Systematic domain-based aggregation of protein structures highlights DNA-, RNA-, and other ligand-binding positions

Abstract

Domains are fundamental subunits of proteins, and while they play major roles in facilitating protein-DNA, protein-RNA and other protein-ligand interactions, a systematic assessment of their various interaction modes is still lacking. A comprehensive resource identifying positions within domains that tend to interact with nucleic acids, small molecules and other ligands would expand our knowledge of domain functionality as well as aid in detecting ligand-binding sites within structurally uncharacterized proteins. Here we introduce an approach to identify perdomain-position interaction "propensities" by aggregating protein co-complex structures by domain and ascertaining how frequently residues mapping to each domain position interact with ligands. We perform this domain-based analysis on ∼82,000 co-complex structures, and infer positions involved in binding DNA, RNA, peptides, ions, or small molecules across 4,120 domains, which we refer to collectively as the InteracDome. Cross-validation testing reveals that ligand-binding positions for 1,327 domains can be con dently modeled and used to identify residues facilitating interactions in ∼60-69% of human genes. Our resource of domain-inferred ligand-binding sites should be a great aid in understanding disease etiology: whereas these sites are enriched in Mendelian-associated and cancer somatic mutations, they are depleted in polymorphisms observed across healthy populations. The InteracDome is available at http://interacdome.princeton.edu.

An integrative approach uncovers genes with perturbed interactions in cancers

“…First, we utilize the set of "confident" domain-ligand interactions from the InteracDome database (v0.3) [33] to identify putative ligand-binding positions. We use the 9,142 domain-ligand interactions across 1,850 domains with 5+ structural instances.…”

“…Each position within a domain is associated with a "binding frequency" between 0 and 1 that corresponds to the fraction of the time residues in this position are found in contact with the ligand of interest when analyzing co-crystal structures. For each human protein, we identify instances of InteracDome domains using HMMER (v2.3.2 and v3.1b2), and require complete, high-scoring domain instances as previously described [33,67,68]. Within a protein, there is a separate track for each domain-ligand instance within it; this track consists of the residues comprising the match states of the domain, and the weights of these residues are the binding frequencies for the ligand in the corresponding domain positions.…”

“…In interaction tracks (red), positions that are more likely to participate in ligand interactions have higher weights (vertical bars). Interaction tracks arise from domain-based binding potential calculations [33] (top two red tracks, each covering the length of the respective domain) or homology modeling [30] (bottom red track, covering the length of the modeled region). Domain tracks (green) specify which residues within a protein are part of a specific domain by 0/1 positional weights; here we have a track for each domain within the sequence.…”

“…Indeed, several cancer driver genes, including TP53 and IDH1, are well 30 known to harbor mutations within their interaction sites [29]. While traditionally interaction sites 31 have been identified directly for the small fraction of human genes with actual or modeled 32 co-complex structures, we have recently developed a domain-based approach that accurately 33 1 detects residues that interact with DNA, RNA, peptides, ions or small molecules across 63% of 34 human genes [33]. A robust computational framework that utilizes this vastly expanded 35 knowledge-base about interaction sites and explicitly integrates it with additional lines of evidence 36 regarding subgene functionality would provide a powerful new approach not only to detect but 37 also to interpret a wide range of mutations driving protein dysfunction in cancer.…”

“…In this work, we have introduced a fast, integrative framework to detect cancer driver genes by 262 uncovering whether somatic mutations across tumors are enriched in sites of different types of 263 functionalities. Our method utilizing this framework, PertInInt, integrates knowledge from the 264 largest set of protein-ligand interaction sites to date [30,33] with additional biological data 265 regarding subgene functionality and whole gene mutability ( Fig. 1).…”