Research ArticleSystems Biology

Therapeutically Targeting ErbB3: A Key Node in Ligand-Induced Activation of the ErbB Receptor–PI3K Axis

See allHide authors and affiliations

Science Signaling  30 Jun 2009:
Vol. 2, Issue 77, pp. ra31
DOI: 10.1126/scisignal.2000352


The signaling network downstream of the ErbB family of receptors has been extensively targeted by cancer therapeutics; however, understanding the relative importance of the different components of the ErbB network is nontrivial. To explore the optimal way to therapeutically inhibit combinatorial, ligand-induced activation of the ErbB–phosphatidylinositol 3-kinase (PI3K) axis, we built a computational model of the ErbB signaling network that describes the most effective ErbB ligands, as well as known and previously unidentified ErbB inhibitors. Sensitivity analysis identified ErbB3 as the key node in response to ligands that can bind either ErbB3 or EGFR (epidermal growth factor receptor). We describe MM-121, a human monoclonal antibody that halts the growth of tumor xenografts in mice and, consistent with model-simulated inhibitor data, potently inhibits ErbB3 phosphorylation in a manner distinct from that of other ErbB-targeted therapies. MM-121, a previously unidentified anticancer therapeutic designed using a systems approach, promises to benefit patients with combinatorial, ligand-induced activation of the ErbB signaling network that are not effectively treated by current therapies targeting overexpressed or mutated oncogenes.


The ErbB/HER family of receptors activates intracellular pathways intricately linked to cellular events common to both embryonic development and cancer (13). Three abnormal mechanisms contribute to tumorigenic ErbB activity: overexpression of receptors, which is often due to gene amplification; constitutive activation due to mutations; and overexpression of ligands. ErbB2 (also known as HER2 or neu)–amplified tumors have been effectively treated with the monoclonal antibody (mAb) against ErbB2, trastuzumab, and with the dual ErbB1-ErbB2 tyrosine kinase inhibitor, lapatinib (4, 5). Likewise, tumors with mutationally activated ErbB1 (also called EGFR, for epidermal growth factor receptor) are treated with the tyrosine kinase inhibitors gefinitib and erlotinib (46). Nonetheless, clinical responses to this first generation of ErbB-targeted therapies have been limited to a small percentage of tumors with these characteristics and may not reflect the full potential of targeting this pathway.

The ErbB signaling network is a complex system comprising four receptors with distinct binding specificities for more than 11 polypeptide ligands (7). Ligand specificity dictates the combinatorial formation of ErbB receptor dimer combinations, which activate intracellular pathways in a dimer-specific manner (8, 9). For example, two members of the ErbB family, ErbB2 and ErbB3, are nonautonomous and signal only in ligand-induced heterodimeric complexes with other ErbB receptors. ErbB2, or the “orphaned receptor,” has no known high-affinity ligand, whereas ErbB3 is catalytically inactive (10). Despite the lack of autonomy, ligand-induced heterodimers containing ErbB2 and ErbB3 can generate potent cellular signals (11). To add to the complexity of the system, ErbB receptor trafficking events are ligand and dimer specific; different ligand-receptor dimer complexes internalize, and recycle or degrade, at different rates (12). One attempt to inhibit a component of the ligand-induced signaling through combinatorial activation of ErbB receptors has been with the antibody against ErbB1, cetuximab, which so far has shown limited efficacy as a single agent in the clinic (13, 14).

Mathematical models have helped elucidate a systems view of ErbB signaling network dynamics. Various computational models describing ErbB signaling have been published that differ in complexity and focus on different biological questions (1519). Early models focused on ErbB1 binding and internalization (20). In 1999, a model by Kholodenko and colleagues indicated that ErbB1 receptor phosphorylation kinetics are controlled by receptor-adaptor molecule interactions that mask phosphotyrosine sites on the receptor from phosphatases (15). The first generation of our large-scale ErbB model predicted that ligand-receptor binding affinity determines early signaling events by controlling initial receptor activity (17). We recently extended this model to include all major interactions between ErbB1, ErbB2, ErbB3, and ErbB4 and the mitogen-activated protein kinase (MAPK) and phosphatidylinositol 3-kinase (PI3K) signaling cascades stimulated by two ligands—epidermal growth factor (EGF) and heregulin 1–β (HRG1-β) (21)—to study the input-output behavior in greater detail. We believe that it represents the most detailed model of the ErbB signaling network to date that was trained on an extensive experimental data set describing lung cancer cell lines. Other computational models have focused on different regulatory events, including crosstalk between the MAPK and PI3K cascades, autocrine loops, and the relationship between receptor dimerization and internalization (16, 22, 23). In 2007, Birtwistle et al. published a computational model of the ErbB signaling network, describing how stimulation of all four ErbB receptors with two ligands—EGF (18) and HRG1-β—leads to activation of two downstream kinases, ERK (extracellular signal–regulated kinase) and AKT [also known as protein kinase B (PKB)]. They found that ErbB2 overexpression transforms transient EGF-induced signaling into sustained signaling and analyzed the crosstalk between the PI3K and MAPK cascades (18).

We previously discussed the idea of applying computational models of biochemical signal transduction pathways to identify drug targets and of using simulations to optimize therapeutics that target complex signaling networks (24). Here, we show that betacellulin (BTC) and HRG1-β are potent inducers of AKT signaling in most tested cell lines obtained from solid tumors. Additionally, we found that these ligands were widely present in tumors obtained from patients. To identify the optimal manner with which to therapeutically inhibit combinatorial, ligand-induced activation of the ErbB receptor–PI3K axis by HRG1-β and EGFR-binding ligands, we used insights gained from a larger computational model (25) to develop and calibrate a simplified model incorporating the ErbB inhibitors cetuximab, lapatinib, and pertuzumab. Sensitivity analysis of the latter model revealed the dominant role of the ErbB3 receptor in AKT activation, suggesting that targeting this key node of the ErbB signaling network may result in therapeutic benefit to cancer patients. We simulated and compared the effects on ErbB3 and AKT phosphorylation of ErbB3 antagonists and other therapeutic agents targeting the ErbB receptor signaling network. Our model predicted that an ErbB3 antagonist would inhibit combinatorial, ligand-induced activation of the ErbB-PI3K network more potently than do current marketed therapeutics. To verify this prediction and explore the therapeutic potential of inhibiting ErbB3, we developed MM-121, a fully human mAb that binds specifically to ErbB3, blocks HRG1-β binding to ErbB3, and inhibits HRG1-β– and BTC-induced AKT signaling. Moreover, MM-121 inhibited multicellular spheroid growth and tumor growth in mouse tumor xenografts, suggesting that it may benefit patients with tumors driven by ligand-induced combinatorial activation of the ErbB receptor signaling network where current therapies fail.


Ligands that bind ErbB1 or ErbB3 trigger differential signaling

To explore the relative effectiveness of a panel of ErbB ligands, we stimulated serum-starved ADRr ovarian cancer cells (26) with various concentrations of ligand and measured receptor and AKT phosphorylation at multiple time points between 2 and 60 min by enzyme-linked immunosorbent assay (ELISA). The ELISA assay allows one to globally capture the ErbB receptor phosphorylation with a pan-phosphotyrosine detection antibody, as the receptors are extensively phosphorylated at multiple residues. Ligands tested included the ErbB3 ligand HRG1-β and the ErbB1 ligands BTC, EGF, transforming growth factor–α (TGFα), amphiregulin (AR), heparin-binding epidermal growth factor (HB-EGF), epigen (EPG), and epiregulin (ER). The time of peak receptor and AKT phosphorylation was ligand specific; therefore, we selected the peak signal for each concentration of ligand at whatever time it occurred to obtain dose-response curves (Fig. 1A). We determined relative ligand effectiveness comparing the median effective concentration (EC50) values and maximum activation for each ligand (Fig. 1A and table S2, A to D). BTC and EGF most potently and effectively stimulated ErbB1 phosphorylation; of all ligands tested, they stimulated the highest observed ErbB1 phosphorylation and had the lowest EC50 values. TGFα, AR, and HB-EGF also stimulated maximal ErbB1 phosphorylation but had higher EC50 values than BTC and EGF. EPG and ER stimulated about 10% of the maximal observed phosphorylation of ErbB1, whereas HRG1-β did not induce any measurable ErbB1 phosphorylation. EGF, BTC, and HB-EGF were the most effective ligands with respect to ErbB2 phosphorylation, whereas AR, TGFα, EPG, and ER stimulated relatively little ErbB2 phosphorylation. HRG1-β most potently induced ErbB3 phosphorylation, followed by BTC and HB-EGF, which induced 30 to 40% of the maximum observed ErbB3 phosphorylation. With respect to AKT phosphorylation, HRG1-β was the most effective ligand, stimulating four times more AKT phosphorylation than did EGF, BTC, AR, or TGFα. EPG and ER elicited some AKT phosphorylation at ligand concentrations above 200 nM. The AKT phosphorylation pattern stimulated by the different ligands resembled the observed phosphorylation pattern of ErbB3 more than it did those of ErbB1 or ErbB2. Together, HRG1-β was the most effective inducer of ErbB3 phosphorylation, and although both EGF and BTC potently stimulated ErbB1 and ErbB2 phosphorylation, BTC stimulated more ErbB3 phosphorylation than did EGF. These insights about the relative ability of ErbB ligands to stimulate the phosphorylation of individual ErbB receptors led us to focus on the ErbB3 ligand HRG1-β and on BTC as a representative ErbB1 ligand.

Fig. 1

HRG1-β and BTC are important ErbB ligands. (A) Serum-starved ADRr cells were stimulated with a seven-point dilution series of different ErbB ligands (EGF, HB-EGF, EPG, AR, BTC, TGFα, ER, and HRG1-β) starting from 250 nM down to 0.34 nM. Phosphorylation status of ErbB1 (pErbB1), ErbB2 (pErbB2), ErbB3 (pErbB3), and AKT (pAKT) were measured at 0, 2, 5, 10, and 60 min after stimulation. The graphs indicate the peak activation achieved by each concentration of each ligand, normalized to the maximal phosphorylation observed for each measured phosphoprotein. Filled circles represent data points and solid lines represent a sigmoidal least-squares fit to the experimental data. (B) Tumor cell lines were seeded at 50,000 cells per well in low-serum medium (1% FCS) in 96-well tissue culture plates, grown overnight, and stimulated with BTC or HRG1-β (100 ng/ml) for 30 min. Lysates were analyzed by pAKT ELISA. The value for the unstimulated control was subtracted from the stimulated signal for each cell line and all signals were subsequently normalized to the highest signal measured for each ligand across the panel of cell lines. (C) Cryogenically preserved primary human tumor samples were lysed and analyzed for HRG1-β and BTC by ELISA. G.I., gastrointestinal.

To further understand the signaling potential of these two ligands, we measured AKT phosphorylation elicited by exogenous HRG1-β and BTC in 54 cell lines from the National Cancer Institute (NCI) panel of human cancer cell lines known as the NCI-60 (27). The NCI-60 includes carcinomas of various origins (kidney, breast, colon, lung, prostate, and ovary), tumors of the central nervous system (CNS), malignant melanomas, leukemias, and lymphomas. We screened all but the leukemia- and lymphoma-derived cell lines and found that exogenous HRG1-β and BTC stimulated AKT in many of these lines (Fig. 1B). We normalized to the maximal amount of AKT phosphorylation measured quantitatively by ELISA for each ligand across the panel of cell lines. HRG1-β and BTC elicited the maximal amount of AKT phosphorylation in MCF7 cells for each ligand, which can be explained by previous findings that AKT is more abundant in MCF-7 cells than it is in other cancer cell lines (28). In cell lines that were responsive to both ligands, the amount of AKT phosphorylation induced by HRG1-β was greater than that induced by BTC, which is in agreement with the dose-response data (Fig. 1A). HRG1-β induced AKT phosphorylation in about two-thirds of the cell lines tested; MCF-7 (breast), MALME3M (melanoma), UACC62 (melanoma), UACC257 (melanoma), SKMEL28 (melanoma), and OvCAR3 (ovarian) showed the strongest induction by HRG1-β, with MALME3M, UACC257 and SKMEL2 (melanoma), TK10 (renal), HCC2998 (colon), and DU145 (prostate) responding only to HRG1-β stimulation. U251, SM539, SNB19, and SF295 (all CNS lines) responded only to BTC stimulation. The effects of exogenous HRG1-β and BTC on AKT phosphorylation could be masked in some cell lines because of endogenous autocrine or paracrine signaling (29, 30); therefore, these data may underestimate the signaling potential of these two ligands.

We analyzed a set of 45 cryogenically frozen primary tumor samples of different tissue origin for HRG1-β and BTC abundance by ELISA (Fig. 1C). BTC and HRG1-β were detectable in all tumor types tested; in particular, BTC was abundant in lung, colon, and kidney tumor samples, and HRG1-β was abundant in ovarian, mammary, and melanoma tumor samples, with the greatest amount of HRG1-β seen in one of the cervical tumor samples. HRG1-β plays an important role in cancer by promoting cell proliferation and migration (31, 32); however, the in vivo role of BTC is unclear (33). Our data show that BTC and HRG1-β are found in many different tumor types. Given that multiple reports have shown the ubiquitous presence of ErbB receptors in different tumor types and normal tissue (34, 35), the widespread presence of ErbB-receptor ligands (Fig. 1C) suggests that many types of tumors are wired to depend on ligand-induced ErbB receptor network activation.

High-density signaling data and simulation provide systems understanding of the ErbB receptor signaling network

To explore the optimal strategy to therapeutically inhibit combinatorial, ligand-induced activation of the ErbB receptor–PI3K axis, we built a computational model of the ErbB receptor signaling network that included the most effective ErbB ligands, as well as known and previously unidentified ErbB inhibitors. We wanted to construct a computational model that was as simple as possible, while still capturing receptor-ligand interactions in sufficient detail, and to explore the insights gained thereby to develop new—potentially better—ErbB receptor–targeted therapies. The protein-protein interactions encompassed by this model are represented schematically in Fig. 2A. The biochemical reactions, kinetic parameters, initial amounts of protein species, and a description of the model assumptions and methods for parameter estimation and sensitivity analysis are given in Materials and Methods and in the Supplementary Materials. Based on our findings that HRG1-β and BTC are present in primary human tumors derived from various tissues and that these ligands stimulate AKT phosphorylation in multiple cell lines, we chose BTC and HRG1-β as representative ErbB1- and ErbB3-binding ligands, respectively, to be implemented into the model. We simplified the model by omitting ErbB4 because ErbB4 abundance is generally very low (often below the limit of detection) in most solid tumor lines (table S1) (36).

Fig. 2

Empirical measurements of ligand-specific ErbB receptor and AKT signaling dynamics. (A) Schematic depiction of the ErbB signaling network showing the receptors ErbB1 to ErbB4, BTC binding to ErbB1 and HRG1-β binding to the ErbB3 receptor, receptor dimerization, dimer internalization, and recycling, and interactions leading to activation of the PI3K-AKT cascade. The computational model is an interpretation of this schematic, using mass action kinetics. Because of the low expression observed in vitro, ErbB4 was omitted in the computational model. (B) Serum-starved ADRr cells were stimulated with a nine-point, dilution series of HRG1-β or BTC starting from 250 nM down to 0.038 or 700 nM down to 0.32 nM, respectively. Phosphorylation of ErbB1, ErbB2, ErbB3, and AKT was measured at 0, 1, 2, 3, 4, 5, 7, 10, 20, 30, 60, and 120 min after stimulation. Not all concentrations were assayed for ErbB2 phosphorylation. Data (filled circles) represent the average of three experiments, expressed as mean ± SD. Model simulation (solid lines) involved conditions identical to that for the experimental data. Sensitive kinetic parameters in the model were calibrated against the experimental data with a genetic algorithm. For each measured phosphoprotein, experimental data and simulation results were normalized to the maximum signal observed.

Experimentally measured dose-time matrices for ErbB1, ErbB2, ErbB3, and AKT phosphorylation in response to HRG1-β and BTC are shown in Fig. 2B. This data set was used to calibrate sensitive kinetic parameters in the model (for instance, dimerization and internalization rates; table S8) using a genetic algorithm to minimize the difference between data and simulation. The availability of the full dose–time matrices for both ligands allowed us to capture temporal and dose-dependent traits and to constrain dimerization and internalization parameters in the computational model. Furthermore, simulations using the calibrated model of ErbB1-3 phosphorylation and AKT phosphorylation for the full dose–time matrices for BTC and HRG1-β were in good agreement with experimental data (Fig. 2B). Because the data set was normalized to the maximal stimulation observed for ErbB1-3 and AKT phosphorylation individually independent of the stimulus, the model captures a number of ligand-specific traits. Transient signaling was observed experimentally with both ligands; with BTC stimulation, phosphorylation of ErbB1, ErbB2, and ErbB3 was maximal within 5 min and returned to near-basal amounts within 30 min of stimulation. In contrast, ErbB3 and AKT phosphorylation persisted as long as 120 min after HRG1-β stimulation; in the model, this is attributed to the slower internalization rates generally thought to be characteristic of HRG1-β–bound dimers compared to EGF-like ligand-bound dimers (37).

No detectable HRG1-β–induced ErbB1 phosphorylation was observed, suggesting that HRG1-β induces formation of ErbB1-ErbB3 heterodimers (and other ErbB1 activation mechanisms) poorly, in agreement with previously published findings (38). The weak ability of HRG1-β to elicit ErbB1 phosphorylation was reproduced in the computational model by constraining the HRG1-β–induced ErbB1-ErbB3 dimerization rate to be 100 times weaker than the ErbB2-ErbB3 dimerization rate. The computer simulations can only recapitulate the experimentally observed ErbB3 phosphorylation dynamics if BTC elicits formation of ErbB1-ErbB3 heterodimers. However, BTC has only been reported as binding to ErbB1 and ErbB4 (39), thus supporting the model implementation of BTC binding to ErbB1 and subsequently inducing ErbB1-ErbB3 heterodimers. With both BTC and HRG1-β stimulation, AKT phosphorylation was more sustained than receptor phosphorylation, a phenomenon only partially captured with the model; inclusion of transcription-mediated feedback mechanisms could be responsible for the observed long-term behavior (18).

ErbB3 is a key node in the ErbB receptor signaling network

Concepts and methods from systems and control analysis such as sensitivity or dynamic metabolic control analysis (40) study how variation of the value of the simulated output of a mathematical model (such as AKT phosphorylation) can be apportioned to different sources of variation in the input of a model (such as variation of ligand or receptor abundance). Positive sensitivity implies that the variation of the input increases the value of output compared, whereas negative sensitivity implies that increasing variation of the input suppresses the value of output compared to the initial amounts. Sensitivity analysis can be applied to identify proteins that control signaling networks: which reactions in complex networks are important, which are not, and which key proteins are important to control the amplitude, duration, and integrated response of the signaling output (41, 42). To identify the key proteins that control signaling induced by either ErbB3- or ErbB1-binding ligands, a sensitivity analysis was performed with respect to the model output, phosphorylated AKT. In Fig. 3A, the normalized time-integrated sensitivity is depicted for species with nonzero initial conditions in the computational model. The following nonzero species were considered in the computational model: receptors ErbB1-3, ligands BTC or HRG1-β, the output kinase AKT, the phospholipid PIP2 (phosphatidylinositol 4,5-bisphosphate), the kinases PI3K and PDK1 (3-phosphoinositide–dependent protein kinase-1), and the phosphatases PP2A (protein phosphatase 2A), RTKpase (receptor tyrosine kinase phosphatase, representing a general phosphatase that dephosphorylates the receptors), and PTEN (phosphatase and tensin homolog).

Fig. 3

Sensitivity analysis of the ErbB model. (A) The normalized time-integrated sensitivity of AKT phosphorylation to each nonzero species was determined by varying the amount of each nonzero species and simulating the time course of phosphorylated AKT in response to 1 nM HRG1-β or BTC, with the calibrated computational model. The normalized sensitivity integrated over the 2 hour time course is shown, with species ranked according to their sensitivity during HRG1-β stimulation. (B and C) The dose-response curves for virtual monoclonal antibodies against ErbB3 (ErbB3 mAb) with different dissociation constants (Kd) that act as ErbB3 antagonists were simulated for a cell expressing 10 times as much ErbB1 and ErbB2 as ErbB3. The ErbB3 mAb was preincubated in the model for 30 min before stimulation with (B) HRG1-β or (C) BTC for 10 min. The phosphorylated AKT signal is normalized to the no-inhibitor control.

For AKT phosphorylation, ErbB3 was identified as the most sensitive node in the network for stimulus with either ligand. AKT phosphorylation was less sensitive to ErbB2 with HRG1-β stimulation, because the total number of ErbB2-ErbB3 heterodimers was constrained by the less abundant receptor, ErbB3 (table S1). For AKT phosphorylation after BTC stimulation, sensitivity analysis identified ErbB3 and ErbB1 as equally sensitive nodes, indicating that BTC induces most of the AKT phosphorylation through ErbB1-ErbB3 heterodimers even though the abundance of ErbB1 is five times that of ErbB3, so that there are far more ErbB1 homodimers than ErbB1-ErbB3 heterodimers. ErbB2 was negatively sensitive when BTC was the stimulus because of competition with ErbB3 for creating ErbB1 heterodimers. Ligand concentration was also a strong determinant of the degree of AKT phosphorylation. Components of the PI3K-AKT pathway were sensitive, but to a lesser degree. The sensitivity of PI3K with respect to phosphorylated AKT was low, because the model assumed that PI3K was in excess of ErbB receptors. Phosphatases that serve as negative regulators within the network by dephosphorylating ErbB receptors, AKT, or phosphatidylinositol 3,4,5-trisphosphate (PIP3) (RTKpase, PP2A, and PTEN, respectively) showed a negative normalized sensitivity as expected. In summary, sensitivity analysis of the computational model identified ErbB3 as the most perturbation-sensitive protein in the signaling network—independent of whether BTC or HRG1-β was the stimulus.

Next, we simulated the effect of a virtual mAb directed against ErbB3 (ErbB3 mAb) in the model to support the finding that ErbB3 serves as a key node for transducing signals triggered by either ErbB3- or ErbB1-binding ligands. The virtual ErbB3 mAb was implemented into the model through mass-action equations with the following characteristics: The virtual ErbB3 mAb blocks ErbB3 dimerization with ErbB2 and ErbB1, inhibits the binding of HRG1-β, and causes cross-linking of ErbB3 receptors as described for bivalent immunoglobulin G (IgG) (43). We examined the effect of the virtual ErbB3 mAb on inhibition of AKT phosphorylation in the presence of HRG1-β or BTC in an in silico modified ADRr cell line where the dimerization partners ErbB1 and ErbB2 are 10 times more abundant than ErbB3 (Fig. 3, B and C). Similar patterns of ErbB receptor abundance are characteristic of some cancer cell lines, including OvCAR8 or ACHN (table S1), and should represent the scenario most difficult for the ErbB3 mAb to inhibit. In addition, we varied the dissociation rate constant (Kd) over four orders of magnitude while keeping the association rate constant at 1 × 105 M−1 s−1 to explore the benefit of improving the affinity. Simulated dose-response curves for phosphorylated AKT with increasing concentrations of virtual ErbB3 mAb that differ in affinity are presented for HRG1-β and BTC stimulation (Fig. 3, B and C). We found that, with HRG1-β stimulation, ErbB3 inhibition blocked AKT phosphorylation entirely in the simulation; with BTC stimulation, ErbB3 inhibition decreased AKT phosphorylation more than 60%, even when ErbB1 and ErbB2 were highly overexpressed in the simulation. In addition, according to the simulation results, improving the Kd of the inhibitor to less than 1 nM did not yield a significant benefit with either HRG1-β and BTC stimulation. The residual AKT phosphorylation observed in the simulation results in Fig. 3C was caused by ErbB1 homodimers and ErbB1-ErbB2 heterodimers.

These simulations show how such biochemical models can be used to identify sensitive targets within a signaling network and to identify design and selection criteria for optimal therapeutic strategy. Thus, on the basis of insights from the computational model, we screened for an antibody that would bind to ErbB3 on cells with a low nanomolar affinity (Fig. 3, B and C), block HRG1-β binding to ErbB3, and block BTC-induced ErbB3 phosphorylation.

MM-121 is a human ErbB3 mAb that functions as an antagonist

Based on the results from the sensitivity analysis, the ErbB3 inhibitor simulations, and experimental data, we identified ErbB3 as a potential therapeutic target and developed MM-121, a fully human IgG2 mAb that binds specifically to ErbB3. The dissociation constant (Kd) for MM-121 was determined to be 769 pM, with an association rate of 1.43 × 105 M−1 s−1 and a dissociation rate of 1.10 × 10−4 s−1 measured by kinetic exclusion assay (KinExA). Using surface plasmon resonance, we found that MM-121 competes with HRG1-β binding to ErbB3.

Immunoblot analysis showed that MM-121 inhibited HRG1-β– and BTC-induced ErbB3 phosphorylation at tyrosine-1289 (Fig. 4A). The ErbB3 phosphorylation induced by BTC and measured by immunoblot was less than that induced by HRG1-β, in agreement with the ELISA measurements (Fig. 2B). In addition, the ability of MM-121 to inhibit ErbB3 phosphorylation was greater with HRG1-β stimulation than with BTC, consistent with the simulation results for the virtual ErbB3 mAb (Fig. 3, B and C). MM-121 inhibited HRG1-β–induced ErbB2-ErbB3 dimer formation in ADRr cells, as determined by immunoprecipitating ErbB2 and probing with an anti-ErbB3 to assess ErbB2-ErbB3 complex formation (Fig. 4B). Neither ErbB2-ErbB3 nor ErbB1-ErbB3 complexes could be detected by coimmunoprecipitation after BTC stimulation. Therefore, we could not experimentally verify the prediction that BTC elicits ErbB3 phosphorylation through the formation of ErbB1-ErbB3 dimers.

Fig. 4

Characterization of MM-121, a potent ErbB3 antagonist. (A) ErbB3 phosphorylation on tyrosine-1289 was measured in serum-starved ADRr cells pretreated with 250 nM MM-121 or buffer (negative control) for 1.5 hours then stimulated with 10 nM HRG1-β or 200 nM BTC for 10 and 4 min, respectively, and measured by immunoblot. (B) Cell lysates from (A) were immunoprecipitated (IP) with an anti-ErbB2 and probed for ErbB3 by immunoblot (WB). (C and D) Serum-starved ADRr cells were preincubated for 30 min with a 10-point serial dilution of MM-121 starting at 2 μM down to 7.6 pM and then stimulated for 10 min with 25 nM HRG1-β (C) or BTC (D). Dose-response curves of MM-121 treatment include ELISA measurements (filled circles) of the readouts pErbB1, pErbB2, pErbB3, and pAKT and simulations (red solid lines) under equivalent conditions. The data represent the mean ± SD of two experiments. Mean values were background-subtracted and normalized to the stimulated ligand control for each readout. Tables include the experimentally determined IC50 values and corresponding 95% confidence intervals as well as simulated IC50 values. N/A indicates the curve-fitting algorithm was unable to produce an acceptable fit.

We compared dose-response experiments of MM-121 inhibition of HRG1-β– or BTC-induced ErbB1, ErbB2, ErbB3, and AKT phosphorylation in ADRr cells to simulation results (Fig. 4, C and D). The kinetic parameters and reaction scheme describing binding of MM-121 to ErbB3 in the simulation were based on the experimentally measured characteristics of MM-121: bivalent binding to ErbB3 (using the monovalent binding constants determined by kinetic exclusion assay), competition with HRG1-β, and inhibition of ErbB1-ErbB3 dimerization after stimulation with BTC. The model correctly predicted the dose-response behavior of MM-121 for phosphorylation of AKT and the receptors with either HRG1-β or BTC stimulation. MM-121 inhibited HRG1-β–stimulated ErbB2 and ErbB3 phosphorylation with median inhibitory concentration (IC50) values of 3.0 and 2.4 nM, respectively, compared to the inhibition of AKT phosphorylation with an IC50 of 6.2 nM (Fig. 4C). The prediction that MM-121 would inhibit BTC-induced AKT phosphorylation without affecting ErbB1 and ErbB2 phosphorylation was experimentally verified. The IC50 values for ErbB3 and AKT phosphorylation after pretreatment with MM-121 and stimulation with BTC were 5.8 and 4.9 nM, respectively, about two times higher than the IC50 values for MM-121 inhibition of HRG1-β–mediated phosphorylation (Fig. 4D). The simulated IC50 values were within a factor of 3 of the actual experimentally observed values (Fig. 4, C and D). These results support the idea that ErbB3 is a key node in the combinatorial, ligand-induced activation of the ErbB receptor signaling network.

The computational model also predicted the dose-response relationship for MM-121 inhibition of ErbB3 and AKT phosphorylation after HRG-β1 or BTC stimulation in OvCAR8 (ovarian) and DU145 (prostate) cells (fig. S1, A and B). Receptor abundance, as experimentally determined for OvCAR8 and DU145 (table S1), was used as an input parameter in the model, with all other parameters held constant. The experimentally observed IC50 values for ErbB3 phosphorylation after pretreatment with MM-121 and stimulation with HRG1-β were 3.1 nM with OvCar8 cells and 1.8 nM with DU145 cells. The observed IC50 values for AKT phosphorylation were two to three times higher, as predicted by the computational model. The experimentally observed IC50 values for ErbB3 and AKT phosphorylation after pretreatment with MM-121 and stimulation with BTC were about 2 nM in both cell lines. The prediction of the experimentally observed partial inhibition of AKT phosphorylation with BTC stimulation in DU145 cells, together with the dose-response curves for pAKT and pErbB3 in OvCAR8 and ADRr cell lines indicate that the computational model can predict dose-response curves of MM-121 in these three cell lines.

The inhibition profile of MM-121 differs from those of cetuximab, lapatinib, pertuzumab, and trastuzumab

The ErbB network has already been targeted by therapeutics; we performed inhibitor dose-response studies with cetuximab, lapatinib, pertuzumab, and trastuzumab in ADRr cells stimulated with either HRG1-β or BTC to compare MM-121 with ErbB inhibitors currently in use. With the exception of trastuzumab, each inhibitor was incorporated into the model with the use of kinetic association and dissociation rates and mechanisms of action obtained from the literature (see Materials and Methods, Supplementary Material, and tables S7 and S8). Trastuzumab was not incorporated into the model because its molecular mechanism of action is poorly understood (44). Overall, the simulated IC50 values corresponded to the experimentally observed IC50 values for ErbB1 and ErbB2 that have been published for lapatinib, cetuximab, and pertuzumab (4547). Lapatinib, a reversible inhibitor of the tyrosine kinase activity of ErbB1 and ErbB2 (45), inhibited HRG1-β– and BTC-induced ErbB3 phosphorylation with IC50 values of 150 and 360 nM, respectively (Fig. 5, B and F). Cetuximab, a mAb that prevents ligand binding to ErbB1 (47), inhibited ErbB3 phosphorylation only in response to BTC stimulation (Fig. 5, C and F). This confirms our observation that HRG1-β signaling is mediated primarily by ErbB2-ErbB3 and validates the prediction that ErbB1 is the main mediator of BTC-induced ErbB3 phosphorylation. Pertuzumab, a mAb that hinders the recruitment of ErbB2 into ErbB ligand complexes (48), inhibited HRG1-β–induced ErbB3 phosphorylation with an IC50 of 3.6 nM (Fig. 5, D and F). In contrast to MM-121, pertuzumab modestly inhibited BTC-induced ErbB3 phosphorylation only at the highest concentrations, providing additional evidence that activation of ErbB3 by BTC is not mediated by ErbB2 but by the ErbB1-ErbB3 heterodimer. The discrepancy between the model simulation and the experimental data indicates that the computational model may not fully capture the mechanism of pertuzumab. Trastuzumab, a humanized mAb directed against the extracellular domain of the tyrosine kinase receptor HER2 that has shown clinical activity in HER2-overexpressing breast cancers, showed only modest inhibition of HRG1-β– or BTC-dependent signaling compared to MM-121 (Fig. 5E).

Fig. 5

MM-121 has an inhibition profile distinct from that of other ErbB inhibitors. (A to E) Serum starved ADRr cells were preincubated for 30 min with serial dilutions of (A) MM-121, (B) lapatinib, (C) cetuximab, and (E) trasuzumab from 2 μM to 7.6 pM or (D) pertuzumab from 100 nM to 1.5 pM and then stimulated for 10 min with 25 nM HRG1-β or BTC. Phosphorylated ErbB3 was measured by ELISA (filled circles). Data represent the mean ± SD of two separate experiments. Mean values were background-subtracted and normalized to the stimulated ligand control for each readout. The response to each inhibitor was simulated under equivalent conditions (solid lines). (F) Representation of experimentally determined IC50 values with corresponding 95% confidence intervals and simulated IC50 values for the dose-response curves shown in (A) to (D).

We also used computer simulations to compare the ability of the different inhibitors to inhibit ErbB1, ErbB2, or ErbB3 phosphorylation in response to stimulation with HRG1-β or BTC (Fig. 6, A and B). In these studies, we plotted the trajectories of the simulated effect of increasing inhibitor concentrations from 0 to 100 nM on receptor phosphorylation. MM-121 completely inhibited ErbB2 and ErbB3 phosphorylation during HRG1-β stimulation, as well as ErbB3 phosphorylation (but not that of ErbB1 and ErbB2) during BTC stimulation. Pertuzumab inhibited ErbB2 phosphorylation in both cases, but inhibited ErbB3 phosphorylation only with HRG1-β stimulation (Fig. 6B). Lapatinib partially inhibited ErbB2 and ErbB3 phosphorylation with HRG1-β stimulation but inhibited phosphorylation of both to a greater extent with BTC stimulation. Lapatinib also inhibited BTC-induced ErbB1 phosphorylation. Cetuximab did not inhibit HRG1-β–induced ErbB2 or ErbB3 phosphorylation, but completely inhibited ErbB1, ErbB2, and ErbB3 phosphorylation in response to BTC. Summarizing the simulation results, MM-121 inhibited ErbB3 phosphorylation to the same extent as pertuzumab during HRG1-β stimulation and cetuximab during BTC stimulation, and thus was the only inhibitor tested that inhibited ErbB3 phosphorylation with low nanomolar IC50 values for both stimuli.

Fig. 6

Simulated inhibitor trajectories on target inhibition. (A and B) The model was used to simulate ErbB1, ErbB2, and ErbB3 phosphorylation after a 30-min preincubation with 0, 0.1, or 100 nM MM-121, lapatinib, cetuximab and pertuzumab followed by 10-min stimulation with either 25 nM HRG1-β (A) or BTC (B). Simulated data are normalized to equivalent stimulation with no inhibitor present. The circle indicates the origin, representing phosphorylation in the absence of inhibitor, and the arrows represent the path of target inhibition with low and high amounts of inhibitor.

MM-121 inhibits spheroid growth and is effective in vivo

Spheroids are generated from a defined number of tumor cells and provide an in vitro 3-dimensional (3D) model that mimics the 3D environment of a tumor and allows one to study drug effects on proliferation or metabolism in a more physiological in vitro setting. We evaluated the effects of MM-121 on ADRr and OvCAR8 cells in a spheroid growth assay (49). MM-121 caused a significant decrease in spheroid area compared to that in untreated cells or control cells treated with isotype IgG2, with the greatest decrease apparent with 100 nM MM-121 (Fig. 7A).

Fig. 7

The effects of MM-121 on spheroid growth and in vivo tumor growth. (A) After 4 days growth, ADRr and OvCAR8 spheroids of similar size were treated with medium (negative control), 0.5 μM IgG2 (isotype control) or 0.01, 0.1, or 1 μM MM-121. Spheroid area was measured before treatment (day 5) and after treatment (day 11). The change in the spheroid area over 6 days as a percentage of the original area on day 1 is shown. Statistical significance was assessed compared to media control. (*P < 0.01 by Student’s t test). Data are the average of five experimental replicates ± SEM. (B to E) ACHN tumor cells (5 × 106) in Matrigel were subcutaneously injected into the right flank of nude mice. Tumors were collected after reaching a tumor volume greater than 200 mm3. Tumor lysates were probed for HRG1-β (B) and BTC (C) by ELISA. (D) ACHN xenograft tumor volume over time for mice administered 100 or 300 μg of MM-121 per injection or PBS control every third day (Q3D). Tumors were measured twice weekly and tumor volume was calculated as π/6 × length × width2 (width being the shortest measurement). Statistical significance is only depicted for the highest-dose group compared to the vehicle control group. Student’s t test was used to assess statistical significance. Data are mean ± SEM; 10 mice were in each group. (E) ACHN xenograft tumor volume over time for MM-121 administered at 600 μg per injection Q3D or at 600, 1000 or 2500 μg per injection weekly (Q7D) compared to PBS control administered Q3D or Q7D. (F) A549 K-Ras–dependent lung xenograft tumor volume over time for MM-121 administered seven times at Q3D until day 22 and monitored after treatment was stopped.

To examine the in vivo efficacy of MM-121, we first identified a cell line that expressed both HRG1-β and BTC when implanted in mice. We found that ACHN xenograft tumors had mean values of 0.42 ± 0.15 pg of HRG1-β per microgram of total protein and 0.1 ± 0.01 pg of BTC per microgram of total protein (Fig. 7, B and C). This is comparable to their abundance in patient tumor samples (Fig. 1B). Next, we conducted an in vivo study in which we treated nude mice with MM-121 every 3 days for 45 days. At 100 μg per injection, we observed a delay in tumor growth, whereas at 300 μg/injection, we saw complete inhibition of tumor growth over the course of 45 days (Fig. 7D). If mice were injected every 7 days with 600 or 1000 μg per injection, tumor growth was slowed, whereas complete inhibition of the tumor growth was only obtained with treatments of 600 μg per injection every 3 days or 2500 μg per injection every 7 days (Fig. 7E).

We also investigated the effect of MM-121 in a K-Ras–dependent A549 lung xenograft model (50. In this study, seven doses of MM-121 were administered every 3 days until day 22 and subsequent tumor growth was monitored until day 45. Tumor growth was significantly slower after the last dose of MM-121 over the course of the following 23 days of observation compared to the untreated vehicle control group (Fig. 7F).


Early interest in the ErbB receptor network focused on the oncogenic role of ErbB1 and ErbB2 (51). Overexpression of ErbB2 in tumor cells, as well as that of ErbB1 variants with deletions in the extracellular domain, and mutations of ErbB1 have all been shown to transform cells in the absence of exogenous ligand (5254). This oncogene paradigm led to the development of therapeutic agents such as erlotinib, lapatinib, and trastuzumab that targeted this aberrant activation. All of these agents are therapeutically effective, albeit with low clinical response rates.

Here, we compare the relative effectiveness of eight ErbB ligands with respect to their ability to stimulate phosphorylation of ErbB1, ErbB2, ErbB3, and AKT in a cell line that expresses modest amounts of all three receptors (table S1). Using the two most effective ligands, BTC and HRG1-β, we studied early signaling dynamics as a function of time and ligand concentration while comparing the absolute signal strength (target phosphorylation) of the two ligands. BTC was previously known to activate only ErbB1 and ErbB4 (55, 56); its importance in activating ErbB3 has not been appreciated, possibly due to its rapid signaling dynamics (Fig. 2B). The physiological role of BTC is largely undefined; however, BTC induces angiogenesis through the MAPK and PI3K signaling cascades (57) and is involved in motility (58) and thus may play a role in promoting tumorigenesis. Our quantification of HRG1-β and BTC in a panel of human tumor samples revealed that both ligands are present to a similar extent and that in some tumors BTC is found in the absence of HRG1-β (Fig. 1C).

Our data from stimulating a large panel of tumor-derived cell lines with HRG1-β and BTC show that ErbB signaling capacity is highly prevalent (Fig. 1B) even in the absence of ErbB2 or ErbB1 overexpression. Furthermore, in addition to overexpression of ErbB2, the response to trastuzumab in breast cancer patients is affected by the status of other ErbB receptors, as well as ligand production (59). Together, these results suggest that there may be two paradigms underlying ErbB receptor–dependent tumor growth: (i) the traditional paradigm of oncogene addiction, which supposes that tumor cells become dependent on a single oncogenic activity and is exemplified by HER2-amplified tumors or tumor dependence on critical mutations; and (ii) a paradigm of combinatorial ligand-induced dependence that does not require the overexpression of the ErbB receptors. Thus, additional cancer patients could benefit from ErbB therapies targeting network activation arising from combinatorial interactions between ligands and receptors (Fig. 1, B and C), which may be more frequent than HER2 overexpression [which accounts for about 20% of all breast cancer cases (60)] or EGFR mutation [which accounts for 10 to 15% of all lung cancer patients (61)].

The computational model described in this report enabled the systematic analysis of sensitive nodes within this combinatorial network. Surprisingly, sensitivity analysis indicated that variation of ErbB3 receptor abundance affected the model output, AKT phosphorylation, to a much greater extent than did variation of ErbB1 and ErbB2, which have previously been the focus of drug development. This suggested a distinct strategy of therapeutically inhibiting ErbB3 for treating tumors activated by combinatorial, ligand-induced activation of the ErbB receptor network.

The computational model, trained using a high-density data set that describes the dynamics and concentration dependence of ErbB receptor phosphorylation, helped define the desired kinetic and biochemical properties of an ErbB3 antagonist; this led to the development of mAb MM-121, an IgG2 that recognizes ErbB3. IgG2 isotype antibodies exert lower amounts of effector functions, such as antibody-dependent cytotoxicity and complement-dependent cytotoxicity, compared to the IgG1 isotype antibodies (62). Therefore, we argue that the in vitro and in vivo effects observed with MM-121 result from the inhibition of ErbB3 signaling. MM-121 inhibits both HRG1-β– and BTC-induced ErbB3 and AKT phosphorylation and thereby inhibits combinatorial, ligand-induced activation of the ErbB receptor network. In the model, MM-121 blocks BTC-induced heterodimerization between EGFR and ErbB3. This is supported by experimental inhibitor studies: MM-121 and cetuximab completely inhibited BTC-dependent ErbB3 phosphorylation in a dose-dependent manner, whereas pertuzumab did not have an inhibitory effect, suggesting that ErbB2 is not involved in BTC-induced ErbB3 phosphorylation (Fig. 5). This inhibition profile differs from that of other ErbB inhibitors used clinically. Indeed, both lapatinib, a small-molecule inhibitor of ErbB2 and ErbB1, and trastuzumab show clinical activity mainly in tumors that overexpress ErbB2, and thus may be less effective in treating tumors driven by combinatorial, ligand-induced activation (59). MM-121, in contrast, effectively treats non–ErbB2-amplified tumors in multiple xenografts (Fig. 7). Thus, MM-121 has the potential to benefit patients with tumors driven by combinatorial, ligand-induced activation resulting from either ErbB1- or ErbB3-binding ligands.

ErbB3 and HRG1-β are thought to be integrally involved in the resistance of several tumor types to chemotherapy (63), antiestrogen therapy (64), trastuzumab (65, 66), and ErbB1-ErbB2 tyrosine kinase inhibitors (67, 68). This compensatory activation may be due in part to the increased production of either HRG1-β or ErbB3 (67). The simulations of signaling dynamics in the ErbB receptor signaling network presented here suggest that ErbB3 and HRG1-β up-regulation could provide a potent mechanism for tumor cells to increase pro-survival signaling through the PI3K pathway.

Traditional drug discovery methods focus on understanding one interaction at a time. However, in this study, we show how a systems approach to drug discovery can be applied to identify a previously unknown, complementary, and potentially superior mechanism of inhibiting the ErbB receptor signaling network. We anticipate that similar approaches can be used to predict optimal combination therapies (69), as well as guide biomarker selection and patient stratification to transform the process of drug discovery and development.


Cell culture

Cell lines were obtained from the NCI’s Developmental Therapeutics Program ( All cell lines were maintained in RPMI 1640 medium (Gibco) supplemented with 10% fetal calf serum (FCS) (hyclone), 2 mM l-glutamine (Gibco), penicillin (100 U/ml), and streptomycin (100 μg/ml; Gibco) and grown in a humidified atmosphere of 5% CO2 and 95% air at 37°C, unless otherwise indicated.

In vitro signaling studies

For in vitro signaling studies measured by ELISA, cells were seeded in 96-well tissue culture plates, grown overnight, then switched to serum-free media for 20 to 24 hours. For time course experiments, serum-starved cells were stimulated with serial dilutions of BTC (R&D Systems, 261-CE-050/CF), EGF (Peprotech, 100-15), HB-EGF (Peprotech, 100-47), TGFα (R&D Systems, 239-A-100), ER (R&D Systems, 1195-EP-025/CF), EPG (Peprotech, 100-51), AR (R&D Systems, 262-AR-100/CF), and HRG1-β (R&D Systems, 377-HB-050/CF) as indicated. For IC50 determinations, serum-starved cells were preincubated with serial dilutions of MM-121, pertuzumab (produced in-house; see Expression of pertuzumab), cetuximab (obtained from the pharmacy), or Lapatinib (custom synthesis by Biomol International L.P., Plymouth Meeting, PA), followed by stimulation with BTC (R&D Systems, 261-CE-050/CF) or HRG1-β (R&D Systems, 377-HB-050/CF) as indicated. After stimulation, cells were placed on ice, washed with cold phosphate-buffered saline (PBS), then lysed for 30 min in cold M-PER Mammalian Protein Extraction Buffer (Thermo Scientific, 78501) supplemented with protease inhibitor cocktail (Sigma-Aldrich, P2714), 1 mM sodium orthovanadate (Sigma-Aldrich, S6508), 5 mM sodium pyrophosphate (Sigma-Aldrich, 221368), 50 μM oxophenylarsine (EMD Biosciences, 521000), and 10 μM bpV(phen) (EMD Biosciences, 203695).

For comparison of ligand-induced AKT phosphorylation in the NCI-60 panel of solid tumor cells lines, MCF7, ADRr, BT549, T47D, HS578T, SF268, U251, SM539, SNB19, SNB75, SF295, SW620, HCT15, HCC2998, KM12, COLO205, HT29, HCT116, A549, H23, H460, HOP62, H226, HOP92, EKVX, H322M, M14, SKMEL28, UACC62, MALME3M, UACC257, SKMEL2, SKMEL5, OvCAR5, SKOV3, OvCAR4, IGROV1, OvCAR8, OvCAR3, DU145, PC3, LNCAP, ACHN, A498, SN12C, CAKI-1, UO31, and TK10 cells were seeded in 96-well plates in low-serum medium (1% FCS) with additional 10 nM Hepes buffer, grown overnight, stimulated with BTC (R&D Systems, 261-CE-050/CF) or HRG1-β (R&D Systems, 377-HB-050/CF) as indicated, then harvested as described above.

Enzyme-linked immunosorbent assay

For analysis of phospho-ErbB1, -ErbB2, -ErbB3, and -AKT, 384-well, black, flat-bottomed polystyrene high-binding plates (Corning, 3708) were incubated overnight at room temperature with capture antibodies against ErbB1 (R&D Systems, AF231), ErbB2 (R&D Systems, MAB1129), ErbB3 (R&D Systems, MAB3481), or AKT (Upstate, 05-591MG). ELISAs were blocked with 2% bovine serum albumin (BSA) in PBS for 1 hour, then incubated with lysates diluted in 2% BSA, 0.1% Tween-20, and PBS for 2 hours at room temperature. After each incubation, plates were washed four times with 0.05% Tween-20 in PBS. ELISAs measuring phospho-ErbB1, -ErbB2 and -ErbB3 were incubated with a pan-phosphotyrosine horseradish peroxidase (HRP)–linked mAb (R&D Systems, HAM1676) for 2 hours at room temperature. ELISAs measuring phospho-AKT were incubated with serine-473–specific, biotinylated, mouse mAb against phospho-AKT (Cell Signaling Technology, 5102) for 2 hours followed by streptavidin-HRP (R&D Systems, DY998) for 30 min at room temperature.

HRG1-β was detected by indirect ELISA. Samples were incubated for 2 hours at 4°C in 96-well Maxisorb high-binding plates (Nunc, 4737111) coated with recombinant histidine-tagged ErbB3 (ErbB3-6His) (70). HRG1-β was detected with biotinylated goat antibody against human HRG1-β (R&D Systems, DY377) for 2 hours at room temperature followed by streptavidin-HRP (R&D Systems, DY998) for 30 min at room temperature. HRG1-β abundance was quantified with a purified recombinant HRG1-β(1-246) protein (R&D, Systems 377-HB). Detection of BTC in tumor cell lysates was carried out with a BTC-specific ELISA kit (R&D Systems, DY261), following the manufacturer’s recommendations.

All ELISAs were visualized with SuperSignal ELISA Pico Chemiluminescent Substrate (Pierce, 37069) and luminescent signal was measured.

Receptor profiling by quantitative FACS

ErbB1, ErbB2, ErbB3, and ErbB4 amounts were quantified with Quantum Simply Cellular anti–human IgG bead standards (Bangs Laboratories, 817) coated with human IgGs in combination with Alexa 647–labeled antibody against ErbB1, cetuximab (obtained from pharmacy), antibody against ErbB2, trastuzumab (obtained from pharmacy), an antibody against ErbB3 (clone B12) (70), and an antibody against ErbB4 (R&D systems, mAb 1131), respectively. Both beads and cells stained with Alexa 647–labeled IgG antibodies were analyzed by flow cytometry (71). Four populations of beads with different antibody-binding capacities were used as standards to quantify the number of surface receptors per cell.

Primary human tumor profiling by ELISA

Forty-five flash-frozen tumors (10 mammary, 6 colon, 16 ovarian, 4 lung, 3 gastrointestinal, 4 melanoma, and 2 cervical) were obtained from four different divisions of the Cooperative Human Tissue Network (CHTN), Midwestern (Columbus, OH), Western (Nashville, TN), Southern (Birmingham, AL), and Eastern (Philadelphia, PA). Fresh-frozen tumors were pulverized in a CryoPrep cryogenic pulverization instrument (Covaris, Woburn, MA), weighed, and lysed for 30 min on ice with lysis buffer [20 mM tris (pH 7.4), 150 mM NaCl, 2 mM EDTA, 1 mM EGTA, and 1% Triton X-100 with freshly added Protease Inhibitor Cocktail Set III (Calbiochem, 539134, and Halt Phosphatase Inhibitor Cocktail (Pierce, 78420)]. Samples were clarified by centrifugation (14,000 rpm) for 10 min then stored at −80°C. Total protein was measured by BCA Assay (Pierce, 23252)

Expression of pertuzumab

The sequence of rhuMab pertuzumab (2C4) was obtained from Adams et al. (48). The DNA sequences for the variable light domain and the variable heavy domain were synthesized by Codon Devices (Cambridge, MA). These sequences were cloned into an in-house vector (pMP 9100K) that contains the constant heavy chain and constant light chain regions. FlpIN Chinese hamster ovary (CHO) cells (Invitrogen) were then transfected with this vector. The transfected adherent cell line was suspension adapted and grown in shaker flasks. Cells were spun down and the supernatant was protein A–purified to obtain full-length pertuzumab.

Antibody selection and expression and characterization of MM-121

MM-121 was derived from a phage display Fab library and screened for high-affinity ErbB3 binders by surface plasmon resonance as described (72). Fab antibodies were reformatted into human IgG2 and expressed in CHO-K1 cells and purified by Protein A affinity chromatography. Solution-based antibody binding kinetics of MM-121 was determined by kinetic exclusion assay. Serial dilutions of ErbB3-6His (70) were made in tubes, keeping the concentration of MM-121 constant at 250 pM. These solutions were incubated for 1 hour and then passed over a column of ErbB3-6His conjugated polymethyl methacrylate beads. MM-121 binding to the beads was detected with Cy5-labeled goat anti-human IgG secondary antibody (2 mg/ml; Jackson ImmunoResearch).

Receptor dimerization by coimmunoprecipitation assay and receptor phosphorylation by immunoblot

Cells growing in six-well culture plates were serum-starved for 2 hours, then preincubated with 250 nM MM-121 or buffer (control) and subsequently stimulated with BTC (R&D Systems, 261-CE-050/CF) or HRG1-β (R&D Systems, 377-HB-050/CF) as indicated. Cells were washed twice with ice-cold PBS, then lysed on ice in buffer containing 25 mM tris (pH 7.5), 150 mM NaCl, 1 mM EDTA, 1.0% Triton X-100, 1.0% CHAPS, 10% (v/v) glycerol containing protease inhibitor cocktail and phosphatase inhibitor cocktail. Receptor dimerization experiments were performed as previously described (73). ErbB2 was immunoprecipitated with trastuzumab (obtained from a pharmacy) covalently coupled to Affi-gel 10 (Bio-Rad, 153-6046). Immunoblots of the immunoprecipitated samples were probed for ErbB3 (Santa Cruz Biotechnology, sc-285) followed by secondary antibody, IRDye800 anti–rabbit IgG (Rockland, 18725). Immunoblots were imaged then reprobed for ErbB2 (Cell Signaling Technology, 2248) to control for loading. Samples were probed for phosphorylated ErbB3 (Cell Signaling Technology, 4791) and detected with IRDye800 anti–rabbit IgG (Rockland, 18725). Actin was used to control for loading (LabVision, MS-1295).

Spheroid growth assay

Spheroids were prepared as previously described (74). Briefly, cells were seeded into hanging drops (1000 cells per drop) on the inverted lids of 96-well plates. The lids were then placed over wells containing water. The hanging drops were shaken in a tissue culture incubator at 37°C and 5% CO2 for 3 hours and then taken off the shaker and incubated at 37°C and 5% CO2. After 4 days, the hanging drops were centrifuged (1 min, 500 RPM) into wells containing 50 μl of 1% agarose in RPMI and 100 μl of RPMI with 10% FBS. The spheroids in each well were imaged and spheroid area was quantified with the use of MetaMorph 7.5 software (Molecular Devices, Downingtown, PA). Spheroids were either left untreated or treated with MM-121 or IgG2 nonspecific control (Sigma 15404) as indicated. Spheroid diameter was measured again after 6 days.

In vivo studies

Tumor xenografts were established by subcutaneous injection of 5 × 106 ACHN cells or 10 × 106 A549 tumor cells diluted 1:1 with Matrigel Basement Membrane Matrix, Growth Factor Reduced (BD Biosciences 354230) in the right flank. Tumors were allowed to reach 150 to 300 mm3 in size before randomization into groups of 10 mice, containing mice with a similar size distribution of tumors. Mice were treated by intraperitoneal injection with MM-121 or vehicle control (PBS) as indicated. Treatment was continued for the duration of the study, except when indicated otherwise. Tumors were measured twice weekly and tumor volume was calculated as π/6 × length × width2, where the width was the shorter measurement. The difference in average tumor volume at the end of the study between PBS- and MM-121–treated mice was assessed by Student’s t test.

Model development and analysis

The computational model was constructed with the use of mass action kinetics describing ligand-induced ErbB receptor homo- and heterodimerization, receptor internalization and degradation, constitutive dimerization, binding of the downstream kinase PI3K, and subsequent activation of AKT (tables S3 to S5). ERK signaling was not included because AKT signaling is not sensitive to species and parameters in the MAPK cascade; therefore, MAPK-PI3K crosstalk is not necessary to describe AKT signaling dynamics (25). Two ligands were implemented in the model: HRG1-β, which binds ErbB3 and ErbB4, and BTC, which binds to ErbB1 and ErbB4, with affinities described in the literature (75).

The reaction scheme and parameters that describe the model are fully outlined in the Supplementary Materials. We initially set all kinetic parameters to commonly used values for association, dissociation, or enzymatic rates (76) as summarized in table S5. Sensitivity analysis was performed to identify the most important parameters to constrain in the model: Normalized, time-integrated sensitivities of each parameter (or species) were calculated by varying that parameter (or species) and simulating the perturbed system output (for example, pErbB3, pErbB2, pErbB1, pAKT).

The normalized sensitivity Sy,x was calculated according to the following equation: Sy,x = ∂ ln yi/∂ ln xj = x(t = 0)j/yi × ∂ yi/∂ xj, where yi is any species (for instance, pErbB3, pAKT) and xj is each nonzero species or parameter value.

The highly sensitive subset of these parameters was selected for estimation (table S6). Model calibration was performed by comparing high-density experimental data that capture dose and time-dependent characteristics of ErbB1, ErbB2, ErbB3, and AKT phosphorylation in ADRr cells with those of equivalent model simulations. A genetic algorithm was used to modify the parameters to minimize the difference between simulation results and the experimental data. Both HRG1-β and BTC signaling data sets were used simultaneously during model optimization, thereby preserving the ligand-specific activation strengths for each target. Multiple parameter estimation runs were performed to ensure only constrained parameters were updated (see Supplementary Materials for more details).

Three clinically relevant ErbB network inhibitors were incorporated into the model with the use of mass action kinetics to describe their putative mechanism of action (table S7): cetuximab, an anti-ErbB1 ligand-blocking antibody that binds to ErbB1 with an association rate of 2.2 × 105 M−1 s−1 and a dissociation rate of 1 × 10−3 s−1 (47), which we confirmed by surface plasma resonance; lapatinib, an ErbB1 and ErbB2 tyrosine kinase small-molecule inhibitor with association rates of 1.28 × 104 and 2.95 × 103 M−1 s−1 for ErbB1 and ErbB2, respectively, and a dissociation rate of 3.83 × 10−5 s−1 (45); and pertuzumab, an anti-ErbB2 dimerization blocking antibody that binds ErbB2 with an association rate of 1.12 × 105 M−1 s−1 and a dissociation rate of 9.5 × 10−4 s−1 (48). An in silico version of MM-121, our proprietary monoclonal anti-ErbB3, was included in the model using characteristics described in the Results section (table S7).

Data normalization and determination of EC50 and IC50 values

Ligand EC50 and inhibitor IC50 values were calculated by least-squares fitting the dose-response data with a sigmoidal curve (GraphPad Prism, La Jolla, CA). For comparison, the simulation results and experimental ELISA data were normalized. The inhibition dose-response data were normalized by first subtracting the unstimulated control (basal signal) and then scaled by the peak asymptote obtained from fitting to a sigmoidal curve (GraphPad Prism); simulation results were scaled by the uninhibited simulation value. Dose-time matrix data were also basal signal subtracted; however, scaling was performed using the maximum signal observed across all time points and ligand stimulation conditions, thus preserving the unique activation strength of each ligand for each readout. The same process was applied to the simulation results.

Supplementary Materials

Supplementary Text: Model Development

Fig. S1. Characterization of MM-121 in OvCar8 and DU145 cell lines.

Table S1. Measured ErbB receptor expression and mutation status for investigated cell lines.

Table S2A. Characterization of ErbB1 phosphorylation dose-response curves for ErbB1 binding ligands and HRG1-β.

Table S2B. Characterization of ErbB2 phosphorylation dose-response curves for ErbB1 binding ligands and HRG1-β.

Table S2C. Characterization of ErbB3 phosphorylation dose-response curves for ErbB1 binding ligands and HRG1-β.

Table S2D. Characterization of AKT phosphorylation dose-response curves for ErbB1 binding ligands and HRG1-β.

Table S3. Initial amounts of nonzero species in the computational model.

Table S4. Summary of biochemical reactions implemented into the computational model using mass action kinetics with corresponding parameters.

Table S5. Description of parameters with values. Parameter number corresponds to the first reaction in which that parameter appears.

Table S6. Sensitive parameters controlling ErbB3, ErbB2, ErbB1, and AKT phosphorylation.

Table S7. Summary of biochemical reactions describing the implementation of the inhibitors: MM-121, cetuximab, pertuzumab, and lapatinib.

Table S8. Inhibitor parameter values.



Stay Connected to Science Signaling

Navigate This Article