Probing the flexibility of large conformational changes in protein structures through local perturbations

Abstract

Protein conformational changes and dynamic behavior are fundamental for such processes as catalysis, regulation, and substrate recognition. Although protein dynamics have been successfully explored in computer simulation, there is an intermediate-scale of motions that has proven difficult to simulate - the motion of individual segments or domains that move independently of the body the protein. Here, we introduce a molecular-dynamics perturbation method, the Rotamerically Induced Perturbation (RIP), which can generate large, coherent motions of structural elements in picoseconds by applying large torsional perturbations to individual sidechains. Despite the large-scale motions, secondary structure elements remain intact without the need for applying backbone positional restraints. Owing to its computational efficiency, RIP can be applied to every residue in a protein, producing a global map of deformability. This map is remarkably sparse, with the dominant sites of deformation generally found on the protein surface. The global map can be used to identify loops and helices that are less tightly bound to the protein and thus are likely sites of dynamic modulation that may have important functional consequences. Additionally, they identify individual residues that have the potential to drive large-scale coherent conformational change. Applying RIP to two well-studied proteins, Dihdydrofolate Reductase and Triosephosphate Isomerase, which possess functionally-relevant mobile loops that fluctuate on the microsecond/millisecond timescale, the RIP deformation map identifies and recapitulates the flexibility of these elements. In contrast, the RIP deformation map of alpha-lytic protease, a kinetically stable protein, results in a map with no significant deformations. In the N-terminal domain of HSP90, the RIP deformation map clearly identifies the ligand-binding lid as a highly flexible region capable of large conformational changes. In the Estrogen Receptor ligand-binding domain, the RIP deformation map is quite sparse except for one large conformational change involving Helix-12, which is the structural element that allosterically links ligand binding to receptor activation. RIP analysis has the potential to discover sites of functional conformational changes and the linchpin residues critical in determining these conformational states.

Figure 1. Structures with intermediate-scale motions.

(A) Triosephosphate Isomerase (TIM) has a mobile loop that covers the active site (orange). The mobile loop has been crystallized in both an open [8tim] (green) and closed [1TPH] (blue) conformations. (B) Dihydrofolate Reductase (DHFR) pos-sesses the Met20-loop that has been crystallized in three different states - open [1RA2] (green), closed [1RX2] (blue) and occluded [1RX7] (purple). There is experimental evidence that the Met20-loop interacts with the adenosine-binding loop, the F–G loop and the G–H loop. (C) α-Lytic protease (αLP) [1SSX] is the control as it is a kinetically-stable protease (catalytic triad in orange) that does not possess any mobile loops. (D) The Estrogen Receptor (ER) has a highly mobile Helix-12 that covers the ligand (red) binding site. ER has been crystallized in a closed [1QKU] (blue) and open [1QKT] (green) conformation. (E) The N-terminal domain of the chaperone HSP90 (HSP90) has a 23 amino acid lid [2IOR] (green) that undergoes a large conformational change to bind ADP [2IOQ] (blue).

Figure 2. Comparison of the RIP method to standard molecular dynamics using methyl-capped amino acids.

(A) The correlation of the average rotational velocities of the χ angles of the 17 amino acids that possess χ angles. Units are in [rad ps⁻¹]. In the graph, the Y-axis standard velocities (extracted from standard molecular-dynamics at 300 K) are plotted against the X-axis RIP velocities (in the RIP protocol the kinetic energy at 300 K are effectively transferred to the χ-angle degrees of freedom). The correlation coefficient is 0.84. Detailed comparison for ILE (which has two χ-angles) of the standard molecular-dynamics simulation to the RIP simulation. (B) The distributions of the average kinetic energy per atom are fairly similar. The differences can be seen in the (C) distribution of the χ1 rotational velocities and (D) distributions of the χ2 rotational velocities.

Figure 3. RIP-induced conformation changes of TIM.

(A) the Cα RMSD response to the perturbation of RIP on Glu128 (red arrow) shown after 2 ps (blue) and at the end of the 10 ps simulation (green). (B) The 10th ps conformation (red) of the TIM structure due to RIP on Glu128 (red spheres), overlaid over the closed state (blue) and open state (green) of the crystal structures. (C) The frequency distribution of Cα RMSD for all residues from the entire set of perturbations of RIP over every residue in TIM.

Figure 4. The global Cα RMSD response of TIM to RIP.

(A) Each column corresponds to the perturbation of the residue marked by the X-value. Intensity represents Cα RMSD deviation for the 10th ps of simulation above background (>6.0 Å). (B) Perturbation strength histogram and structural linchpins. Structural linchpins are defined if the number of residues with Cα RMSD >6Å induced by a residue is greater than 3σ above the mean. The structural linchpins are mapped onto the structure as sticks. Scale is 0 (white) to 30+ (red). (C) Conversely, local regions that are highly susceptible to perturbation can be identified via a local flexibility histogram and mapped onto the structure. Scale is 0 (white) to 15+ (red).

Figure 5. Ensemble of the last conformation of the trajectories of perturbations induced by RIP for (A) TIM, (B) DHFR, (C) αLP, (D) ER, (E) HSP90.

Red corresponds to an RMSD deviation of 15 Å or more. These ensembles are used to generate the flexibility histograms.

Figure 6. RIP perturbations of Dihydrofolate Reductase (DHFR).

(A) RIP deformation map. (B) Structural linchpins and perturbation strength histogram. (C) Local flexibility mapped on structure and in histogram. Colors are as in Figure 4.

Figure 7. RIP perturbations of α-lytic Protease (αLP).

(A) RIP deformation map. (B) Structural linchpins and perturbation strength histogram. (C) Local flexibility mapped on structure and in histogram. Colors are as in Figure 4.

Figure 8. RIP perturbations of Estrogen Receptor (ER).

(A) RIP deformation map. (B) Structural linchpins and perturbation strength histogram. (C) Local flexibility mapped on structure and in histogram. Colors are as in Figure 4.

Figure 9. Conformational change in Helix-12 of the Estrogen Receptor ligand-binding domain.

(A) Overlay of the crystal structures showing Helix-12 in the closed conformation (blue) and the open conformation (green), which indicates the pivot point between the two structures. In the RIP simulations, perturbation on Trp-83 (red spheres) induces a large conformational change in Helix-12 (red), from the starting conformation of the closed structure (blue), where the hinge of the perturbed motion corresponds to the pivot point of the crystal structures.

Figure 10. RIP perturbations of HSP90.

(A) RIP deformation map. (B) Structural linchpins and per-turbation strength histogram. (C) Local flexibility mapped on structure and in histogram. Colors are as in Figure 4.

Figure 11. Comparison of the calculated flexibilities to various experimental data.

For each graph, an experimental measure (grey) is compared to the flexibility calculated from a structure (labeled on the top left of graph) using RIP (blue), ANM (red) or CONCOORD (green). RMSDf is calculated as the mean of the Cα RMSD of the crystal structures if they are found in different states as shown in Figure 1. (A) Flexibility of TIM in the open conformation compared to the RMSDf. (B) Flexibility of TIM in the closed conformation. Flexibilities from the DHFR occluded structure, compared to (C) RMSDf, (D) S2 parameters and (E) Rex factors. (F) Flexibility of ER in the closed conformation compared to the RMSDf. (G) Flexibility of HSP90 compared to the RMSDf.