Peptide-Protein Binding Investigated by Far-IR Spectroscopy and Molecular Dynamics Simulations

Abstract

Molecular dynamics (MD) simulations and far-infrared (far-IR) spectroscopy were combined to study peptide binding by the second PDZ domain (PDZ1) of MAGI1, which has been identified as an important target for the Human Papilloma Virus. PDZ1 recognizes and binds to the C-terminal end of the E6 protein from high-risk Human Papilloma Virus. The far-IR spectra of two forms of the protein, an unbound APO form and a HOLO form (where the PDZ1 is bound to an 11-residue peptide derived from the C terminus of HPV16 E6), were obtained. MD simulations were used to determine the most representative structure of each form and these were used to compute their respective IR spectra by normal mode analysis. Far-UV circular dichroism spectroscopy was used to confirm the secondary structure content and the stability through temperature-dependent studies. Both the experimental and calculated far-IR spectra showed a red shift of the low-frequency peaks upon peptide binding. The calculations show that this is coincident with an increased number of hydrogen bonds formed as the peptide augments the protein β-sheet. We further identified the contribution of surface-bound water molecules to bands in the far-IR and, through the calculations, identified potential pathways for allosteric communication. Together, these results demonstrate the utility of combining far-IR experiments and MD studies to study peptide binding by proteins.

Introduction

Protein-protein interactions play a fundamental role in mediating a vast number of physiological processes. These versatile interactions involve recognition of relatively short spans of sequence that might be found in different regions, such as loops, disordered regions, and terminal ends. Such interactions are important mediators of signaling and regulation. Peptide-protein interactions are also important targets for drug development; understanding their interactions can aid in the development of specially designed peptides. In addition, although fundamental in nature and necessary for the proper functioning of specific organisms, such interactions can be assumed by proteins not native to the human organism giving rise to a variety of diseases, including cancer. Examples include the proteins of the human papillomaviruses. The papillomavirus E6 protein is a key player in the development of cervical tumors (1).

Peptide binding by proteins has traditionally been studied by NMR spectroscopy, x-ray crystallography and other biophysical techniques. Adding to the arsenal of biophysical techniques, we develop here a physical chemical approach that combines far infrared (far-IR) spectroscopy and molecular dynamics simulations for the study of peptide binding. Infrared (IR) spectroscopy is a powerful technique for the study of proteins, but these studies have been largely restricted to the mid-IR range where changes in secondary structure due to some physiological event can be quantified or protonation changes are monitored in combination with difference techniques (2). Its steady-state application in the very low-frequency, far-IR domain (ω < 600 cm⁻¹) has been to the study of peptides and proteins with the goal of trying to obtain structural information (3, 4, 5), whereas its time-dependent application in this frequency domain has been used to study isolated protein dynamics or protein-water interactions (6, 7, 8, 9). Concerning protein/peptide binding, the application of far-IR spectroscopy remains largely unexplored, but likely contains a plethora of information. It has been assumed that vibrational modes in this frequency range are collective in nature and play crucial roles in biological function. Moreover, it has been shown that the far-IR region of the vibrational spectrum of biological systems are also related to the vibration of the hydrogen bonds, involving both intra- and intermolecular hydrogen bonding in peptides, proteins, or lipid bilayers (10, 11). These experiments are relatively fast and use little sample in comparison to other techniques, lending itself to experiments that can probe a large series of potential inhibitors and drugs against specific diseases. The interpretation of the data, however, is not well established yet, and theoretical approaches are needed before further opening of the field of inhibitor-related studies by far-IR methods.

In this study, we use far-IR spectroscopy to probe the low-frequency domain to obtain information pertaining to protein/peptide interactions. These experiments were complemented by molecular dynamics simulations and normal mode analysis studies. We focus on PDZ domain proteins, which are one of the most abundant protein-protein interaction domains in eukaryotes playing essential roles in cellular processes such as protein trafficking, transport, and signal transduction (12, 13). Consisting of a short globular domain of ∼90 residues organized in five or six β-sheets and two or three α-helices, multiple copies can be found within PDZ-containing proteins (14). They specifically recognize short linear amino-acid motifs, mainly at the extreme C-terminal end of proteins, and in particular, the four-residue motif X-T/S-X-V/L (where X can be any residue type) (15). It has also been suggested that PDZ domains may participate in other recognition processes, such as the binding to internal motifs of the same protein, to other PDZ domains and to small lipids.

We focus here on the second PDZ domain (PDZ1) of MAGI1, a member of the MAGUK protein family. MAGUK proteins play a role in the assembly of multiprotein complexes on the inner surface of the plasma membrane at regions of cell-cell contact. MAGI1 acts as a scaffolding protein (16). However, the second PDZ domain (PDZ1) of this protein has been identified as an important target for the high-risk Human Papilloma Virus (HPV) (17). PDZ1 recognizes and binds to the C-terminal end of the E6 protein from high-risk HPV such as types 16 and 18 (18). This PDZ domain is organized into five β-sheets and two α-helices (19). The binding interactions between PDZ1 and the C terminus of HPV16 E6 have been the subject of several studies (19, 20). Results from biophysics experiments have suggested that peptide binding may elicit an allosteric response from PDZ1 (19, 21). Moreover, a disorder-to-order transition of the C-terminal end of MAGI1 PDZ1 was observed by NMR measurement (21). These regions of the PDZ domains are of high importance as they are crucial for the recognition of ligands, and short extensions of these parts often involve critical modulations in both the structure and the function of the domains (14, 22).

In this work, we investigated peptide binding to the MAGI1 PDZ1 domain by an approach that combines all-atom molecular dynamics (MD) simulations and IR spectroscopy. The PDZ1 domain used in this study is formed by 129 amino acids encompassing residues 456–580 from human MAGI1 (UniProt: Q96QZ7), containing the core of the PDZ domain as identified by sequence alignments (15) and two extensions in N- and C termini of 12 and 25 residues, respectively. These extensions have been shown to stabilize the construct (23). In the NMR structures, the first four N-terminal residues originate from the cleavage site with the fusion partner. We studied the two forms of the protein: an (unbound) APO form and a HOLO form where the PDZ1 is bound to an 11-residue peptide derived from the C terminus of HPV16 E6 (19). We investigated changes in dynamics of PDZ1 upon peptide binding, detectable in the far-IR spectrum and by the MD simulations. Far-UV circular dichroism (CD) spectroscopy was used to confirm the secondary structure contents obtained by MD simulations of the APO form. The stability of the secondary structure elements was shown through temperature-dependent CD studies.

The use of MD simulations was motivated by their atomistic resolution, which allows one to understand and interpret the experimental IR data. We used MD simulations to determine the most representative structure of each form, APO and HOLO, and we used these structures to compute the IR spectra by normal mode analysis (NMA) (24, 25). The results demonstrated that the two forms of PDZ1 share very similar profiles for the frequencies ω > 600 cm⁻¹ and that the small deviations, particularly those located in the amide region of the spectra, are related to the difference of percentage of secondary structure between the APO and HOLO forms. In contrast, we observed that the largest differences between the APO and HOLO forms are found in the far-IR region of our computed IR spectra. Both theory and experiment showed the same trends in peak shifts in the far-IR region. We related each major IR peak to local atomic fluctuations and the entire far-IR region to collective motions. We observed that changes in the IR spectra reflect changes in the correlated motions of MAGI1. The network of correlated motions suggests that ligand binding influences the structural dynamics of the terminal ends, providing a mechanism for communication to other PDZ domains. Among the differences observed in the simulations were hydrogen bonding patterns upon ligand binding, which were also suggested by the experimental NMR studies (19, 21).

Materials and Methods

Sample preparation

MAGI-1 PDZ1. Samples of the PDZ1 domain of MAGI-1 were expressed and purified as described in Fournane et al. (26) in 20 mM phosphate buffer (pH 6.8) complemented by 200 mM NaCl.

Synthetic peptide. The 11-residue peptide derived from the C terminus of HPV16 E6 was obtained by chemical peptide synthesis (Peptide Synthesis Service, IGBMC; http://www.igbmc.fr/technologies/6/team/63/). Before use, the peptide was dissolved in ultrapure water. The solution was neutralized with 0.1 M NaOH and passed through a Sephadex G-25 column (PD-10 desalting columns; GE Healthcare, Little Chalfont, UK). The desalted fractions containing the peptide were lyophilized and dissolved in the same buffer as PDZ1.

Infrared experiments

Mid-IR spectra of the MAGI1 PDZ1 were obtained on a Vertex 70 spectrometer (Bruker, Billerica, MA) equipped with a liquid-nitrogen-cooled MCT detector for the 4000–1000 cm⁻¹ and a KBr beamsplitter. 5 × 256 scans with a 2-cm⁻¹ resolution were averaged in the mid-IR. Samples were deposited on a diamond ATR unit (Harrick Scientific Products, Pleasantville, NY) and measured as films. Protein stability was determined by probing the amide I mode, which is specific for secondary structure (data not shown). The far-IR experiments were carried out on the Vertex 70 (Bruker) with an FIR dTGS detector, a mercury lamp, and an Si beamsplitter. Typically, 128 scans at a resolution of 4 cm⁻¹ were averaged in the far-IR. A scan velocity of 2.5 kHz was used.

Data is presented as a plot of absorbance versus wavenumber. No deconvolution process was applied in the far-IR, because the number of overlapping modes is completely unknown.

CD experiments

CD experiments were recorded on a model No. J-815 spectropolarimeter (Jasco, Easton, MD) equipped with an automatic six-position Peltier thermostated cell holder. The instrument was calibrated with 10-camphorsulphonic acid. Samples (65 μL) were prepared at 40 μM in 20 mM phosphate buffer (pH 6.8) complemented by 200 mM NaCl and 2 mM TCEP. Far-UV CD data were collected in the 190–270 nm range using a 0.1-mm path-length cell (Quartz-Suprasil; Hellma, Southend-on-Sea, UK) at 30.00 ± 0. 01°C. Spectra were acquired using a continuous scan rate of 100 nm/min and are presented as an average of six successive scans. Response time and bandwidth were 1.0 s and 2 nm, respectively. The spectra were systematically corrected by subtracting the solvent spectrum obtained under identical conditions. The spectra have been deposited in the Protein Circular Dichroism Data Bank (http://pcddb.cryst.bbk.ac.uk) under the CD0005960000 (MAGI-1 PDZ1) identification numbers (27).

Molecular dynamics simulations

Structure preparation. The NMR structures of the MAGI1 PDZ1 in both a HOLO form, where PDZ1 is bound to an 11-residue peptide derived from the C terminus of the human papillomavirus E6 (16E6ctL0/V), and an APO form, were obtained from the Protein Data Bank (www.rcsb.org) (28) (PDB: 2KPL and 2KPK, respectively (19)). The notation “L0/V” signifies that the C-terminal Leu of the native peptide was mutated to a Val. For the sake of clarity, the residues of the PDZ1 domain were renumbered to start at 1, whereas in the experimental structures, the numbering starts at −3. The peptide with the sequence RSSRTRRETQV is numbered backward from V₋₀ to R₋₁₀ to conform to the numbering employed in the PDZ field. Model 1 from each set of NMR structures was used as the initial configuration for the molecular dynamics simulations. The protonation states of His residues were predefined in the NMR structures and used without modification. Using the CHARMM program (29, 30) and the CHARMM all-atom force field, version 27 (31, 32), a first energy minimization that consisted of 500 steps using the steepest-descent method followed by 5000 steps of Adapted Basis Newton-Raphson minimization method was realized to eliminate strong steric contacts before system solvation. A distance-dependent dielectric with an ε of 4.0 was used in this phase of system preparation.

Molecular dynamics simulation protocol

Explicit solvent molecular dynamics simulations were done using the NAMD program (33) and the CHARMM27 (31) all-atom force field with the CMAP corrections (32) at 300 K and at ambient pressure in the NVE ensemble. After the energy minimization described above, a cubic box of equilibrated TIP3P water molecules (34, 35) was centered on the protein center of mass, with a final box length of ∼75 Å per side. Waters overlapping the protein complex were removed. Na⁺ and Cl⁻ ions were added to neutralize the system and to create a salt concentration of 0.15 M. Simulations were carried out under periodic boundary conditions and the long-range electrostatic interactions were treated with the particle mesh Ewald algorithm (36). All bonds between hydrogen and heavy atoms were constrained using the SHAKE algorithm (37), and an atom-based switching function with a cutoff of 12 Å was applied to the van der Waals nonbonded interactions. An integration time steps of 1 fs was used for all simulations.

A combined energy minimization-molecular dynamics protocol was used to prepare the solvated system for the MD simulations. The water molecules were first relaxed around the fixed protein by 1000 steps of conjugate gradient (CG) energy minimization, followed by heating to 600 K over 23 ps, 250 steps of CG, and finally heating over 25 ps to reach the temperature of 300 K. Next, the positional constraints on the protein were removed and the entire system was subject to 2000 steps of CG, heating to 300 K over 15 ps, and a last equilibration consisting of a short molecular dynamics simulation of 1 ns was performed. Finally, a production run of 100 ns was carried out. This protocol was repeated three times for each form (APO and HOLO) of the MAGI1 PDZ1 domain. In total, six MD simulations were done where different initial velocities were used to create different initial conditions. It was shown by Caves et al. (38) that multiple simulations starting from different initial conditions yield a better conformational sampling than one single, but longer simulation. The coordinates were saved every 0.5 ps and extracted every 1 ps for the data analysis. All snapshots from the trajectories were reoriented on the core of the protein (residues n = 14–104) by using the first frame as reference to remove the global translations and rotations. The three runs of each form were concatenated in one single MD trajectory of 300-ns duration to facilitate the analysis.

Analysis of the trajectories

For each simulation, the root-mean-square coordinate difference (RMSD) and the radius of gyration (RGYR) were calculated considering only the Cα backbone atoms; the backbone atomic root-mean-square fluctuations (RMSfs) were calculated and averaged by residue. The RMSfs were calculated from the molecular simulations over the time frames from 0 to 300 ns.

Harmonic analysis

To characterize the low-frequency, collective motions of PDZ1, as well as changes in these motions as a function of ligand binding, NMA was performed. The NMA was carried out for selected conformations (one for each simulated form of the protein) using the VIBRAN module of the CHARMM program (29, 30). All the modes were calculated in this analysis (3 × N atoms) corresponding to 6270 Cartesian displacement modes for the APO form and 7030 modes the wild-type HOLO form. The first six modes corresponding to the global translational and rotational modes were removed from our analysis.

To characterize the flexibility of the protein as a function of individual normal modes or a sum of modes, the atomic fluctuations were computed using the VIBRAN module of the CHARMM program (29, 30). These fluctuations correspond to the local flexibility of the protein in the harmonic approximation and can be compared to crystallographic B-factors (39, 40) and to the RMSfs obtained from analysis of the family of NMR structures.

Cross-correlation coefficients C_ij (41) between residues assess the nature of interresidue motions, that is, whether relative motions between the residues i and j are correlated or anticorrelated. Cross-correlation coefficients can be calculated from normal modes, quasiharmonic modes, as well as from molecular dynamics simulations following the equation:

$$C_{ij}=\frac{{\sum }_{i=1}^N\left(r_i-{\overline{r}}_i\right){\sum }_{j=1}^N\left(r_j-{\overline{r}}_j\right)}{\sqrt{{\sum }_{i=1}^N{\left(r_i-{\overline{r}}_i\right)}^2{\sum }_{j=1}^N{\left(r_j-{\overline{r}}_j\right)}^2}},$$ (1)

where r_i is the displacement from the mean position of the residue i. From the C_ij correlation coefficients, which are organized as a matrix, a cross-correlation map was calculated using a color-coded 2D representation. In this representation, C_ij = 1 identifies correlated motions and C_ij = −1 anticorrelated motions. These values give us additional information concerning the global collective motions of PDZ. In this work, we calculated the cross-correlation coefficients only from the NMA.

Computational IR spectra

Using structures extracted from the MD runs after being identified as representative of highly probable structures through the analysis of the free energy surface, the IR spectra were computed from the normal modes and the dipole derivatives associated at each mode (24). Each IR spectrum was represented as function of the integrated intensity, Γ_k, which has units of the molar absorptivity, calculated from the dipole derivative values (42) by using the following equation:

$${\Gamma }_k=\frac{N_0{\pi }^2}{3c^2{\epsilon }_0{\omega }_k}{\left(\frac{\partial p}{\partial Q_k}\right)}^2,$$ (2)

where N₀ is Avogadro’s number, ε₀ is the permittivity of vacuum, c is the speed of light, p is the dipole moment, Q_k is the normal coordinate, and ω_k is the respective frequency. The IR spectra were then smoothed by multiplying the intensity of each peak by a Gaussian function with a full width at half-maximum of 20 cm⁻¹.

Results and Discussion

Experimental mid- and far-IR spectra

The experimental far-IR spectra of the APO (peptide-free) and the HOLO (peptide-bound) forms were recorded by following the protocol described in Materials and Methods. The results are shown in the Fig. 1, where one observes that the two measured far-IR spectra have a similar shape. There is one very intense and broad peak that appears in the frequency range [100–200] cm⁻¹ and a second peak that is present in the frequency range [500–550] cm⁻¹. The signals at 154 and 515 cm⁻¹ observed in the HOLO form are shifted to higher frequencies (164 and 532 cm⁻¹) in the APO form spectrum. Previous works, including isothermal titration calorimetry (19), surface plasmon resonance (26), and NMR spectroscopy (19, 21) have all shown that this PDZ domain readily binds to this peptide, so this red shift in the far-IR is likely to be a signature of this interaction. Note that both spectra show small artifacts from the Si-beamsplitter at ∼670 cm⁻¹.

Figure 1.

Shown here are the experimental IR spectra of the APO form (black line) and of the HOLO form (gray line) of PDZ1 from MAGI1. (A) Far-IR; (B) mid-IR.

Molecular dynamics simulations

The molecular dynamics simulations of the APO and HOLO forms were run according to the procedure described in Materials and Methods. The Cα atomic fluctuations calculated from the simulations were compared to those computed from the NMR ensembles for the both APO (PDB: 2KPK) and HOLO (PDB: 2KPL) forms (see Fig. 2, A and B, respectively). The NMR data consists of an ensemble of 20 structures for each of the two forms and the results are shown in Fig. 2 B. Interestingly, one observes that, from the NMR structures, the fluctuations of the Cα atoms of the N- and C- terminal parts of the HOLO form are larger than those of the same region of the APO form. This observation is in contrast with our MD results, for which the local fluctuations of the Cα atoms were found to be more important in the APO form than the HOLO. We believe this difference arises from larger uncertainties concerning structures of flexible terminal ends resulting from the lack of experimental NOEs. The NMR study (19) showed, however, that there is little change in the picosecond-nanosecond motion for residues of the N-terminal end between the APO and the HOLO structures, but significant deviations were observed for the C-terminal residues 105 < n < 120. Large deviations in the picosecond-nanosecond motion suggest that the C-terminal end of the HOLO form is considerably more restricted compared to the APO form of the PDZ1 domain of MAGI, in agreement with our simulation results.

Figure 2.

Given here is a comparison of the RMSfs of the Cα atoms as a function of sequence computed from the MD simulations (A) and from the 20 NMR-PDB structures (B) the APO form (black triangles) and the wild-type HOLO form (red spheres). In both panels, the positions of the β-sheets, the α-helices, and the ligand are indicated by light gray, dark gray, and blue-gray stripes, respectively. To see this figure in color, go online.

Free-energy landscape and clustering from molecular dynamics simulations show differences in dynamic behavior between the different forms

To assess the conformational landscape of the PDZ1 protein in its different forms (APO- and HOLO-), an effective 2D free-energy landscape (FEL) was constructed from the molecular dynamics simulations. The 2D FEL was based on the values of the RMSD and the RGYR computed from the simulations and the relationship as follows:

$$\text{FEL}\left(\text{RMSD},\text{RGYR}\right)=-k_{\text{B}}T\text{ln}\left[P\left(\text{RMSD},\text{RGYR}\right)\right],$$ (3)

where k_B and T are the Boltzmann constant and the temperature, respectively; and P is the joint probability of a structure having the values of RMSD and RGYR (43). The FELs were computed after concatenating the MD trajectories of each form. The results are shown in Fig. 3 for both the APO- and HOLO-forms simulated in this study.

Figure 3.

Given here are effective free-energy 2D landscapes built from the RMSD and the RGYR values computed from the concatenated MD runs of the APO form (A) and of HOLO form (B) of MAGI1 PDZ1. The black line contours were plotted every 1 k_BT U. To see this figure in color, go online.

Comparison of the FELs reveals that the APO-form explores a larger conformational space than the HOLO-form. Multiple local minima in the APO FEL were found and the most populated minimum corresponds to structures having large RMSD and RGYR values with respect to the initial starting NMR structure. These large values were due to the variability of the N- and C-terminal ends. This is consistent with the experiments, which show that the 20 NMR structures maintain a relatively stable core and more largely variable terminal ends. The associated dynamics determined by the NMR studies support the conclusion that in the unbound protein, the PDZ1 C-terminal tail is likely in a disordered state (19, 21).

From the MD simulations, we calculated the secondary structure content of both forms averaged over the entire ensemble of structures using the COOR SEC module of the CHARMM program. We found an average of 9.7 and 9.1% α-helix content and 21.6 and 22.5% β-sheet content for the APO and HOLO forms, respectively. Experimentally, the percentages of secondary structure elements were obtained for the APO form by far-UV CD spectroscopy. Spectral deconvolution was done using the CDPro software suite (44) and the SMP56 reference database run with the CONTIN method (see Fig. 4). CONTIN fits the CD spectra by comparison to a linear combination of spectra of the database of proteins with known conformations. This led to (14 ± 2)% for α-helix, (33 ± 4)% for β-sheet, and (52 ± 4)% for other secondary structure content. By simple comparison, we found these values to be in good agreement with those calculated from the MD simulations.

Figure 4.

Shown here is the far-UV circular-dichroism spectrum of PDZ1 domain in 20 mM phosphate buffer (pH 6.8) with 200 mM NaCl and 2 mM TCEP. The CD signal is expressed in mean residue ellipticity (deg.cm².dmol⁻¹). The spectrum is corrected by subtracting the solvent spectrum obtained under identical conditions and is presented as an average of six successive scans. Experimental data and fitting results (see the main text) are shown in shaded spheres and solid line, respectively. The residuals are represented with ovals and display a random distribution.

For the HOLO form, we observed a more restrained sampling of conformational space. The HOLO-FEL has one large and deep minimum corresponding to structures close to the experimental ones (19), and several small, less populated local minima with higher RMSD values. The structure of the HOLO form has, on average, a smaller RGYR value than the APO form. This reflects the observation that the HOLO form was found in the MD simulations to be more compact than the APO form. Visual examination of the structures shows that this is due to the stabilization of the C-terminal end through its interactions with the bound peptide, an observation that is also made from the NMR structures. This result is consistent with thermal melting curves recorded by CD (Fig. S1). As seen by the shift of the melting temperature from ∼65 to ∼75°C, the presence of the peptide increases the thermal stability of the domain. In addition, the CD denaturation profiles show that the start of the unfolding transitions for the APO and HOLO forms occur at ∼60 and ∼70°C, respectively, and that below those temperatures, particularly in the temperature range of the MD simulations, the folded ratios do not display any change. This indicates that the secondary structures of the two forms are stable at least within the 15–55°C range. This excludes the possibility that the distinct behaviors observed for the APO and HOLO forms are due to differences in thermal behavior.

More specifically, the largest well of the HOLO-FEL contains structures showing a localized and stable interaction between the peptide ligand and the C-terminal tail of the PDZ domain (see Fig. 5 A). Residues S116, L117, and V118 are in close proximity to residues T₋₆ and R₋₇ of the ligand in the representative structure extracted from well 1 of the HOLO-FEL. This suggests that the main interactions are through the positively charged ligand residue R₋₇ and the side chains of protein residues L117 and V118 (see Fig. 5 A).

Figure 5.

Comparison is given of the representative structures extracted from the wells of the FEL built on the RMSD and RGYR values of the HOLO form. The structures are shown in cartoon representation. The experimental structure corresponding to model 1 of PDB: 2KPL is shown in a black cartoon structure as comparison. The C-terminal region (n > 102) is highlighted by using a large radius cartoon representation and different color, depending on the well: well 1 in red (A), well 2 in green (B), well 3 in blue (C), well 4 in magenta (D), and well 5 in cyan (E). Indicated by transparent cartoon are core residues n = 18–102. The ligand is indicated by a large ribbon representation and by using a different color than the one used for the C-terminal region of the protein: orange for well 1 (A), yellow for well 2 (B), cyan for well 3 (C), red for well 4 (D), and blue for well 5 (E). The residues involved in the interaction between the PDZ1 C-terminal end and the ligand are indicated by a stick representation. To see this figure in color, go online.

Structures extracted from wells 2–5 have higher RMSD and RGYR values and reveal a mode of interaction that is not seen in the specific experimental structure used to initiate the simulations, nor in the representative structure from well 1 (see Fig. 5), but is observed in several other structures of the experimental NMR ensemble. This interaction comprises the stabilization of the PDZ C-terminal tail by residues D112 and D113, depending on the well. In the representative structures extracted from wells 2 and 5, the C-terminal tail is stabilized by interactions between peptide residue R₋₇ and D112 of the protein (see Fig. 5, B and E). The structure from well 3 shows interactions between residues D112 and D113, which interact with the side chain of the peptide residue R₋₇ (see Fig. 5 C). For the representative structure from well 4, the C-terminal tail of the protein is stabilized by interactions between the peptide residue R₋₇ and protein residue D112 and between peptide residue R₋₅ and protein residue P114 through the backbone atoms O (P114) and HN (R₋₅) (Fig. 5 D).

These different modes of interaction found in the representative structures from the deepest wells of the HOLO FEL arise from alternative conformations of the C-terminal tail (n > 115) of the protein and explain their larger RMSD values. Referring to Fig. 5, one can see that the C-terminal end in the representative structure of well 4 (Fig. 5 D) perpendicularly overlaps the ligand, whereas in the representative structure of the well 3 (Fig. 5 C), it interacts with the core of the protein, specifically with β2 and β3; see Fig. S2 for notation of secondary structural elements. This stabilization might be important in the context of the multidomain PDZ construct and can contribute to stabilizing the relative orientation of the following domain.

Normal mode calculations

The structures populating the deepest minimum well of the FEL of each form of PDZ1 were extracted from the MD simulations, generating groups of 33,580, and 111,192 structures of the APO and the HOLO forms, respectively. These two groups of structures were then subjected to an RMSD-based clustering to find the most representative structure of each MD run. A cutoff of 1 Å within the clusters was used. The water molecules within a distance of 2.6 Å of any protein atom of each representative structure were kept in the coordinate files, resulting in a thin water layer of 45 and 65 molecules around the APO and HOLO forms, respectively. The IR spectra of the APO and HOLO forms of the MAGI1 PDZ1 domain were calculated by Eq. 2 following a normal mode analysis for each structure following the protocol presented in Materials and Methods.

Far and middle IR spectra of the APO and HOLO forms

Using the protocol described in the Materials and Methods, the IR spectra of the APO and HOLO forms of MAGI1 PDZ1 were calculated. The resulting spectra are shown in Fig. 6 up to a frequency of ω = 2000 cm⁻¹. Overall, the spectra show largely similar global profiles as a function of frequency that can be organized in three regions: a low-frequency region [0:600] cm⁻¹, an intermediate region [600:1200] cm⁻¹, and the amide region [1200:2000] cm⁻¹. The low-frequency region (0 < ω < 600 cm⁻¹) corresponds to what has been long considered to be the collective vibrational modes of the protein (45, 46, 47) and is characterized by the largest variations in peak intensity between the two MAGI1 PDZ1 states (APO and HOLO). The calculated spectra show two high and wide peaks with several local maxima depending on the structures, which are followed by a strong decrease of the IR intensity above ω > 320 cm⁻¹. The intermediate ([600:1200] cm⁻¹) and amide band ([1200:2000] cm⁻¹) regions of the spectra show less variation between the different forms. Instead, they show similar peak shapes and positions in the two spectra. The main deviations in these regions are the values of the peak intensities, which vary between the APO and HOLO forms; this can be particularly well observed in the amide band region of the spectra ([1200:2000] cm⁻¹). The similarity in the frequency profiles of the spectrum computed from the two structures of the PDZ1 domain in two different forms is consistent with a recently published NMA study that attests to the universality of the vibrational spectra of globular proteins (48). Similar observations can be made from the experimental spectra; peptide binding seems to affect the low-frequency region more than the mid- and amide regions of the spectrum.

Figure 6.

(A) Shown here is a comparison of the far-IR spectra computed from the most representative structure extracted from the concatenated MD runs of the APO (red) and the HOLO (black) forms of MAGI1 PDZ1 by using NMA and Eq. 2. Nearby water molecules were included in the calculations. (B) corresponds to a zoom-in of the computed IR spectra in the far-IR region (ω < 700 cm⁻¹). The computed spectra of both forms in the absence of water molecules are shown by the dashed red and thin solid black curves as a comparison to the APO (red) and HOLO (black) curves, respectively. To see this figure in color, go online.

Global assessment of the effect of water

The interpretation and comparison of the low-frequency domain of the IR-spectrum is complicated by the influence of water molecules (49). To demonstrate the contribution of water to the calculated IR spectra of the PDZ1 domain, calculations were carried out both in the presence and in the absence of water molecules (compare the curves in Fig. 6), as well as on a single water cell (data not shown). Our results confirm that the motions of water molecules contribute significantly to the low-frequency vibrational modes in this region of the IR spectra and the normal mode calculations carried out on a single water cell reveal a large intense peak between 0 < ω < 440 cm⁻¹ with the maximum of intensity at ω = 290 cm⁻¹. These results are in agreement with recently published calculations of IR spectroscopic data using an ab initio dipole moment surface of water (50) and experimental measurement of water IR spectrum (51). Previous calculations have shown that water molecule motions can be strongly coupled to those of the protein (52), whereas time-dependent far-IR spectroscopy studies have shown that there is a dynamical water solvation shell around the protein (53). More recent studies suggest that the influence of protein on water extends well beyond the generally accepted two-to-three water solvation shell (9).

The comparison of the computed IR spectra of hydrated and nonhydrated APO and HOLO protein indicates a strong contribution of the water molecules to the intensities of the two first peaks (Figs. 1 and 6). Our results further show that the intensity of the peak depends on the number of water molecules surrounding the protein, as the structure with the largest number of water molecules, i.e., the wild-type HOLO form, has the largest intensity (see Figs. 1 and 6). To further investigate this, we equalized the number of waters in the APO and HOLO forms by randomly removing water molecules from the HOLO form. We then reran the normal mode analysis and the results, shown in Supporting Material (Fig. S11), do not significantly change the position of the peaks, but their intensity.

Interestingly, the peak at ω = 1743 cm⁻¹ also shows a strong increase in intensity in the presence of water. We can suppose that this peak (at ω = 1743 cm⁻¹), activated in presence of the water molecules, is related to the bending motion of the TIP3P water model molecules (35). The influence on the lower frequencies could be due to librational motions of water molecules hydrogen bound to the protein.

Interpretation of the IR peaks calculated by normal modes local fluctuations

Mid-IR: amide I band

In the interest of confirming the relevance of our computed IR spectra and to compare them to the experimental results, we analyzed the [1200:2000] cm⁻¹ region. This region of the IR spectrum is related to C=O vibrational stretching and its variation, due largely to backbone couplings, reflects, in part, interactions such as hydrogen bonding that occur in specific secondary structural motifs. Variations in the C=O stretching for the amide I band are often used for secondary structure analysis (2).

The two IR peaks at ω = 1613 and 1686 cm⁻¹ are present in both spectra and their values are close to the experimental value of the amide I band (2). There seems to be little influence of water on these particular peaks. A greater intensity for the complex is observed, which could reflect the somewhat higher percentage of secondary structure, as the peptide ligand increases the β-strand content.

To investigate the vibrational motions underlying the calculated IR peaks, we examined the respective atomic fluctuations as a function of frequency. The normal modes were selected by their frequency corresponding to the peak of interest. Frequencies within an interval of ±20 cm⁻¹ were taken to be consistent with the experimental resolution. In this way, the fluctuations of the Cα atoms were calculated by-residue for the principal IR peaks of the spectra of both the APO and HOLO forms, and their values were then projected onto the 3D structures of the protein and visualized using the PyMol program (54). The results are shown in Figs. S2 and S3. One can see that the peak at ω = 1613 cm⁻¹ is related to vibration of the antiparallel β-sheets (see Fig. S2) and the peak at ω = 1686 cm⁻¹ to the vibration of the α-helices (see Fig. S3). This shift of the amide-I band to higher frequencies for α-helices with respect to β-sheets is consistent with what is generally observed experimentally (2).

Far-IR region shows more variation upon peptide binding

The far-IR region of the spectra, i.e., the frequency domain ω < 600 cm⁻¹, is more complex, with larger differences between the APO and HOLO forms. From the calculated spectrum of the APO form, the lowest frequency peak is found at 131 cm⁻¹, which is slightly lower than the lowest corresponding experimental peak at 164 cm⁻¹ (compare Figs. 1 and 6 B). Addition of the peptide, both experimentally and computationally, leads to a red shift of the lowest frequency to 98 and to 154 cm⁻¹ in the calculated and the experimental spectrum, respectively, reflecting what is essentially a softening of the low-frequency mode by the addition of the peptide to form the complex. Similar effects have been seen in other cases, such as protein binding of heme groups (55) and water binding to BPTI (56). Such effects have also been associated to hydrogen bonding (57). Experimentally, the lowest frequency peak for the HOLO form is slightly broader than that of the APO form. In the calculated spectrum of the HOLO form, a second peak appears at higher frequency, whereas in the calculated spectrum of the APO form, it appears more as a shoulder to the peak at 131 cm⁻¹.

For both APO and HOLO forms, a broad peak centered around ω = 300 cm⁻¹ is observed (see the Fig. 6). Several other local maxima are found in the frequency interval of 320 < ω < 450 cm⁻¹ and their positions are variable as a function of the peptide binding. These peaks are not observed in the experimental spectra, but as our calculations show, water appears to have particularly large effect on these peaks and the experimental spectra were taken for dry films of PDZ1 containing only residual water. Even in the absence of water, a peak remains in this region, but unpublished experimental data measured down to 20 K by one of us suggests that the intensity of this peak strongly increases with decreasing temperature. Recall that a normal mode analysis is effectively done at 0 K. Near 500 cm⁻¹, small intensity peaks at ω = 502 and 550 cm⁻¹ for the APO form and two local peaks at ω = 488 and 532 cm⁻¹ for the HOLO form are apparent in the calculated spectra. We relate these peaks to the second small peak in the experimental far-IR spectra found at ω = 515 and 532 cm⁻¹ for the HOLO and the APO forms, respectively. As for the low-frequency peaks, peptide binding leads to a red shift of these peaks around this frequency in both the calculated and experimental spectra.

Atomic fluctuations of the far-IR spectral region

Here we investigate the atomic level motions that underlie the major peaks of the far-IR spectra. Comparing the computed IR spectra of the APO and the HOLO forms, differences between the two are observed in the region [0:200] cm⁻¹ (see Figs. 1 and 6). As was done for the peaks in the mid-IR region, the modes were selected by their corresponding frequency in the IR spectra over an interval of ±20 cm⁻¹ to adhere to the experimental resolution.

Concerning the APO form, the largest Cα fluctuations of the vibrational mode at the lowest frequency (ω = 131 cm⁻¹) are observed for the glycine residues and, in particular, residues n = 37, 38, 54, 60, 65, 104, as well as for L45 (the results are shown in the Fig. S4. These residues are located, for the most part, at the connection between secondary structure elements and loops. Some of these glycine residues (n = 60, 65, and 104) are surface exposed and interact with the surrounding water molecules through hydrogen bonds, whereas residues G37, G38, and L45 localize to form a flexible cluster.

The peak at ω = 502 cm⁻¹ reveals residue fluctuations different from those of the lower frequency ω = 131 cm⁻¹ peak (see Fig. S5). At ω = 502 cm⁻¹, a more varied distribution of residues is found, many of which are implicated in H-bonds to either other amino acids or to water molecules as their side chains point out toward the solution (see Fig. S5). That the water molecules influence the spectra is confirmed by the shift of the peak at ω = 502 cm⁻¹ in the IR spectrum computed from the structure without the water (see Fig. 6 B).

HOLO forms shows shifted peaks with respect to the APO form

Upon peptide binding, we see a red shift of the lowest frequency peak from ω = 131 cm⁻¹ in the APO form to ω = 98 cm⁻¹ for the HOLO form. Interestingly, one observes that the motions of the residues implicated in this lowest frequency mode are also glycine residues, in particular, Gly37 and Gly65, as in the APO form (see Fig. S6).

At higher frequencies, the second peak of the IR spectra of the HOLO form (ω = 166 cm⁻¹) shows significant involvement of the ligand in the vibrational mode (Fig. S7). This second peak is not present in the APO form. In the experimental spectrum, this peak is not clearly visible, although differences from the APO form are apparent.

The analysis of the fluctuations of the atoms of the HOLO structure at ω = 307 cm⁻¹ reveals that only the three glycine residues, G37, G54, and G94, show high fluctuations and that the major motions reflected by these normal modes are those of the water molecules surrounding the protein complex (see Fig. S8). The same residues fluctuate in the APO structure at ω = 305 cm⁻¹. Other residues that have nonnegligible fluctuations are D109 and D112, which have their respective side chains bridged by a water molecule and residues G17, I47, and G65 (Fig. S9). These results could explain why there was a significant contribution of water to this particular mode. Its absence in the experimental spectra could reflect the nearly dry nature of the sample; although some residual water molecules are likely present in the film samples, the amount is probably less than in the theoretical structures.

For the peak at ω = 488 cm⁻¹ of the HOLO structure, the important fluctuations involve more isolated residues of the β-sheet and some residues in the vicinity of the peptide (Fig. S10). With respect to the APO form, the differences are slight and this is probably reflected in the red shift of the experimental peak in this frequency range, again reflecting the increased number of hydrogen bonds.

Correlated motions suggest a pathway for information transfer

Through the comparison of the measured far-IR spectra and those calculated following our molecular dynamics protocol, we have established that the atomic motions in our simulations represent well the low-frequency dynamics of both the APO and HOLO forms of the MAGI1 PDZ1. This now allows us to examine the collective correlated motions in this region of the spectrum and elucidate the effects of peptide binding. Many studies of low-frequency modes and correlated motions have often been focused on the very lowest frequency modes of the spectral density, which were not probed here experimentally (45, 58, 59, 60). Rather, we include the higher frequency modes in the far-IR that we characterized experimentally. Using the normal modes and Eq. 1 (see Materials and Methods), we computed the correlated motions of the PDZ1 Cα atoms for the representative APO and HOLO structures. For these calculations, we included all normal modes having frequencies in the far-IR region of the spectra (i.e., ω < 600 cm⁻¹).

The correlation maps of the Cα atoms are shown in the Fig. 7 for the APO and HOLO forms. Their comparison reveals that they are highly similar. The regions having strong (anti-) correlated motions are the closely packed secondary structural elements, such as the β-sheets and α-helices. The motions of β1 are correlated to β5 due to their direct interaction in the structured protein. The motions of β5 are, in turn, propagated to β4, which has its local fluctuations correlated to those of β3. Finally, the fluctuations of the β3 are correlated to β2. Significant differences between the Cα atom-correlation maps of the APO and HOLO forms are located in the terminal tails, which show an increased correlation upon peptide binding. This suggests that the terminal regions become more ordered upon binding as the ligand joins the underlying correlation network of the PDZ domain through its β-sheet interactions. The peptide ligand is strongly (anti-) correlated to almost all secondary structural elements of the PDZ1 domain, in part through the hydrogen bonds formed between the peptide and the protein. In Fig. 7 B, H-bonds are indicated (green triangles) between eight residues of the peptide (n = R₋₁₀, R₋₇, S₋₅, R₋₄, E₋₃, T₋₂, Q₋₁, and V₀) and residues of the core of the protein located in the β2, β3, as well as several residues of the N- and C-termini. The hydrogen bonds in the C-terminal tail of the APO form reflect a structure that is more folded upon itself as opposed to the tail of the HOLO form, which shows a more extended structure as a result of its interaction with the peptide. The same observation is made from the family of NMR structures. Peptide binding alters the motions, and therefore the couplings of the protein termini. This is evident from the enhancement in correlated motions between the N- and C-terminal ends (see Fig. 7 B), but also through the β-sheets, as shown through the changes in motions coupling the N-terminal end with β2, β3, and the ligand.

Figure 7.

Comparison of the fluctuation correlation maps for the Cα atoms computed from NMA by considering all the normal modes up to ω ≤ 600 cm⁻¹ of the APO (A) and the HOLO (B) forms. In (A) and (B), the position of the α-helices, the β-sheets, and of the ligand is indicated by sticks, single-headed arrows, and a double-headed arrow, respectively. The triangles indicate positions of all hydrogen bonds, backbone, and sidechains included. To see this figure in color, go online.

This is consistent with the NMR observation that the PDZ domain responds globally to peptide binding. Indeed, the experimental results showed differences in the R₂/R₁ relaxation ratios between the APO and the HOLO form, particularly for the terminal ends (21). NMR studies also showed other manifestations of binding, as indicated by their change in chemical shifts. Small clusters of residues, although distant from the binding site of the peptide, were affected (19); of particular interest are the changes in chemical shift for Thr64 and Gly65, which are located in loop α1–β4. From the correlation maps, we see that loop α1−β4 is strongly correlated to many of the β-sheet structural elements, and in particular, to β3, which contains Lys48/Ser49, two residues that also show significant changes in their chemical shifts upon peptide binding; β3 is also strongly correlated to the peptide ligand via β2, as well as to β5. In both the APO and HOLO forms, all of the β-sheet elements are correlated. Other long-range effects were observed for the residues of α1, which showed a significant change in hydrogen/deuterium exchange rates upon peptide binding. Interestingly, addition of the peptide ligand significantly alters the correlated motions of α1 with respect to the loop β1–β2. In the absence of ligand, the motions are strongly anticorrelated, yet upon peptide binding, the motions become strongly correlated. The region experienced significant changes in H/D exchange rates, which may reflect the significantly different correlated motions. Referring to Fig. 2, we see that α1 becomes less flexible upon peptide binding. These results suggest that there can be multiple pathways for the transmission of binding information in PDZ1. Also of note, many of the amino acids implicated in the correlated motions were singled out through the fluctuation analysis of the calculated far-IR spectra.

This suggests that the β-sheets of the MAGI1 PDZ1 provide a means for the transfer of binding information through changes in their relative correlated motions. Although previous studies have shown correlated motions of secondary structural elements in PDZ domains (61, 62), the studies here, as they were done on PDZ domains with extended termini, suggest a possible physiological consequence of the information transfer. Involved in regulating cellular signaling pathways, PDZ domains are often found in multidomain scaffold proteins. Multiple PDZ domains are often sequential, and changes in the dynamics of termini could represent one means by which PDZ domains transfer information from one domain to another.

Conclusions

In this work, we present a combined MD/IR approach to the study of ligand binding by the PDZ1 domain of MAGI1, focusing in particular on the far-IR region of the spectrum. CD spectra were determined to validate the secondary structure populations. We first demonstrated the feasibility of using far-IR experiments to measure the effect of peptide binding by a given protein, in this case, the binding of an 11-residue peptide derived from the C-terminal end of the HPV E6 protein. The experiments showed that upon peptide binding, the lowest frequency peak undergoes a red shift suggesting a softening of the vibrational mode. This is due to an increased number of hydrogen bonds in the ligand-bound complex, which was observed in the previous NMR experiments (19) and further confirmed by the molecular dynamics simulations presented here.

To further understand this shift, the far-IR spectrum was analyzed through computational analysis of the same constructs used in the experiments. We first carried out extensive molecular dynamics simulations of both the APO and HOLO forms, which allowed us to generate free energy surfaces based on the distribution of RMSD and radius of gyration. From the simulations, we found that the distribution of conformations for the HOLO protein was more compact than for the APO form, and we identified a number of alternative, but well populated, conformations. From the primary free energy wells on these surfaces, the most representative structures were extracted and used for the normal mode analysis and subsequent calculation of the IR spectra.

The computational results demonstrated that, in the far-IR domain, the IR spectra displays large variations between the APO and HOLO forms. The effects were less significant in the mid-IR region of the spectra, which is likely due to the strong similarity of the protein core despite peptide binding. We confirmed this latter point through an analysis of the atomic fluctuations, which showed that the mid-IR region of the spectrum reflects hydrogen bonding in secondary structural elements, and the changes in the spectra coincide with the augmentation of β-sheet content in the protein/peptide complex. In the far-IR region, the computational spectroscopic data followed the same trend as that observed in the experiments, that is, peptide binding resulted in a broadening and a red shift of the lowest frequency peak with respect to the APO PDZ1 domain. This softening of the lowest frequency peak corresponded to an increase in hydrogen bonding, in particular with the ligand peptide, which joins the existing β-sheet. In the calculated spectrum, a second peak appears that largely implicated peptide binding. The principal peaks were examined in more detail and we were able to show that the lowest frequency peak displays largely global, collective motions that implicate flexible Gly residues in the PDZ1 domain. Higher frequencies in the far-IR spectrum reflect more localized motions related to the β-sheets. We were able to identify modes where specific water molecules were implicated.

We further demonstrated that the far-IR region reflects the correlated motions of the β-sheets propagated through the H-bond network and we identified pathways for information transfer that could signal the binding of the peptide. We confirm through the simulations that, in accord with the NMR studies, hydrogen bonding occurs between the peptide and protein, but also, the simulations show that a significant number of water molecules also remain hydrogen bound to the protein and they can affect the spectrum. We also distinguished modes that are more affected by water than others. By studying the extended PDZ1 domain, we were able to show that ligand binding affects the global network of correlated motions and by several pathways, can lead to the coupling of the N- and C-terminal ends, which themselves may play an important role in the transfer of binding information of other domains in the context of entire multidomain scaffold proteins.

We showed in this work that two approaches are complementary; they showed similar trends and therefore demonstrated the strong efficacy and the utility of the far-IR spectroscopy to detect and study the binding of ligands on proteins and the use of molecular dynamics simulations for their interpretation. This approach is easily generalizable to other protein complexes and has relatively few limitations, in that the far-IR spectroscopy is not limited by sample size and the computations, in theory, have no limitations. However, a practical limitation might be dependent on the computer power made available to a particular study, but this is becoming less of an issue as computer power continues to increase. The sensitivity to binding affinities remains to be explored. In the future, we expect to study the effects of point mutations to better understand their effects on the structural dynamics arising from the collective modes.

Author Contributions

Y.C. designed research, performed research, analyzed data, and wrote the manuscript. Y.N. designed research, performed CD experiments and analyzed data, contributed to protein purification, and contributed to manuscript writing. J.R. performed protein purification. P.H. designed research and performed far-IR experiments and contributed to manuscript writing. R.H.S. designed research, analyzed data, and wrote the manuscript.

Editor: Monika Fuxreiter.