Investigation of the dynamics of the viral immediate-early protein 1 in different conformations and oligomerization states

The viral immediate-early protein 1 (IE1) is crucial for efficient replication of cytomegalovirus (CMV). A recent crystal structure of the IE1 protein from rhesus CMV revealed that the protein exhibits a novel fold and crystallizes in two slightly different dimeric arrangements. Molecular dynamics simulations and energetic analyses performed in this study show that both dimers are stable and allowed us to identify a common set of five residues that appear particularly important for dimer formation. These residues are distributed over the entire dimer interface and do not form a typical hot spot for protein interactions. In addition, the dimer interface of IE1 proved to include a high portion of hydrophilic interactions pointing toward the transient nature of dimer formation. Characterization of monomeric and dimeric IE1 revealed three sequentially discontinuous dynamic domains that exhibit correlated motion within the domain and are simultaneously anti-correlated to the adjacent domains. The hinge motions observed between the dynamic domains increase the shape complementarity to the coiled–coil region of tripartite motif proteins, suggesting that the detected dynamics of IE1 might be physiologically important by enabling a better interaction with its cellular target molecules.

Recently, the core region (residues 42-393) of rhesus cytomegalovirus IE1 protein was crystallized and revealed a novel protein fold (Scherer et al., 2014). The structure of IE1 is composed of 11 α-helices that align in an up-and-down manner forming an overall elongated shape, which can be coarsely subdivided in an upper and lower head region as well as a stalk region in the middle (Figure 1(A)).
Interestingly, IE1 crystallized in two slightly different dimeric forms, which exhibit an overall similar antiparallel arrangement of the subunits (Figure 1(A)). One of the dimers is formed by two monomers of virtually identical backbone geometry and is therefore termed symmetric dimer (IE1 sy ). The second dimer is composed of two monomers that differ in their backbone conformation and was termed asymmetric dimer (IE1 as ). A comparison of the backbone topology of the individual monomers found in the crystal structures shows that they differ by approximately 2-3 Å (Figure 1(B)-(D)) causing also variations of the atomic contacts in the dimer interfaces.
In the present work, MD simulations were applied to study conformational stability and interface properties of the two IE1 dimers observed in the crystal. To address the dynamics of the monomeric IE1 building blocks, the stability of the IE1 fold was also analyzed for all different conformations of the monomeric chains present in the crystal. Our study revealed the presence of three dynamic domains in IE1 that are connected by hinge regions. Interestingly, the same dynamic domains are still present in dimeric IE1, although the overall flexibility is significantly decreased upon dimerization. Furthermore, the observed dynamics also increases the shape complementarity to the proteins of the TRIM family, suggesting the relevance of these motions for the IE1 recognition of cellular targets.

Methods
Computational investigations were based on the two crystal structures of the IE1 core region (residues M42-N393), namely the asymmetric IE1 as and the symmetric IE1 sy dimer (Scherer et al., 2014). Missing amino acids were separately modeled with ModLoop, which relies on a protocol consisting of the conjugate gradient minimization and molecular dynamics simulation with simulated annealing (Fiser, Do, & Šali, 2000;Fiser & Sali, 2003). Furthermore, unresolved side chains of amino acids were added with Swiss-PdbViewer 4.04 (Guex & Peitsch, 1997) choosing the most favorable rotamer. The final models were analyzed with WHATCHECK (Hooft, Vriend, Sander, & Abola, 1996) revealing neither stereochemical outliers nor clashes for the modeled segments or side chains. For simulation of the IE1 monomers, chains A and B of the IE1 as dimer, as well as one chain of the IE1 sy dimer, were extracted (in this study referred to as IE1 as A, IE1 as B, and IE1 sy C). N-and C-termini were capped with acetyl and N-methylamide, respectively.
All MD simulations were performed in explicit solvent with AMBER11 (Case et al., 2010) using the ff99SB force field (Hornak et al., 2006) and the particle mesh Ewald summation method to handle long-range electrostatic interactions (Darden, York, & Pedersen, 1993). The protonation state of titratable amino acids was calculated by PROPKA 3.0 (Olsson, Søndergaard, Rostkowski, & Jensen, 2011), and charges were neutralized by adding sodium counter ions. An octahedral water box with TIP3P water model (Jorgensen, Chandrasekhar, Madura, Impey, & Klein, 1983) was generated with at least 10 Å distance between any atom of the solute and the end of the periodic box. In addition, the SHAKE procedure was applied to constrain all covalent bonds involving hydrogens (Miyamoto & Kollman, 1992;Ryckaert, Ciccotti, & Berendsen, 1977). All systems in the present work were minimized using the AMBER program SANDER in three subsequent steps in order to relax the initial structures and avoid inappropriately high energies assigned to the atoms. This was followed by a 25 ps equilibration of the systems to 310 K based on a previously established protocol (Dirauf, Meiselbach, & Sticht, 2010;Meiselbach, Horn, Harrer, & Sticht, 2007;Wartha, Horn, Meiselbach, & Sticht, 2005). Non-bonded interactions were calculated with a cutoff of 10 Å and updated every 15 steps. In the production phase, the  (residues 42-74, 112-151, 207-236, and 280-314), and a lower head region (residues 152-206 and 315-393). The overall topology of the IE1 sy dimer is highly similar and mainly differs in the backbone conformation of the monomeric chains (see panels B-D). (B-D) Pairwise RMSD comparison of the IE1 chains forming the IE1 as and the IE1 sy dimers. Views differ by a rotation of~180°around vertical axis with respect of view in (A). The two chains of the IE1 as dimer, which differ in their backbone conformation by an RMSD of 2.5 Å, are colored gray and red, respectively. The two chains of IE1 sy dimer are identical. Thus, only superimpositions for one chain of IE1 sy (green) with the chains of the IE1 as subunits (C, D) are shown. The RMSD of the superposition in (C) is 2.9 Å and in (D) 2.2 Å. systems were simulated either for 50 ns (dimers) or 100 ns (monomers) with a time step of 2 fs.
For analyses, snapshots of the simulation data obtained during the production stage were extracted every 20 ps resulting in 2500 snapshots for the 50 ns simulations and 5000 snapshots for the 100 ns simulation. Analyses were performed using PTRAJ and CPPTRAJ implemented in AMBER tools (Roe & Cheatham, 2013). All root mean square deviation (RMSD) calculations were limited to the backbone atoms (C, C α , and N) of IE1 excluding residues N66-K81, which are partially unresolved in the crystal structure and also very flexible during simulation. Structure comparison between IE1 and TRIM proteins was performed using PDBefold (Krissinel & Henrick, 2004).
For calculation of the solvent accessible surface area (SASA), the linear combination of pairwise overlaps implemented in CPPTRAJ was applied (Weiser, Shenkin, & Still, 1999). The buried accessible surface area (BASA) was calculated as follows: where SASA Acomplex depicts the SASA of protein A in complex and SASA Afree the SASA of protein A in the absence of any other molecule from the same simulation. The same type of calculation was also done for individual residues. In order to calculate the protein-protein interface area (PPIA) in a dimer complex, the average of the BASA of proteins A and B was determined: For calculation of the BASA of hydrophobic residues, the BASA calculation and method were only applied to residues alanine, proline, valine, leucine, isoleucine, methionine, phenylalanine, and tryptophan (Tsai, Lin, Wolfson, & Nussinov, 1997). To save computational costs, all SASA calculations were performed on snapshots taken every 200 ps. For calculations of the binding free energy ΔG b , the MM/GBSA method implemented in AMBER 11 was applied. GB model 2 was chosen in combination with recommended bondi settings for atomic radii (Bondi, 1964;Onufriev, Bashford, & Case, 2004). In solution, the binding free energy is the sum of the molecular mechanics (MM) interaction energy (E MM ) and a solvation term G sol : Figure 2. Secondary structure of the individual chains forming the IE1 as dimer (A, B) and the IE1 sy dimer (C, D) over the simulation time. α-helix is depicted red, 3 10 -helix blue, π-helix yellow, and the H-bonded turn green. β-sheets were not detected.
ΔE MM is calculated with the SANDER module of AMBER and represents the MM interaction energy between protein and its ligand. It comprises bond, angle, and torsion-angle energies as well as electrostatic and van der Waals energies. The solvation term ΔG sol is composed of polar and non-polar energy contributions: The electrostatic term uses the implicit GB solvent model. Nonpolar contributions were calculated as a function of the SASA: where γ = 0.0072 kcal mol −1 Å −2 and b = 0.00 kcalmol −1 . Dielectric constants for solute and solvent were set to 1 and 80, respectively. MM/GBSA energetic analyses were performed over a timeframe of 25 ns for the IE1 dimers complexes (25-50 ns) with snapshots taken at a time interval of 20 ps. The relative change in free energy of binding (ΔΔG b ) for the selected IE1 alanine and glycine mutations was calculated by a single trajectory in silico alanine scan. In the scan, the structures from the trajectory were mutated by truncating the side chains of the residues one at a time. For the calculation, we followed the standard protocol described by Massova and Kollman (1999), which assumes that the entropy of the mutant and the wild type does not differ significantly, and therefore, entropic contributions can be neglected in the calculation of ΔΔG b .

Results and discussion
Conformational stability of dimeric IE1 MD simulations were performed to compare the dynamic properties of the asymmetric (IE1 as ) and symmetric Figure 3. Conformational stability of the IE1 as and IE1 sy dimer. The two chains of the IE1 as dimer are colored black and red, respectively. The two chains of the IE1 sy dimer are colored green and yellow, respectively. (A) RMSD was calculated for each protomer with the equilibrated structure as reference. (B) Backbone RMSF of each chain from the dimer simulation. Structure of (C) IE1 as and (D) IE1 sy dimers at the beginning of the simulation and after 50 ns.
(IE1 sy ) dimer observed in the IE1 crystals. Monitoring the secondary structure over simulation time revealed that the α-helical structure of IE1 is maintained during simulation indicating the overall conformational stability of both crystal forms (Figure 2(A)-(D)). To investigate the stability of the individual chains in more detail, the RMSD of each monomer within the dimers was calculated. RMSD analysis reveals slightly higher deviations from the starting structure in both IE1 sy chains compared to IE1 as with values up to~5 Å (Figure 3(A)). Visualization of IE1 sy during simulation shows that these rather high RMSD values originate from a more pronounced bending of the helices at each end of the stalk region ( Figure 3(D), Figure S2 (C), (D)). In summary, both IE1 systems maintain their helical elongated fold and form stable homodimers.
The next analysis focused on the differences of intermolecular interaction between the two IE1 crystal forms. For that purpose, the mean PPIA created due to dimerization was calculated over simulation time. The average IE1 as dimer interface of 2875 Å 2 is over 400 Å 2 larger than the 2430 Å 2 interface in the IE1 sy dimer. This is in line with the higher interaction energy of −151 kcal/mol for the IE1 as dimer compared to −112 kcal/mol for the IE1 sy dimer. Taken together, the larger PPIA and stronger interaction energy indicate that the chains in IE1 as form a more stable dimer. This is an interesting finding since the IE1 as dimer is asymmetrical suggesting that IE1 might also form heterodimers that do not exhibit symmetry of their binding interface. Table 1. Accessibility of individual residues in the IE1 dimer interface. Listed are all residues that have an average buried accessible surface area (BASA) > 30 Å 2 in one of the IE1 dimers. Corresponding residues in IE1 of HCMV are listed. Amino acids, which were selected for an in silico alanine scan, are listed in Figure 4. Interface properties of dimeric IE1 Having examined the overall stability of the two IE1 dimers by MD simulations, the dimer interaction and interface were analyzed in more detail. In order to identify amino acids that are important for dimerization, the BASA of each residue in the IE1 dimers was calculated revealing its contribution to the dimer interface. Table 1 lists the mean BASA of those residues with the highest contribution to the dimer interface area. As expected from previous analyses, there are differences between the IE1 as and the IE1 sy dimer. For instance, K300 yields a BASA of~95 Å 2 in the IE1 as dimer, whereas it contributes only 41 Å 2 to the dimer interface in the IE1 sy system. However, there are also residues that have high BASA values in both dimers. Out of these 35 detected amino acids, 15 residues are evolutionary conserved thus representing candidate hotspots for protein-protein interactions (Guharoy & Chakrabarti, 2010;Ma, Elkayam, Wolfson, & Nussinov, 2003;Valdar & Thornton, 2001). Interestingly, out of those 15 residues, only three amino acids have entirely hydrophobic side chains, namely A239, L290, and L311 highlighting the overall polar character of the interface. The 15 conserved residues were selected for further energetic analysis in an in silico alanine scan, in which all residueswith exception of A239were mutated to alanine in a post-processing manner; A239 was mutated to glycine. In the IE1 as dimer, mutations were performed for chains A and B, since they are conformationally slightly different from each other and can therefore yield different results. In the IE1 sy dimer, the mutations were only generated in one chain because of the similarity between both chains. Although the overall energy values differ considerably for the individual chains investigated, there is a common set of five residues that exhibit the strongest effect on binding affinity: K238, Q252, K300, L311, and R325 ( Figure 4).
Visualizing the 15 residues considered for the in silico alanine scan shows that they do not cluster at one surface patch but are rather distributed along the vertical axis of the dimer ( Figure 5). Interestingly, this applies also to the five residues with the highest effect upon substitution. Therefore, the dimerization of IE1 seems to depend on several residues that are not located in one cluster.
Large interfaces like observed for IE1 are generally predominantly hydrophobic and only contain few polar interactions (Keskin, Tsai, Wolfson, & Nussinov, 2004;Ma et al., 2003;Tsai et al., 1997). However, in the IE1 dimers, the hydrophobic portion of 35.9 and 33.7% for the IE1 as and the IE1 sy dimer, respectively, is rather low for a protein-protein interaction with such a large interface area (Tsai et al., 1997). In addition, polar interfaces are rather characteristic for weak homodimers Figure 4. In silico alanine scan performed for selected residues in the IE1 dimer systems. Shown is the average ΔΔG for the 15 mutations in chain A (black) and chain B (red) of the IE1 as dimer. For the symmetric IE1 sy dimer, only one subunit was analyzed (green). Figure 5. Localization of residues mutated in the in silico alanine scan. Chain A and chain B of the IE1 as dimer are colored white and gray, respectively. Residues, for which mutation to A/G was performed, are shown as space filled in chain A; the five residues for which mutation to A/G resulted in the strongest effect on binding affinity are highlighted with orange carbon atoms. (Nooren & Thornton, 2003;Zhanhua, Gan, lei, Sakharkar, & Kangueane, 2005), which is in line with recent in vitro experiments showing that IE1 has only a weak tendency for self-interaction (Scherer et al., 2014). Taken together, this suggests that IE1 may exist in a monomerdimer equilibrium at lower concentrations in solution and that dimer formation might be favored at high concentrations in a crystal environment.

Dynamics of monomeric IE1 and comparison to the dimeric forms
The observations above suggest that monomeric IE1 may represent the physiologically relevant IE1 species; therefore, its structural characterization should help to understand the mechanism of IE1 action. In addition, an investigation of the dynamics of monomeric IE1 should be helpful to explain how it can give rise to the formation of two distinct dimer structures.
To gain more insight into the conformational stability of monomeric IE1, we investigated its molecular dynamics. Three simulations were performed to account for the differences of IE1 conformation in the crystal structures: Since IE1 as is composed of asymmetric protomers, both chains were considered, termed IE1 as A and IE1 as B. Of the symmetric IE1 sy dimer only, one protomer was simulated, termed IE1 sy C. Each of these monomer conformations served as initial structure for one MD simulation thus allowing for a variation of the starting conditions. An analysis of the RMSD values reveals that monomeric IE1 generally exhibits a higher conformational Figure 6. Conformational stability of three IE1 monomers. (A) RMSD for IE1 as A (black), IE1 as B (red), and IE1 sy C (green) is calculated vs. their respective crystal structure. (B) Backbone RMSF of the IE1 monomers during simulation. Coloring as in (A). RMSF of each monomer is color-coded and mapped on the respective crystal structure of (C) IE1 as A, (D) IE1 as B, and (E) IE1 sy C. flexibility compared to the dimers (Figure 6(A)). This effect is most pronounced in the second half of the IE1 as A simulation. Remarkably, an RMSF analysis (Figure 6(B)) reveals that regions with high RMSF values are frequently not adjacent to each other in sequence but separated by amino acids stretches with lower fluctuations. Apart from the very flexible loop N66-K81, mapping of the RMSF values on each structure reveals that in all three monomer simulations high fluctuations can be ascribed to the tips of the upper head region and the border of the stalk region (Figure 6(C)-(E)). This dynamics can also be observed in a superposition of snapshots taken every 20 ns of each monomer (Figure 7): Especially, IE1 as A bends at the upper and lower border of the stalk region explaining the high RMSD and RMSF values (Figure 7 (A)). Similar motions can also be detected for IE1 as B and IE1 sy C; however, they are not as pronounced as in IE1 as A. In summary, a bending of the elongated shape at distinct hinge-like regions is observed in all three simulations leading to variations in their RMSD and RMSF values. Despite this considerable dynamics of monomeric IE1, those residues found to be important for dimer formation (K238, Q252, K300, L311, and R325) remain solvent accessible over the entire simulation time and form only weak and transient intramolecular interactions.
Next, we investigated whether the bending motions observed above occur in a correlated fashion. A correlation analysis of C α atoms reveals a rather regular pattern for all monomers with correlated and anticorrelated regions alternating with each other (Figure 8(A)-(C)). Notably, even though RMSD and RMSF analysis showed some differences between IE1 as A, IE1 as B, and IE1 sy C, all three simulations reveal a very similar correlation pattern indicating similar motions. Grouping correlating amino acids together results in two corresponding regions that can be mapped in the structure (Figure 8(D)). One region is composed of the entire stalk, whereas the other region contains parts of the upper and lower head region. This analysis shows that amino acids within these three sequentially discontinuous dynamic domains exhibit correlated motions within the domain and are simultaneously anticorrelated to the adjacent domain.
For a more detailed understanding of the motions, backbone dihedral angles ϕ and ψ of individual amino acids located in the hinge region between the upper head and middle stalk region were analyzed (Figure 9(A)). While D116 exhibits only some slight variation of its dihedral angles over time, the scattered plot of K238 reveals high backbone variations of this amino acid in the adjacent helix especially in IE1 as B. This is congruent with the observation that amino acids A230-I240 do not show a helical conformation in the simulation of the IE1 dimers ( Figure 2). The second adjacent helix (residues D274-N286), which also exhibits a certain degree of Figure 7. Sampling of the IE1 monomers. Overlay of snapshots taken every 20 ns from the simulation of (A) IE1 as A, (B) IE1 as B, (C) and IE1 sy C. Snapshots are color-coded according to time from red to blue. Hinge-like regions in IE1 as A are highlighted with arrows.
flexibility in the dimer, shows some differences between the monomers: While in IE1 as A backbone flexibility is detected for residue D274-E279, variations in the dihedral angles in the IE1 as B and IE1 sy C simulation occur for residues F282-N286. This indicates that not one short specific amino acid stretch is responsible for the observed flexibility in IE1 but rather the whole region between D274 and N286 (Figure 9(B)).
These differences in the exact site of the hinge are also reflected by the fact that the three monomer conformations simulated do not interconvert over the simulation time ( Figure S1). Nevertheless, it appears particularly remarkable in the light of this finding that all three simulations reveal highly similar patterns of correlated motion (Figure 8), indicating that the overall dynamics is not critically affected by the exact site of the hinge. The lack of interconversion on our simulation time also indicates that the distinct monomer conformations are rather long-lived and stable. Such conformationally distinct monomers might also dimerize, thereby explaining the structural differences detected between the dimeric IE1 forms in the crystal.
The pronounced motions detected for monomeric IE1 render it interesting to compare its dynamics to that of the IE1 dimers to assess the degree of flexibility lost upon dimerization. The high flexibility of the N-terminal segment (residues 60-90) is rather unchanged upon dimerization. However, the dimers exhibit a clearly reduced flexibility of residues 240-280 that constitute a hinge region in the monomer (Figures 3(B), 6 (B)). This difference becomes also evident from an overlay of the monomeric and dimeric structures over simulation time Figure 8. Correlation analysis of the IE1 monomers. Correlation (red) and anti-correlation (blue) of C α atoms is color-coded and shown for (A) IE1 as A, (B) IE1 as B, and (C) IE1 sy C. (D) Crystal structure of IE1 as B is shown in gray with the main correlating regions colored red (residues 80-105, 160-195, 245-272, and 328-362) and blue (residues 42-64, 115-151, 202-239, 281-321, and 370-377). (Figure 7, S2) and is also reflected by the overall RMSD values (Figures 3(A), 6(A)). Interestingly, despite this considerable difference in the magnitude of the hinge motions, the pattern of correlated motions detected in the monomer simulations ( Figure 8) is still preserved in the dimers ( Figure S3). Thus, although the magnitude of the hinge motion is significantly reduced upon dimerization, the same three correlated subdomains are still present.

Biological implication of IE1 dynamics
The recent study by Scherer et al. already noted a certain degree of structural similarity between monomeric IE1 and the coiled-coil domain (CCD) of TRIM factors, which forms a dimeric helical structure. It was speculated that IE1 might either pair up with the TRIM-CCD by means of extensive helix-helix interactions or alternatively might substitute one of the monomers, thereby disrupting the dimeric TRIM-CCD structure (Scherer et al., 2014).
In order to distinguish between these two alternative modes of interaction, we performed a more extensive structure comparison that relies on the different IE1 conformations sampled during the MD simulation of monomeric IE1 sy C ( Figure 10). From an overlay of different IE1 conformations with the crystal structure of the dimeric TRIM25 CCD (PDB: 4LTB), we investigated whether IE1 exhibits a distinct structural similarity to one of the subunits or rather an overall structural similarity to the dimeric CCD.
We noted that the structural similarity increases over simulation time and results in two different types of structural overlay exemplified by the snapshots after 30 ns and 50 ns (Figure 10(D), (E)). After 30 ns, there are 85 structurally equivalent residues, which exclusively belong to one of the TRIM25 subunits (Figure 10(D)). This observation would support a mechanism in which IE1 directly interferes with TRIM dimer formation. However, this mode of interaction would require that IE1 disrupts the tight TRIM dimer interface, which is formed by two long interdigitating coiled-coil regions, rendering this type of interaction rather unlikely.
A clue for an alternative mode of IE1 action comes from the overlay detected after 50 ns of simulation time. The respective superimposition has 132 structurally equivalent residues, which origin from both chains of the TRIM dimer (Figure 10(E)). This overall structural similarity appears particularly interesting in the light of the fact that the dimeric CCDs are involved in the Figure 9. Analysis of residues in hinge-like region. (A) Ramachandran plots of four residues in hinge-like region. ϕ and ψ angles are extracted for each residue every 200 ps from IE1 as A (black), IE1 as B (red), and IE1 sy C (green) monomer simulation. (B) Crystal structure of IE1 as B is colored from N-to C-terminus in blue to red. Residues considered for Ramachandran plot in (A) are shown as space filled. Carbon atoms of D116, K238, Q277, and H283 are colored orange, green, magenta, and gray, respectively. formation of higher-order multimers for several TRIM proteins (Li et al., 2014). Based on this observation, it is tempting to speculate that IE1 mimics the dimeric CCD due to its shape similarity, thereby interfering with the formation of higher-order multimers.
Interestingly, IE1 was recently shown to interact not only with PML, but also with other proteins of the TRIM family, namely TRIM5α (Scherer et al., 2014) and TRIM33 (Martínez & Tang, 2014). Thus, the observed structure and dynamics of the helical IE1 topology likely reflect a strategy for the interaction with the CCD of multiple TRIM proteins, thereby enabling a better viral replication.

Supplementary material
The supplementary material for this article is available online at http://dx.doi.10.1080/07391102.2015.1065204.