a noncompartmental model: maximum concentration (Cmax), AUC (0–336 hours), AUC (0–infinity), half-life (t½), clearance (CL), volume of distribution at steady state (Vss).
Safety and toxicity
Six- to 8-week-old female BALB/c mice and SD rats were injected intraperitoneally or via slow (2 minutes) intravenous injection, respectively, with a single dose of 10D1F at the indicated concentrations. Blood samples were obtained at −24 and 96 hours postinjection for mice and at 24, 6, 24, 96, 168, and 336 hours for the rats. Biochemistry parameters were analyzed using Vetscan VS2 chemistry analyzer using Abaxis Comprehensive Diagnostic Profile rotor and hematologic analysis was performed using Vetscan HM5 analyzer. Gross necropsy (lesions, organ abnormalities, ascites, and hemorrhage) was performed at the end of the study.
IHC
Arrays of frozen normal and malignant human tissue cryosections were from US Biomax (FMC282d). Slides were dried in a desiccator for 10 minutes and fixed in 100% acetone for 10 minutes at room temperature. Endogenous peroxidase was blocked with 3% (v/v) H2O2 for 15 minutes at room temperature. Slides were then blocked with 10% goat serum for 30 minutes at room temperature, and incubated with 0.124 mg/mL (827 nmol/L) primary antibody (10D1F in a mouse IgG2a backbone) overnight at 4°C and HRP polymer–conjugated goat anti-mouse secondary antibody for 30 minutes at room temperature. Slides were developed using Bond Mixed DAB Refine for 5 minutes at room temperature, followed by rinse in DI H2O to stop the reaction. Following IHC, slides were counterstained with hematoxylin for 5 minutes at room temperature, rinsed, dehydrated, and mounted in synthetic mounting media. Sections were scanned using a Leica SCN scanner at ×20 magnification.
Results
Existing anti-HER3 neutralizing antibodies attempt to inhibit ligand binding (class 1, e.g., seribantumab) or trap the receptor in an inactive conformation (class 2, e.g., elgemtumab) to prevent heterodimerization of HER3 with its binding partners (Supplementary Table S1). However, these fail to completely suppress HER3 activation as both modes of HER3 activation, ligand-dependent and ligand independent, are not fully blocked (Fig. 1A). Furthermore, although many of these are IgG1 antibodies and thus capable of inducing depletion of tumor cells through antibody-dependent cellular cytotoxicity (ADCC), this has been insufficient to differentiate their clinical efficacy. We reasoned that a more effective strategy to abrogate pathway activation would be the prevention of all HER3 heterodimerization using a novel molecular mechanism of action that would directly block the dimerization interface of HER3, irrespective of conformational state.
Development of an anti-HER3 antibody against the HER3 dimerization interface
Raising antibodies that bind to the heterodimerization interface of HER3 is challenging due to the restricted surface area and the low immunogenicity of this region. Therefore, computational structural analysis and modeling were used to predict HER3 specific epitopes within the dimerization interface that would be accessible in both conformations and conserved across mammalian orthologs. The selected target region was located on the dimerization interface of HER3 subdomain II and spatially separated from the ligand-binding site located between subdomains I and III (Fig. 1B).
mAbs against the target region were raised by immunization of mice and subsequent isolation of hybridoma clones using Hummingbird's Rational Antibody Discovery Platform. One clone, 10D1P, was selected for further development based on preliminary binding affinity (Kd 52.6 nmol/L), specificity within the EGFR family and species cross-reactivity to HER3 orthologs (Supplementary Fig. S1). 10D1P was subsequently humanized and affinity matured (Supplementary Methods). The final antibody, 10D1F, was selected from among the optimized variants based on its in vitro physicochemical and functional properties, including increased binding affinity, thermal stability (>70°C), and low aggregation potential after freeze-thaw (below and Fig. 2).
10D1F binds HER3 with high affinity and specificity
10D1F demonstrated high affinity to human HER3 (Kd of ∼1 pmol/L), a >1,000-fold improvement over the parental antibody, and no cross-reactivity with EGFR or HER2 (Fig. 2A and B). There was no difference observed for the binding affinity in the presence or absence of NRG1, that is, to a more open or closed conformation of HER3. This contrasted to that observed with examples of ligand blockers (class 1) and conformational change blockers (class 2) where binding for both was decreased in the presence of NRG1 (Supplementary Fig. S1E–S1F). Consistently, 10D1F bound with high specificity to HER3-overexpressing HEK293T cells, but not wild-type cells (Fig. 2C). Moreover, although there was a 10-fold difference in affinity between human/cyno and rodent HER3 for the parental antibody, 10D1F exhibited identical binding to all HER3 orthologs (Fig. 2D).
10D1F binds to a unique epitope in the dimerization interface of HER3, distinct from other anti-HER3 antibodies
To verify the binding site of 10D1F, we conducted an epitope binning experiment using class 1 (seribantumab) and class 2 (elgemtumab) anti-HER3 antibodies. The epitope of the seribantumab has previously been mapped to domain I of HER3, within the ligand-binding site, whereas the epitope of elgemtumab has been mapped to a discontinuous epitope across the interaction surface of HER3 subdomains II and IV, hypothesized to lock the HER3 in an inactive conformation (Figs. 1B and 3). 10D1F did not compete with class 1 or class 2 antibodies for binding to human HER3 ectodomain (Fig. 3A).
To further elucidate the 10D1F-binding region, a peptide-based epitope mapping study was conducted (Supplementary Methods). This identified the consensus sequence CFGPNPNQCC to have the strongest interaction with 10D1F, and thus likely to contain multiple residues that constitute the 10D1F epitope (Supplementary Fig. S2). This sequence maps to the desired target region in the dimerization interface (Fig. 3C). It should be noted, however, that the observed interaction with the linear peptides was weak, suggesting that 10D1F binds a discontinuous epitope involving additional residues outside (yet topologically close) of this sequence.
Together, these results demonstrate that 10D1F binds a novel epitope on the domain II dimerization interface, topologically distinct from the epitopes of other anti-HER3 antibodies.
10D1F inhibits HER3 heterodimerization
To assess the ability of 10D1F to prevent HER3 heterodimerization with EGFR and HER2, we conducted plate-based in vitro dimerization assays. 10D1F inhibited HER3 dimerization with EGFR (Fig. 3C) and HER2 (Fig. 3D) in a dose-dependent manner, with significantly higher potency than the anti-HER3 class 1 and class 2 antibodies. Notably, the anti-HER3 class 2 antibody did not cause any significant inhibition of HER3 heterodimerization.
10D1F demonstrates superior inhibition of proliferation in a panel of cancer cell lines
Given the unique binding region and resulting superior inhibition of heterodimerization, it was hypothesized that 10D1F would show more potent inhibition of HER3-mediated cell growth and survival pathways than other anti-HER3 antibodies. To this end, we measured the effect of 10D1F on the proliferation of a panel of cancer cell lines (gastric: N87, SNU16; SCCHN: FaDu; lung: A549, HCC95; kidney: ACHN; ovarian: OvCar8; thyroid: BCPAP). Although the genetic background and expression profiles of these cell lines are unique, they were selected as examples of cancer cell lines where HER3 activation of the PI3K pathway is likely to be important to growth and proliferation. Furthermore, based on mRNA and protein expression data, these cell lines should reflect the main classes of alternative HER3 activation. We included cell lines with high NRG1, expected to show predominantly ligand-driven HER3 activation (FaDu, A549, HCC95), those with low NRG1 and high HER2/EGFR for which HER3 activation is expected to be driven by HER2/EGFR (N87, SNU16), and those where both HER2/EGFR and NRG1 are high and where HER3 activation could be driven by both mechanisms (OvCAR8, ACHN). In addition, one cell line harboring the BRAF mutation V600E and known to be resistant to BRAF inhibitors was selected (BCPAP), which also showed the highest EGFR, HER2, and HER3 protein expression (Supplementary Table S2). As a monotherapy, 10D1F effectively inhibited the proliferation of all cell line models in a dose-dependent manner. Furthermore, 10D1F inhibited proliferation more potently than anti-HER3 class 1 or class 2 antibodies as well as selected anti-EGFR or anti-HER2 antibodies (Fig. 4A; Supplementary Fig. S3A). Remarkably, 10D1F also potently inhibited the proliferation of the BRAF V600E–mutant line BCPAP (Fig. 4A; ref. 41).
10D1F inhibits HER3 phosphorylation and pathway activation
Given its effects on in vitro proliferation of cancer cell lines, we hypothesized that 10D1F would effectively inhibit dimerization-induced receptor phosphorylation and downstream signaling. To test this, we measured the levels of total and phosphorylated HER family receptors as well as downstream PI3K/AKT/mTOR pathway and MAPK/ERKK pathway intermediates by Western blot analysis in a subset of the cancer cell lines that responded to 10D1F previously (A549, N87, FaDu, and OvCar8). Cells were treated with 10D1F (or HER3 class 1 and class 2 antibodies) at the IC50 for each cell line, as determined during the proliferation inhibition assay, for 30 minutes (N87, FaDu, and OvCar8) or 4 hours (A459) prior to stimulating HER3 with NRG1. Consistent with the observed blockade of receptor heterodimerization, 10D1F completely abrogated phosphorylation of HER3 in FaDu and OvCar8, and substantially attenuated the phosphorylation in A549 and N87 (Fig. 4B; Supplementary Fig. S3B). In addition, 10D1F decreased phosphorylation of HER2 in A549, N87, and OvCar8, did not alter the levels of pan-HER3 in N87 and FaDu, and only slightly decreased pan-HER3 levels in A549 and OvCar8. Furthermore, 10D1F markedly reduced the levels of phosphorylated AKT in A549, N87, FaDu, and OvCar8, and phosphorylated mTOR in N87, FaDu, and OvCar8, indicating downstream inhibition of the PI3K/AKT/mTOR pathway. In all cell lines, 10D1F showed the greatest inhibition of cell pathways, in comparison with the HER3 class 1 or class 2 antibodies.
10D1F induces ADCC but does not cause significant HER3 internalization
Some anti-HER family antibodies may have other mechanisms of action beyond neutralization. Trastuzumab, for example, can deplete HER2-expressing cells by ADCC due to interaction of the antibody Fc domain with activatory Fcγ receptors on natural killer (NK) cells (42), and can also inhibit signaling indirectly by triggering internalization and degradation of the receptor, thus reducing the concentration of activated receptors at the cell surface (40).
Most anti-HER3 antibodies, such as elgemtumab, are capable of inducing ADCC as they are IgG1 isotype antibodies similar to 10D1F (Supplementary Table S1). Some, such as seribantumab, are IgG2 antibodies and are not expected to induce ADCC. This can be demonstrated in vitro using isolated NK cells cocultured with HER3-overexpressing cells. 10D1F and elgemtumab stimulated ADCC to a similar level, while seribantumab did not (Supplementary Fig. S4).
The ability of anti-HER3 antibodies to drive internalization of HER3 is less clear. To further investigate the mechanism of action of 10D1F, we evaluated whether 10D1F is internalized upon binding to cell-surface expressed HER3. We used a selection from our panel of cancer cell lines (HCC95, N87, OvCar8), and treated these cells with 10D1F, anti-HER2, or anti-HER3 class 1 and class 2 antibodies tagged with pH sensitive dye that only fluoresces after the antibody has been internalized and trafficked to the low-pH endosome (Supplementary Methods). Internalization was measured for 24 hours by live fluorescence microscopy (Supplementary Fig. S4). Whereas the anti-HER2 antibody was clearly internalized by N87 cells (high HER2 expression), only the class 2 anti-HER3 antibody was efficiently internalized, and by only one of the HER3-expressing lines, OvCar8. 10D1F was not internalized to detectable levels in any of the tested cancer cell lines. Conversely, all anti-HER3 antibodies were significantly internalized by a recombinant HER3-overexpressing cell line. Alongside the minimal effects on total HER3 protein observed by Western blot analysis, these results indicate that the downregulation of the PI3K/AKT/mTOR pathway observed with 10D1F was not the result of internalization of HER3 from the cell surface.
10D1F inhibits tumor growth in mouse xenograft models
The results of the in vitro proliferation inhibition assay suggested that 10D1F would be effective as a single agent across HER3-driven tumors. To test the monotherapy efficacy of 10D1F in vivo, we conducted tumor growth inhibition studies in murine cell-derived xenograft (CDX) models of tumors with high NRG1 (FaDu, A549), high HER2/EGFR (N87), or high NRG1 and HER2/EGFR (OvCar8). Female NCr nude mice were subcutaneously implanted and treated either once a week (FaDu, OvCar8) or twice a week (N87, A549) with vehicle control (PBS) or 25 mg/kg 10D1F and other anti-HER family antibodies. 10D1F outperformed both class 1 and class 2 anti-HER3 antibodies as well as other relevant anti-HER family antibodies in high NRG1 CDX models, demonstrating complete inhibition of tumor growth in A549, FaDu, and OvCar8 CDX models (Fig. 5). Remarkably, 10D1F also showed significant antitumor efficacy (64% TGI) in the high HER2/EGFR N87 model in contrast to the class I and class 2 anti-HER3 antibodies, which failed to show any effect. Further support for the broad efficacy of 10D1F's mechanism of blocking HER3 dimerization was seen in additional tumor models treated with the parental antibody, 10D1P, where potent efficacy was observed in very high NRG1 (HCC95), HER2/EGFR high (SNU16), or high HER2/EGFR and NRG1 (ACHN) models (Supplementary Fig. S5A). Of note, although the majority of CDX experiments used NCr Nude mice with competent NK cells and active ADCC, 10D1P was also tested in an NPG mouse background without NK cells and ADCC and demonstrated identical results (Supplementary Fig. S5A, bottom right). Furthermore, examining the tumors from treated mice (FaDu and OvCar8) confirmed that the PI3K pathway was indeed robustly and durably suppressed (Supplementary Fig. S5B, c.f. Fig. 4B).
10D1F exhibits favorable pharmacokinetic and safety profiles
To confirm the tumor specificity and evaluate the potential for off-target binding of 10D1F in healthy tissues, we examined the staining pattern of 10D1F to human malignant and normal tissue cryosections by IHC (Fig. 6). In agreement with the specificity noted in previous ELISAs, 10D1F exhibited preferential binding to cancerous tissue, with marginal or no cross-reactivity to healthy tissue.
Therapeutic antibodies must possess half-lives in plasma compatible with appropriate dosing regimens and must demonstrate minimal toxicity to normal tissues. To determine the pharmacokinetic parameters of 10D1F in rodents, NCr nude mice or SD rats were administered with a single dose (25 mg/kg, i.p.) or multiple doses (25, 50, 100, and 250 mg/kg, i.v.) of 10D1F, respectively, and antibody concentration in blood was measured at 0, 0.5, 6, 24, 96, 163, and 336 hours after administration. 10D1F exhibited a half-life of more than 10 days in both mice and rats (Fig. 6).
To examine the potential for adverse effects, acute dose toxicity studies were conducted. BALB/c mice were treated with increasing doses (25, 50, 100, and 250 mg/kg, i.p.) of 10D1F and hematology and biochemistry profiles were obtained 96 hours posttreatment. There were no signals of toxicity observed in the hematologic or biochemical parameters, nor behavioral or gross anatomic differences, at any of the doses tested (Supplementary Tables S3). To further explore the safety profile of 10D1F, SD rats were treated with the same doses, and the hematology and biochemistry profiles were recorded at 0, 6, 24, 96, 168, and 336 hours posttreatment. With the exception of a mild increase in ALP at the highest doses, which was not observed in the mice, there were again no behavioral or gross anatomic differences observed, and no significant toxicity signals noted for hematologic or biochemical parameters (Supplementary Table S3).
Discussion
There has been significant interest in HER3 as a therapeutic target to treat cancer progression and acquired resistance to MAPK inhibitors including EGFR and HER2-targeted antibodies and TKIs. Several antibody-based strategies have been pursued to suppress HER3-mediated signaling, which include (i) inhibiting heterodimerization between HER2 and HER3 (pertuzumab; ref. 43); (ii) blocking NRG1 binding (e.g., seribantumab; ref. 38); and (iii) trapping HER3 in the closed inactive conformation (e.g., elgemtumab; ref. 40). However, none of these strategies have been completely effective in inhibiting HER3-mediated signaling; pertuzumab does not address the promiscuity of HER3 to form heterodimers with EGFR, seribantumab fails to prevent ligand-independent dimerization, and elgemtumab does not efficiently trap HER3 in the inactive conformation. In contrast to these previous approaches, we hypothesized that directly blocking the heterodimerization interface of HER3 in all conformations would provide a broader and more efficacious mechanism for inhibiting HER3 activation, and prevent tumors from escaping by overexpressing a heterodimer partner and promoting ligand-independent heterodimerization, the leading hypothesis for why the ligand blocking (class 1) anti-HER3 antibodies failed (26, 27).
Here, we report an anti-HER3 antibody, 10D1F, that selectively binds with high affinity and specificity at a binding site on the domain II heterodimerization interface, which is unique and distant from previously described anti-HER3 antibodies and available in all conformations. 10D1F was found to inhibit HER3 heterodimerization with EGFR and HER2 more effectively than other anti-HER3 antibodies and demonstrated superior efficacy over other HER-targeted agents in inhibiting tumor growth of both in vitro and in vivo models of cancers with ligand-dependent and ligand independent HER3 activation. As little evidence of 10D1F internalization was observed, and identical efficacy of 10D1 was shown in a HER3 CDX model using a NSG-like mouse (with no ADCC), our data strongly support the hypothesis that the primary mechanism of action of 10D1F is the inhibition of HER3 heterodimerization, and that this is a more effective strategy than previous approaches to targeting HER3. Intriguingly, while some internalization of a class 2 HER3 antibody was seen in the OvCar8 cell line, other cell lines demonstrated minimal internalization, independent of class/binding site. Indeed, this inconsistent internalization has also been noted for the anti-HER3 antibody patritumab (the antibody component of the antibody–drug conjugate, U3-1402; ref. 44). The much lower tumor overexpression observed for HER3 compared with HER2 (Supplementary Table S2), and lack of significant and consistent internalization, casts doubt on HER3 as a reliable tumor target for delivery of a cytotoxic payload.
All previous attempts to inhibit HER3 have, at best, exerted only partial inhibition but failed to sustainably block downstream signaling. In contrast, we show that 10D1F either fully or substantially reduces phosphorylation of HER3 and AKT in NRG1, HER2/EGFR, and NRG1 + HER2/EGFR–driven cell line models without significantly affecting the levels of pan-HER3. These results show that effective inhibition of heterodimerization by 10D1F causes a more potent and sustained downregulation of the PI3K/AKT/mTOR signaling pathway, which in turn leads to superior efficacy of this antibody. Notably, 10D1F as a single agent potently inhibits the growth of BCPAP cells that carry the constitutively activating BRAF mutation V600E in the MAPK/ERKK pathway and are known to be resistant to BRAF V600E inhibition. These cells also express the highest levels of HER3, EGFR, and HER2. These findings suggest that activation of the PI3K/AKT/mTOR pathway by HER3 is a critical mechanism for driving cell proliferation in these cells and has important therapeutic implications as it suggests that 10D1F, either as monotherapy, or in combination with BRAF inhibitors, may be beneficial in patients with drug-resistant BRAF mutations.
Increased levels of the HER3 ligand NRG1 and genomic rearrangements involving NRG1 fusions are also associated with resistance to trastuzumab in breast cancer and certain types of lung carcinomas (45). 10D1F demonstrated potent efficacy in the NRG1 high models, including SCCHN (FaDu) and lung (A549, and HCC95). Notably, HCC95 was found to express the highest levels of NRG1 mRNA among a panel of 67 lung cancer cell lines, due to a gene amplification (45). Although HCC95 does not harbor an NRG1 fusion, it is thus a useful preclinical model to evaluate the efficacy of drugs in highly NRG1-driven cancers. Our in vitro tumor growth inhibition data clearly show that 10D1F is more effective than existing class 1 and class 2 anti-HER3 antibodies in HCC95 and other NRG1-driven models, suggesting additional clinical benefit over existing therapies.
Alongside its superior efficacy, 10D1F demonstrated favorable specificity, pharmacokinetic and toxicologic profiles, with no overt toxicities detected in hematology or biochemistry parameters in rodent models.
The data presented in this study positions 10D1F as a promising candidate therapeutic to treat patients with solid tumors driven by HER3 heterodimers. Of note, our data strongly suggest that BRAF V600E mutant, and tumors resistant to TKI, trastuzumab, and cetuximab may be suitable indications for clinical assessment.
Disclosure of Potential Conflicts of Interest
D. Thakkar is the principal scientist (Head of Discovery) at and has ownership interest (including patents) in Hummingbird Bioscience. Z. Wu has ownership interest (including patents) in Hummingbird Bioscience and is a coinventor of the patent for humanized anti-HER3 antibody. K.H. Paszkiewicz is the chief technology officer at and has ownership interest (including patents) in Hummingbird Bioscience. P.J. Ingram is the CEO at and has ownership interest (including patents) in Hummingbird Bioscience. J.D. Boyd-Kirkup is the chief scientific officer at and has ownership interest (including patents) in Hummingbird Bioscience. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: D. Thakkar, M.M. Taguiam, P.J. Ingram, J.D. Boyd-Kirkup
Development of methodology: D. Thakkar, M.M. Taguiam, P.J. Ingram, J.D. Boyd-Kirkup
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Thakkar, V. Sancenon, S. Guan
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Thakkar, V. Sancenon, M.M. Taguiam, S. Guan, E. Ng, K.H. Paszkiewicz, P.J. Ingram, J.D. Boyd-Kirkup
Writing, review, and/or revision of the manuscript: D. Thakkar, V. Sancenon, K.H. Paszkiewicz, P.J. Ingram, J.D. Boyd-Kirkup
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Thakkar, M.M. Taguiam, K.H. Paszkiewicz
Study supervision: D. Thakkar, P.J. Ingram, J.D. Boyd-Kirkup
Other (engineering of antibodies): Z. Wu
Acknowledgments
The authors would like to thank all members of the Hummingbird Bioscience team, especially Sabrina Ng, Shalini Paliwal, Shani Ajumal, Akila Sadasivam, Rahmat Hidayat, Raihanah Ayob, and Michelle Su, for their assistance during the development of the molecule and data acquisition for this study. The authors would also like to thank Raymond Price for assistance during the preparation of this manuscript. All work was fully funded by Hummingbird Bioscience.
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.