Conformational flexibility of the ErbB2 ectodomain and trastuzumab antibody complex as revealed by molecular dynamics and principal component analysis
- First Online:
- Cite this article as:
- Franco-Gonzalez, J.F., Cruz, V.L., Ramos, J. et al. J Mol Model (2013) 19: 1227. doi:10.1007/s00894-012-1661-3
- 569 Views
Human epidermal growth factor receptor 2 (ErbB2) is a transmembrane oncoprotein that is over expressed in breast cancer. A successful therapeutic treatment is a monoclonal antibody called trastuzumab which interacts with the ErbB2 extracellular domain (ErbB2-ECD). A better understanding of the detailed structure of the receptor-antibody interaction is indeed of prime interest for the design of more effective anticancer therapies. In order to discuss the flexibility of the complex ErbB2-ECD/trastuzumab, we present, in this study, a multi-nanosecond molecular dynamics simulation (MD) together with an analysis of fluctuations, through a principal component analysis (PCA) of this system. Previous to this step and in order to validate the simulations, we have performed a detailed analysis of the variable antibody domain interactions with the extracellular domain IV of ErbB2. This structure has been statically elucidated by x-ray studies. Indeed, the simulation results are in excellent agreement with the available experimental information during the full trajectory. The PCA shows eigenvector fluctuations resulting in a hinge motion in which domain II and CH domains approach each other. This move is likely stabilized by the formation of H-bonds and salt bridge interactions between residues of the dimerization arm in the domain II and trastuzumab residues located in the CH domain. Finally, we discuss the flexibility of the MD/PCA model in relation with the static x-ray structure. A movement of the antibody toward the dimerization domain of the ErbB2 receptor is reported for the first time. This finding could have important consequences on the biological action of the monoclonal antibody.
KeywordsExtracellular ErbB2 receptorHerceptinMolecular dynamicsPrincipal component analysisTrastuzumab
The human epidermal growth factor receptors (EGFR) HER1 (ErbB1, EGFR), HER2 (ErbB2), HER3 (ErbB3) and HER4 (ErbB4) belong to the family of receptor tyrosine kinase proteins. These receptors are engaged in the regulation of many processes such as cell proliferation, differentiation and apoptosis. Loss of regulation of these receptors has a great impact in a number of human diseases, such as cancer [1, 2].
The ErbB1, ErbB3 and ErbB4 receptors exhibit a similar tethered structure of the extracellular ligand-binding region when being in the inactive state. Upon ligand activation, a conformational rearrangement from the tethered (inactive) to the extended conformation (active) takes place that permits the homo- and heterodimerization of the receptor through domain II interactions. These active receptor dimers are involved in signaling activity and regulatory protein processes [11, 12]. By contrast, the ErbB2 receptor is characterized by an extended conformation, without ligand binding activation, ready to form dimers with other receptors through an exposed dimerization arm that is located in domain II [3, 8, 12–14]. Thus, over-expression of ErbB2 leads to EGFR receptor activation in tissue culture, while over-expression of other EGFR receptors is not active unless a ligand is added .
Trastuzumab, also known as its commercial name Herceptin, currently constitutes a part of the immunotherapy treatment of advanced breast cancers, i.e., those with extensive metastasis, and in general, solid tumors over-expressing ErbB2. Different clinical trials confirm the efficiency of the antibody as an anticancer treatment. The monoclonal antibody trastuzumab can bind to ErbB2-ECD domain . The trastuzumab-fab that binds to domain IV of the ErbB2-ECD categorizes this site as a possible target for anticancer therapies. Additionally, some authors claim, based on these structural studies, that trastuzumab is not effective in blocking dimerization of ErbB2 with ligand activated EGFR or ErbB3 [16, 17]. However it has been reported that ligand independent ErbB2/ErbB3 complex is disrupted by trastuzumab . On the other hand, the Pertuzumab antibody, which binds to domain II (dimerization arm), has shown to be effective to disrupt the ErbB2/ErbB3 ligand activated complex but ineffective for the independent ligand specie. In the light of these findings, it has been hypothesized that the ligand independent interaction between ErbB2/ErbB3 is different from the ligand-induced dimerization. Thus, the domain II-domain II interfaces may not mediate in the ligand-independent complex . In contrast, very recent single-molecule force spectroscopy studies suggest a mechanism of blocking of the heterodimerization of ErbB2/trastuzumab and ErbB3 receptors even in presence of the heregulin (HRG) ligand .
Several computational studies based on computer simulations have tackled the structure and interactions between transmembrane ErbB2 domains in lipidic bilayer models [19–24] and in tyrosine kinase domain activation [25–27]. However, computational studies of the interaction of ErbB2 ectodomain (ErbB2 ECD) with antibodies, such as trastuzumab, are less considered. Wang et al. have studied the binding regions of ErbB2 ECD with inhibitory (trastuzumab) and non-inhibitory (HF) monoclonal antibodies using a combination of site-directed mutagenesis, docking and short molecular dynamics simulations. They concluded that the inhibitory trastuzumab antibody binds to domain IV (C-terminal region) of the ECD and that the non-inhibitory HF antibody recognizes domain II (N-terminal region) . In other study, the 3-D structure of an auto-inhibitor (herstatin)/ErbB2 ECD complex has been proposed using molecular docking methods. The binding site of herstatin of the ErbB2 ECD domain was proposed to be at the S1 domain (here domain II). That observation was verified by inmunoprecipitation, confocal microscopy and fluorescence resonance energy transfer experiments . Very recently, Fuentes et al. have published a 20 ns MD study and a fluctuation analysis of the interaction between ErbB2 and a combination of trastuzumab and pertuzumab antibodies . Their simulations throw light on two important aspects of the interaction: on one hand, the fluctuations in domain II are enhanced by the trastuzumab binding, and on the other hand, the existence of a cooperative mechanism between these two antibodies and the ErbB2 ECD that could avoid the homo and heterodimerization of ErbB2 with other members of the EGFR family.
In our work we performed a long 170 ns molecular dynamics (MD) simulations of the ErbB2 ECD/trastuzumab complex (ErbB2/trastuzumab) to elucidate details of the interaction between its components using as starting point the x-ray crystal structure . Additionally, an analysis of the large scale fluctuations has been performed using the principal component analysis (PCA). It should be mentioned here the experimental finding that the variable trastuzumab domains bind to domain IV (juxtamembrane domain) of ErbB2 ECD. These interactions have been largely conserved along the MD simulation. However, large fluctuations are observed which allow the formation of novel contacts between the dimerization arm of domain II ErbB2 and the trastuzumab residues in the CH domain. To our best knowledge this interaction has not been yet reported.
The aminoacid sequence intervals are 1–165, 166–310, 311–480 and 481–607 for domain I, II, III and IV of the ErbB2-ECD structure, respectively. The aminoacid sequence intervals of the variable domains of the heavy chain (VH) and light chain (VL) of trastuzumab are 1–119 and 1–117, respectively. And finally, the aminoacid sequence intervals of the constant domains of the trastuzumab heavy (CH) and light (CL) chains are 127–220 and 114–214 respectively.
The initial model of the ErbB2 ectodomain and trastuzumab-fab (extra-ErbB2/trastuzumab) was directly taken from the 3-D crystal structure deposited in the Protein Data Bank server (PDB code: 1N8Z) . The missing residues 102–110, 303–305, 361–364 and 581–590 have been modeled based on homologue sequences using the PRODAT database implemented in Sybyl 8.0 . The loop fragment that gave the best geometric fit, based on the homology score and RMS fit, was automatically incorporated into the model [31, 32]. The side chains were built residue by residue, one at a time, using the rotamer library of Sybyl using a scan angle of 30 ° and VDW factor of 0.9. The selected side chain conformation for each modeled residue is the one that presents the fewest VDW contacts with the rest of the molecule. Finally, the structure was relaxed for 2500 steps using the steepest descent minimization algorithm as implemented in GROMACS 4.5.3 . The PROCHECK analysis of the added fragments reports a Ramachandran plot with the following statistics: 64.3 % in most favored regions, 21.4 % in additional allowed regions, 14.3 % in generously allowed regions and 0 % in disallowed regions. From that point of view the structure of the added fragments seems to be quite reasonable.
We assumed that the pKa of the individual amino acid residues at physiological pH does not change when assembled into the protein receptor. Thus, histidine (H) residues remained neutral; lysine (L) and arginine (R) were protonated and aspartic (D) and glutamic (E) acids were deprotonated. The resulting total charge for the complex was −10 e units. The system was solvated by 60,717 water molecules and 10 Na+ ions have been added to yield an electrically neutral system. Periodic boundary conditions were applied along the three dimensions. The initial rectangular box lengths were 13.9 nm, 12.5 nm and 11.5 nm respectively. The system was equilibrated in a 2 ns NPT-MD simulation with position restraint for all protein atoms.
The OPLS force field [34–36] for protein and the SPC model  for water were used along the whole work. Short range repulsion-dispersion interactions were smoothly truncated at 10 Å. The particle mesh Ewald (PME) method [38, 39] was used to calculate long range electrostatic interactions, with a maximum grid spacing of 2.5 Å and using fourth-order (cubic) interpolation for the fast Fourier transforms. The temperature was kept constant at 300 K by coupling the protein, the ions and the solvent independently to an external bath using the Berendsen algorithm  with a coupling constant of 0.2 ps.
We used isotropic scaling for the pressure (1 bar). A coupling constant of 1.0 ps and a compressibility of 4.5 × 10-5 bar-1 were used in the Berendsen algorithm . The dynamics were run using the velocity Verlet integrator, with a time step of 2 fs and bonds constrained conditions using the LINCS algorithm .
Production dynamics was performed at constant pressure and temperature (NPT ensemble) releasing all constraints on the heavy atoms during 170 ns and accumulating the trajectory frames every 10 ps. All minimizations, restrained and unrestrained MD runs were performed with GROMACS 4.5.3 . Molecular graphics have been drawn using the VMD 1.8.7 package .
Principal component analysis, hydrogen bonds and contact maps
Principal component analysis (PCA) is a method that takes the trajectory of long MD simulations and calculates the dominant modes in the motion of the molecule. Thus, the configurational space is reduced, containing few relevant collective degrees of freedom in which long range fluctuation can be studied [43, 44]. A PCA diagonalizes the covariance matrix of the atom fluctuations from their average trajectory. In this framework, the larger eigenvalues correspond to eigenvectors which explain most of the variance of the atomic fluctuations. The ordering of these eigenvalues gives rise to a small set of modes that capture most of the protein’s fluctuations. We have performed a PCA analysis in order to identify the lowest frequency motions occurring in the ErbB2/trastuzumab complex. Along this work, we make use of the first three eigenvectors, which were projected along the MD trajectory. The g_covar and g_anaeig tools in the GROMACS package were used to perform the PCA analysis.
Hydrogen bond (HB) is considered to exist when both distance between the donor (D) and the acceptor (A) is less than 0.30 nm and the hydrogen-donor-acceptor (HDA) angle is lower than 30 °.
The contact maps show the smallest distance between any pair of atoms belonging to two different residues. The output is a symmetrical matrix of smallest distances between all residues. Plotting these matrices for different time-frames is useful to analyze changes in the structure, and particularly hydrogen bond networks and hydrophobic contacts.
The root-mean square fluctuation for each residue has been calculated using the g_rmsf tool from GROMACS.
The change of secondary structure elements during the simulation was monitored using the program define secondary structure of proteins (DSSP) .
Results and discussion
Stability analysis of the full MD trajectory
The time evolution of the Cα-RMSD (root mean square deviation) for the ErbB2-ECD receptor has been used to track the equilibration and any possible reorganization of present domains in the whole complex (see Fig. S2 in the Additional information section). The Cα-RMSD relaxes over the first 70 ns to a value of around 0.41 ± 0.03 nm (averaged over the interval 70–170 ns). These structural changes are mainly due to domains IV and II of the ErbB2-ECD, showing the largest values (Cα-RMSD(70ns-170ns) = 0.27 ± 0.02 and 0.31 ± 0.04 for domains IV and II, respectively) . On the other hand, Cα-RMSD values for the domains I and III keep stable around 0.12 ± 0.02 nm along the whole trajectory. As shown in Scheme 1 and Fig. 1, the domains II and IV are more exposed to the interaction with the VL and CH domains of the antibody structure, respectively.
Principal component analysis
The PCA allows the projection of the complex protein dynamics on a set of collective modes which can be ordered from the largest to smallest contributions of the protein fluctuation variance, as measured by the eigenvalues of the covariance matrix [43, 44]. The largest eigenvalue corresponds to the slowest motion, and so forth.
These eigenvalues involve large motions of the domain II and CH motions which can be confirmed by visualizing the distances between the center of mass of the different domains in ErbB2 and trastuzumab moieties. The approach of II and CH domains is evidenced by a decrease of more than 2 nm of the center of mass distance (CMD). On the other side, the CMD between domain IV in ErbB2 and trastuzumab domains are kept nearly constant throughout the full dynamics and close to those found in the x-ray structure.
In summary, these principal component eigenvector fluctuations result in a hinge motion in which domain II and CH domains approach each other, allowing the formation of some interactions between the dimerizarion arm in the domain II of ErbB2 protein and the trastuzumab residues located in the CH domain. These interactions will be discussed in the next section.
A detailed analysis of the inter-domain hydrogen bonds and electrostatic interactions between ErbB2 and trastuzumab proteins
Domain IV-VH/VL interactions. Comparison with experimental structure
Domain II-CH interactions. A possible explanation about the absence of these interactions in the crystal packing structure
On the other hand, the binding free energy has been calculated according to the molecular mechanics-Poisson-Boltzmann surface area (MM-PBSA) method. The “in-silico” binding energy ΔGbind is quite large (−285.0 kcal mol-1) in comparison with the reported experimental values between −12.4 and −14.0 kcal mol-1 [50, 51]. In this sense, Fuentes et al.  reported an “in-silico” value of ΔGbind = −1144.6 kcal mol-1 using molecular mechanics-generalized-Born surface area (MM-GBSA) approximation without entropic terms. Thus, it seems clear that entropic terms are needed in order to improve the binding energy between the apo-ErbB2 protein and the trastuzumab ligand. Details of the MM-PBSA calculation are given as Supplementary information.
A molecular dynamics and PCA study of the flexibility of the complex between the extracellular domain (ectodomain) of the ErbB2 receptor and the trastuzumab antibody has been presented. The initial structure was prepared from the crystal structure reported by Cho et al. (PDB code: 1N8Z) . From this study the following conclusions can be drawn. Firstly, both secondary structure for the full complex and putative interactions between domain IV of ErbB2 and variable domains (VH and VL) of trastuzumab are well conserved along the molecular dynamics trajectory. Secondly, a hinge move approaching the domain II to the trastuzumab component is revealed by combining the MD trajectory and principal component analysis of the largest eigenvectors. This global motion allows the interaction between the dimerization arm of the ErbB2-ECD domain II and sub-domain CH of the antibody. The effect of this interaction on the heterodimerization of ErbB2 and other EGFR receptors is under study in our group. In addition, we have observed some differences between the MD simulation and the x-ray structure attributed to the crystal packing. Thus, the monomer packing in the crystalline cell hinders the hinge move discussed above due to the presence of two other nearby ErbB2/trastuzumab complexes, thus preventing the interaction between the dimerization arm and the CH domain of its own trastuzumab protein. In any case, we expect that these results are useful to identify the underlying interaction mechanism between receptor and antibody, which could help to design new therapeutic antibodies. The observed interaction of the antibody with the dimerization arm could provide us with new clues to design possible modifications of the antibody that may then enhance its therapeutic power. An effective blockade of the dimerization arm would impinge the ErbB dimerization and consequently would lead to the interruption of the signaling cascade. The simultaneous effect on both ErbB2 domains II and IV exerted by a modified trastuzumab would be of great interest in the treatment of ErbB2 over-expressed tumors.
Thanks are due to the, Comision Interministerial de Ciencia y Tecnologia (CICYT) (MAT2009-12364 and MAT2012-36341 projects) for financial support. The authors also acknowledge Secretaria General Adjunta de Informatica- Consejo Superior de Investigaciones Cientificas (SGAI-CSIC) for technical support during the simulations. One of us (J.R) thanks for financial support through the Ramon y Cajal program, contract RYC-2011-09585. Very fruitful conversations with Dr. Rafael Nuñez during the discussion of literature experimental details are gratefully appreciated.