Reads meet rotamers: structural biology in the age of deep sequencing

Building a Hybrid Physical-Statistical Classifier for Predicting the Effect of Variants Related to Protein-Drug Interactions

A key issue in drug design is how population variation affects drug efficacy by altering binding affinity (BA) in different individuals, an essential consideration for government regulators. Ideally, we would like to evaluate the BA perturbations of millions of single-nucleotide variants (SNVs). However, only hundreds of protein-drug complexes with SNVs have experimentally characterized BAs, constituting too small a gold standard for straightforward statistical model training. Thus, we take a hybrid approach: using physically based calculations to bootstrap the parameterization of a full model. In particular, we do 3D structure-based docking on ∼10,000 SNVs modifying known protein-drug complexes to construct a pseudo gold standard. Then we use this augmented set of BAs to train a statistical model combining structure, ligand and sequence features and illustrate how it can be applied to millions of SNVs. Finally, we show that our model has good cross-validated performance (97% AUROC) and can also be validated by orthogonal ligand-binding data.

Intensification: A Resource for Amplifying Population-Genetic Signals with Protein Repeats

Large-scale genome sequencing holds great promise for the interpretation of protein structures through the discovery of many, rare functional variants in the human population. However, because protein-coding regions are under high selective constraints, these variants occur at low frequencies, such that there is often insufficient statistics for downstream calculations. To address this problem, we develop the Intensification approach, which uses the modular structure of repeat protein domains to amplify signals of selection from population genetics and traditional interspecies conservation. In particular, we are able to aggregate variants at the codon level to identify important positions in repeat domains that show strong conservation signals. This allows us to compare conservation over different evolutionary timescales. It also enables us to visualize population-genetic measures on protein structures. We make available the Intensification results as an online resource (http://intensification.gersteinlab.org) and illustrate the approach through a case study on the tetratricopeptide repeat.

Localized structural frustration for evaluating the impact of sequence variants

Population-scale sequencing is increasingly uncovering large numbers of rare single-nucleotide variants (SNVs) in coding regions of the genome. The rarity of these variants makes it challenging to evaluate their deleteriousness with conventional phenotype-genotype associations. Protein structures provide a way of addressing this challenge. Previous efforts have focused on globally quantifying the impact of SNVs on protein stability. However, local perturbations may severely impact protein functionality without strongly disrupting global stability (e.g. in relation to catalysis or allostery). Here, we describe a workflow in which localized frustration, quantifying unfavorable local interactions, is employed as a metric to investigate such effects. Using this workflow on the Protein Databank, we find that frustration produces many immediately intuitive results: for instance, disease-related SNVs create stronger changes in localized frustration than non-disease related variants, and rare SNVs tend to disrupt local interactions to a larger extent than common variants. Less obviously, we observe that somatic SNVs associated with oncogenes and tumor suppressor genes (TSGs) induce very different changes in frustration. In particular, those associated with TSGs change the frustration more in the core than the surface (by introducing loss-of-function events), whereas those associated with oncogenes manifest the opposite pattern, creating gain-of-function events.

Fragment Screening of RORγt Using Cocktail Crystallography: Identification of Simultaneous Binding of Multiple Fragments

The retinoic-acid-related orphan receptor γ t (RORγt), as a master regulator of Th17 cell pathology, has become an attractive target for small-molecule drug discovery for the treatment of Th17-cell-related autoimmune diseases. A crystallographic fragment screening was carried out for RORγt using the ligand binding domain. An overall hit rate of 5.5 % was obtained by screening 384 compounds in 96 cocktails. Five distinct hotspots were identified, and four regions of anchoring polar interactions were observed. In addition, significant induced fit was found for the binding of several fragments. Strikingly, a simultaneous binding of three fragments was revealed which presents interesting features including π-π stacking, multiple hydrogen bonds to the protein, and significant induced fit. Overall, the results offer a complete mapping of the ligand binding pocket and provide valuable inspiration in structure-based design for RORγt lead generation and optimization. The crystallographic screening also resulted in fragment hits that bind at the surface away from the ligand binding pocket. This surface site is near the plausible dimer interface by analogy with other nuclear receptor systems, which can provide initial hints to explore alternative ways to modulate RORγt through protein-protein interactions.

Identifying Allosteric Hotspots with Dynamics: Application to Inter- and Intra-species Conservation

The rapidly growing volume of data being produced by next-generation sequencing initiatives is enabling more in-depth analyses of conservation than previously possible. Deep sequencing is uncovering disease loci and regions under selective constraint, despite the fact that intuitive biophysical reasons for such constraint are sometimes absent. Allostery may often provide the missing explanatory link. We use models of protein conformational change to identify allosteric residues by finding essential surface pockets and information-flow bottlenecks, and we develop a software tool that enables users to perform this analysis on their own proteins of interest. Though fundamentally 3D-structural in nature, our analysis is computationally fast, thereby allowing us to run it across the PDB and to evaluate general properties of predicted allosteric residues. We find that these tend to be conserved over diverse evolutionary time scales. Finally, we highlight examples of allosteric residues that help explain poorly understood disease-associated variants.