A flexible volumetric comparison of protein cavities can reveal patterns in ligand binding specificity

Abstract

Conformational flexibility is an underlying cause of error in all comparisons of protein structure. Using flexible representations, some comparison algorithms can identify subtle functional similarities among distantly related proteins even when they exhibit different backbone conformations. The same techniques are not designed to identify subtle variations among closely related proteins that might cause differences in specificity. In such cases, molecular flexibility obscures structural details that influence the specific recognition of similar but non-identical ligands.To enhance the analysis of ligand binding specificity, this paper presents FAVA (Flexible Aggregate Volumetric Analysis), a conformationally robust tool for comparing similar binding cavities with different binding preferences. FAVA examines a large number of conformational samples to characterize local flexibility using Constructive Solid Geometry. Using molecular dynamics simulations as a source for conformational samples, we used FAVA to analyze a nonredundant sample of serine protease and enolase structures. Different snapshots from the same proteins exhibited significant variations in binding cavity shape. Nonetheless, analysis with FAVA revealed subfamilies with different binding preferences. FAVA also identified amino acids associated with differences in binding preferences, predicting established experimental results. These results illustrate a new approach to flexible comparison that uses sampled conformational data. It reveals that detailed comparisons of very similar proteins, such as those within small ligand binding cavities, are possible even in the presence of conformational flexibility. Identifying influences on specificity in this manner points to new applications of protein engineering and drug design. * equal contribution.

Predicting protein-ligand binding specificity based on ensemble clustering

“…We define the shape of each sampled ligand binding cavity as {tl,t 2 , ... ,tn} using VASP [17]. Following our earlier work [27], we compute the average intersection volume of each amino acid I between the sampled amino acid 'i of Ti and the sampled ligand binding cavity tj for all pairs of samples i and j. The large average intersection volume indicates that I frequently changes the shape of the binding cavity.…”

“…Considerable variation of conformational binding cavities All these observations demonstrate considerable variations of binding cavities of the same protein. The cavity variations create errors for flexible structure comparisons, preventing accurate predictions of ligand binding specificity when pro tein conformational samples are considered [27]. The protein binding cavity varied because of the motion of adjacent amino acids, leading to the motivation to identify structural motifs for binding specificity prediction.…”

“…However, this method did not provide direct comparison between conformations of different proteins. A second method, FAVA [27], defined the three dimensional region that is accessible to most conformational cavities of one protein as the frequent region. Frequent regions could be clustered to produce categorizations of ligand binding cavities, but because the frequent region does not necessarily exist in highly flexible binding sites, we examine methods here based on atomic comparisons.…”

Variational Bayesian clustering on protein cavity conformations for detecting influential amino acids

“…Given the conformational samples {A1, A2, ..., AN } of protein structure A and the binding ligand l, we define the shape of the sampled ligand binding cavities as {a1, a2, ..., aN } following our earlier work [22], leading to the volumetric representation for surface properties of binding cavities [13]. Then, we measure the volumetric distance, D(ai, aj), between two protein sampled cavities using Tanimoto Coefficient [26]:…”