Computational Drug Discovery and Design

conformational ensembles generated from the MD simulated

trajectories.

CCij¼

hiCBðÞi CBðÞj hiCBðÞi hiCBðÞj

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðÞδCBðÞi^2

rDE



ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðÞδCBðÞj^2

rDEð 6 Þ

whereh...iindicates an ensemble average, thushCB(i)irefers to the

average betweenness centrality for the residue i. The cross-

correlation matrices for the apo and agonist-bound forms showed

that the signatures of correlation between residues are scattered all

over the structure. In this map, high signal residue pairs indicate

that the two residues are linked to each other either via direct

contact or long-range communication. In order to identify the

residue pairs with long-range cross-correlation, the pairs which

showed high cross-correlation values (CCij0.5) but far from

each other, i.e., which have minimum path over six edges (dij>6),

were extracted.

Surprisingly, in the agonist-bound form, there are significant

long-range cross-correlations between the extracellular ligand

binding site and cytoplasmic G-protein binding site. Between the

correlated residues, minimum path seems to pass through the

residues with high betweenness centralities, including the impor-

tant microswitches. The results correlate well with the function of

GPCRs, i.e., coordinated domain coupling [43]: an agonist bind-

ing induces a significant conformational change in the region where

G-protein may bind. On the contrary, such long-range cross-corre-

lation was not observed in the simulated trajectories of the apo

structure. The long-range coupling between the ligand binding site

and G-protein binding site for the agonist-bound form is also

grasped by computing the mean square fluctuation using structural

ensembles. In a recent review, Unal et al. proposed that the intrin-

sically flexible extracellular and intracellular regions are functionally

coupled and this coupling is mediated by TM helix structures

[44]. It is assumed that when a domain engages a ligand, the

intrinsic disorder of receptor structure is decreased, and it coopera-

tively influences the conformation of the neighboring domain

[44]. Our computational analysis gave a more detailed picture of

this model, which could be valuable in drug discovery (seeNote 6).

3 Conclusions

We have presented a computational protocol for the mapping of

allosteric signal flow in GPCRs using conventional MD simulations

and network analysis. Through network analysis, we identified the

hotspots accountable for agonistic activity of A2AAR and the cross-

correlation between the agonist- and G-protein binding sites were

Molecular Dynamics Approach for Investigation of GPCR Allostery 465