Allosteric Communication Networks in Proteins Revealed through Pocket Crosstalk Analysis

Abstract

The detection and characterization of binding pockets and allosteric communication in proteins is crucial for studying biological regulation and performing drug design. Nowadays, ever-longer molecular dynamics (MD) simulations are routinely used to investigate the spatiotemporal evolution of proteins. Yet, there is no computational tool that can automatically detect all the pockets and potential allosteric communication networks along these extended MD simulations. Here, we use a novel and fully automated algorithm that examines pocket formation, dynamics, and allosteric communication embedded in microsecond-long MD simulations of three pharmaceutically relevant proteins, namely, PNP, A2A, and Abl kinase. This dynamic analysis uses pocket crosstalk, defined as the temporal exchange of atoms between adjacent pockets, along the MD trajectories as a fingerprint of hidden allosteric communication networks. Importantly, this study indicates that dynamic pocket crosstalk analysis provides new mechanistic understandings on allosteric communication networks, enriching the available experimental data. Thus, our results suggest the prospective use of this unprecedented dynamic analysis to characterize transient binding pockets for structure-based drug design.

Figure 1

(A) Representation of the merging and splitting matrix F, calculated using all the detected pockets. The matrix F allows retrieving information on merging and splitting events. For each frame along the MD trajectory, pockets at time t are compared with pockets at time t – Δt, using the Jaccard index. In this example, at time t, the pockets 1, 2, 3, and 4 (in rows) have been detected and stored. At this point, the Jaccard index is computed with all pockets detected in the previous frame at times t – Δt, i.e., with pockets 1, 2, and 3 (in columns). Moving from t – Δt to t, this example shows that pocket 1 split into two pockets, forming the new pocket 4. Concomitantly, pockets 2 and 3 merged, forming a larger pocket that is still identified as pocket 3, according to its Jaccard index. (B) Schematic example of the conversion of the aggregate merging/splitting statistics N into an undirected network graph. In the matrix N, the off-diagonal red numbers indicate the frequency of the merging and splitting events, which is then reflected by the size of the edge connecting two pockets.

Figure 2

(A) Localization of the main pockets computed for the PNP X-ray structure 3K8O. On the right, the orthosteric ligand DADMe-ImmH located in the orthosteric pocket (in orange), as in the X-ray structure, and in the prebinding pocket (in yellow), as found in our MD simulations. (B) Volume over time of the three orthosteric sites pID 4, pID 9, and pID 12. The volumes have been smoothed employing a Gaussian filter.

Figure 3

(A) Time persistency of residues that define the orthosteric pocket (blue stem, pID 15) and the allosteric pocket (red stem, pID 22). (B) Volume plot of pID 15 and pID 22 in selected frames and representation of merging and splitting events.

Figure 4

Pocket crosstalk analysis reveals that when pID 15 merges with the allosteric pID 22, a small channel is formed (Figure 3). We used adiabatic biased simulations to characterize the passage of a sodium ion through the transient channel detected by our algorithm (see Supporting Information). (A) The green spheres indicate the pathway followed by the sodium ion along the simulations, toward the extracellular site. The two conserved residues Trp246 and Asp52 are shown. In panel B we show the narrowest section of the channel, with the gating Trp246 partially flipped so as to allow ion passage.

Figure 5

Representation of the dynamical behavior of the ATP pockets during MD simulations of the KD_in (a, b) and KD_out systems (c–f). In KD_in, the ATP pocket pID 5 coexists with the nearby pID 3 for 63.0% of the simulations (a), while pID 5 is the only pocket for 35% of the simulations (b). In KD_out, the ATP pocket coexists with pID 3 and pID 28 for 11% of the MD simulations (c). It is the only emerging pocket for 38% of the simulations (d), and it coexists with pID 3 for 20% of the simulations (e) and with pID 28 for 23% of the simulations (f).

Figure 6

Networks of the most persistent pockets found in the KD_in, T315I-KD_in, Myr/KD_in, and Myr/T315I-KD_in trajectories. Each pocket (i.e., network node) is represented as a sphere, with the different colors indicating the pocket’s persistency. The pockets are connected via black lines (i.e., network edges). The width of each edge is proportional to the communication frequency. The networks connect the ATP and the myristate binding sites in all systems except T315I-KD_in. We performed our analysis considering only pockets having a persistency of at least 20% and above, along the simulation time.