A unique binding mode of the eukaryotic translation initiation factor 4E for guiding the design of novel peptide inhibitors

Cyfip1 is a noncanonical 4E binding protein

The full-length structures of eIF4E with its binding partners (4E-BPs and eIF4G) are not available. However, structures of complexes of eIF4E with 4E-BP- and eIF4G-derived peptides have been very valuable to the field.11, 34-36 In all of these examples the canonical binding sequence YXXXL is found to be conserved among different isoforms and species [Fig. 2(A)]. Also generally conserved is the short α-helical scaffold on which the binding motif is assembled. Even though it has a very different length, structure33 and a degenerated eIF4E binding sequence, CYFIP1 has been clearly demonstrated to belong to the 4E-BP family [Fig. 2(A)].23, 26, 29 In an effort to elucidate the mode of interaction between CYFIP1p and eIF4E and compare it to those of 4E-BP- and eIF4G-derived peptides, we conducted MD simulations of these complexes.

Based on the alignment reported in Figure 2(A) and the available set of structures of eIF4E complexes, we have selected five complexes amenable to modeling as listed in Table 1. These comprise two 4E-BP-derived peptides and three eIF4G-derived peptides.11, 34-36 Superposition of the starting conformation of the peptides shows a conserved binding mode among all isoforms, with respect to both, the positioning of the α-helix into the binding site and the orientation of the side chains of the conserved residues [Fig. 2(B)]. The complexes display substantial stability when subjected to 50 ns of MD with explicit solvation, as illustrated by the main chain RMSD trajectories, which are all within 0.3 nm of the starting structure [Fig. 2(C)]. The peptides that showed the greatest degree of conformational variation were 1RF8 and 4AZA [Fig. 2(C), yellow and violet lines]. Their distinct behavior can be rationalized by the fact that 1RF8 is an NMR derived structure34 possibly reflective of a higher degree of flexibility and 4AZA is an example of a synthetic peptide21 likely to display unique dynamics relative to the endogenous eIF4G-derived peptides. In any case, the simulation results obtained in this work for 4AZA agree with the original structural effort and modeling characterization.21 The stability of the peptides has also been analyzed in terms of persistence of secondary structure elements. In all of the cases investigated, the conformation of the helix is maintained during the MD trajectory, as indicated by the number of residues in the helix, which ranged between 4 and 8 [Fig. 2(D)].

Based on the RMSD and helix content analysis we have selected the complex with 4E-BP1 (Table 1)35 as the most stable to employ as a starting template model for the CYFIP1p-eIF4E complex [Fig. 3(A)]. The 100 ns simulation of this complex resulted in a completely different behavior than the other peptides. This is clearly illustrated by the corresponding main chain RMSD trajectory profile [Fig. 3(B)], which is monotonically increasing without reaching a clear plateau. CYFIP1p underwent a consistent drift away from the initial conformation aligned to the 1WKW structure and, during this process, it also underwent a nearly complete loss of secondary structure, which was only partially reacquired in the last 10 ns of simulation [Fig. 3(C)].

The cause for both the conformational drift and unfolding are the repulsive electrostatic forces between the side chains of CYFIP1p and eIF4E residues in the initial conformation based on the structure of 4E-BP1 bound to eIF4E. Specifically, the proximity of the guanidinium group of Arg186 to the amino group of lysine at position 3 of CYFIP1p, causes an electrostatic repulsion between the two residues, as indicated by the continuous increase in the atomic distance of the centers of mass of the guanidinum and ammonium groups of the opposing residues (Supporting Information Fig. S1, gray line). At the same time a strong interaction is observed between residues Leu1 and Leu5 of CYFIP1p, establishing a hydrophobic interaction that stabilizes the new conformation of the peptide (Supporting Information Fig S1, black line). These two events together cause the unfolding of the short α-helix, the crucial secondary structure element involved in the interactions with eIF4E.

Binding of CYFIP1p to eIF4E is incompatible with the canonical binding mode of 4E-BP and eIF4G peptides

Cluster analysis of the simulation trajectories based on structural similarity was conducted to establish whether a conserved and preferred binding mode is present. The results show that for the 4E-BP and eIF4G peptides a major family of structures, containing more than 80% of total conformations sampled, can be detected (Supporting Information Fig. S2A). Furthermore, superposition of the centroids of their major clusters shows a conserved binding mode for the complexes, with the only exception of the synthetic peptide that is positioned in a slightly different conformation. Interestingly, despite undergoing unfolding, CYFIP1p is also characterized by a unique binding mode, as evidenced by the existence of a main cluster with more than 80% population (Supporting Information Fig. S2B).

In the case of 4E-BP and eIF4G peptides, the presence of the canonical sequence gives rise to a conserved profile of interaction with eIF4E (Table 2), where not only residues Y1, L6, and 7 but also the variant residues (X2-X5, X8-X11) always interact with the same partners [Fig. 4(A) and Table 2]. Among these interactions, many are already found in the X-ray or NMR starting configurations, reported in bold in Table 2, while others appear along the simulation time due to the rearrangement of side chains (Table 2).

Table 2. Residue-Residue Contacts Established Between the Peptides and eIF4E in the Computational Models

Peptide's residue	Position	eIF4E residue	Peptide's residue	Position	eIF4E residue
4E-BPs and eIF4G peptides			CYFIP1 peptide
					MD model	BEDAM model
Y	Y1	His37, Pro38, Leu39, Gly139	L	L1		Ile138, Gly139
D/S/G	X2		D	X2
R/P	X3	Glu132, Leu135, Arg186	K	K3	Asp143, Asp144	Glu132, Glu140
K/E/T	X4		R	R4
F	X5	His37	L	L5		Val69, Trp73, Ile138
L	L6	Val69, Trp73, Leu135, Ile138, Gly139	R	R6	Glu140, Ser141, Asp143	Glu132, Leu135
L/M	7	Trp73, Leu135	S	S7
D/E/G/Q	X8		E	E8
F/C/R	X9	Val69, Glu70, Trp73	C	C9		Trp73, Leu135
K/R/Q	X10	Trp73, Tyr76, Ile78, Glu128	K	K10
N/D/F	X11		N	N11

• ^aInteractions conserved among MD and X-ray or NMR structures are highlighted in bold.
• Contacts occurring in more than 50% of the trajectory snapshots are listed. A contact is defined as an interatomic distance smaller than 4 Å.

The lack of conservation in the canonical sequence and the nearly complete loss of secondary structure, cause the disappearance of most of the interactions of CYFIP1p with eIF4E as observed in the simulations of the other peptides, with only 2 residues out of 11, K3, and R6, maintaining contact with eIF4E via long range electrostatic interactions [Fig. 4(B) and Table 2]. This data suggest that the starting configuration of CYFIP1p-eIF4E complex based on the 4E-BP1-eIF4E structure is not stable and it is unlikely to represent a suitable conformation for binding, making us hypothesize a different binding mode for CYFIP1p when compared with other 4E-BPs and eIF4Gs. To confirm this hypothesis we have performed binding energy distribution analysis method (BEDAM) simulations (see below and Methods) to model the formation of the CYFIP1p-eIF4E complex.

CYFIP1p interacts with eIF4E through a unique binding mode

We have employed the BEDAM to search for a likely binding mode of CYFIP1p to eIF4E. BEDAM is a well-established protocol to estimate protein-ligand binding free energies.38 One key element of the method is a hybrid potential energy function, which depends on an alchemical coupling parameter λ (see Methods). At λ = 1 the energy function represents the physical associated state of the complex whereas λ = 0 represents the uncoupled state whereby the ligand and receptor do not interact. The intermediate states at represent “alchemical” unphysical states in which receptor and ligand partially interact. This series of alchemical states composes a thermodynamic path, which connects, in a thermodynamic sense, the two end states allowing the calculation of the free energy of binding. The aim of the present calculations is to use this infrastructure, normally used to estimate binding free energies, as a conformational sampling device to locate stable bound states.39 Quantitative binding free energy estimates of eIF4E complexes, while desirable, are not yet feasible due to the long simulation times required to reach convergence.

To maximize conformational sampling throughput, we employed not one but many MD threads in parallel41 at multiple thermodynamic states (144 threads here, see Methods). In addition to exploring conformational space, each thread is also allowed to travel in alchemical space according the so-called Hamiltonian Replica Exchange algorithm.37, 42, 43 This is an important aspect of the computational protocol and key to accelerate sampling of conformations of the complex. The transition between two stable conformations of the complex hindered by receptor-ligand contacts may be quickly achieved by a BEDAM thread that temporarily acquires a small values of λ where the ligand can move freely relative to the receptor. In this way, conformational transitions, which would normally occur on a timescale of microseconds or longer, can be made to occur on a nanosecond timescale amenable to computer simulation.44 In this application, thermodynamic states also include high temperature states.45, 46 The idea is that, while exploration in the alchemical dimension enhances sampling of intermolecular degrees of freedom (the position of the ligand relative to the receptor), access to high temperatures enhances sampling of intramolecular degrees of freedom (such as the unfolding and folding of CYFIP1p).

While the replica exchange strategy provides a level of conformational sampling unmatched by standard methods, it imposes some limitations in terms of system size.47 Accordingly, as described in detail in the Methods section, for these calculations we employ a minimal model of the eIF4E receptor (see Methods and Supporting Information Fig. S4). In addition, we imposed restraints to limit the region of conformational search. We believe that the results obtained are not significantly affected by these choices. To ensure unbiased predictions, we started the BEDAM-based conformational search by positioning the CYFIP1 peptide away from the canonical binding site of 4E-BPs. We used as a starting conformation a pose predicted by protein-protein docking experiments for full-length CYFIP132 (Supporting Information Fig. S3A) further discussed below.

This starting structure the CYFIP1 peptide, while residing on the same face of eIF4E, is significantly removed from the canonical binding site (Supporting Information Fig. S3A) and displays a set of residue-residue contact distinct from the final model discussed below. From this starting structure, the BEDAM simulations identified a unique and novel binding mode of CYFIP1p. This is within the general area of the canonical eIF4E binding site shared with 4E-BPs and eIF4Gs (Fig. 5) that caused unfolding of CYFIP1p in the brute-force MD computational experiment above. Clustering of the structures sampled by BEDAM at λ = 1 (bound state) and room temperature (T = 300 K) gave rise to 21 clusters (Supporting Information Fig. S3B). The two most populated clusters (55 and 10% of the total structures and nearly all of the structures generated in the second half of the trajectory, see Supporting Information Fig. S3B) were found to correspond to CYFIP1p bound in the canonical site of eIF4E.

MD threads in which CYFIP1p migrated to the canonical binding site remained bound there for the remainder of the simulation where they established strong interactions with the receptor [Fig. 6(A,B)]. We think that the small fluctuations displayed in the latter part of the trajectory reflects simulation converegence and a high level of stability of this binding mode. In contrast, the BEDAM trajectories obtained using an identical procedure for three CYFIP1 mutants previously assayed23 to probe the binding hot spots of the CYFIP1-eIF4E interaction, were not observed to bind in the canonical binding site of eIF4E (results not shown). To further confirm the reliability of the structural prediction for wild-type CYFIPp, we subjected the conformation of the eIF4E-CYFIPp complex closest to the main cluster center to 50 ns to MD with explicit solvation using the same protocol used for the simulations of eIF4E-4E-BP complexes started from crystal structures above. As illustrated by the RMSD profile obtained for this system [Fig. 6(C)], the conformation of the eIF4E-CYFIPp complex remains close to the predicted binding mode for the duration of the MD run (see also Supporting Information Fig. S5). This behavior is in contrast with that observed for CYFIPp placed in the canonical binding site of eIF4E by homology with 4E-BP complexes, in which the CYFIPp peptide suffered rapid unfolding [Fig. 4(B)]. It must be therefore concluded that, according to this measure (stability within 50 ns of MD), the binding mode of the eIF4E-CYFIPp complex proposed here is as thermodynamically stable as those from the crystal structure models of 4E-BPs.

Overall, these findings indicate that the BEDAM conformational search produced structures corresponding to the actual, stable conformational states of the complex of wild-type CYFIP1p with eIF4E. In these conformations, CYFIP1p maintains the α-helical fold conformation characteristic of the full-length protein. Furthermore, it is in a similar position and interacts with the same binding motifs of eIF4E as the 4E-BPs and eIF4G peptides. The binding mode of CYFIP1p is, never the less, distinct from those of 4E-BPs and eIF4G peptides.

Superposition of the centroid of the main cluster with those obtained for the simulations of 4E-BPs and 4Gs previously discussed, show that, relative to the consensus motif (Fig. 2), the CYFIP1 peptide is rotated on its axis counterclockwise by ∼45 [Fig. 5(A)]. In addition it is tilted longitudinally to the protein surface by ∼20°, bringing the N-terminal end of the peptide downward along the axis of the central helix of eIF4E from the point of view of Figure 5(A,B). This novel binding mode is a direct consequence of the distinct physical characteristics of CYFIP1p, which has a degenerated canonical sequence motif. Analysis of the interactions established with eIF4E shows that CYFIP1p has a tripartite structure of interactions with the canonical binding site residues of eIF4E [Figs. 5(C) and 6(D), Table 2, and Supporting Information PDB file CYFIP1p-eIF4E.pdb]. Specifically, L5 establishes a hydrophobic cluster with Val69, Trp73, and Ile138; K3 and R6 establish an electrostatic network with Glu132 stabilized by the hydrophobic shielding of Leu135; Asn11 establishes a hydrophilic cluster with His33 and His37 [Fig. 5(C)]. These interactions are made possible thanks to the new orientation acquired by CYFIP1p and are not observed in the other peptides [Fig. 4(A) and Table 2].

Explicit solvent MD simulation procedure

The initial coordinates for the 4E-BPs-eIF4E and 4Gs-eIF4E complexes have been taken from X-ray or NMR derived structures (PDB ID: 1WKW, 1EJH, 3AM7, 4AZA, and 1RF8)11, 21, 34-36 (Table 1). The initial conformation of CYFIP1p has been extracted from the crystal structure of the WRC (PDB ID 3P8C),33 residues 721–734 (numbering of human protein, uniprot Q7L576) and superimposed to the structure of the Homo sapiens 4E-BP1 with code 1WKW using the Maestro program (Schrödinger, NY). Structures have been prepared using the protein preparation facility of Maestro. The complexes have been placed in a hydration box of ∼90 Å with about 20,000 TIP3P water molecules. All systems were neutralized adding the proper number of Na⁺ or Cl⁻ counterions as needed. MD simulations have been conducted with the Desmond program (version 3.6, D.E. Shaw Research, NY)50 using the OPLS AA force field51 (version 2005)52 and Ewald summation for long-range electrostatic interactions on graphical processing units.

Trajectories have been collected and converted to Gromacs format to perform further analysis. Trajectory analysis has been performed with Gromacs 4.653 and in-house written code. Clustering has been performed on the main chain atoms of the peptides with the Gromos method using a cut-off of 0.17 nm and taking the structure of eIF4E as a reference.

BEDAM protocol

The BEDAM38, 44 is a computational protocol routinely used for the estimation of protein-ligand binding free energies. The method, its statistical mechanics basis and its applications are thoroughly discussed in published works.38, 44, 54-61 Briefly, the BEDAM method defines a dimensionless potential energy function

(1)

where λ is an alchemical progress parameter ranging from 0, corresponding to the uncoupled state of the complex, to 1, corresponding to the coupled state of the complex and β is the inverse temperature. is the potential energy of the complex when receptor and ligand are uncoupled, that is as if they were separated at infinite distance from each other. The quantity b(x), the binding energy, is defined as the change in effective potential energy of the complex for bringing the receptor and ligand rigidly from infinite separation to the given conformation, x, of the complex. The system is propagated by multi-dimensional replica exchange MD37 with λ and β as exchange parameters. The purpose of the sampling along λ is to enhance mixing of conformations along the alchemical pathway while high temperatures enhance sampling of internal molecular degrees of freedom at each alchemical stage. BEDAM employs an implicit solvent representation of the system (here the OPLS/AGBNP2 model).51, 62

In this application for each peptide we employed 144 replicas resulting from all possible combinations of 18 λ values distributed between 0 and 1 and eight temperatures distributed between 300 and 417 K. Replica exchange was conducted using the Asynchronous Replica Exchange software module (https://github.com/ComputationalBiophysicsCollaborative/AsyncRE) running on the Brooklyn College WEB computational grid. Periodically replicas were allowed to exchange λ and β parameters, allowing each replica to move in conformational space as well as alchemical and temperature space. The average simulation period between parameter exchanges was set to 25 ps. Conformations were saved on disk according to the same period. On average, each replica underwent ∼450 exchange periods resulting in 10 ns of simulation time per replica on average or ∼1.5 μs of overall simulation time per peptide. The conformations and binding energies of the replicas at λ = 1 and T = 300 K at each time point were extracted and analyzed.

To enhance computational throughput the BEDAM parallel conformational search employed a reduced model of the eIF4E receptor (Supporting Information Fig. S4). Specifically, we modeled only the canonical binding site region and regions around it composed of residues 68 − 77, 122 − 142, and 185 − 189 of eIF4E. The C-α atoms of this construct were restrained by a spherically symmetric harmonic potential with force constant 0.3 kcal/mol Ų. The dangling ends of simulated chains were capped with standard N-methylacetamide and acetyl capping groups. The CYFIP1-derived peptides were not restrained with the exception of a flat-bottom harmonic potential tether centered at 8 Å distance with 8 Å tolerance between the C-α atom of Glu132 of eIF4E and the C-α atom of Lys725 of CYFIP1 [position 3 in Fig. 2(A)]. The imposition of a restraining tether between the ligand and receptor is required by the alchemical statistical mechanics theory on which BEDAM is based.54, 55, 63 The size of the tolerance parameter has been adjusted to minimize its effect.