- Email: [email protected]
Structure-based prediction of host–pathogen protein interactions
Available online at www.sciencedirect.com
ScienceDirect Structure-based prediction of host–pathogen protein interactions Rachelle Mariano1 and Stefan Wuchty2,3,4 The discovery, validation, and characterization of proteinbased interactions from different species are crucial for translational research regarding a variety of pathogens, ranging from the malaria parasite Plasmodium falciparum to HIV-1. Here, we review recent advances in the prediction of host– pathogen protein interfaces using structural information. In particular, we observe that current methods chiefly perform machine learning on sequence and domain information to produce large sets of candidate interactions that are further assessed and pruned to generate final, highly probable sets. Structure-based studies have also emphasized the electrostatic properties and evolutionary transformations of pathogenic interfaces, supplying crucial insight into antigenic determinants and the ways pathogens compete for host protein binding. Advancements in spectroscopic and crystallographic methods complement the aforementioned techniques, permitting the rigorous study of true positives at a molecular level. Together, these approaches illustrate how protein structure on a variety of levels functions coordinately and dynamically to achieve host takeover. Addresses 1 Brigham & Women’s Hospital, Harvard Medical School, Harvard University, Cambridge, MA, United States 2 Dept. of Computer Science, Univ. of Miami, Coral Gables, FL, United States 3 Center for Computational Science, Univ. of Miami, Coral Gables, FL, United States 4 Sylvester Cancer Center, Univ. of Miami, Miami, FL, United States Corresponding author: Mariano, Rachelle ([email protected])
Current Opinion in Structural Biology 2017, 44:119–124 This review comes from a themed issue on Sequences and topology Edited by Ramanathan Sowdhamini and Kenji Mizuguchi
http://dx.doi.org/10.1016/j.sbi.2017.02.007 0959-440X/ã 2017 Elsevier Ltd. All rights reserved.
The accurate determination of protein–protein interactions between bacterial, viral, and parasitic pathogens and their human hosts harbors great medicinal potential, as these discoveries could be used to target specific diseaserelated interfaces with minimal disruption of the underlying human interaction network. Indeed, elucidating www.sciencedirect.com
host–pathogen protein–protein interactions (HP-PPIs) for therapeutics drives their intensive computational and experimental study, and rapidly improving approaches have generated valuable high-fidelity HP-PPI candidates. Benchside high-throughput methods often bear a considerable proportion of false positives when applied to HP-PPI prediction. Moreover, exogenous expression of pathogenic proteins remains difficult, and most results must be translated across evolutionarily distant species. Computational inference of HP-PPIs can identify small subsets of highly probable interactions for informed experimental follow-up by techniques such as nuclear magnetic resonance microscopy (NMR) and crystallography. Combined, these methods allow researchers to not only ascertain how a pathogenic protein interacts with its host on a molecular scale, but also how such interactions function in a larger cellular network.
Computational HP-PPI prediction based on sequence and domain information: homology-based approaches While first limited to intraspecies interactions, sequence similarity-based approaches have since extended to interspecies PPI prediction. Since high primary sequence similarity implies an interaction – an interolog – these methods map known interaction interface sequences onto homologous or orthologous pairs of sequences in different organisms. For example, such sequence comparisons yielded HP-PPIs between Plasmodium falciparum (P. falciparum) [1–3] and Helicobacter pylori (H. pylori) [4] and their human host. Interolog screens benefit from their straightforward execution as well as abundant protein sequence information from which to mine data [5]. As the interologous proteins should demonstrate at least 80% sequence similarity, the ability to correctly determine HP-PPIs from interologs rapidly decreases with evolutionary distance. Additionally, pathogens are locked in a biological ‘arms race’ with their hosts, and their proteins may experience rapid changes in sequence that affect the fidelity of interolog screens [6]. Interolog screens also have a penchant to generate a huge amount of false positive hits. Therefore, further computational investigation of potential hits involves filtering based on the cellular localization, biological functions, and expression profiles of putative HP-PPIs to significantly improve the quality of potential HP-PPI candidates [1–4]. Machine learning approaches using derived sequencebased features have also procured possible HP-PPIs. Shen et al. [7] represented sequences of interacting Current Opinion in Structural Biology 2017, 44:119–124
120 Sequences and topology
proteins as numerical profiles of the occurrence of aminoacid triplets. To predict interactions between human proteins, they utilized support vector machine algorithms (SVM) that were trained by carefully picked positive and negative training sets of protein interactions. This approach was similarly applied to interactions between human proteins and the malaria parasite P. falciparum [8]. Such a representation reduces the size of the feature space but may impair the quality of results. All approaches that predict HP-PPIs via supervised machine learningwhether with sequence or higher order information- need appropriate positive and negative training sets to robustly classify interacting proteins. Yu et al. [9] showed that the choice of non-interactions in the training data greatly impacts the accurate identification of interacting vs. non-interacting pairs. Tools balancing negative example selection have been recently developed to combat this issue. In particular, Eid et al. [10] developed a dissimilarity-random-sampling algorithm for the determination of unlikely occurring interactions between human host and pathogen proteins. The authors sampled highly dissimilar protein sequences from other viruses compared to interacting proteins of the virus in question to generate a negative training set. These training sets trained a SVM with prediction accuracies up to 86% [10], suggesting that the skilled choice of negative training sets drives the reliability of predicted HP-PPIs.
their human host [4]. DDIs have also been combined with protein sequence k-mers and topological properties of host proteins in a human protein interaction network to predict host–pathogen interactions using a SVM [16]. Notably, the use of DDIs allowed prioritization of proteins with extracellular or trans-membrane domains to assess interactions driving invasion and intracellular signaling [4,16]. Domain-based prediction also assisted in the identification of common functional features that allow pathogens to interact with more than one host [17].
Domain-based approaches
Motif and integration-based approaches
Computational inference of HP-PPIs often combines primary sequence similarity with higher order structural information from motifs and domains to increase prediction accuracy [11]. A protein domain is usually defined as a conserved part of a protein’s sequence and three-dimensional structure that mediates the protein’s biological functions while folding and evolving independently [12]. Since domain–domain interactions (DDIs) are largely considered to drive PPIs, numerous studies have used known intra-species DDIs as a basis for the prediction of HP-PPIs. Furthermore, unbiased approaches to elucidate significant predictive features of HP-PPIs have repeatedly emphasized their role [9]. In particular, out of 44 descriptors involving amino acid frequencies of host and pathogen sequences, protein–domain associations appeared to have the highest predictive effect when used with SVM and random forest (RF) algorithms [13]. In a different study, DDIs were included in an 18-dimensional vector and combined with topological sequence and functional characteristics to predict interactions between proteins of HIV-1 using different variations of neural network methods [14] that outperformed RFs [15].
Motif–domain and motif–motif interactions have also gained traction as foundations for HP-PPI prediction, as short linear motifs have proved vital for host–pathogen protein binding. Evans et al. annotated short eukaryotic linear motifs (ELMs) in HIV-1 proteins and used human counter domains that interact with these ELMs to generate an HIV-1 and human interactome [20]. SeguraCabrera et al. [21] combined motif information from 3D interaction databases with stringent filters to create an infectome representing HP-PPIs of 5 viruses with the human host, integrating surface accessibility and structural information. Although these studies predicted HIV1 HP-PPIs with similar techniques, their results differed as a consequence of discrepancies in filtering and motif definition. Although stringent filters were applied to secure biologically meaningful results, both studies provided a plethora of interactions [21]. This is a standing issue in computational HP-PPI prediction, as results depend on preferred methods and suffer from persistence of false positive hits.
Combining domain-based data with primary sequence homology searches of interacting domains allowed the large-scale detection of hypothetical interactions between proteins of H. pylori, HIV-1, and Salmonella with Current Opinion in Structural Biology 2017, 44:119–124
DDIs can also be employed separately from primary sequences to derive HP-PPIs. Itzhaki et al. integrated sets of protein interactions from various organisms with verified protein–domain profiles, assuming that intraspecies DDIs similarly connect in HP-PPIs [18]. The authors modeled the probability that proteins with certain domains interact in a Bayesian framework and generated a protein interaction network between P. falciparum and the human host. Remarkably, interactions thus obtained featured significant co-expression of involved pathogen and human proteins, illustrating the high probability of interaction that DDI-based HP-PPIs can achieve. Liu et al. [19] used an expectation maximization algorithm to find expression-correlated interactions between P. falciparum and human erythrocytes that were subsequently verified using expression data.
Although primary and secondary sequence integration enhances computational HP-PPI predictions, auxiliary data are increasingly used to curb the impact of false positives. In particular, current approaches are assimilating domain, sequence, and ELM data with gene ontology (GO) features, graph topological properties, and gene co-expression data to train HP-PPI classifiers, instead www.sciencedirect.com
Prediction of host–pathogen interactions Mariano and Wuchty 121
of using such features as post hoc filters and verification [5,22–24]. For example, Tastan et al. [15] used such information to deduce interactions between HIV-1 and the human host, thereby finding an interaction between the HIV-1-specific protein tat and the human vitamin D receptor that has been experimentally validated. Coelho et al. predicted the human-microbial oral interactome [25] by incorporating DDIs, protein sequence features, and GO annotations. While such integrative approaches appear promising, the massive dearth of fully annotated host–pathogen interactions has prompted researchers to also develop tools to transfer data from other host–pathogen systems [26,27]. In particular, Ksirsagar et al. [27] built models to infer interactions between proteins of the bacterial species Salmonella typhimurium and the plant host Arabidopsis thaliana for which no experimentally determined PPIs exist. To cope with the massive absence of data, the authors inferred PPIs from known interactions between human and Salmonella through homology searches and graph-based alignments, as well as from the utilization of various kernel methods to predict potential PPIs from many small source data sets that were obtained from other host–pathogen systems.
PPI prediction based on 3-D structural information Structure-structure interaction (SSI) methods complement existing high-confidence interactions by increasing the functional coherence and pathway coverage of the underlying interaction network [28]. Although a number of approaches predict PPIs based on the 3-D structure of the participating proteins [29], such techniques have found relatively little attention for HP-PPI prediction mainly as a result of a relative lack of 3-D structures and protein interactions. Still, a small number of studies have employed structural similarities to detect HP-PPIs. Davis et al. presented the first instance of using 3-D structural homology for interspecies PPI-prediction [30]. Their protocol first scanned the host and pathogen genomes for proteins with similarity to known protein complexes, assessed these putative interactions using structural information, and filtered the remaining interactions based on biological context for several human pathogens. Doolittle and Gomez used the structural similarity of 9 HIV-1 proteins to human proteins with known interactions, functional data from RNAi studies, and GO cellular component annotation to generate 502 interaction predictions between HIV and human proteins [33]. Mycobacterium tuberculosis (TB)-human interactions elucidated through SSI revealed putative host protein targets with known roles in HIV replication, thus supplying mechanistic insight into the intensified disease phenotype that occurs during TB-HIV co-infection [34]. de Chassey et al. [31] used siRNA gene silencing to confirm human and influenza NS1 protein interactions derived from SSI www.sciencedirect.com
methods by measuring the effect of interactor knockdown on viral replication. They concluded that 10/26 of their final candidates directly mediated this process [31].
Structure and evolution While structure-driven approaches appear as powerful predictors of host–pathogen interaction, data suggest that pathogens evolve protein interfaces to achieve binding without sequence or structural similarity to native interactors [6,32]. Franzosa and Xia [6] showed that exogenous interfaces (HP-PPIs) overlap with, mimic, and compete with endogenous (host–host) interfaces. The endogenous interfaces mimicked by viral proteins tend to participate in multiple interactions that are transient and regulatory in nature. While endogenous interfaces evolve more slowly than the rest of the protein surface, exogenous interfaces—including many sites of endogenous-exogenous overlap—tend to evolve faster, consistent with the aforementioned evolutionary ‘arms race’ [6]. Furthermore, structural comparative modeling of human pathogenic and non-pathogenic Ebola virus revealed that pathogenic residues did not localize to known interaction interfaces and instead clustered to distal regions [33]. Investigating molecular mimicry, Doolittle and Gomez predicted interactions between the human host and the Dengue fever virus, assuming that pathogen proteins can replace structurally similar host proteins in intra-species host protein interactions [34]. Apart from featuring rapidly evolving sequences, secreted effector proteins of pathogens also harbor long intrinsically disordered regions that may assist in functional hijacking through mimicry of host proteins and immune escape through dynamic conformational changes. While an increasingly important structural characteristic of pathogen proteins, we refer the reader to [35–37] that extensively review such topics.
Surface electrostatics and epitope prediction Electrostatic features of interacting protein surfaces play a crucial role in finding epitopes, which are the antigenic determinants of pathogens that are recognized by the immune system. Antibodies recognize a variety of foreign proteins through a relatively limited and homogeneous reservoir of sequences and structures. In a structure-based computational analysis, Peng, Jian and Yang [38] found that antigen-binding sites of antibodies show an enrichment of aromatic residues surrounded by short hydrophilic sequences that permits promiscuous non-covalent interactions through contact with backbone and sidechain carbons. Such features allow a sparse number of host antigens to recognize numerous pathogen antigens through shared physicochemical properties. These electrostatic characteristics also play a role in the binding of gp120, an HIV-1 protein, to either co-receptor CCR5 or CXCR4 of human CD4 cells in a choice associated with differential rates of AIDS progression. Heider et al. used hydrophobicity and electrostatic potential information Current Opinion in Structural Biology 2017, 44:119–124
122 Sequences and topology
Figure 1
organism
training set A
B E
sequence
similarity
F
C
G
D
host
pathogen
interologs
host pathogen
SVM, naïve Bayes, random forests,…
pathogen p host h
domain-domain interactions
functions
3-D interactions
literature
host pathogen
co-expression o-express
auxiliary data
…
Current Opinion in Structural Biology
Common features of computational workflows for the prediction of interactions between proteins of pathogens and hosts. In a first step, sequence and structure based information is gathered to generate sets of candidate host–pathogen interactions. Subsequently, machine learning approaches are used to predict high confidence interactions by combining appropriate sets of positive and negative training sets and auxiliary biological information that characterize potential host–pathogen interactions.
from patient V3 loops of gp120 viral protein sequences to accurately predict co-receptor CCR5/CXCR4 tropism [39]. Furthermore, differential surface electrostatic potentials of a given protein from assorted viruses of the same family can lead to altered binding to host proteins, as recently suggested by crystallization of NS1 from Zika virus and its comparison to existing flaviviral structures [40]. Electrostatic attributes have already been integrated in several computational protein interaction tools, where the analysis of surface energetics and epitope prediction have generalized to binary and higher-order complexes [41].
Molecular-level findings regarding HP-PPIs Several recent works have employed NMR and related techniques to derive HP-PPIs de novo and to complement static crystallographic information with dynamic functional data. For example, solution state NMR determined the interaction between the Chikungunya a-virus nsP3 protein with the SH3 domain of amphiphysin-2, Current Opinion in Structural Biology 2017, 44:119–124
emphasizing the ways surface charge distribution, intrinsic disorder, and host mimicry realize specific binding [42 ]. New insights regarding the interaction between the human protein Cyclophilin A (CypA) and the HIV-1 capsid protein (CA) exemplify how NMR coupled with molecular dynamics simulations (MDs) can dramatically enrich a HP-PPI with a pre-existing structure. For example, Lu et al. [43] used magic-angle spinning NMR to examine the CypA-binding loop in wild type and escape mutant CAs, demonstrating that solely decreasing the loop’s mobility correlated with both escape mutations and wild type CypA binding. A combination of cryo-electron microscopy, MDs, and solid-state NMR illustrated how individual CypA proteins bridge different CA hexamers through a novel interface to stabilize the HIV-1 capsid lattice [44]. These findings coupled with the assessment of the CA lattice’s stability yielded a model in which successful un-coating was facilitated by a low CypA to CA ratio. Additionally, NMR recently demonstrated how the www.sciencedirect.com
Prediction of host–pathogen interactions Mariano and Wuchty 123
transmembrane domain of the HIV-1 envelope spike protein stabilized its interaction in a lipid bilayer-like environment with important implications for vaccine development [45].
Conclusions Here, we have reviewed how diverse levels of protein structural information work in concert to produce putative HP-PPIs across a variety of pathogenic species and their human host (Figure 1). We refer the reader to other recent reviews regarding experimental and computational HPPPI prediction for increased perception of the field [5,22,23,29]. In general, computational workflows mine previously published sequence and structural information to generate candidate sets of HP-PPIs that are further vetted using auxiliary biological information, either in silico or in vitro. While current computational tools suffer from the dearth of full HP-PPI annotation, imputation has made it possible to address interaction questions and achieve high fidelity results in novel systems where structural and functional data are sparse. Finally, studies revealing new molecular insight are continuously forthcoming, facilitating detailed mechanistic studies of HP-PPIs.
References and recommended reading Papers of particular interest, published within the period of review, have been highlighted as: of special interest of outstanding interest 1.
Lee SA, Chan CH, Tsai CH, Lai JM, Wang FS, Kao CY, Huang CY: Ortholog-based protein–protein interaction prediction and its application to inter-species interactions. BMC Bioinform 2008, 9(Suppl. 12):S11.
2.
Krishnadev O, Srinivasan N: A data integration approach to predict host–pathogen protein–protein interactions: application to recognize protein interactions between human and a malarial parasite. In Silico Biol 2008, 8:235-250.
3.
Pavithra SR, Kumar R, Tatu U: Systems analysis of chaperone networks in the malarial parasite Plasmodium falciparum. PLoS Comput Biol 2007, 3:1701-1715.
4.
Tyagi N, Krishnadev O, Srinivasan N: Prediction of protein– protein interactions between Helicobacter pylori and a human host. Mol Biosyst 2009, 5:1630-1635.
5.
Nourani E, Khunjush F, Durmus S: Computational approaches for prediction of pathogen–host protein–protein interactions. Front Microbiol 2015, 6:94.
6.
Franzosa EA, Xia Y: Structural principles within the human– virus protein–protein interaction network. Proc Natl Acad Sci U S A 2011, 108:10538-10543. Based on 3-D structural information this paper shows that exogenous interfaces (HP-PPIs) evolve faster, mimic and compete with endogenous interfaces. 7.
Shen J, Zhang J, Luo X, Zhu W, Yu K, Chen K, Li Y, Jiang H: Predicting protein–protein interactions based only on sequences information. Proc Natl Acad Sci U S A 2007, 104:4337-4341.
8.
Wuchty S: Computational prediction of host–parasite protein interactions between P. falciparum and H. sapiens. PLoS One 2011, 6:e26960.
www.sciencedirect.com
9.
Yu J, Guo M, Needham CJ, Huang Y, Cai L, Westhead DR: Simple sequence-based kernels do not predict protein– protein interactions. Bioinformatics 2010, 26:2610-2614.
10. Eid FE, ElHefnawi M, Heath LS: DeNovo: virus–host sequencebased protein–protein interaction prediction. Bioinformatics 2016, 32:1144-1150. 11. Pawson T, Nash P: Assembly of cell regulatory systems through protein interaction domains. Science 2003, 300:445-452. 12. Hleap J, Blouin C: The semantics of the modular architecture of protein structures. Curr Prot Pept Sci 2016, 17:62-71. 13. Barman RK, Saha S, Das S: Prediction of interactions between viral and host proteins using supervised machine learning methods. PLoS One 2014, 9:e112034. 14. Qi Y, Tastan O, Carbonell JG, Klein-Seetharaman J, Weston J: Semi-supervised multi-task learning for predicting interactions between HIV-1 and human proteins. Bioinformatics 2010, 26:i645-652. 15. Tastan O, Qi Y, Carbonell JG, Klein-Seetharaman J: Prediction of interactions between HIV-1 and human proteins by information integration. Pac Symp Biocomput 2009, 51:6-527. 16. Dyer MD, Murali TM, Sobral BW: Supervised learning and prediction of physical interactions between human and HIV proteins. Infect Genet Evol 2011, 11:917-923. One of the first papers that predicted interactions between proteins of P. falciparum and the human host by modeling domain interactions in a Bayesian framework. 17. Schleker S, Garcia-Garcia J, Klein-Seetharaman J, Oliva B: Prediction and comparison of Salmonella–human and Salmonella–Arabidopsis interactomes. Chem Biodivers 2012, 9:991-1018. 18. Itzhaki Z: Domain–domain interactions underlying herpesvirus–human protein–protein interaction networks. PLoS One 2011, 6:e21724. 19. Liu X, Huang Y, Liang J, Zhang S, Li Y, Wang J, Shen Y, Xu Z, Zhao Y: Computational prediction of protein interactions related to the invasion of erythrocytes by malarial parasites. BMC Bioinform 2014, 15:393. 20. Evans P, Dampier W, Ungar L, Tozeren A: Prediction of HIV-1 virus-host protein interactions using virus and host sequence motifs. BMC Med Genom 2009, 2:27. 21. Segura-Cabrera A, Garcia-Perez CA, Guo X, Rodriguez-Perez MA: A viral–human interactome based on structural motif–domain interactions captures the human infectome. PLoS One 2013, 8: e71526. 22. Durmus S, Cakir T, Ozgur A, Guthke R: A review on computational systems biology of pathogen–host interactions. Front Microbiol 2015, 6:235. 23. Edwards RA, McNair K, Faust K, Raes J, Dutilh BE: Computational approaches to predict bacteriophage–host relationships. FEMS Microbiol Rev 2016, 40:258-272. 24. Arnold R, Boonen K, Sun MG, Kim PM: Computational analysis of interactomes: current and future perspectives for bioinformatics approaches to model the host–pathogen interaction space. Methods 2012, 57:508-518. 25. Coelho ED, Arrais JP, Matos S, Pereira C, Rosa N, Correia MJ, Barros M, Oliveira JL: Computational prediction of the human– microbial oral interactome. BMC Syst Biol 2014, 8:24. 26. Kshirsagar M, Carbonell J, Klein-Seetharaman J: Techniques to cope with missing data in host–pathogen protein interaction prediction. Bioinformatics 2012, 28:i466-i472. 27. Kshirsagar M, Schleker S, Carbonell J, Klein-Seetharaman J: Techniques for transferring host–pathogen protein interactions knowledge to new tasks. Front Microbiol 2015, 6:36. This work applied homology and statistical methods to ‘transfer’ interaction information from other host–pathogen systems to predict interactions between A. thaliana and S. typhimurium. Current Opinion in Structural Biology 2017, 44:119–124
124 Sequences and topology
28. Tyagi M, Hashimoto K, Shoemaker BA, Wuchty S, Panchenko AR: Large-scale mapping of human protein interactome using structural complexes. EMBO Rep 2012, 13:266-271.
38. Peng HP, Lee KH, Jian JW, Yang AS: Origins of specificity and affinity in antibody–protein interactions. Proc Natl Acad Sci U S A 2014, 111:E2656-2665.
29. Keskin O, Tuncbag N, Gursoy A: Predicting protein–protein interactions from the molecular to the proteome level. Chem Rev 2016, 116:4884-4909.
39. Heider D, Dybowski JN, Wilms C, Hoffmann D: A simple structure-based model for the prediction of HIV-1 co-receptor tropism. BioData Min 2014, 7:14.
30. Davis FP, Barkan DT, Eswar N, McKerrow JH, Sali A: Host pathogen protein interactions predicted by comparative modeling. Protein Sci 2007, 16:2585-2596.
40. Song H, Qi J, Haywood J, Shi Y, Gao GF: Zika virus NS1 structure reveals diversity of electrostatic surfaces among flaviviruses. Nat Struct Mol Biol 2016, 23:456-458.
31. de Chassey B, Meyniel-Schicklin L, Aublin-Gex A, Navratil V, Chantier T, Andre P, Lotteau V: Structure homology and interaction redundancy for discovering virus–host protein interactions. EMBO Rep 2013, 14:938-944. Based on structural information interactions between proteins of the Influenza virus and the human host were predicted and confirmed by using siRNA gene silencing screens by measuring the effect of interactor knockdown on viral replication.
41. Peri C, Morra G, Colombo G: Surface energetics and protein– protein interactions: analysis and mechanistic implications. Sci Rep 2016, 6:24035. This study experimentally validates their computational predictions of hotspot residues using alanine scanning. Analysis provides poignant evidence that aromatic sidechains (especially Tyrosine) contacting the epitope backbone complemented by short hydrophilic sequences drive diverse antibody–antigen bonding.
32. Garamszegi S, Franzosa EA, Xia Y: Signatures of pleiotropy: economy and convergent evolution in a domain-resolved map of human-virus protein–protein interaction networks. PLoS Pathog 2013, 9:e1003778.
42. Tossavainen H, Aitio O, Hellman M, Saksela K, Permi P: Structural basis of the high affinity interaction between the alphavirus nonstructural protein-3 (nsP3) and the SH3 domain of amphiphysin-2. J Biol Chem 2016, 291:16307-16317. This work illustrates how dynamic conformational changes can lead to pathogenic proteins having nanomolar affinity for and superior binding to host proteins.
33. Cong Q, Pei J, Grishin NV: Predictive and comparative analysis of Ebola virus proteins. Cell Cycle 2015, 14:2785-2797. 34. Doolittle JM, Gomez SM: Mapping protein interactions between Dengue virus and its human and insect hosts. PLoS Negl Trop Dis 2011, 5:e954. 35. Guven-Maiorov E, Tsai CJ, Nussinov R: Pathogen mimicry of host protein–protein interfaces modulates immunity. Semin Cell Dev Biol 2016, 58:136-145 http://dx.doi.org/10.1016/j. semcdb.2016.06.004. 36. Marin M, Uversky VN, Ott T: Intrinsic disorder in pathogen effectors: protein flexibility as an evolutionary hallmark in a molecular arms race. Plant Cell 2013, 25:3153-3157. 37. Brown CJ, Johnson AK, Dunker AK, Daughdrill GW: Evolution and disorder. Curr Opin Struct Biol 2011, 21:441-446.
Current Opinion in Structural Biology 2017, 44:119–124
43. Lu M, Hou G, Zhang H, Suiter CL, Ahn J, Byeon IJ, Perilla JR, Langmead CJ, Hung I, Gor’kov PL et al.: Dynamic allostery governs cyclophilin A-HIV capsid interplay. Proc Natl Acad Sci U S A 2015, 112:14617-14622. 44. Liu C, Perilla JR, Ning J, Lu M, Hou G, Ramalho R, Himes BA, Zhao G, Bedwell GJ, Byeon IJ et al.: Cyclophilin A stabilizes the HIV-1 capsid through a novel non-canonical binding site. Nat Commun 2016, 7:10714. 45. Dev J, Park D, Fu Q, Chen J, Ha HJ, Ghantous F, Herrmann T, Chang W, Liu Z, Frey G et al.: Structural basis for membrane anchoring of HIV-1 envelope spike. Science 2016, 353:172-1175.
www.sciencedirect.com
ScienceDirect Structure-based prediction of host–pathogen protein interactions Rachelle Mariano1 and Stefan Wuchty2,3,4 The discovery, validation, and characterization of proteinbased interactions from different species are crucial for translational research regarding a variety of pathogens, ranging from the malaria parasite Plasmodium falciparum to HIV-1. Here, we review recent advances in the prediction of host– pathogen protein interfaces using structural information. In particular, we observe that current methods chiefly perform machine learning on sequence and domain information to produce large sets of candidate interactions that are further assessed and pruned to generate final, highly probable sets. Structure-based studies have also emphasized the electrostatic properties and evolutionary transformations of pathogenic interfaces, supplying crucial insight into antigenic determinants and the ways pathogens compete for host protein binding. Advancements in spectroscopic and crystallographic methods complement the aforementioned techniques, permitting the rigorous study of true positives at a molecular level. Together, these approaches illustrate how protein structure on a variety of levels functions coordinately and dynamically to achieve host takeover. Addresses 1 Brigham & Women’s Hospital, Harvard Medical School, Harvard University, Cambridge, MA, United States 2 Dept. of Computer Science, Univ. of Miami, Coral Gables, FL, United States 3 Center for Computational Science, Univ. of Miami, Coral Gables, FL, United States 4 Sylvester Cancer Center, Univ. of Miami, Miami, FL, United States Corresponding author: Mariano, Rachelle ([email protected])
Current Opinion in Structural Biology 2017, 44:119–124 This review comes from a themed issue on Sequences and topology Edited by Ramanathan Sowdhamini and Kenji Mizuguchi
http://dx.doi.org/10.1016/j.sbi.2017.02.007 0959-440X/ã 2017 Elsevier Ltd. All rights reserved.
The accurate determination of protein–protein interactions between bacterial, viral, and parasitic pathogens and their human hosts harbors great medicinal potential, as these discoveries could be used to target specific diseaserelated interfaces with minimal disruption of the underlying human interaction network. Indeed, elucidating www.sciencedirect.com
host–pathogen protein–protein interactions (HP-PPIs) for therapeutics drives their intensive computational and experimental study, and rapidly improving approaches have generated valuable high-fidelity HP-PPI candidates. Benchside high-throughput methods often bear a considerable proportion of false positives when applied to HP-PPI prediction. Moreover, exogenous expression of pathogenic proteins remains difficult, and most results must be translated across evolutionarily distant species. Computational inference of HP-PPIs can identify small subsets of highly probable interactions for informed experimental follow-up by techniques such as nuclear magnetic resonance microscopy (NMR) and crystallography. Combined, these methods allow researchers to not only ascertain how a pathogenic protein interacts with its host on a molecular scale, but also how such interactions function in a larger cellular network.
Computational HP-PPI prediction based on sequence and domain information: homology-based approaches While first limited to intraspecies interactions, sequence similarity-based approaches have since extended to interspecies PPI prediction. Since high primary sequence similarity implies an interaction – an interolog – these methods map known interaction interface sequences onto homologous or orthologous pairs of sequences in different organisms. For example, such sequence comparisons yielded HP-PPIs between Plasmodium falciparum (P. falciparum) [1–3] and Helicobacter pylori (H. pylori) [4] and their human host. Interolog screens benefit from their straightforward execution as well as abundant protein sequence information from which to mine data [5]. As the interologous proteins should demonstrate at least 80% sequence similarity, the ability to correctly determine HP-PPIs from interologs rapidly decreases with evolutionary distance. Additionally, pathogens are locked in a biological ‘arms race’ with their hosts, and their proteins may experience rapid changes in sequence that affect the fidelity of interolog screens [6]. Interolog screens also have a penchant to generate a huge amount of false positive hits. Therefore, further computational investigation of potential hits involves filtering based on the cellular localization, biological functions, and expression profiles of putative HP-PPIs to significantly improve the quality of potential HP-PPI candidates [1–4]. Machine learning approaches using derived sequencebased features have also procured possible HP-PPIs. Shen et al. [7] represented sequences of interacting Current Opinion in Structural Biology 2017, 44:119–124
120 Sequences and topology
proteins as numerical profiles of the occurrence of aminoacid triplets. To predict interactions between human proteins, they utilized support vector machine algorithms (SVM) that were trained by carefully picked positive and negative training sets of protein interactions. This approach was similarly applied to interactions between human proteins and the malaria parasite P. falciparum [8]. Such a representation reduces the size of the feature space but may impair the quality of results. All approaches that predict HP-PPIs via supervised machine learningwhether with sequence or higher order information- need appropriate positive and negative training sets to robustly classify interacting proteins. Yu et al. [9] showed that the choice of non-interactions in the training data greatly impacts the accurate identification of interacting vs. non-interacting pairs. Tools balancing negative example selection have been recently developed to combat this issue. In particular, Eid et al. [10] developed a dissimilarity-random-sampling algorithm for the determination of unlikely occurring interactions between human host and pathogen proteins. The authors sampled highly dissimilar protein sequences from other viruses compared to interacting proteins of the virus in question to generate a negative training set. These training sets trained a SVM with prediction accuracies up to 86% [10], suggesting that the skilled choice of negative training sets drives the reliability of predicted HP-PPIs.
their human host [4]. DDIs have also been combined with protein sequence k-mers and topological properties of host proteins in a human protein interaction network to predict host–pathogen interactions using a SVM [16]. Notably, the use of DDIs allowed prioritization of proteins with extracellular or trans-membrane domains to assess interactions driving invasion and intracellular signaling [4,16]. Domain-based prediction also assisted in the identification of common functional features that allow pathogens to interact with more than one host [17].
Domain-based approaches
Motif and integration-based approaches
Computational inference of HP-PPIs often combines primary sequence similarity with higher order structural information from motifs and domains to increase prediction accuracy [11]. A protein domain is usually defined as a conserved part of a protein’s sequence and three-dimensional structure that mediates the protein’s biological functions while folding and evolving independently [12]. Since domain–domain interactions (DDIs) are largely considered to drive PPIs, numerous studies have used known intra-species DDIs as a basis for the prediction of HP-PPIs. Furthermore, unbiased approaches to elucidate significant predictive features of HP-PPIs have repeatedly emphasized their role [9]. In particular, out of 44 descriptors involving amino acid frequencies of host and pathogen sequences, protein–domain associations appeared to have the highest predictive effect when used with SVM and random forest (RF) algorithms [13]. In a different study, DDIs were included in an 18-dimensional vector and combined with topological sequence and functional characteristics to predict interactions between proteins of HIV-1 using different variations of neural network methods [14] that outperformed RFs [15].
Motif–domain and motif–motif interactions have also gained traction as foundations for HP-PPI prediction, as short linear motifs have proved vital for host–pathogen protein binding. Evans et al. annotated short eukaryotic linear motifs (ELMs) in HIV-1 proteins and used human counter domains that interact with these ELMs to generate an HIV-1 and human interactome [20]. SeguraCabrera et al. [21] combined motif information from 3D interaction databases with stringent filters to create an infectome representing HP-PPIs of 5 viruses with the human host, integrating surface accessibility and structural information. Although these studies predicted HIV1 HP-PPIs with similar techniques, their results differed as a consequence of discrepancies in filtering and motif definition. Although stringent filters were applied to secure biologically meaningful results, both studies provided a plethora of interactions [21]. This is a standing issue in computational HP-PPI prediction, as results depend on preferred methods and suffer from persistence of false positive hits.
Combining domain-based data with primary sequence homology searches of interacting domains allowed the large-scale detection of hypothetical interactions between proteins of H. pylori, HIV-1, and Salmonella with Current Opinion in Structural Biology 2017, 44:119–124
DDIs can also be employed separately from primary sequences to derive HP-PPIs. Itzhaki et al. integrated sets of protein interactions from various organisms with verified protein–domain profiles, assuming that intraspecies DDIs similarly connect in HP-PPIs [18]. The authors modeled the probability that proteins with certain domains interact in a Bayesian framework and generated a protein interaction network between P. falciparum and the human host. Remarkably, interactions thus obtained featured significant co-expression of involved pathogen and human proteins, illustrating the high probability of interaction that DDI-based HP-PPIs can achieve. Liu et al. [19] used an expectation maximization algorithm to find expression-correlated interactions between P. falciparum and human erythrocytes that were subsequently verified using expression data.
Although primary and secondary sequence integration enhances computational HP-PPI predictions, auxiliary data are increasingly used to curb the impact of false positives. In particular, current approaches are assimilating domain, sequence, and ELM data with gene ontology (GO) features, graph topological properties, and gene co-expression data to train HP-PPI classifiers, instead www.sciencedirect.com
Prediction of host–pathogen interactions Mariano and Wuchty 121
of using such features as post hoc filters and verification [5,22–24]. For example, Tastan et al. [15] used such information to deduce interactions between HIV-1 and the human host, thereby finding an interaction between the HIV-1-specific protein tat and the human vitamin D receptor that has been experimentally validated. Coelho et al. predicted the human-microbial oral interactome [25] by incorporating DDIs, protein sequence features, and GO annotations. While such integrative approaches appear promising, the massive dearth of fully annotated host–pathogen interactions has prompted researchers to also develop tools to transfer data from other host–pathogen systems [26,27]. In particular, Ksirsagar et al. [27] built models to infer interactions between proteins of the bacterial species Salmonella typhimurium and the plant host Arabidopsis thaliana for which no experimentally determined PPIs exist. To cope with the massive absence of data, the authors inferred PPIs from known interactions between human and Salmonella through homology searches and graph-based alignments, as well as from the utilization of various kernel methods to predict potential PPIs from many small source data sets that were obtained from other host–pathogen systems.
PPI prediction based on 3-D structural information Structure-structure interaction (SSI) methods complement existing high-confidence interactions by increasing the functional coherence and pathway coverage of the underlying interaction network [28]. Although a number of approaches predict PPIs based on the 3-D structure of the participating proteins [29], such techniques have found relatively little attention for HP-PPI prediction mainly as a result of a relative lack of 3-D structures and protein interactions. Still, a small number of studies have employed structural similarities to detect HP-PPIs. Davis et al. presented the first instance of using 3-D structural homology for interspecies PPI-prediction [30]. Their protocol first scanned the host and pathogen genomes for proteins with similarity to known protein complexes, assessed these putative interactions using structural information, and filtered the remaining interactions based on biological context for several human pathogens. Doolittle and Gomez used the structural similarity of 9 HIV-1 proteins to human proteins with known interactions, functional data from RNAi studies, and GO cellular component annotation to generate 502 interaction predictions between HIV and human proteins [33]. Mycobacterium tuberculosis (TB)-human interactions elucidated through SSI revealed putative host protein targets with known roles in HIV replication, thus supplying mechanistic insight into the intensified disease phenotype that occurs during TB-HIV co-infection [34]. de Chassey et al. [31] used siRNA gene silencing to confirm human and influenza NS1 protein interactions derived from SSI www.sciencedirect.com
methods by measuring the effect of interactor knockdown on viral replication. They concluded that 10/26 of their final candidates directly mediated this process [31].
Structure and evolution While structure-driven approaches appear as powerful predictors of host–pathogen interaction, data suggest that pathogens evolve protein interfaces to achieve binding without sequence or structural similarity to native interactors [6,32]. Franzosa and Xia [6] showed that exogenous interfaces (HP-PPIs) overlap with, mimic, and compete with endogenous (host–host) interfaces. The endogenous interfaces mimicked by viral proteins tend to participate in multiple interactions that are transient and regulatory in nature. While endogenous interfaces evolve more slowly than the rest of the protein surface, exogenous interfaces—including many sites of endogenous-exogenous overlap—tend to evolve faster, consistent with the aforementioned evolutionary ‘arms race’ [6]. Furthermore, structural comparative modeling of human pathogenic and non-pathogenic Ebola virus revealed that pathogenic residues did not localize to known interaction interfaces and instead clustered to distal regions [33]. Investigating molecular mimicry, Doolittle and Gomez predicted interactions between the human host and the Dengue fever virus, assuming that pathogen proteins can replace structurally similar host proteins in intra-species host protein interactions [34]. Apart from featuring rapidly evolving sequences, secreted effector proteins of pathogens also harbor long intrinsically disordered regions that may assist in functional hijacking through mimicry of host proteins and immune escape through dynamic conformational changes. While an increasingly important structural characteristic of pathogen proteins, we refer the reader to [35–37] that extensively review such topics.
Surface electrostatics and epitope prediction Electrostatic features of interacting protein surfaces play a crucial role in finding epitopes, which are the antigenic determinants of pathogens that are recognized by the immune system. Antibodies recognize a variety of foreign proteins through a relatively limited and homogeneous reservoir of sequences and structures. In a structure-based computational analysis, Peng, Jian and Yang [38] found that antigen-binding sites of antibodies show an enrichment of aromatic residues surrounded by short hydrophilic sequences that permits promiscuous non-covalent interactions through contact with backbone and sidechain carbons. Such features allow a sparse number of host antigens to recognize numerous pathogen antigens through shared physicochemical properties. These electrostatic characteristics also play a role in the binding of gp120, an HIV-1 protein, to either co-receptor CCR5 or CXCR4 of human CD4 cells in a choice associated with differential rates of AIDS progression. Heider et al. used hydrophobicity and electrostatic potential information Current Opinion in Structural Biology 2017, 44:119–124
122 Sequences and topology
Figure 1
organism
training set A
B E
sequence
similarity
F
C
G
D
host
pathogen
interologs
host pathogen
SVM, naïve Bayes, random forests,…
pathogen p host h
domain-domain interactions
functions
3-D interactions
literature
host pathogen
co-expression o-express
auxiliary data
…
Current Opinion in Structural Biology
Common features of computational workflows for the prediction of interactions between proteins of pathogens and hosts. In a first step, sequence and structure based information is gathered to generate sets of candidate host–pathogen interactions. Subsequently, machine learning approaches are used to predict high confidence interactions by combining appropriate sets of positive and negative training sets and auxiliary biological information that characterize potential host–pathogen interactions.
from patient V3 loops of gp120 viral protein sequences to accurately predict co-receptor CCR5/CXCR4 tropism [39]. Furthermore, differential surface electrostatic potentials of a given protein from assorted viruses of the same family can lead to altered binding to host proteins, as recently suggested by crystallization of NS1 from Zika virus and its comparison to existing flaviviral structures [40]. Electrostatic attributes have already been integrated in several computational protein interaction tools, where the analysis of surface energetics and epitope prediction have generalized to binary and higher-order complexes [41].
Molecular-level findings regarding HP-PPIs Several recent works have employed NMR and related techniques to derive HP-PPIs de novo and to complement static crystallographic information with dynamic functional data. For example, solution state NMR determined the interaction between the Chikungunya a-virus nsP3 protein with the SH3 domain of amphiphysin-2, Current Opinion in Structural Biology 2017, 44:119–124
emphasizing the ways surface charge distribution, intrinsic disorder, and host mimicry realize specific binding [42 ]. New insights regarding the interaction between the human protein Cyclophilin A (CypA) and the HIV-1 capsid protein (CA) exemplify how NMR coupled with molecular dynamics simulations (MDs) can dramatically enrich a HP-PPI with a pre-existing structure. For example, Lu et al. [43] used magic-angle spinning NMR to examine the CypA-binding loop in wild type and escape mutant CAs, demonstrating that solely decreasing the loop’s mobility correlated with both escape mutations and wild type CypA binding. A combination of cryo-electron microscopy, MDs, and solid-state NMR illustrated how individual CypA proteins bridge different CA hexamers through a novel interface to stabilize the HIV-1 capsid lattice [44]. These findings coupled with the assessment of the CA lattice’s stability yielded a model in which successful un-coating was facilitated by a low CypA to CA ratio. Additionally, NMR recently demonstrated how the www.sciencedirect.com
Prediction of host–pathogen interactions Mariano and Wuchty 123
transmembrane domain of the HIV-1 envelope spike protein stabilized its interaction in a lipid bilayer-like environment with important implications for vaccine development [45].
Conclusions Here, we have reviewed how diverse levels of protein structural information work in concert to produce putative HP-PPIs across a variety of pathogenic species and their human host (Figure 1). We refer the reader to other recent reviews regarding experimental and computational HPPPI prediction for increased perception of the field [5,22,23,29]. In general, computational workflows mine previously published sequence and structural information to generate candidate sets of HP-PPIs that are further vetted using auxiliary biological information, either in silico or in vitro. While current computational tools suffer from the dearth of full HP-PPI annotation, imputation has made it possible to address interaction questions and achieve high fidelity results in novel systems where structural and functional data are sparse. Finally, studies revealing new molecular insight are continuously forthcoming, facilitating detailed mechanistic studies of HP-PPIs.
References and recommended reading Papers of particular interest, published within the period of review, have been highlighted as: of special interest of outstanding interest 1.
Lee SA, Chan CH, Tsai CH, Lai JM, Wang FS, Kao CY, Huang CY: Ortholog-based protein–protein interaction prediction and its application to inter-species interactions. BMC Bioinform 2008, 9(Suppl. 12):S11.
2.
Krishnadev O, Srinivasan N: A data integration approach to predict host–pathogen protein–protein interactions: application to recognize protein interactions between human and a malarial parasite. In Silico Biol 2008, 8:235-250.
3.
Pavithra SR, Kumar R, Tatu U: Systems analysis of chaperone networks in the malarial parasite Plasmodium falciparum. PLoS Comput Biol 2007, 3:1701-1715.
4.
Tyagi N, Krishnadev O, Srinivasan N: Prediction of protein– protein interactions between Helicobacter pylori and a human host. Mol Biosyst 2009, 5:1630-1635.
5.
Nourani E, Khunjush F, Durmus S: Computational approaches for prediction of pathogen–host protein–protein interactions. Front Microbiol 2015, 6:94.
6.
Franzosa EA, Xia Y: Structural principles within the human– virus protein–protein interaction network. Proc Natl Acad Sci U S A 2011, 108:10538-10543. Based on 3-D structural information this paper shows that exogenous interfaces (HP-PPIs) evolve faster, mimic and compete with endogenous interfaces. 7.
Shen J, Zhang J, Luo X, Zhu W, Yu K, Chen K, Li Y, Jiang H: Predicting protein–protein interactions based only on sequences information. Proc Natl Acad Sci U S A 2007, 104:4337-4341.
8.
Wuchty S: Computational prediction of host–parasite protein interactions between P. falciparum and H. sapiens. PLoS One 2011, 6:e26960.
www.sciencedirect.com
9.
Yu J, Guo M, Needham CJ, Huang Y, Cai L, Westhead DR: Simple sequence-based kernels do not predict protein– protein interactions. Bioinformatics 2010, 26:2610-2614.
10. Eid FE, ElHefnawi M, Heath LS: DeNovo: virus–host sequencebased protein–protein interaction prediction. Bioinformatics 2016, 32:1144-1150. 11. Pawson T, Nash P: Assembly of cell regulatory systems through protein interaction domains. Science 2003, 300:445-452. 12. Hleap J, Blouin C: The semantics of the modular architecture of protein structures. Curr Prot Pept Sci 2016, 17:62-71. 13. Barman RK, Saha S, Das S: Prediction of interactions between viral and host proteins using supervised machine learning methods. PLoS One 2014, 9:e112034. 14. Qi Y, Tastan O, Carbonell JG, Klein-Seetharaman J, Weston J: Semi-supervised multi-task learning for predicting interactions between HIV-1 and human proteins. Bioinformatics 2010, 26:i645-652. 15. Tastan O, Qi Y, Carbonell JG, Klein-Seetharaman J: Prediction of interactions between HIV-1 and human proteins by information integration. Pac Symp Biocomput 2009, 51:6-527. 16. Dyer MD, Murali TM, Sobral BW: Supervised learning and prediction of physical interactions between human and HIV proteins. Infect Genet Evol 2011, 11:917-923. One of the first papers that predicted interactions between proteins of P. falciparum and the human host by modeling domain interactions in a Bayesian framework. 17. Schleker S, Garcia-Garcia J, Klein-Seetharaman J, Oliva B: Prediction and comparison of Salmonella–human and Salmonella–Arabidopsis interactomes. Chem Biodivers 2012, 9:991-1018. 18. Itzhaki Z: Domain–domain interactions underlying herpesvirus–human protein–protein interaction networks. PLoS One 2011, 6:e21724. 19. Liu X, Huang Y, Liang J, Zhang S, Li Y, Wang J, Shen Y, Xu Z, Zhao Y: Computational prediction of protein interactions related to the invasion of erythrocytes by malarial parasites. BMC Bioinform 2014, 15:393. 20. Evans P, Dampier W, Ungar L, Tozeren A: Prediction of HIV-1 virus-host protein interactions using virus and host sequence motifs. BMC Med Genom 2009, 2:27. 21. Segura-Cabrera A, Garcia-Perez CA, Guo X, Rodriguez-Perez MA: A viral–human interactome based on structural motif–domain interactions captures the human infectome. PLoS One 2013, 8: e71526. 22. Durmus S, Cakir T, Ozgur A, Guthke R: A review on computational systems biology of pathogen–host interactions. Front Microbiol 2015, 6:235. 23. Edwards RA, McNair K, Faust K, Raes J, Dutilh BE: Computational approaches to predict bacteriophage–host relationships. FEMS Microbiol Rev 2016, 40:258-272. 24. Arnold R, Boonen K, Sun MG, Kim PM: Computational analysis of interactomes: current and future perspectives for bioinformatics approaches to model the host–pathogen interaction space. Methods 2012, 57:508-518. 25. Coelho ED, Arrais JP, Matos S, Pereira C, Rosa N, Correia MJ, Barros M, Oliveira JL: Computational prediction of the human– microbial oral interactome. BMC Syst Biol 2014, 8:24. 26. Kshirsagar M, Carbonell J, Klein-Seetharaman J: Techniques to cope with missing data in host–pathogen protein interaction prediction. Bioinformatics 2012, 28:i466-i472. 27. Kshirsagar M, Schleker S, Carbonell J, Klein-Seetharaman J: Techniques for transferring host–pathogen protein interactions knowledge to new tasks. Front Microbiol 2015, 6:36. This work applied homology and statistical methods to ‘transfer’ interaction information from other host–pathogen systems to predict interactions between A. thaliana and S. typhimurium. Current Opinion in Structural Biology 2017, 44:119–124
124 Sequences and topology
28. Tyagi M, Hashimoto K, Shoemaker BA, Wuchty S, Panchenko AR: Large-scale mapping of human protein interactome using structural complexes. EMBO Rep 2012, 13:266-271.
38. Peng HP, Lee KH, Jian JW, Yang AS: Origins of specificity and affinity in antibody–protein interactions. Proc Natl Acad Sci U S A 2014, 111:E2656-2665.
29. Keskin O, Tuncbag N, Gursoy A: Predicting protein–protein interactions from the molecular to the proteome level. Chem Rev 2016, 116:4884-4909.
39. Heider D, Dybowski JN, Wilms C, Hoffmann D: A simple structure-based model for the prediction of HIV-1 co-receptor tropism. BioData Min 2014, 7:14.
30. Davis FP, Barkan DT, Eswar N, McKerrow JH, Sali A: Host pathogen protein interactions predicted by comparative modeling. Protein Sci 2007, 16:2585-2596.
40. Song H, Qi J, Haywood J, Shi Y, Gao GF: Zika virus NS1 structure reveals diversity of electrostatic surfaces among flaviviruses. Nat Struct Mol Biol 2016, 23:456-458.
31. de Chassey B, Meyniel-Schicklin L, Aublin-Gex A, Navratil V, Chantier T, Andre P, Lotteau V: Structure homology and interaction redundancy for discovering virus–host protein interactions. EMBO Rep 2013, 14:938-944. Based on structural information interactions between proteins of the Influenza virus and the human host were predicted and confirmed by using siRNA gene silencing screens by measuring the effect of interactor knockdown on viral replication.
41. Peri C, Morra G, Colombo G: Surface energetics and protein– protein interactions: analysis and mechanistic implications. Sci Rep 2016, 6:24035. This study experimentally validates their computational predictions of hotspot residues using alanine scanning. Analysis provides poignant evidence that aromatic sidechains (especially Tyrosine) contacting the epitope backbone complemented by short hydrophilic sequences drive diverse antibody–antigen bonding.
32. Garamszegi S, Franzosa EA, Xia Y: Signatures of pleiotropy: economy and convergent evolution in a domain-resolved map of human-virus protein–protein interaction networks. PLoS Pathog 2013, 9:e1003778.
42. Tossavainen H, Aitio O, Hellman M, Saksela K, Permi P: Structural basis of the high affinity interaction between the alphavirus nonstructural protein-3 (nsP3) and the SH3 domain of amphiphysin-2. J Biol Chem 2016, 291:16307-16317. This work illustrates how dynamic conformational changes can lead to pathogenic proteins having nanomolar affinity for and superior binding to host proteins.
33. Cong Q, Pei J, Grishin NV: Predictive and comparative analysis of Ebola virus proteins. Cell Cycle 2015, 14:2785-2797. 34. Doolittle JM, Gomez SM: Mapping protein interactions between Dengue virus and its human and insect hosts. PLoS Negl Trop Dis 2011, 5:e954. 35. Guven-Maiorov E, Tsai CJ, Nussinov R: Pathogen mimicry of host protein–protein interfaces modulates immunity. Semin Cell Dev Biol 2016, 58:136-145 http://dx.doi.org/10.1016/j. semcdb.2016.06.004. 36. Marin M, Uversky VN, Ott T: Intrinsic disorder in pathogen effectors: protein flexibility as an evolutionary hallmark in a molecular arms race. Plant Cell 2013, 25:3153-3157. 37. Brown CJ, Johnson AK, Dunker AK, Daughdrill GW: Evolution and disorder. Curr Opin Struct Biol 2011, 21:441-446.
Current Opinion in Structural Biology 2017, 44:119–124
43. Lu M, Hou G, Zhang H, Suiter CL, Ahn J, Byeon IJ, Perilla JR, Langmead CJ, Hung I, Gor’kov PL et al.: Dynamic allostery governs cyclophilin A-HIV capsid interplay. Proc Natl Acad Sci U S A 2015, 112:14617-14622. 44. Liu C, Perilla JR, Ning J, Lu M, Hou G, Ramalho R, Himes BA, Zhao G, Bedwell GJ, Byeon IJ et al.: Cyclophilin A stabilizes the HIV-1 capsid through a novel non-canonical binding site. Nat Commun 2016, 7:10714. 45. Dev J, Park D, Fu Q, Chen J, Ha HJ, Ghantous F, Herrmann T, Chang W, Liu Z, Frey G et al.: Structural basis for membrane anchoring of HIV-1 envelope spike. Science 2016, 353:172-1175.
www.sciencedirect.com