Interactome evolution: insights from genome-wide analyses of protein-protein interactions

Wiring Between Close Nodes in Molecular Networks Evolves More Quickly Than Between Distant Nodes

As species diverge, a wide range of evolutionary processes lead to changes in protein-protein interaction (PPI) networks and metabolic networks. The rate at which molecular networks evolve is an important question in evolutionary biology. Previous empirical work has focused on interactomes from model organisms to calculate rewiring rates, but this is limited by the relatively small number of species and sparse nature of network data across species. We present a proxy for variation in network topology: variation in drug-drug interactions (DDIs), obtained by studying drug combinations (DCs) across taxa. Here, we propose the rate at which DDIs change across species as an estimate of the rate at which the underlying molecular network changes as species diverge. We computed the evolutionary rates of DDIs using previously published data from a high-throughput study in gram-negative bacteria. Using phylogenetic comparative methods, we found that DDIs diverge rapidly over short evolutionary time periods, but that divergence saturates over longer time periods. In parallel, we mapped drugs with known targets in PPI and cofunctional networks. We found that the targets of synergistic DDIs are closer in these networks than other types of DCs and that synergistic interactions have a higher evolutionary rate, meaning that nodes that are closer evolve at a faster rate. Future studies of network evolution may use DC data to gain larger-scale perspectives on the details of network evolution within and between species.

From interaction networks to interfaces, scanning intrinsically disordered regions using AlphaFold2

The revolution brought about by AlphaFold2 opens promising perspectives to unravel the complexity of protein-protein interaction networks. The analysis of interaction networks obtained from proteomics experiments does not systematically provide the delimitations of the interaction regions. This is of particular concern in the case of interactions mediated by intrinsically disordered regions, in which the interaction site is generally small. Using a dataset of protein-peptide complexes involving intrinsically disordered regions that are non-redundant with the structures used in AlphaFold2 training, we show that when using the full sequences of the proteins, AlphaFold2-Multimer only achieves 40% success rate in identifying the correct site and structure of the interface. By delineating the interaction region into fragments of decreasing size and combining different strategies for integrating evolutionary information, we manage to raise this success rate up to 90%. We obtain similar success rates using a much larger dataset of protein complexes taken from the ELM database. Beyond the correct identification of the interaction site, our study also explores specificity issues. We show the advantages and limitations of using the AlphaFold2 confidence score to discriminate between alternative binding partners, a task that can be particularly challenging in the case of small interaction motifs.

Primate protein-ligand interfaces exhibit significant conservation and unveil human-specific evolutionary drivers

Despite the vast phenotypic differences observed across primates, their protein products are largely similar to each other at the sequence level. We hypothesized that, since proteins accomplish all their functions via interactions with other molecules, alterations in the sites that participate in these interactions may be of critical importance. To uncover the extent to which these sites evolve across primates, we built a structurally-derived dataset of ~4,200 one-to-one orthologous sequence groups across 18 primate species, consisting of ~68,000 ligand-binding sites that interact with DNA, RNA, small molecules, ions, or peptides. Using this dataset, we identify functionally important patterns of conservation and variation within the amino acid residues that facilitate protein-ligand interactions across the primate phylogeny. We uncover that interaction sites are significantly more conserved than other sites, and that sites binding DNA and RNA further exhibit the lowest levels of variation. We also show that the subset of ligand-binding sites that do vary are enriched in components of gene regulatory pathways and uncover several instances of human-specific ligand-binding site changes within transcription factors. Altogether, our results suggest that ligand-binding sites have experienced selective pressure in primates and propose that variation in these sites may have an outsized effect on phenotypic variation in primates through pleiotropic effects on gene regulation.

Growing ecosystem of deep learning methods for modeling protein-protein interactions

Numerous cellular functions rely on protein-protein interactions. Efforts to comprehensively characterize them remain challenged however by the diversity of molecular recognition mechanisms employed within the proteome. Deep learning has emerged as a promising approach for tackling this problem by exploiting both experimental data and basic biophysical knowledge about protein interactions. Here, we review the growing ecosystem of deep learning methods for modeling protein interactions, highlighting the diversity of these biophysically informed models and their respective trade-offs. We discuss recent successes in using representation learning to capture complex features pertinent to predicting protein interactions and interaction sites, geometric deep learning to reason over protein structures and predict complex structures, and generative modeling to design de novo protein assemblies. We also outline some of the outstanding challenges and promising new directions. Opportunities abound to discover novel interactions, elucidate their physical mechanisms, and engineer binders to modulate their functions using deep learning and, ultimately, unravel how protein interactions orchestrate complex cellular behaviors.

Targeting stressor-induced dysfunctions in protein-protein interaction networks via epichaperomes

Diseases are manifestations of complex changes in protein-protein interaction (PPI) networks whereby stressors, genetic, environmental, and combinations thereof, alter molecular interactions and perturb the individual from the level of cells and tissues to the entire organism. Targeting stressor-induced dysfunctions in PPI networks has therefore become a promising but technically challenging frontier in therapeutics discovery. This opinion provides a new framework based upon disrupting epichaperomes - pathological entities that enable dysfunctional rewiring of PPI networks - as a mechanism to revert context-specific PPI network dysfunction to a normative state. We speculate on the implications of recent research in this area for a precision medicine approach to detecting and treating complex diseases, including cancer and neurodegenerative disorders.

In search of a Drosophila core cellular network with single-cell transcriptome data

Along with specialized functions, cells of multicellular organisms also perform essential functions common to most if not all cells. Whether diverse cells do this by using the same set of genes, interacting in a fixed coordinated fashion to execute essential functions, or a subset of genes specific to certain cells, remains a central question in biology. Here, we focus on gene coexpression to search for a core cellular network across a whole organism. Single-cell RNA-sequencing measures gene expression of individual cells, enabling researchers to discover gene expression patterns that contribute to the diversity of cell functions. Current efforts to study cellular functions focus primarily on identifying differentially expressed genes across cells. However, patterns of coexpression between genes are probably more indicative of biological processes than are the expression of individual genes. We constructed cell-type-specific gene coexpression networks using single-cell transcriptome datasets covering diverse cell types from the fruit fly, Drosophila melanogaster. We detected a set of highly coordinated genes preserved across cell types and present this as the best estimate of a core cellular network. This core is very small compared with cell-type-specific gene coexpression networks and shows dense connectivity. Gene members of this core tend to be ancient genes and are enriched for those encoding ribosomal proteins. Overall, we find evidence for a core cellular network in diverse cell types of the fruit fly. The topological, structural, functional, and evolutionary properties of this core indicate that it accounts for only a minority of essential functions.

Are transient protein-protein interactions more dispensable?

Protein-protein interactions (PPIs) are key drivers of cell function and evolution. While it is widely assumed that most permanent PPIs are important for cellular function, it remains unclear whether transient PPIs are equally important. Here, we estimate and compare dispensable content among transient PPIs and permanent PPIs in human. Starting with a human reference interactome mapped by experiments, we construct a human structural interactome by building three-dimensional structural models for PPIs, and then distinguish transient PPIs from permanent PPIs using several structural and biophysical properties. We map common mutations from healthy individuals and disease-causing mutations onto the structural interactome, and perform structure-based calculations of the probabilities for common mutations (assumed to be neutral) and disease mutations (assumed to be mildly deleterious) to disrupt transient PPIs and permanent PPIs. Using Bayes' theorem we estimate that a similarly small fraction (<~20%) of both transient and permanent PPIs are completely dispensable, i.e., effectively neutral upon disruption. Hence, transient and permanent interactions are subject to similarly strong selective constraints in the human interactome.

Interactome of Arabidopsis Thaliana

More than 95,000 protein–protein interactions of Arabidopsis thaliana have been published and deposited in databases. This dataset was supplemented by approximately 900 additional interactions, which were identified in the literature from the years 2002–2021. These protein–protein interactions were used as the basis for a Cytoscape network and were supplemented with data on subcellular localization, gene ontologies, biochemical properties and co-expression. The resulting network has been exemplarily applied in unraveling the PPI-network of the plant vacuolar proton-translocating ATPase (V-ATPase), which was selected due to its central importance for the plant cell. In particular, it is involved in cellular pH homeostasis, providing proton motive force necessary for transport processes, trafficking of proteins and, thereby, cell wall synthesis. The data points to regulation taking place on multiple levels: (a) a phosphorylation-dependent regulation by 14-3-3 proteins and by kinases such as WNK8 and NDPK1a, (b) an energy-dependent regulation via HXK1 and the glucose receptor RGS1 and (c) a Ca2+-dependent regulation by SOS2 and IDQ6. The known importance of V-ATPase for cell wall synthesis is supported by its interactions with several proteins involved in cell wall synthesis. The resulting network was further analyzed for (experimental) biases and was found to be enriched in nuclear, cytosolic and plasma membrane proteins but depleted in extracellular and mitochondrial proteins, in comparison to the entity of protein-coding genes. Among the processes and functions, proteins involved in transcription were highly abundant in the network. Subnetworks were extracted for organelles, processes and protein families. The degree of representation of organelles and processes reveals limitations and advantages in the current knowledge of protein–protein interactions, which have been mainly caused by a high number of database entries being contributed by only a few publications with highly specific motivations and methodologies that favor, for instance, interactions in the cytosol and the nucleus.

Design and synthesis of a new orthogonally protected glutamic acid analog and its use in the preparation of high affinity polo-like kinase 1 polo-box domain - binding peptide macrocycles

Targeting protein - protein interactions (PPIs) has emerged as an important area of discovery for anticancer therapeutic development. In the case of phospho-dependent PPIs, such as the polo-like kinase 1 (Plk1) polo-box domain (PBD), a phosphorylated protein residue can provide high-affinity recognition and binding to target protein hot spots. Developing antagonists of the Plk1 PBD can be particularly challenging if one relies solely on interactions within and proximal to the phospho-binding pocket. Fortunately, the affinity of phospho-dependent PPI antagonists can be significantly enhanced by taking advantage of interactions in both the phospho-binding site and hidden "cryptic" pockets that may be revealed on ligand binding. In our current paper, we describe the design and synthesis of macrocyclic peptide mimetics directed against the Plk1 PBD, which are characterized by a new glutamic acid analog that simultaneously serves as a ring-closing junction that provides accesses to a cryptic binding pocket, while at the same time achieving proper orientation of a phosphothreonine (pT) residue for optimal interaction in the signature phospho-binding pocket. Macrocycles prepared with this new amino acid analog introduce additional hydrogen-bonding interactions not found in the open-chain linear parent peptide. It is noteworthy that this new glutamic acid-based amino acid analog represents the first example of extremely high affinity ligands where access to the cryptic pocket from the pT-2 position is made possible with a residue that is not based on histidine. The concepts employed in the design and synthesis of these new macrocyclic peptide mimetics should be useful for further studies directed against the Plk1 PBD and potentially for ligands directed against other PPI targets.

Protein sequence-to-structure learning: Is this the end(-to-end revolution)?

The potential of deep learning has been recognized in the protein structure prediction community for some time, and became indisputable after CASP13. In CASP14, deep learning has boosted the field to unanticipated levels reaching near-experimental accuracy. This success comes from advances transferred from other machine learning areas, as well as methods specifically designed to deal with protein sequences and structures, and their abstractions. Novel emerging approaches include (i) geometric learning, that is, learning on representations such as graphs, three-dimensional (3D) Voronoi tessellations, and point clouds; (ii) pretrained protein language models leveraging attention; (iii) equivariant architectures preserving the symmetry of 3D space; (iv) use of large meta-genome databases; (v) combinations of protein representations; and (vi) finally truly end-to-end architectures, that is, differentiable models starting from a sequence and returning a 3D structure. Here, we provide an overview and our opinion of the novel deep learning approaches developed in the last 2 years and widely used in CASP14.

Mutation Edgotype Drives Fitness Effect in Human

Missense mutations are known to perturb protein-protein interaction networks (known as interactome networks) in different ways. However, it remains unknown how different interactome perturbation patterns (“edgotypes”) impact organismal fitness. Here, we estimate the fitness effect of missense mutations with different interactome perturbation patterns in human, by calculating the fractions of neutral and deleterious mutations that do not disrupt PPIs (“quasi-wild-type”), or disrupt PPIs either by disrupting the binding interface (“edgetic”) or by disrupting overall protein stability (“quasi-null”). We first map pathogenic mutations and common non-pathogenic mutations onto homology-based three-dimensional structural models of proteins and protein-protein interactions in human. Next, we perform structure-based calculations to classify each mutation as either quasi-wild-type, edgetic, or quasi-null. Using our predicted as well as experimentally determined interactome perturbation patterns, we estimate that >∼40% of quasi-wild-type mutations are effectively neutral and the remaining are mostly mildly deleterious, that >∼75% of edgetic mutations are only mildly deleterious, and that up to ∼75% of quasi-null mutations may be strongly detrimental. These estimates are the first such estimates of fitness effect for different network perturbation patterns in any interactome. Our results suggest that while mutations that do not disrupt the interactome tend to be effectively neutral, the majority of human PPIs are under strong purifying selection and the stability of most human proteins is essential to human life.

Modulating Protein-Protein Interactions by Cyclic and Macrocyclic Peptides. Prominent Strategies and Examples

Cyclic and macrocyclic peptides constitute advanced molecules for modulating protein-protein interactions (PPIs). Although still peptide derivatives, they are metabolically more stable than linear counterparts, and should have a lower degree of flexibility, with more defined secondary structure conformations that can be adapted to imitate protein interfaces. In this review, we analyze recent progress on the main methods to access cyclic/macrocyclic peptide derivatives, with emphasis in a few selected examples designed to interfere within PPIs. These types of peptides can be from natural origin, or prepared by biochemical or synthetic methodologies, and their design could be aided by computational approaches. Some advances to facilitate the permeability of these quite big molecules by conjugation with cell penetrating peptides, and the incorporation of β-amino acid and peptoid structures to improve metabolic stability, are also commented. It is predicted that this field of research could have an important future mission, running in parallel to the discovery of new, relevant PPIs involved in pathological processes.

Use of Integrative Interactomics for Improvement of Farm Animal Health and Welfare: An Example with Fescue Toxicosis

Rapid scientific advances are increasing our understanding of the way complex biological interactions integrate to maintain homeostatic balance and how seemingly small, localized perturbations can lead to systemic effects. The ‘omics movement, alongside increased throughput resulting from statistical and computational advances, has transformed our understanding of disease mechanisms and the multi-dimensional interaction between environmental stressors and host physiology through data integration into multi-dimensional analyses, i.e., integrative interactomics. This review focuses on the use of high-throughput technologies in farm animal research, including health- and toxicology-related papers. Although limited, we highlight recent animal agriculture-centered reports from the integrative multi-‘omics movement. We provide an example with fescue toxicosis, an economically costly disease affecting grazing livestock, and describe how integrative interactomics can be applied to a disease with a complex pathophysiology in the pursuit of novel treatment and management approaches. We outline how ‘omics techniques have been used thus far to understand fescue toxicosis pathophysiology, lay out a framework for the fescue toxicosis integrome, identify some challenges we foresee, and offer possible means for addressing these challenges. Finally, we briefly discuss how the example with fescue toxicosis could be used for other agriculturally important animal health and welfare problems.

Association study based on topological constraints of protein-protein interaction networks

The non-random interaction pattern of a protein–protein interaction network (PIN) is biologically informative, but its potentials have not been fully utilized in omics studies. Here, we propose a network-permutation-based association study (NetPAS) method that gauges the observed interactions between two sets of genes based on the comparison between permutation null models and the empirical networks. This enables NetPAS to evaluate relationships, constrained by network topology, between gene sets related to different phenotypes. We demonstrated the utility of NetPAS in 50 well-curated gene sets and comparison of association studies using Z-scores, modified Zʹ-scores, p-values and Jaccard indices. Using NetPAS, a weighted human disease network was generated from the association scores of 19 gene sets from OMIM. We also applied NetPAS in gene sets derived from gene ontology and pathway annotations and showed that NetPAS uncovered functional terms missed by DAVID and WebGestalt. Overall, we show that NetPAS can take topological constraints of molecular networks into account and offer new perspectives than existing methods.

Estimating dispensable content in the human interactome

Protein-protein interaction (PPI) networks (interactome networks) have successfully advanced our knowledge of molecular function, disease and evolution. While much progress has been made in quantifying errors and biases in experimental PPI datasets, it remains unknown what fraction of the error-free PPIs in the cell are completely dispensable, i.e., effectively neutral upon disruption. Here, we estimate dispensable content in the human interactome by calculating the fractions of PPIs disrupted by neutral and non-neutral mutations. Starting with the human reference interactome determined by experiments, we construct a human structural interactome by building homology-based three-dimensional structural models for PPIs. Next, we map common mutations from healthy individuals as well as Mendelian disease-causing mutations onto the human structural interactome, and perform structure-based calculations of how these mutations perturb the interactome. Using our predicted as well as experimentally-determined interactome perturbation patterns by common and disease mutations, we estimate that <~20% of the human interactome is completely dispensable.

The fraction of protein-protein interactions (PPIs) that can be disrupted without fitness effect is unknown. Here, the authors model how disease-causing mutations and common mutations carried by healthy people perturb the interactome, and estimate that <20% of human PPIs are completely dispensable.

Dependency Between Protein-Protein Interactions and Protein Variability and Evolutionary Rates in Vertebrates: Observed Relationships and Stochastic Modeling

Recent developments in sequencing and growth of bioinformatics resources provide us with vast depositories of protein network and single nucleotide polymorphism data. It allows us to re-examine, on a larger and more comprehensive scale, the relationship between protein–protein interactions and protein variability and evolutionary rates. This relationship has remained far from unambiguously resolved for quite a long time, reflecting shifting analysis approaches in the literature, and growing data availability. In this study, we utilized several public genomic databases to investigate this relationship in human, mouse, pig, chicken, and zebrafish. We observed strong non-linear relationship patterns (tending towards convex decreasing function shapes) between protein variability and the density of corresponding protein–protein interactions across all five species. To investigate further, we carried out stochastic simulations, modeling the interplay between protein connectivity and variability. Our results indicate that a simple negative linear correlation model, often suggested (or tacitly assumed) in the literature, as either a null or an alternative hypothesis, is not a good fit with the observed data. After considering different (but still relatively simple, and not overfitting) simulation models, we found that a convex decreasing protein variability–connectivity function (specifically, exponential decay) led to a much better fit with the real data. We conclude that simple correlation models might be inadequate for describing protein variability–connectivity interplay in vertebrates; they often tend towards false negatives (showing no more than marginal linear or rank correlation where there are in fact strong non-random patterns).

PPI-Detect: A support vector machine model for sequence-based prediction of protein-protein interactions

The prediction of peptide-protein or protein-protein interactions (PPI) is a challenging task, especially if amino acid sequences are the only information available. Machine learning methods allow us to exploit the information content in PPI datasets. However, the numerical codification of these datasets often influences the performance of data mining approaches. Here, we introduce a procedure for the general-purpose numerical codification of polypeptides. This procedure transforms pairs of amino acid sequences into a machine learning-friendly vector, whose elements represent numerical descriptors of residues in proteins. We used this numerical encoding procedure for the development of a support vector machine model (PPI-Detect), which allows predicting whether two proteins will interact or not. PPI-Detect (https://ppi-detect.zmb.uni-due.de/) outperforms state of the art sequence-based predictors of PPI. We employed PPI-Detect for the analysis of derivatives of EPI-X4, an endogenous peptide inhibitor of CXCR4, a G-protein-coupled receptor. There, we identified with high accuracy those peptides which bind better than EPI-X4 to the receptor. Also using PPI-Detect, we designed a novel peptide and then experimentally established its anti-CXCR4 activity. © 2019 Wiley Periodicals, Inc.