Carboxyl Group Footprinting Mass Spectrometry and Molecular Dynamics Identify Key Interactions in the HER2-HER3 Receptor Tyrosine Kinase Interface* ♦

Background: HER2 and HER3 receptor tyrosine kinases form potent oncogenic signaling dimers. Results: Carboxyl group footprinting and molecular dynamics reveal changes in the HER2-HER3 dimer interface and the HER2 activation loop. Conclusion: HER2 and HER3 form asymmetric heterodimers in a single configuration. The HER2 unphosphorylated activation loop can assume an active conformation. Significance: This study provides the first structural characterization of HER2-HER3 kinase dimers. The HER2 receptor tyrosine kinase is a driver oncogene in many human cancers, including breast and gastric cancer. Under physiologic levels of expression, HER2 heterodimerizes with other members of the EGF receptor/HER/ErbB family, and the HER2-HER3 dimer forms one of the most potent oncogenic receptor pairs. Previous structural biology studies have individually crystallized the kinase domains of HER2 and HER3, but the HER2-HER3 kinase domain heterodimer structure has yet to be solved. Using a reconstituted membrane system to form HER2-HER3 kinase domain heterodimers and carboxyl group footprinting mass spectrometry, we observed that HER2 and HER3 kinase domains preferentially form asymmetric heterodimers with HER3 and HER2 monomers occupying the donor and acceptor kinase positions, respectively. Conformational changes in the HER2 activation loop, as measured by changes in carboxyl group labeling, required both dimerization and nucleotide binding but did not require activation loop phosphorylation at Tyr-877. Molecular dynamics simulations on HER2-HER3 kinase dimers identify specific inter- and intramolecular interactions and were in good agreement with MS measurements. Specifically, several intermolecular ionic interactions between HER2 Lys-716-HER3 Glu-909, HER2 Glu-717-HER3 Lys-907, and HER2 Asp-871-HER3 Arg-948 were identified by molecular dynamics. We also evaluated the effect of the cancer-associated mutations HER2 D769H/D769Y, HER3 E909G, and HER3 R948K (also numbered HER3 E928G and R967K) on kinase activity in the context of this new structural model. This study provides valuable insights into the EGF receptor/HER/ErbB kinase structure and interactions, which can guide the design of future therapies.

The HER2 receptor tyrosine kinase is a driver oncogene in many human cancers, including breast and gastric cancer. Under physiologic levels of expression, HER2 heterodimerizes with other members of the EGF receptor/HER/ErbB family, and the HER2-HER3 dimer forms one of the most potent oncogenic receptor pairs. Previous structural biology studies have individually crystallized the kinase domains of HER2 and HER3, but the HER2-HER3 kinase domain heterodimer structure has yet to be solved. Using a reconstituted membrane system to form HER2-HER3 kinase domain heterodimers and carboxyl group footprinting mass spectrometry, we observed that HER2 and HER3 kinase domains preferentially form asymmetric heterodimers with HER3 and HER2 monomers occupying the donor and acceptor kinase positions, respectively. Conformational changes in the HER2 activation loop, as measured by changes in carboxyl group labeling, required both dimerization and nucleotide binding but did not require activation loop phosphorylation at Tyr-877. Molecular dynamics simulations on HER2-HER3 kinase dimers identify specific inter-and intramolecular interactions and were in good agreement with MS measurements. Specifically, several intermolecular ionic interactions between HER2 Lys-716-HER3 Glu-909, HER2 Glu-717-HER3 Lys-907, and HER2 Asp-871-HER3 Arg-948 were identified by molecular dynamics. We also evaluated the effect of the cancerassociated mutations HER2 D769H/D769Y, HER3 E909G, and HER3 R948K (also numbered HER3 E928G and R967K) on kinase activity in the context of this new structural model. This study provides valuable insights into the EGF receptor/HER/ ErbB kinase structure and interactions, which can guide the design of future therapies.
The epidermal growth factor receptor (EGFR) 3 /ErbB family of receptor tyrosine kinases consists of four transmembrane receptor tyrosine kinases (EGFR/ErbB1, HER2/ErbB2/neu, HER3/ErbB3, and HER4/ErbB4) that play important roles in breast, lung, and other cancer types. Kinase activity is initiated by the binding of growth factor ligands (i.e. epidermal growth factor, neuregulin, etc.) to the extracellular domain of one or more EGFR family monomers and their subsequent homo-or heterodimerization. Dimerization of the extracellular domain brings the cytoplasmic kinase domains together, allowing the formation of asymmetric kinase domain dimers required for kinase activation (1,2). The formation of different dimer pairs initiated by specific ligands allows a small number of proteins to affect several potential signaling pathways. The HER2-HER3 dimer pair represents a uniquely potent combination within this system. The HER2 receptor does not bind ligand. HER3, due to numerous substitutions in its catalytic domain, has significantly lower kinase activity and can only signal as a member of a heterodimer (3). Despite their individual limitations, the HER2-HER3 dimer forms the most potent receptor pair of the ErbB family in terms of cellular proliferation and transformation (4).
HER2 is overexpressed in 20 -30% of breast cancers (5) and is the preferred heterodimerization partner of all other ErbB family members, especially HER3 (6 -8). Therefore, understanding the inter-and intramolecular interactions of HER2 het-erodimers is a high priority. In the case of most high resolution structures of ErbB family kinase domains, the structures are of homodimers (1, 9 -12). To date, the structure of an ErbB family kinase heterodimer has yet to be solved. Although crystallography has provided atomic level resolution of these kinases, it is limited in its potential to characterize dynamic structural events in a membrane environment.
Several mass spectrometry (MS)-based footprinting strategies have been developed to provide insights into protein structure. These approaches include hydrogen/deuterium exchange (H/DX) (13) and hydroxyl radical labeling (14). H/DX labels amide hydrogens, whereas hydroxyl radical labeling modifies aromatic, aliphatic, and sulfur-containing side chains. MS footprinting strategies require less protein material than traditional structural methods and have been successfully employed in the characterization of protein-protein interactions, oligomerization, and folding dynamics (15)(16)(17)(18). These techniques suffer some limitations when investigating membrane-associated proteins, such as extensive back-exchange (in the case of H/DX) or nonspecific oxidation (in the case of hydroxyl radical labeling) resulting from the extensive post-labeling sample cleanup required to remove lipids. However, recent improvements show promise for the implementation of these strategies for membrane proteins (19,20).
An alternative to HD/X and hydroxyl radical footprinting is labeling carboxylic acid side chains (Glu and Asp) with the 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC)-mediated incorporation of glycine ethyl ester (GEE) (21,22). This carboxyl group footprinting reaction is efficient in biologically relevant buffer systems and can tolerate the extensive postlabeling purification required for studying membrane protein structural dynamics. It can also be utilized to rapidly characterize several experimental conditions in parallel. We previously utilized this strategy to examine the dimer interface of the HER4/ErbB4 homodimer, determine its dimer association constant on a lipid membrane, and characterize the conformational changes resulting from activation loop tyrosine phosphorylation (23).
In this work, we utilize carboxyl group footprinting MS to examine the effect of dimerization and phosphorylation on recombinantly expressed HER2 and HER3 kinase domains. In vitro, the kinase domain interactions are weak and do not occur readily in solution, but they occur readily when the kinases are anchored to a surface. Using a previously described in vitro model system, dimerization was achieved using an N-terminal His 6 tag to orient the kinase domain monomers on the surface of a nickel-chelating liposome, increasing the local concentration and facilitating protein-protein interaction (1,24). Carboxyl group footprinting confirms the orientation of the heterodimer with HER2 and HER3 occupying the acceptor and donor positions, respectively. Additionally, we observed conformational changes in the HER2 activation loop upon heterodimerization and the addition of either ATP or its nonhydrolyzable analog AMP-PNP. We observed similar activation loop conformational changes in the nonreceptor tyrosine kinase Src using carboxyl group footprinting MS. We performed molecular dynamics simulations on a model of the HER2-HER3 asymmetric dimer and identified several inter-and intramolecular interactions. Finally, we performed functional analyses on footprinted residues and cancer-associated mutations, evaluating them within the context of the new structural model. These results provide the first structural insights into an ErbB family heterodimer and advance the understanding of ErbB family kinase activation.

EXPERIMENTAL PROCEDURES
Protein Expression and Liposome Preparation-HER2 and HER3 kinase domains were expressed as described previously (24). Briefly, HER2 kinase domain (residues 704 -1029) and HER3 kinase domain (residues 675-1001) were cloned into pFastBacHT vector. Site-directed mutagenesis was performed using the QuikChange II method (Agilent), and constructs were confirmed by DNA sequence analysis. Recombinant baculovirus was made following the Bac-to-Bac protocol (Invitrogen). Spodoptera frugiperda Sf9 cells were infected with baculovirus at one multiplicity of infection and harvested 48 h post-infection. His 6 -tagged HER2 kinase domain was purified using Ni-NTA beads (Qiagen). His 6 -tagged HER3 kinase-domain was purified using Ni-NTA beads followed by gel filtration chromatography. Numbering of the HER2 and HER3 residues is based on their crystal structures (3PP0 and 3LMG, respectively). Liposomes were prepared as described previously (24) using the reverse phase method detailed by Szoka et al. Protein Dimerization and Phosphorylation-Proteins were kept in stock solutions containing 20 mM Tris-HCl, 150 mM NaCl, and 1 mM dithiothreitol (DTT), pH 8.0. The HER2 stock solution also contained 0.01% Tween 20. Dimerization experiments were performed by equilibrating HER2 and HER3 in a 1:1 ratio with nickel liposomes on ice for 15 min. Protein phosphorylation was initiated by adding ATP and MgCl 2 to the dimer solution (final concentration 100 M ATP and 10 mM MgCl 2 ). Evaluation in the presence of a nonhydrolysable ATP mimetic was performed similarly, but with 100 M AMP-PNP instead of ATP. Samples were then subjected to carboxyl group footprinting after incubating for 5 min on ice after addition of ATP and MgCl 2 .
Carboxyl Group Footprinting Reaction and GeLC-Mass Spectrometry-Labeling reactions and subsequent LC-MS/MS analyses were performed in triplicate for all conditions. Protein samples (10 l of 2.5 M) in 20 mM Tris-HCl, 100 mM NaCl, pH 7.5, was combined with 0.5 l of 2 M GEE. The reaction was initiated with the addition of 0.5 l of 50 mM EDC. Reactions were quenched after 10 min with the addition of 11 l of 1 M ammonium acetate. Immediately after quenching the reaction, 5ϫ SDS-PAGE sample-loading buffer, containing DTT as the reducing agent, was mixed with each sample, and the mixture was boiled for 5 min. The samples were loaded onto a 10% SDS-polyacrylamide gel and detected by SimplyBlue SafeStain (Invitrogen). In-gel digestion was performed as per Shevchenko et al. (26). Tryptic peptides were separated by a reverse-phase Eksigent 1Dϩ nano-liquid chromatography system (Eksigent) directly coupled to an Orbitrap Velos-Pro mass spectrometer (Thermo Fisher Scientific). The resulting RAW files were subjected to chromatographic alignment, peak quantification, and peak list generation using Progenesis LC-MS (Nonlinear Dynamics). Peptides were identified in MASCOT server 2.3 (Matrix Science) using a custom database containing Src in addition to the HER2 and HER3 kinase domain construct sequences and a list of common contaminants added to the Escherichia coli K12 database (4,280 total sequences) (27). Cysteine carbamidomethylation was considered as a fixed modification, although custom-built GEE modifications (ϩ57.021464 Da and ϩ85.052764 Da specific to glutamic and aspartic acid), methionine oxidation, and phosphorylation of serine, threonine, and tyrosine residues were all considered as variable modifications. Other considerations for the MASCOT search include the following: enzyme, trypsin; maximum number of missed cleavages, 2; peptide mass tolerance ϭ 10 ppm; fragment ion tolerance 0.8 Da; instrument type, ESI-trap. Residue level quantification was achieved using Progenesis LC-MS to quantify feature extracted ion chromatograms in addition to an Excel-Based Visual Basic Application described previously (28).
Antibodies and Western Blotting-Blot for Src phosphorylation was performed by incubation in 1:2,000 dilution of 4G10 anti-Tyr(P) antibody (Upstate Biotechnology, Lake Placid, NY) followed by a 1:10,000 dilution of ECL peroxidase-labeled antimouse antibody (Amersham Biosciences). Data shown are representative of at least two experiments.
Homology Modeling and Molecular Dynamics-A HER2-HER3 heterodimeric structural model was built using Modeler (29). Initially, individual models for the HER2 monomer were built with 3PP0 x-ray structure as the template and HER3 monomer with 3LMG as the template. The heterodimer was then built by replacing the HER2 donor kinase in the HER2 homodimer model with HER3. This is used as the starting structure for Molecular Dynamics simulations. A total of three simulations each of 100 ns duration was carried out for both the heterodimer and individual monomers. Simulations were carried out using NAMD (30,31) in explicit water under standard room temperature (300 K) and pressure (1 atm) conditions in an NPT ensemble. AMBER (32) forcefield was used for the simulations. SASA and H-bond analysis were carried out for the last 25 ns of simulations using Visual Molecular Dynamics (33).
Kinase Assays-Radiometric kinase assays were performed as per Ref. 24. Briefly, wild type and mutant HER2 and HER3 constructs were evaluated in solution (monomers) or when attached to nickel liposomes to induce dimerization. Assays were performed with 100 M of the HER2 optimized peptide substrate biotin-GGMEDIYFEFMGGKKK (34) in 30 mM Tris, pH 7.5, 10 mM MgCl 2 , 100 M Na 3 VO 4 , 2 mM DTT, 100 M ATP, 1 Ci of [␥-32 P]ATP, 60 mM NaCl, 5% glycerol, 0.005% Tween 20 in 25-l total reaction volume. The reactions were initiated with the addition of 1.25 M kinase, allowed to react for 8 min at 30°C, and quenched with EDTA. Avidin (100 g) was added, and samples were transferred to centrifugal filtra-tion units with 30,000 molecular weight cutoff (Millipore). Samples were washed three times with 0.5 M sodium phosphate, 0.5 M NaCl, pH 8.5. Retained 32 P was measured by scintillation counting. For the evaluation of kinase activity on liposomes, kinases were incubated on liposomes on ice for 15 min before initiating the reaction. The total lipid concentration was 0.5 mg/ml. For heterodimerization reactions, kinases were mixed prior to the addition of liposomes. Statistical comparisons were performed using the Student's t test, and results are representative of at least two independent experiments.

RESULTS
Carboxyl Group Footprinting Mass Spectrometry-We recombinantly expressed and purified His 6 -tagged HER2 and HER3 kinase domains and formed dimers by binding them to liposomes containing 5 mol % nickel-chelating lipid (Ni-NTA-DOGS). We performed carboxyl group footprinting experiments on four conditions as follows: kinase monomers in solution, kinase heterodimers on liposome, heterodimers on liposome with the addition of ATP, and heterodimers on liposome with the addition of the nonhydrolyzable ATP analog, AMP-PNP, and we examined the relative extents of modification of aspartic and glutamic acid residues with the EDC-mediated addition of GEE (supplemental Table S1). Carboxyl group footprinting using GEE labeling produces a mass shift of 85 Da (or 57 Da after ester hydrolysis) that can be localized and quantified using LC-MS/MS (Fig. 1).
Footprinting between HER2 and HER3 Monomers and Heterodimer-The asymmetric dimer model for ErbB family kinase activation postulates that specific residues on the C-terminal lobe of one kinase monomer contact N-terminal lobe residues of another monomer (1). In this arrangement, the former kinase monomer is often referred to as the "donor" kinase and the latter as the "acceptor." Given the divergence of the HER3 kinase domain's N-lobe relative to the rest of the ErbB family, it is predicted to be a poor acceptor monomer (1,7,9). However, HER3 retains significant homology in its C-lobe with the rest of the ErbB family and can therefore serve as a donor monomer in heterodimers. We combined equimolar amounts of HER2 and HER3 in the presence of liposomes that facilitate dimer formation, and we compared the extents of carboxyl group footprinting to the monomers in solution. Decreased levels of GEE labeling were observed on several residues at the HER2 acceptor interface (Glu-717 and Glu-719) and ␣C-helix (Glu-766, Asp-769, and Glu-770) ( Fig. 2A) and additionally at the HER3 donor interface (Asp-903, Glu-906, Glu-909, Asp-920, and Asp-932) (Fig. 2B). Using previously published crystal structures of HER2 and HER3 kinase domains, we generated a three-dimensional model of a HER2-HER3 asymmetric heterodimer. Specifically, the C-lobe region of HER3 kinase domain structure (Protein Data Bank (PDB) code 3LMG) was aligned to the C-lobe region of the donor monomer in the HER2 kinase domain homodimer structure (PDB code 3PP0) based on the significant homology between these domains across all EGFR/ErbB family members (9,10). This model, shown in Fig. 2C, highlights GEE-labeled acidic residues colored based on their extents of modification when dimerized relative to monomers in solution. The carboxyl group footprinting MS results support the model as evidenced by residues having decreased GEE labeling (Fig. 2C, highlighted in red) being located at the dimer interface. We do not detect changes in GEE labeling on HER2 donor interface residues ( Fig. 2A, E936, E939, E992, and D993) or at the HER3 acceptor interface (Fig. 2B, E687 and E689). Based on the ana-lytical reproducibility, the threshold of detection is ϳ10%. Therefore, if HER2 or HER3 homodimers or heterodimers in the opposite configuration were present, they are minor species representing less than 10% of the population. The HER2-HER3 heterodimer with HER3 as donor and HER2 as accep-  tor is the major species, accounting for greater than 90% of the dimer population.
Activation Loop Conformation Changes in Presence of Nucleotide-Differential extents of GEE labeling were also observed for acidic residue side chains that are not implicated in intermolecular interactions. The activation loop consists of residues 863-884 in HER2 and 833-854 in HER3, each beginning with the highly conserved DFG motif and extending to the WMALE motif. As shown in Fig. 3A, HER2 Asp-863 in the DFG motif exhibited increased GEE labeling compared with its monomeric state when dimerized on the nickel liposomes, and this increased further after the addition of ATP or the nonhydrolyzable ATP analog AMP-PNP. Although no significant change was detected with dimerization alone, the five acidic residues (Asp-871, Asp-873, Glu-874, Glu-876, and Asp-880) along the activation loop of HER2 showed increased GEE labeling when dimerized on the nickel liposomes and incubated with ATP or AMP-PNP (Fig. 3A). In contrast, the DFG aspartate in HER3 (Asp-833) and other acidic residues along the HER3 A-loop (Asp-838, Asp-843, Asp-844, and Glu-851) did not show any statistically significant changes in GEE labeling (Fig. 3B). The relative levels of autophosphorylation were also measured for HER2 and HER3 in each experimental condition. High levels of phosphorylation were only detected for HER2 Tyr-877 when in the presence of ATP (Fig.   3C), but the increases in GEE labeling of the HER2 A-loop occurred with either ATP or AMP-PNP addition. These results suggest that the HER2 A-loop adopts an open conformation (Fig.  3D) independently of Tyr-877 phosphorylation, but with the requirement for both dimerization and nucleotide binding.
Footprinting of Src Activation Loop-To ensure that the A-loop conformational changes observed in HER2 kinase domain were in fact the result of adopting open versus closed conformations (Fig.  3D), we also performed carboxyl group footprinting experiments on the cytoplasmic tyrosine kinase Src. Src is not dependent on dimerization for activation and so could be examined with the addition of ATP, AMP-PNP, or buffer without the need for liposomes. As indicated in Fig. 4, we saw a similar increase in GEE labeling of acidic side chains in the Src A-loop with the addition of ATP, correlating with detection of increased phosphorylation of the A-loop autophosphorylation site (Tyr-416) by mass spectrometry and Western blot (Fig. 4B). However, the same increase in GEE labeling of A-loop residues did not occur when Src was incubated with AMP-PNP (Fig. 4A)

Molecular Dynamics of HER2-HER3 Asymmetric Heterodimer-To investigate what interactions may play a role in dimerization
and stabilize the open conformation of the A-loop, we subjected our three-dimensional model of the HER2-HER3 kinase domain heterodimer (Fig. 2C) and their respective kinase domain monomer structures to molecular dynamics simulations. Homology modeling of HER2-HER3 heterodimer was accomplished using the HER2 homodimer crystal structure as template. 100-ns simulations of the heterodimer and individual monomers were performed using NAMD with AMBER forcefield parameters. A probe radius of 1.4 Å was used to calculate the SASA for selected acidic residues corresponding to those that were identified using carboxyl group footprinting MS. The analysis was performed on 50 snapshots corresponding to the last 25 ns of the 100-ns simulation for both dimer and monomer model systems.
A comparison of the SASA change between the monomer and heterodimer and the GEE footprinting measurement of the same residues is illustrated in Fig. 5. Most of the changes in SASA for residues corresponding to the HER2 acceptor interface (Glu-717 and Glu-719), ␣C-helix (Glu-766 and Asp-769), and the donor interface (Glu-936, Glu-992, and Glu-993) agree well with the GEE footprinting analysis (Fig. 5A). In the case of Glu-744, the residue is predominantly solvent-accessible in the simulated heterodimer, but in monomer simulations the residue is involved in partial interactions with Lys-777. A similar rationale causes the discrepancy with respect to simulation of SASA changes with Glu-975. Likewise, we see strong agreement of SASA calculations from the MD simulation with the carboxyl group footprinting MS data for HER3 (Fig. 5B). There are no significant changes in SASA at the HER3 acceptor interface (Glu-689, Glu-712, Glu-714, and Asp-739), although decreases in SASA are observed at the HER3 donor interface (Asp-903, Glu-909, and Asp-920). The MD simulation shows that Asp-932 is near the HER3 donor interface but is still solvent-accessible in the heterodimer. Differences between the MD and chemical footprinting measurements may result from the lack of the lipid surface and the His tag construct in silico or from the difference in the time scale of the simulation (100 ns) versus the duration of the carboxyl group footprinting experiment (minutes). However, despite these differences, the changes in SASA observed by MD agrees well with the carboxyl group footprinting results.
The MD simulation on the HER2-HER3 heterodimer identified several persistent H-bonds (Table 1 and Fig. 6). At the HER2-HER3 dimerization interface, there are two persistent interactions, including HER2-Lys-716 to HER3-Glu-909 (Fig.  6A) in addition to HER2-Glu-717 to HER3-Lys-907. We also observe several intramolecular salt bridges and hydrogen bonds involved in coordination of the catalytic loop with the ␥-phosphate of ATP (Fig. 6, B and C) as well as stabilize the open conformation of the unphosphorylated HER2 activation loop; however, these interaction are more dynamic and variable, involving Asp-769 and Tyr-772 and Asp-871 on Her2 and Asp-920, Arg-948, and Arg-951 on HER3 (Fig. 6D).
Functional Analysis of Footprinted Residues and Known Cancer-associated Mutations-We evaluated the effects of residues involved in HER2-HER3 dimerization. These analyses utilized in vitro kinase assays in which recombinant His 6 -tagged wild type or mutant HER2 and HER3 constructs were assayed as  monomers or attached to nickel liposomes to induce dimer formation. Furthermore, several interface residues have known cancer-associated mutations, including HER2 D769H and D769Y, HER3 E909G, and HER3 R948K (38 -41).
We measured kinase activity of HER2 wild type, D769A, D769H, and D769Y (Fig. 7A). As we have shown previously, the attachment of the HER2 and HER3 kinase domains to nickel liposomes resulted in increased specific activity relative to the kinases in solution alone (24). The specific activity of HER2 D769A was greatly reduced both as monomers and heterodimers with HER3 WT. HER2 D769H and D769Y showed 3-fold more specific activity in heterodimers, in addition to increased activity as monomers.
We also assessed the kinase activity of mutations of two HER3 residues relative to wild type. HER3 Glu-909 was determined to participate in the dimer interface (Fig. 2B). We assessed the activity of several mutations of these side chains, including E909G, E909A, and E909K (Fig. 7B). E909G, also numbered E928G when including the signal peptide, has been identified as a cancer-associated mutation (42)(43)(44)(45). Wild type HER3 kinase domain showed a 5.7-fold increase in kinase activity when dimerized with HER2 relative to HER2 monomer activity. To determine whether the acidic side chain was necessary for activation, we assessed the activity of HER3 E909K, which resulted in increased activation of HER2 relative to HER3 wild type, a 10-fold increase over HER2 monomer activity. HER3 E909A showed a further increase in HER2 activation (20-fold), although E909G showed the greatest increase with a 32-fold increase in HER2 activation.
In the molecular dynamics simulation, HER3 Arg-948 was implicated in the intermolecular stabilization of the HER2 activation loop (Fig. 6D). An R948K mutation has likewise been observed in cancer cell lines derived from many different tissues (44). HER3 R948K gave a marginal increase in activation of HER2 (6.8-fold relative to HER2 in solution) (Fig. 7C) relative to HER3 WT. However, the increase was not statistically significant. Eliminating the basic side chain with an R948A mutation produced a modest yet statistically significant decrease in HER2 activation.

DISCUSSION
This study provides the first structural characterization of HER2-HER3 asymmetric heterodimers on a lipid membrane surface. Carboxyl group footprinting MS confirms the asymmetric dimer model in which HER3 and HER2 are configured as the donor and acceptor kinases, respectively. A-loop motion, associated with HER2 activation, is not exclusively governed by tyrosine phosphorylation, but it occurs as a result of dimerization and nucleotide binding. Determining protein structure and motion is an important aspect of characterizing protein complexes and informing the design of inhibitors. Molecular dynamics simulations of the HER2-HER3 dimer model support carboxyl group footprinting MS results and identify key salt bridges at the heterodimer interface. MD has implicated several intra-and intermolecular interactions that provide additional stability to the open, nonphosphorylated A-loop conformation. Combining existing crystal structures with the solution-based information obtained with carboxyl group footprinting MS and the atomistic information provided by MD simulations produces a powerful platform for characterizing protein structural dynamics.
HER2 has been shown to be the preferred heterodimerization partner for other ErbB family members, especially HER3 (6 -8). Our analysis of an equimolar mixture of HER2 and HER3 also suggests that heterodimer formation is preferred over the formation of HER2-HER2 or HER3-HER3 homodimers. The carboxyl group footprinting experiments performed on HER2-HER3 dimers confirm the hypothesis that HER3 functions as the donor kinase and HER2 as the acceptor in an asymmetric configuration. Decreased GEE labeling was observed on acidic residues located near the dimer interface of HER3's C-lobe and HER2's N-lobe, including decreased GEE labeling of HER2 Glu-717 and HER3 Glu-909, which in MD simulations are involved in salt bridge interaction with HER3 Lys-907 and HER2 Lys-716, respectively (Fig. 6). Analogous salt bridge interactions were previously observed in HER4 homodimers (23). Hydrophobic interactions also play a key role at the dimerization interface, and we will further explore those interactions in the future (46).
The A-loop regulates access of target substrates to the catalytic loop, which facilitates phosphoryl transfer. Upon kinase activation, the A-loop undergoes a dramatic conformational shift in which it extends to uncover the kinase substrate-binding pocket (Fig. 3D) (36,47). In many protein kinases, including the Src family of tyrosine kinases, the A-loop extension is stabilized by phosphorylation of a tyrosine residue located within the loop itself and is required for kinase activation (48). The study of the Src family kinase Lck crystal structure, in which Tyr-394 on the activation loop was phosphorylated, shows the A-loop in an open and extended conformation with a hydrogen bonding network extending from the DFG aspartate to the phosphotyrosine, including interactions with the catalytic loop and ␣C-helix (49). However, the requirement for A-loop tyrosine phosphorylation in EGFR and HER2 is still controversial (50 -55). Previous molecular dynamics studies of homologymodeled HER2, based on the EGFR crystal structure, similarly suggested that a hydrogen bond network secured the open A-loop conformation upon tyrosine phosphorylation (46). The presence of several aspartic and glutamic acid residues along the activation loop makes carboxyl group footprinting MS a powerful technique for monitoring its conformational changes. Carboxyl group footprinting analysis of Src revealed a marked increase in the GEE labeling of A-loop acidic residues in the presence of ATP, indicative of the open A-loop observed in crystallographic studies. We observe an almost identical increase in HER2 A-loop modification when dimerized with HER3 in the presence of ATP, suggesting a similar open and extended A-loop conformation as that observed in Src. In contrast to Src, however, we also observe increased A-loop labeling with the addition of AMP-PNP, a nonhydrolyzable ATP analog. We thus conclude that the open A-loop conformation in HER2 is dependent on dimerization and nucleotide binding, but it does not require A-loop tyrosine phosphorylation.
Molecular dynamics simulations were able to explore interactions that play a role in stabilizing the open conformation of the unphosphorylated HER2 activation loop. Several H-bond FIGURE 7. Effect of HER2 and HER3 kinase domain mutations on specific activity. A, specific activities of HER2 WT and mutants Her2 D769A/D769H/ D769Y in solution and dimerized with HER3 WT. B, HER3 WT and mutants HER3 E909G/E909A/E909K with fold-change calculated relative to HER2 WT alone with no liposomes. C, HER3 WT and mutants HER3 R948A/K with foldchange calculated relative to HER2 WT alone with no liposomes. Student's t test were performed for pairwise comparison of HER3 WT with either R948K or R9489A.
interactions are implicated in the open conformation. Interactions between the HER2 ␣C-helix and the A-loop, including Glu-766 -Arg-868, and interactions involving Lys-753 and Asp-863 with the coordination with Mg 2ϩ and ATP in the catalytic pocket have been previously described and also appear in our simulations (1,56). We also observe H-bond interactions involving Asp-845, Arg-849, and Asn-850 in the catalytic loop that have been implicated in coordination and transfer of the ␥-phosphate of ATP (Fig. 6C) (46,56,57). MD simulations also revealed novel interactions that stabilize the unphosphorylated A-loop of HER2. A persistent backbone-backbone interaction between Tyr-877 and Phe-899 is likely to be further stabilized by the parallel-displaced -stacking of the two aromatic side chains, and interaction is not accounted for by the simulation that would provide 0.5-1.0 kcal/mol more stability (Fig. 6E) (58). We found interactions between several groups of residues, Asp-769, Tyr-772, and Asp-871 on HER2 with Asp-920, Arg-948, and Arg-951 on HER3, provide a possible mechanism for intermolecular stabilization of the open and extended activation loop (see, for example, Fig. 6D). Near the C-terminal end of the activation loop, several H-bonds may form between the extended A-loop and the C-lobe of the kinase domain. The implication of these interactions on the regulation of kinase activity warrants further study.
The development of this new HER2-HER3 dimer structural model provides a new framework for interpreting mutation data. The HER2 kinase domain mutations D769H and D769Y have been observed in breast, lung, gastric, and colorectal cancers (38 -40). We recently reported that D769H/D769Y were activating mutations in breast cancer and possessed greater specific activity when assayed as HER2 homodimers (38). The kinase assay data performed in this work show that both D769H and D769Y are more activating relative to wild type in the context of the HER2-HER3 heterodimer (Fig. 7A). Furthermore, D769A results in a loss of kinase activity. Potential interpretations of these results include increased hydrophobic interaction between HER2 D769H/D769Y and HER3 Ile-919 and Met-923 or intramolecular interactions with the HER2 A-loop. Our MD simulations show the interaction of Asp-769 with the amide proton of HER2 Leu-869 in the A-loop, providing additional stability to the open conformation. Mutation to histidine or tyrosine increases the side chain length and hydrophobicity, which could provide additional stability to the open activation loop through interactions with the HER2 Leu-869 side chain. This additional stability accounts for the observed increases in D769H/D769Y monomer and dimer activity relative to wild type (Fig. 7A). In contrast, replacing the aspartate with a shorter nonpolar alanine would result in a loss of those stabilizing interactions. The corresponding loss of kinase activity for D769A supports this reasoning.
Our molecular dynamics simulations suggest the interaction of HER3 Arg-948 with HER2 Asp-871 plays a role in the stabilization of the open and extended HER2 A-loop (Fig. 6D). We assessed the effect of this interaction on dimer activity by generating a HER3 R948A mutant. Compared with wild type HER3, R948A showed a modest but statistically significant decrease in kinase activity (Fig. 7C). We also assessed the effect of a R948K mutant, which has been observed in several cancer cell lines.
R948K showed a slight increase in activity, which was not statistically significant. R948K is a conservative substitution and may be a natural polymorphism with no oncogenic consequences.
HER3 Glu-909 participates in the heterodimer interface (Figs. 2B and 5), and MD simulation suggested Glu-909 forms a salt bridge with HER2 Lys-716 (Fig. 6A). HER3 E909G has recently been identified as an oncogenic mutation in gastric and breast cancers, showing increased Akt and ERK signaling in the presence of HER2 (45). We created HER3 mutants E909K, E909A, and E909G to assess how changes at this location affected the specific activity of the HER2-HER3 dimer. Interestingly, all three mutants displayed increased activation of HER2 relative to wild type HER3, with E909G showing the greatest effect (Fig. 7B). Reversing the charge of Glu-909 by mutating to lysine would disrupt the HER3 Glu-909-HER2 Lys-716 salt bridge, but it allows the formation of other ionic interactions with the neighboring HER2 Glu-717. In contrast, mutation of HER3 Glu-909 to alanine or glycine disrupts this charge interaction and allows the dimerization interface to shift toward the HER2 ␣C-helix, favoring increased activation of HER2. These results suggest that Glu-909 is an important modulator of activity at the HER2-HER3 dimer interface and that mutations at that site are activating through an allosteric mechanism.
In conclusion, this study extends our understanding of the structural interactions that contribute to the HER2-HER3 asymmetric dimer using both direct measurements by carboxyl group footprinting MS and MD simulations. We have also demonstrated that HER2 activation can occur independent of tyrosine phosphorylation on its activation loop. Given the importance of HER2-HER3 interaction in several cancer types and the clinical interest in mutations of HER2 and other members of the ErbB family (38), the ability to rapidly evaluate and understand their consequences is critical. This study provides valuable insights into ErbB family kinase structure and interactions, which can guide the design of future therapies.