Prediction of protein–protein interactions between fungus (Magnaporthe grisea) and rice (Oryza sativa L.)

Prediction of protein–protein interactions between fungus (Magnaporthe grisea) and rice (Oryza... Abstract Rice blast disease caused by the fungus Magnaporthe grisea (M. grisea) is one of the most serious diseases for the cultivated rice Oryza sativa (O. sativa). A key factor causing rice blast disease and defense might be protein–protein interactions (PPIs) between rice and fungus. In this research, we have developed a computational pipeline to predict PPIs between blast fungus and rice. After cross-prediction by interolog-based and domain-based method, we achieved 532 potential PPIs between 27 fungus proteins and 236 rice proteins. Accuracy of jackknife test, 10-fold cross-validation test and independent test for these PPIs were 90.43, 93.85 and 84.67%, respectively, by using support vector machine classification method. Meanwhile, the pathogenic genes of blast fungus were enriched in the predicted PPIs network when compared with 1000 random interaction networks. The rice regulatory network was downloaded and divided into 228 subnetworks with over six nodes, and the top seven subnetworks affected by blast fungus through PPIs were investigated. The results indicated that 34 upregulated and 12 downregulated master regulators in rice interacting with the fungus proteins in response to the infection of blast fungus. The common master regulators in rice in response to the infection of M. grisea, Xanthomonas oryzae pv.oryzae and rice stripe virus were analyzed. The ubiquitin proteasome pathway was the common pathway in rice regulated by these three pathogens, while apoptosis signaling pathway was induced by fungus and bacteria. In summary, the results in this article provide insight into the process of blast fungus infection. rice (Oryza sativa L.), blast fungus (M. grisea), interspecies interaction, master regulators Introduction Rice blast, bacterial blight and stripe, which are caused by Magnaporthe grisea, Xanthomonas oryzae pv.oryzae (Xoo) and rice stripe virus (RSV), respectively, are three of the most widespread and devastating rice diseases, especially rice blast. The rice plant is susceptible to the blast fungus in all of its growth period, with infection affecting leaves, nodes, panicles and roots [1]. Annual yield losses of rice destroyed by blast are sufficient to feed 60 million people [2]. In addition, the blast fungus is also harmful to wheat and other small grains and leads to a large reduction in yield [3]. Strategies such as fungicides, resistant cultivars, agronomical practices and biotechnological methods have been developed to overcome this disastrous disease [4]. However, use of plant host-resistant ability is the most effective and economic way to control the diseases [5]. Thus, it is essential to explore the pathogenicity of blast fungus and the resistant ability of rice. The infection of M. grisea is considered as a hemibiotrophic process, and it must go through the important biotrophic phase in its early stages [6, 7]. There have been minutely cytological interpretations of occurring events during rice blast infection [8–10]. During the lengthy warfare between pathogens and hosts, plants have evolved elaborate defense systems, including pathogen-associated molecular patterns (PAMPs), pathogen-triggered immunity (PTI) and effector-triggered immunity (ETI) system [11]. The molecular mechanisms of rice PTI and ETI to resist the infection of blast fungus have been partially illustrated. The genes underlying rice plant defenses comprise a substantial portion of the host genome through the direct interactions with pathogen proteins or the initiation of rice plant defense responses to the infection [12]. It is critical to understand the protein–protein interaction (PPI) network (i.e. interactome) between plant and pathogen for studying the molecular basis of pathogenesis [13, 14]. In the past decade, the genomes of rice (Oryza sativa) and blast fungus (M. grisea) have been sequenced and well annotated [1, 15]. This enables the pathosystem between rice and blast fungus to be a model of the study of plant–microbe interactions. Although many biochemical techniques have been developed to verify PPIs, the experimental identification of PPIs between plant and pathogen is still a time-consuming and challenging work [16]. Till now, few pairs of PPIs between rice and blast fungus have been identified experimentally, which is insufficient to elucidate the molecular mechanism of pathogenicity [17–19]. Experimental methods are the ideal tools for validating PPIs, but these should be more efficient for large-scale PPI prediction [20]. With the massive accumulation of genomic data and known PPIs, computational methods to predict PPIs as an alternative to biochemical techniques have been improved with high performance recently. These include predictors based on genomic data [21, 22], protein structure [23], domain information [24, 25], protein sequence [26] and Gene Ontology (GO) annotation semantic similarity [27]. Most of the predictors follow the common principle that text mining is used to extract the information from the known PPIs in the biomedical literatures [28]. The method based on interolog, a PPI predictor based on protein sequence similarity, has been used widely to predict PPIs on different scales. He et al. [13] used the interolog-based method to predict PPIs in blast fungus genome, while Wang et al. [29] constructed the interacting network of heat shock protein 70 s (Hsp70s) in rice genome. Meanwhile, PPI predictors based on protein domain information and support vector machine (SVM) predictors based on protein structural information have been used alone or complementary to sequence similarity predictors to predict intraspecies PPIs [30, 31] and achieve high performance [32]. To the best of our knowledge, computational efforts to predict interspecies PPIs have been hardly reported. Li et al. [30] successfully used the method based on interolog and a domain-base method to construct PPI network between Ralstonia solanacearum and Arabidopsis thaliana. The regulatory network in host plants affected by pathogen infection through the PPIs information could be developed to understand the influence of pathogen proteins to the host [33]. The ClusterViz program in Cytoscape was introduced to analyze the regulatory network [34]. Master regulators, genes at the top of a regulatory hierarchy, usually regulate lots of downstream genes [35]. Lefebvre et al. [36] developed a tool, Viper, to analyze master regulators in the regulatory network. In this research work, PPIs between blast fungus and rice were crosswise predicted using the method based on interolog and the domain-based method. Then, the potential PPIs were further confirmed by using SVM and the enrichment of pathogenic proteins. After that, we analyzed the regulatory network in rice in response to the infection of blast fungus through the predicted PPI network. Materials and methods Data sets A total of 11 054 protein sequences in blast fungus genome were downloaded from the M. grisea database [1] (http://www.broadinstitute.org/annotation/genome/magnaporthe_grisea/MultiHome.html), while 66 338 protein sequences in rice genome were obtained from MSU Rice Genome Annotation Project Database [15] (ftp://ftp.plantbiology.msu.edu/pub/data/Eukaryotic_Projects/o_sativa/annotation_dbs/pseudomolecules/version_7.0/all.dir/). To perform the method based on interolog, 635 008 experimentally verified PPIs were firstly downloaded from five public databases, including Biomolecular Interaction Network Database [37] (BIND), Molecular INTeraction database [38] (MINT), Database of Interacting Proteins [39] (DIP), The Arabidopsis Information Resource [40] (TAIR) and IntAct [41]. Overall, 23 845 host–pathogen PPIs were obtained from Host–Pathogen Interaction Database [42] (HPIDB; http://agbase.msstate.edu/hpi/main.html). The microarray data of rice in response to three biotic stresses, including M. grisea, Xoo and RSV infection, were downloaded from Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/). The GEO Accession for rice under M. grisea, Xoo and RSV infection was GSE7256, GSE19844 and GSE11025, respectively. RiceNet data under the biotic stress, including every nodes and links, were obtained from http://www.functionalnet.org/ricenet. Identification of secreted and membrane proteins The proteins in M. grisea were inferred to be transmembrane when the number of predicted transmembrane helices was >1 by using TMHMM [43]. According to the methods introduced by Kawahara et al. [44], SignalP [45] and WoLFPSORT [46], predictors were used to identify secretory proteins with the default criterion. The proteins containing signal peptide predicted by SignalP and locating extracellular predicted by WoLFPSORT were deduced to be secreted proteins. Prediction of PPIs between blast fungus and rice using the method based on interolog The potential PPIs between blast fungus and rice were predicted using the method based on interolog (Figure 1). Briefly, each protein in M. grisea and O. sativa was first searched against BIND, MINT, DIP, TAIR and IntAct database to identify homologs with 10−5E-value, 30% sequence identify and 40% aligned sequence length coverage cutoff. Proteins in M. grisea and O. sativa were BLASTed against pathogenic proteins and host proteins in HPIDB, respectively, to identify their homologs with the same criterion. If the corresponding homologs of one protein pair from M. grisea and O. sativa had at least one interaction in the above databases, the protein pair is predicted to interact. Figure 1. View largeDownload slide Prediction pipeline of the potential PPIs between blast fungus and rice. Figure 1. View largeDownload slide Prediction pipeline of the potential PPIs between blast fungus and rice. Verification of PPIs between blast fungus and rice using domain-based method The potential PPIs between blast fungus and rice were first verified by using the domain-based method (Figure 1). Each protein in the potential PPIs was searched against the Pfam database (www.pfam.com) with 10−5E-value and 90% sequence identify. When an interacting Pfam domain was found in the interacting protein pair, the potential PPI between blast fungus and rice was further identified. After that, the PPIs predicted by the method based on interolog and verified by the domain-based method were intersected into a PPI network between blast fungus and rice by using Cytoscape [47]. Those PPIs with blast fungus proteins predicted to be non-membrane or nonsecreted ones were removed from the established PPI network. Three different methods based on SVMs and the enrichment of pathogenic proteins were used to confirm the above predicted PPIs in the network. Following Shen et al. [32], the sequence information of every predicted PPI was extracted to assess the prediction effectiveness using 10-fold crossover test, independent test [48] and jackknife test [49]. For 10-fold crossover test, we randomly selected proteins in blast fungus and rice genome to construct a negative data set with the same size of the predicted PPIs. Then, the negative set and the predicted PPIs data were mixed together and divided into 10 sub-data sets. One of these sub-data sets and the other nine sub-data sets were used as test and training data set, respectively, to detect the mean index with 10 repetitions. In the independent test, the size of negative data set was 20 times the size of predicted PPIs. The negative data set and the predicted PPI pairs were also merged together and divided into three subsets, two of which was used as training data set and the other one was used as test data set. Because the jackknife test was regarded as the most objective and effective approach to evaluate the accuracy (ACC) of various predictors [50, 51], the predictive performance was further examined by the jackknife test. Moreover, fungus proteins in the predicted PPIs should match the pathogenic proteins in Phi-base [52]. Simultaneously, we built 1000 random PPI networks containing the above predicted 523 PPIs and calculated the P-value and significance of comparison by the following formula:   P=nnp≥iN (1) where np is the number of M. grisea pathogenic proteins found in every random PPI network, and i is the number of M. grise pathogenic proteins in the predicted interactome; nnp≥i, denotes the number of random PPI networks containing more pathogenic proteins than in the predicted interactome; N is the total number of random PPI networks. Analysis on the PPI network topology Two topological parameters (i.e. degree and betweenness) for each protein in the PPI network were calculated using NetworkAnalyzer [53]. In a PPI network, each protein is represented as a node. The degree of a node is simply defined as the number of interactions that a node has. The betweenness is a centrality measure of a node in a network. The betweenness centrality of a node n can be calculated as the formula.   Cb(n)=∑s≠t≠n(σst(n)/σst) (2) where s and t are nodes in the network that are different from n, σstdenotes the number of shortest paths from s to t and σst( n) is the number of shortest paths from s to t that pass through the node n. Analysis of the impact of fungus proteins on rice regulatory network The regulatory networks in rice in response to biotic stresses were downloaded from http://www.functionalnet.org/ricenet. The ClusterViz program in Cytoscape, which is based on the FAG-EC algorithm [34], was used to cluster rice regulatory networks into different functional modules. The first 10 largest networks named Clusters 1–10 were selected to analyze the impact of pathogenic genes in blast fungus on the functional modules in rice. The 236 rice proteins involved in the above predicted PPIs were first searched against these 10 clusters. If a match is not found then, the downstream proteins of the 236 rice proteins in the predicted PPI networks were searched. After that, GO enrichment of each cluster was determined using the Fisher exact test followed by the false discovery rate (FDR) correction with P < 0.05. Analysis of master regulators in rice in response to blast fungus infection First, the original microarray data were normalized using Robust Multi-Array in oneChannelGUI running by R program and the ID of probe was transformed into gene ID in MSU Rice Genome Annotation Project Database by BioMart. When a gene corresponded to more than one probe, the expression of this gene was represented by a probe which had a largest coefficient of variation (V) calculated as the formula:   V=σE (3) where σ is the SD of the expression of a probe in all samples, and E is the average of the expression of a probe in all samples. The rowTtest in viper was used to analyze the differentially expressed genes (DEGs) in the microarrays under three biotic stresses with P < 0.05 corrected by Bonferroni. DEGs, and RiceNet data were submitted to viper, and msviper was used to search master regulators with P < 0.05 corrected by Bonferroni. The degree of a regulator was set as at least 20, and there would be some false-positive regulators, which were defined as shadow regulators [36]. To remove them, the function of shadow in viper was used with the default criterion. The synergy of master regulators was obtained by using msviperSynergy with the default criterion. Analysis of GO enrichment and metabolic pathway The PANTHER database was used to analyze the GO enrichment of all the subnetworks with P < 0.05 corrected by Bonferroni [54]. PANTHER GO slim was applied in GO annotations, while metabolic pathway of every gene was obtained from PANTHER database directly. Results Prediction of PPIs between M. grisea and O. sativa After predicting PPIs using the method based on interolog and verifying by using domain-based method, we achieved 523 potential PPIs between M. grisea and O. sativa, which is shown in Figure 2 and Supplementary Table S1. The methods based on SVM and enrichment of pathogenic proteins were used to confirm the above PPIs between M. grisea and O. sativa. Compared with the random 1000 PPI networks, the pathogenic proteins were significantly enriched in the predicted PPIs (empirical P-value = 0). Additionally, ACC of jackknife test, 10-fold cross test and independent test based on SVM were 90.43, 93.85 and 84.67%, respectively (Table 1). Table 1. Classification results of PPIs in jackknife test, 10-fold crossover test and independent test Test  ACC (%)  Sp (%)  Sn (%)  AUC  Jackknife test  90.43  87.19  93.68  0.904  10-folds crossover test  93.85  95.67  85.41  0.955  Independent test  84.67  84.57  94.26  0.935  Test  ACC (%)  Sp (%)  Sn (%)  AUC  Jackknife test  90.43  87.19  93.68  0.904  10-folds crossover test  93.85  95.67  85.41  0.955  Independent test  84.67  84.57  94.26  0.935  Sp, Sn and AUC denote specificity, sensitivity and area under curve, respectively. Table 1. Classification results of PPIs in jackknife test, 10-fold crossover test and independent test Test  ACC (%)  Sp (%)  Sn (%)  AUC  Jackknife test  90.43  87.19  93.68  0.904  10-folds crossover test  93.85  95.67  85.41  0.955  Independent test  84.67  84.57  94.26  0.935  Test  ACC (%)  Sp (%)  Sn (%)  AUC  Jackknife test  90.43  87.19  93.68  0.904  10-folds crossover test  93.85  95.67  85.41  0.955  Independent test  84.67  84.57  94.26  0.935  Sp, Sn and AUC denote specificity, sensitivity and area under curve, respectively. Figure 2. View largeDownload slide PPIs between M. grisea and O. sativa. Each node represents a protein, and each edge denotes an interaction. Red and blue nodes are M. grisea and O. sativa proteins, respectively. Figure 2. View largeDownload slide PPIs between M. grisea and O. sativa. Each node represents a protein, and each edge denotes an interaction. Red and blue nodes are M. grisea and O. sativa proteins, respectively. As shown in Figure 2, 27 fungus proteins and 236 rice proteins were involved in the PPIs between M. grisea and O. sativa. This result indicated that one fungus protein averagely had 19 interacting partners from rice, and one rice protein interacted approximately with two pathogen proteins. In total, 137 of 236 O. sativa proteins that interacted with pathogen proteins could be found in the regulatory networks in rice. The average degree and betweenness of proteins in rice regulatory network and the pathogen-targeted proteins in rice regulatory network were 64.02, 1.39E-04 and 164.15, 1.43E-04, respectively. This meant that the potential pathogen-targeted proteins had a higher degree as well as a larger betweenness than other proteins in the network (Wilcoxon rank-sum test, P = 3.431e-14 and 3.158e-14). It was worth to notice here that we focused on the topology of proteins inside rice regulatory network, which were target by blast fungus through PPI, not the PPIs between blast fungus and rice [55]. To determine the functional module of rice proteins interacting with pathogen proteins, GO enrichment of these proteins was analyzed. The results showed that GO terms proteolysis (GO:0006508), vesicle-mediated transport (GO:0016192), exocytosis (GO:0006887), protein transport (GO:0015031), protein kinase activity (GO:0004672) and kinase activity (GO:0016301) were enriched significantly in these 236 rice proteins (Supplementary Table S2). Furthermore, the results in the analysis of GO functional classification implied that these potential pathogen-targeted rice proteins mainly participated in metabolic process, ligand–receptor binding, response to stimulus and function of membrane (Supplementary Figure S1). Modular analysis of regulatory networks in rice in response to biotic stresses By using the FAG-EC algorithm, the regulatory networks in rice in response to biotic stresses were clustered into 228 subnetworks, each of which contained more than six nodes. The subnetworks with maximum and minimum nodes was named Clusters 1 and 228, respectively. We found that Cluster 1 contained 45 potential pathogen-targeted proteins, while Cluster 2, Cluster 3, Cluster 5, Cluster 6, Cluster 7, Cluster 9 and Cluster 10 connect with the predicted PPIs network through one rice protein (Figure 3A). The results also indicated that GO annotations were significantly enriched in the above seven clusters except for Cluster 10 by using the Fisher exact test followed by the FDR correction with P < 0.05 (Supplementary Table S2). Figure 3. View largeDownload slide The sub-regulatory networks in rice affected by blast fungus proteins. (A) Seven main clusters. (B) Rice genes connected to multiple functional subnetworks. Red nodes denote fungus proteins, pink nodes denote fungus interactors and blue nodes denote rice proteins. Figure 3. View largeDownload slide The sub-regulatory networks in rice affected by blast fungus proteins. (A) Seven main clusters. (B) Rice genes connected to multiple functional subnetworks. Red nodes denote fungus proteins, pink nodes denote fungus interactors and blue nodes denote rice proteins. Cluster 1 was the main functional module of rice regulatory networks. There were 3684 genes in the cluster, which was much more than the others. According to the GO enrichment and PANTHER metabolic pathway analysis, we found that 143 GO terms (Supplementary Table S2) and several metabolic pathway were enriched in Cluster 1 (Table 2). The first three enriched pathways were ubiquitin proteasome pathway, apoptosis signaling pathway and general transcription regulation. In addition, those 45 potential pathogen-targeted proteins in Cluster 1 had a higher degree as well as a larger betweenness than other proteins in the network (Wilcoxon rank-sum test, P = 2.156e-10 and 3.221e-10), indicating that it should play an important role in the rice regulatory network. Table 2. PANTHER pathway enrichment analysis in Cluster 1 PANTHER pathway  P-value  Ubiquitin proteasome pathway  2.20E-17  Apoptosis signaling pathway  5.24E-14  General transcription regulation  7.12E-14  Transcription regulation by bZIP transcription factor  1.03E-12  Cell cycle  2.95E-09  Huntington disease  6.06E-09  Inflammation mediated by chemokine and cytokine signaling pathway  1.98E-08  DNA replication  6.98E-08  Tryptophan biosynthesis  1.93E-07  Cytoskeletal regulation by Rho GTPase  2.29E-05  Adenosine triphosphate synthesis  8.81E-05  De novo purine biosynthesis  5.12E-04  Tricarboxylic acid cycle  1.24E-02  Lysine biosynthesis  4.31E-02  De novo pyrimidine deoxyribonucleotide biosynthesis  3.27E-20  PANTHER pathway  P-value  Ubiquitin proteasome pathway  2.20E-17  Apoptosis signaling pathway  5.24E-14  General transcription regulation  7.12E-14  Transcription regulation by bZIP transcription factor  1.03E-12  Cell cycle  2.95E-09  Huntington disease  6.06E-09  Inflammation mediated by chemokine and cytokine signaling pathway  1.98E-08  DNA replication  6.98E-08  Tryptophan biosynthesis  1.93E-07  Cytoskeletal regulation by Rho GTPase  2.29E-05  Adenosine triphosphate synthesis  8.81E-05  De novo purine biosynthesis  5.12E-04  Tricarboxylic acid cycle  1.24E-02  Lysine biosynthesis  4.31E-02  De novo pyrimidine deoxyribonucleotide biosynthesis  3.27E-20  Table 2. PANTHER pathway enrichment analysis in Cluster 1 PANTHER pathway  P-value  Ubiquitin proteasome pathway  2.20E-17  Apoptosis signaling pathway  5.24E-14  General transcription regulation  7.12E-14  Transcription regulation by bZIP transcription factor  1.03E-12  Cell cycle  2.95E-09  Huntington disease  6.06E-09  Inflammation mediated by chemokine and cytokine signaling pathway  1.98E-08  DNA replication  6.98E-08  Tryptophan biosynthesis  1.93E-07  Cytoskeletal regulation by Rho GTPase  2.29E-05  Adenosine triphosphate synthesis  8.81E-05  De novo purine biosynthesis  5.12E-04  Tricarboxylic acid cycle  1.24E-02  Lysine biosynthesis  4.31E-02  De novo pyrimidine deoxyribonucleotide biosynthesis  3.27E-20  PANTHER pathway  P-value  Ubiquitin proteasome pathway  2.20E-17  Apoptosis signaling pathway  5.24E-14  General transcription regulation  7.12E-14  Transcription regulation by bZIP transcription factor  1.03E-12  Cell cycle  2.95E-09  Huntington disease  6.06E-09  Inflammation mediated by chemokine and cytokine signaling pathway  1.98E-08  DNA replication  6.98E-08  Tryptophan biosynthesis  1.93E-07  Cytoskeletal regulation by Rho GTPase  2.29E-05  Adenosine triphosphate synthesis  8.81E-05  De novo purine biosynthesis  5.12E-04  Tricarboxylic acid cycle  1.24E-02  Lysine biosynthesis  4.31E-02  De novo pyrimidine deoxyribonucleotide biosynthesis  3.27E-20  In Cluster 2, 21 GO terms were significantly enriched and 78 genes accounting for 60.93% of all genes were related to metabolic processes (Supplementary Table S2). Blast fungus showed impact on the metabolic process in rice through interacting with Cluster 2, such as hydrolase activity (P = 1.99E-23) and transport (P = 2.84E-18). Cluster 3 showed enrichment in regulatory activity, including regulation of catalytic activity, regulation of molecular function and kinase regulator activity. We note that Cluster 3 is a typical scale-free network (Figure 3A). The central regulatory gene, LOC_Os05g34770, had a high degree and was regulated by pathogenic genes indirectly through four rice proteins. Meanwhile, we found that Clusters 5, 6 and 7 mainly overrepresented peptidase activity, ubiquitin-protein ligase activity and oxidoreductase activity, respectively (Supplementary Table S2). In Cluster 9, receptor-mediated endocytosis, endocytosis and vesicle-mediated transport were enriched, indicating that rice plants could take advantage of those functions to resist pathogens (Supplementary Table S2). From the above seven clusters, 10 genes exist in more than one cluster, and these can be considered as the bottleneck of the network, which might be involved in the multiple cellular processes (Figure 3B). These 10 genes were LOC_Os02g06640, LOC_Os06g40560, LOC_Os01g62500, LOC_Os02g48290, LOC_Os02g10640, LOC_Os01g73530, LOC_Os05g44310, LOC_Os02g54340, LOC_Os06g09290 and LOC_Os03g16920. Master regulators in rice in response to pathogen infection Viper was used to search the master regulators in the regulatory network in rice under the infection of blast fungus. Based on the microarray data and the regulatory networks, we found 1676 (935 upregulated and 741 downregulated) master regulators in rice in response to the infection of M. grisea, among which 34 upregulated and 12 downregulated genes were also those predicted to be pathogen-targeted proteins (Supplementary Table S3). The results in PANTHER pathway analysis indicated that the upregulated master regulators mainly participated in ubiquitin proteasome pathway and apoptosis signaling pathway, while seven proteins were classified as Hsp70s family chaperone. Although there were no enriched PANTHER pathways in downregulated master regulators, Hsp70s family chaperone and hydrolase were also involved. Furthermore, synergistic effect analysis manifested that eight pairs of synergistic genes in the rice regulatory network respond to the infection of blast fungus, and they could regulate the expression of downstream genes (Table 3). Table 3. Synergistic genes in rice in response to the infection of blast fungus Synergistic genes  Predicted score  LOC_Os01g49330-LOC_Os01g70580  1.000000000  LOC_Os01g74190-LOC_Os02g39410  0.001138238  LOC_Os03g13170-LOC_Os03g51690  0.011223978  LOC_Os03g13170-LOC_Os03g55540  0.101105938  LOC_Os03g13170-LOC_Os04g01590  0.012648366  LOC_Os03g51690-LOC_Os04g01590  0.206007994  LOC_Os03g63420-LOC_Os04g42930  0.681000000  LOC_Os01g70580-LOC_Os01g74190- LOC_Os02g39410  0.007490552  Synergistic genes  Predicted score  LOC_Os01g49330-LOC_Os01g70580  1.000000000  LOC_Os01g74190-LOC_Os02g39410  0.001138238  LOC_Os03g13170-LOC_Os03g51690  0.011223978  LOC_Os03g13170-LOC_Os03g55540  0.101105938  LOC_Os03g13170-LOC_Os04g01590  0.012648366  LOC_Os03g51690-LOC_Os04g01590  0.206007994  LOC_Os03g63420-LOC_Os04g42930  0.681000000  LOC_Os01g70580-LOC_Os01g74190- LOC_Os02g39410  0.007490552  Table 3. Synergistic genes in rice in response to the infection of blast fungus Synergistic genes  Predicted score  LOC_Os01g49330-LOC_Os01g70580  1.000000000  LOC_Os01g74190-LOC_Os02g39410  0.001138238  LOC_Os03g13170-LOC_Os03g51690  0.011223978  LOC_Os03g13170-LOC_Os03g55540  0.101105938  LOC_Os03g13170-LOC_Os04g01590  0.012648366  LOC_Os03g51690-LOC_Os04g01590  0.206007994  LOC_Os03g63420-LOC_Os04g42930  0.681000000  LOC_Os01g70580-LOC_Os01g74190- LOC_Os02g39410  0.007490552  Synergistic genes  Predicted score  LOC_Os01g49330-LOC_Os01g70580  1.000000000  LOC_Os01g74190-LOC_Os02g39410  0.001138238  LOC_Os03g13170-LOC_Os03g51690  0.011223978  LOC_Os03g13170-LOC_Os03g55540  0.101105938  LOC_Os03g13170-LOC_Os04g01590  0.012648366  LOC_Os03g51690-LOC_Os04g01590  0.206007994  LOC_Os03g63420-LOC_Os04g42930  0.681000000  LOC_Os01g70580-LOC_Os01g74190- LOC_Os02g39410  0.007490552  Finally, we compared the different master regulators in rice in response to the infection of M. grisea, Xoo and RSV. There were 1053 and 715 master regulators in rice under the infection of Xoo and RSV, respectively. Taken into account of 1676 master regulators in rice under the infection of M. grisea, total 150 master regulators were common in rice in response to these three kinds of infection, 341 common in M. grisea and RSV, 587 in M. grisea and Xoo, while 235 in RSV and Xoo (Figure 4). Figure 4. View largeDownload slide The Venn diagram of the master regulators in rice in response to the infection of three pathogens. MGG, XOO and RSV denote M. grisea, Xoo and RSV, respectively. Figure 4. View largeDownload slide The Venn diagram of the master regulators in rice in response to the infection of three pathogens. MGG, XOO and RSV denote M. grisea, Xoo and RSV, respectively. Discussion Considering the critical role of PPIs in the blast fungus infection process, a protocol pipeline including three methods was used sequentially to predict and verify the PPIs between M. grisea and O. sativa in this article. First, the potential PPIs between M. grisea and O. sativa were predicted by using the method based on interolog. This method has been widely used in biological research work, such as prediction of PPIs in rice genome [56] and interacting network of Hsp70s in rice [29]. However, most previous researches predicted PPIs in intraspecies genome by using the experimentally identified PPIs as templates. In this research, experimentally identified PPIs including 63 PPIs between host and pathogen were downloaded and used as templates to predict PPIs between interspecies, M. grisea and O. sativa. Second, the domain-based method was applied as a complementary method to confirm the predicted PPIs between M. grisea and O. sativa. Recently, PPIs between R. solanacearum and A. thaliana were predicted by using the above two methods [30]. Third, SignalP and WolFPSORT were subsequently used to identify the secretory proteins in M. grisea, which were involved in the above predicted PPIs because the secretory proteins in pathogens were considered to interact with receptors in plants [30]. Only the cross-verified PPIs in these three methods were selected for the downstream analysis. Finally, we achieved 523 PPIs between M. grisea and O. sativa, among which 470 PPIs were predicted by the 63 PPIs templates between host and pathogen. The 523 PPIs between M. grisea and O. sativa were further confirmed. Although some methods including the similarity of GO annotation [56], subcellular localization and gene co-expression [29] had been used to evaluate the PPIs in intraspecies, they might not be adapted for PPIs between interspecies. In this study, we used the two methods that were introduced by Shen et al. and He et al., respectively [13, 32], to confirm the PPIs between M. grisea and O. sativa. The method of Shen et al. [32] was applied to investigate whether the protein sequences in PPI pairs could be statistically classified, while the method of He et al. [13] was used to figure out the enrichment of pathogenic proteins in the PPIs. The results in these two methods all supported the above predicted PPIs between M. grisea and O. sativa. In summary, 523 PPIs between M. grisea and O. sativa achieved in this research work were cross-verified by using three methods and were confirmed again by two strategies. The results in GO annotations analysis on the rice proteins involved in 523 PPIs between M. grisea and O. sativa showed that the most enrichment of GO annotation of these proteins was related to the function of vesicle-mediated transport and exocytosis, which could drive focal and/or nondirectional secretion of antimicrobial cocktails to resist microbial pathogens infection [57]. Alternatively, pathogenic proteins in fungus interacted with these vesicle-mediated transport and exocytosis-related proteins intercept the secretion machinery by blocking vesicle formation from intracellular membranes [57]. This might be one of the models showing the balance between immune responses of rice plants and counter defense of blast fungus. Receptor kinase function was enriched in the rice proteins involved in the 523 PPIs between blast fungus and rice. Receptor kinases in rice plants could recognize the conservative pathogenic molecules and subsequently active intracellular signaling pathways [58]. In summary, during the infection process of blast fungus, the kinase receptor-mediated recognition of rice plants detected the presence of blast fungus and triggered immune responses including activating the vesicle-associated exocytosis pathways. Analysis using microarrays on the regular network of rice in response to blast fungus infection showed that 935 master regulators were upregulated by blast fungus infection, among which 34 proteins interacted with blast fungus proteins. In total, 7 of those 34 master regulators belonged to the Hsp70s family, which could help hosts to resist environmental stresses [59]. This was consistent with previous research work where Hsp70s was upregulated in rice under high temperature [60], and was involved in macromolecular translocation, carbohydrate metabolism, innate immunity, photosystem II repair and regulation of kinase activities [29]. Meanwhile, another 2 of the 34 master regulators, LOC_Os01g08400.1 and LOC_Os01g73300.1, were annotated as SNARE proteins, which mediated the exocytosis in plant cell and the fusion between intracellular vesicles and cytomembrane [61]. Our results suggest that Hsp70s and SNARE proteins in rice were upregulated by the infection of blast fungus and then induce the downstream immune system in rice. In this research work, GO and PANTHER were used to explore the biological functions of subnetworks in rice, which were affected by blast fungus through the PPIs. In Cluster 1, there were several immune-related metabolic pathways, including the protein enzyme and apoptosis signaling pathways. Meanwhile, we found that the ubiquitin proteasome pathway was enriched, respectively, in rice under the infection of M. grisea, Xoo or RSV, further illustrating that ubiquitin proteasome pathway was the common response of rice plants to the infection of fungi, bacteria and viruses. It was reported that ubiquitin proteasome pathway was the key step in the process of plant immune response [62], while apoptosis signaling pathway was the final presentation forms of plant ETI pathway [11]. The results in this study implied that the pathogens inhibit the rice plant immune process through PPIs by disrupting hosts’ ubiquitination [63]. We found 10 bottleneck proteins related to multiple subnetworks, which might contribute to the regulation of metabolic process [64]. In total, 4 of 10 bottleneck proteins including LOC_Os06g4056, LOC_Os06g09290, LOC_Os02g54340 and LOC_Os02g10640, which were participating in ubiquitin protease pathway, were all connected to Cluster 5. The Cluster 5 subnetwork was related to proteolysis, peptidase and hydrolase activity. Furthermore, LOC_Os02g54340 and LOC_Os02g10640 were simultaneously connected to Cluster 6, which accounted for ubiquitin-protein ligase. These indicated that the interactions between fungus proteins and the bottleneck proteins showed impact on hydrolase and peptidase activity of rice plants, which might also be connected to immune metabolism [58]. Therefore, these 10 bottleneck proteins in rice might be the infecting target of blast fungus. Conclusion Using two well-known PPI prediction methods, 523 potential PPIs between M. grisea and O. sativa were predicted. We used jackknife test, 10-fold crossover test and independent test based on SVM to further confirm the PPIs with ACC of 90.34, 93.85 and 84.67%, respectively. Then, the regulatory network of rice, RiceNet data, was clustered into 228 subnetworks with over six nodes and the top seven clusters affected by blast fungus through PPIs was analyzed. In addition, 46 master regulators interacting with the fungus proteins, including 34 upregulated and 12 downregulated regulators, were observed under the infection of M. grisea. Our results uncover the interacting mechanism between M. grisea and O. sativa, and provide researchers a framework of validation for future experimental work. Key Points According to the classical gene-for-gene system, the key factor causing rice blast disease and defense might be PPIs between rice and fungus. After cross-verified by using three methods and confirmed again by two strategies, 523 PPIs between blast fungus and rice were achieved in this research work. The regulatory networks in rice in response to biotic stresses were divided into 228 subnetworks with over six nodes and the top seven subnetworks affected by blast fungus through PPIs were investigated. The result indicated that 34 upregulated and 12 downregulated master regulators in rice interacting with the fungus proteins in response to the infection of blast fungus. Supplementary Data Supplementary data are available online at http://bib.oxfordjournals.org/. Shiwei Ma is a PhD student at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Qi Song is an MSc student at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Huan Tao is a Lecturer at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Andrew Harrison is a Senior Lecturer at Department of Mathematical Sciences, University of Essex, UK. Shaobo Wang is an MSc student at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Wei Liu is an Associate Professor at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Shoukai Lin is a Lecturer at Fujian Provincial Key Laboratory of Ecology-toxicological Effects and Control for Emerging Contaminants, Putian University, China Ziding Zhang is a Professor of bioinformatics at the College of Biological Sciences, China Agriculture University, China. Shoukai Lin is a Lecturer at Fujian Provincial Key Laboratory of Ecology-toxicological Effects and Control for Emerging Contaminants, Putian University, China Yufang Ai is an Associate Professor at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Huaqin He is a Professor of bioinformatics at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Acknowledgement The authors thank the anonymous referees whose constructive comments were helpful in improving the quality of this work. Funding Natural Science Foundation of China and Fujian (grant numbers 31270454, 81502091 and 2013J01077, 2014N5006), Innovative Foundation of FAFU (grant numbers CXZX2017132 and CXZX2017303) and Fujian-Taiwan Joint Innovative Centre for Germplasm Resources and Cultivation of Crop (grant number 2015-75. FJ 2011 Program). References 1 Dean RA, Talbot NJ, Ebbole DJ, et al.   The genome sequence of the rice blast fungus Magnaporthe grisea. Nature  2005; 434: 980– 6. Google Scholar CrossRef Search ADS PubMed  2 Parker D, Beckmann M, Enot DP, et al.   Rice blast infection of Brachypodium distachyon as a model system to study dynamic host/pathogen interactions. Nat Protoc  2008; 3: 435– 45. Google Scholar CrossRef Search ADS PubMed  3 Xue MF, Yang J, Li ZG, et al.   Comparative analysis of the genomes of two field isolates of the rice blast fungus Magnaporthe oryzae. PLoS Genet  2012; 8: e1002869. Google Scholar CrossRef Search ADS PubMed  4 Ribot C, Hirsch J, Balzergue S, et al.   Susceptibility of rice to the blast fungus, Magnaporthe grisea. J Plant Physiol  2008; 165: 114– 24. Google Scholar CrossRef Search ADS PubMed  5 Hulbert SH, Webb CA, Smith SM, et al.   Resistance gene complexes: evolution and utilization. Annu Rev Phytopathol  2001; 39: 285– 312. Google Scholar CrossRef Search ADS PubMed  6 Tao Y, Xie Z, Chen W, et al.   Quantitative nature of Arabidopsis responses during compatible and incompatible interactions with the bacterial pathogen Pseudomonas syringae. Plant Cell  2003; 15: 317– 30. Google Scholar CrossRef Search ADS PubMed  7 Kankanala P, Czymmek K, Valent B. Roles for rice membrane dynamics and plasmodesmata during biotrophic invasion by the blast fungus. Plant Cell  2007; 19: 706– 24. Google Scholar CrossRef Search ADS PubMed  8 Talbot NJ, Ebbole DJ, Hamer JE. Identification and characterization of MPG1, a gene involved in pathogenicity from the rice blast fungus Magnaporthe grisea. Plant Cell  1993; 5: 1575– 90. Google Scholar CrossRef Search ADS PubMed  9 Xu JR, Zhao X, Dean RA. From genes to genomes: a new paradigm for studying fungal pathogenesis in Magnaporthe oryzae. Adv Genet  2007; 57: 175– 218. Google Scholar PubMed  10 Tucker SL, Talbot NJ. Surface attachment and pre-penetration stage development by plant pathogenic fungi. Annu Rev Phytopathol  2001; 39: 385– 417. Google Scholar CrossRef Search ADS PubMed  11 Jones JD, Dangl JL. The plant immune system. Nature  2006; 444: 323– 9. Google Scholar CrossRef Search ADS PubMed  12 Bishop JG, Dean AM, Mitchell-Olds T. Rapid evolution in plant chitinases: molecular targets of selection in plant-pathogen coevolution. Proc Natl Acad Sci USA  2000; 97: 5322– 7. Google Scholar CrossRef Search ADS PubMed  13 He F, Zhang Y, Chen H, et al.   The prediction of protein-protein interaction networks in rice blast fungus. BMC Genomics  2008; 9: 519. Google Scholar CrossRef Search ADS PubMed  14 Kim JG, Park D, Kim BC, et al.   Predicting the interactome of Xanthomonas oryzae pathovar oryzae for target selection and DB service. BMC Bioinformatics  2008; 9: 41. Google Scholar CrossRef Search ADS PubMed  15 Kawahara Y, de la Bastide M, Hamilton JP, et al.   Improvement of the Oryza sativa Nipponbare reference genome using next generation sequence and optical map data. Rice  2013; 6: 4. Google Scholar CrossRef Search ADS PubMed  16 Bogdanove AJ. Protein-protein interactions in pathogen recognition by plants. Plant Mol Biol  2002; 50: 981– 9. Google Scholar CrossRef Search ADS PubMed  17 Jia Y, McAdams SA, Bryan GT, et al.   Direct interaction of resistance gene and avirulence gene products confers rice blast resistance. Embo J  2000; 19: 4004– 14. Google Scholar CrossRef Search ADS PubMed  18 Kanzaki H, Yoshida K, Saitoh H, et al.   Arms race co-evolution of Magnaporthe oryzae AVR-Pik and rice Pik genes driven by their physical interactions. Plant J  2012; 72: 894– 907. Google Scholar CrossRef Search ADS PubMed  19 Cesari S, Thilliez G, Ribot C, et al.   The rice resistance protein pair RGA4/RGA5 recognizes the Magnaporthe oryzae effectors AVR-Pia and AVR1-CO39 by direct binding. Plant Cell  2013; 25: 1463– 81. Google Scholar CrossRef Search ADS PubMed  20 Coelho ED, Arrais JP, Matos S, et al.   Computational prediction of the human-microbial oral interactome. BMC Syst Biol  2014; 8: 24. Google Scholar CrossRef Search ADS PubMed  21 Barker D, Pagel M. Predicting functional gene links from phylogenetic-statistical analyses of whole genomes. PLoS Comput Biol  2005; 1: e3. Google Scholar CrossRef Search ADS PubMed  22 Najafabadi HS, Salavati R. Sequence-based prediction of protein-protein interactions by means of codon usage. Genome Biol  2008; 9: R87. Google Scholar CrossRef Search ADS PubMed  23 Aloy P, Russell RB. Interrogating protein interaction networks through structural biology. Proc Natl Acad Sci USA  2002; 99: 5896– 901. Google Scholar CrossRef Search ADS PubMed  24 Huang C, Morcos F, Kanaan SP, et al.   Predicting protein-protein interactions from protein domains using a set cover approach. IEEE/ACM Trans Comput Biol Bioinform  2007; 4: 78– 87. Google Scholar CrossRef Search ADS PubMed  25 Wang RS, Wang Y, Wu LY, et al.   Analysis on multi-domain cooperation for predicting protein-protein interactions. BMC Bioinformatics  2007; 8: 391. Google Scholar CrossRef Search ADS PubMed  26 Zhang YQ, Zhang DL, Mi G, et al.   Using ensemble methods to deal with imbalanced data in predicting protein-protein interactions. Comput Biol Chem  2012; 36: 36– 41. Google Scholar CrossRef Search ADS PubMed  27 Maetschke SR, Simonsen M, Davis MJ, et al.   Gene Ontology-driven inference of protein-protein interactions using inducers. Bioinformatics  2012; 28: 69– 75. Google Scholar CrossRef Search ADS PubMed  28 Jaeger S, Gaudan S, Leser U, et al.   Integrating protein-protein interactions and text mining for protein function prediction. BMC Bioinformatics  2008; 9 (Suppl 8): S2. Google Scholar CrossRef Search ADS PubMed  29 Wang Y, Lin S, Song Q, et al.   Genome-wide identification of heat shock proteins (Hsps) and Hsp interactors in rice: Hsp70s as a case study. BMC Genomics  2014; 15: 344. Google Scholar CrossRef Search ADS PubMed  30 Li ZG, He F, Zhang ZD, et al.   Prediction of protein-protein interactions between Ralstonia solanacearum and Arabidopsis thaliana. Amino Acids  2012; 42: 2363– 71. Google Scholar CrossRef Search ADS PubMed  31 Ng SK, Zhang Z, Tan SH. Integrative approach for computationally inferring protein domain interactions. Bioinformatics  2003; 19: 923– 9. Google Scholar CrossRef Search ADS PubMed  32 Shen J, Zhang J, Luo X, et al.   Predicting protein-protein interactions based only on sequences information. Proc Natl Acad Sci USA  2007; 104: 4337– 41. Google Scholar CrossRef Search ADS PubMed  33 Tully JP, Hill AE, Ahmed HM, et al.   Expression-based network biology identifies immune-related functional modules involved in plant defense. BMC Genomics  2014; 15: 421. Google Scholar CrossRef Search ADS PubMed  34 Wang J, Li M, Deng Y, et al.   Recent advances in clustering methods for protein interaction networks. BMC Genomics  2010; 11 (Suppl 3): S10. Google Scholar PubMed  35 Chan SS, Kyba M. What is a master regulator? J Stem Cell Res Ther  2013; 3: 114. Google Scholar PubMed  36 Lefebvre C, Rajbhandari P, Alvarez MJ, et al.   A human B-cell interactome identifies MYB and FOXM1 as master regulators of proliferation in germinal centers. Mol Syst Biol  2010; 6: 377. Google Scholar CrossRef Search ADS PubMed  37 Liu T, Lin Y, Wen X, et al.   BindingDB: a web-accessible database of experimentally determined protein-ligand binding affinities. Nucleic Acids Res  2007; 35: D198– 201. Google Scholar CrossRef Search ADS PubMed  38 Licata L, Briganti L, Peluso D, et al.   MINT, the molecular interaction database: 2012 update. Nucleic Acids Res  2012; 40: D857– 61. Google Scholar CrossRef Search ADS PubMed  39 Salwinski L, Miller CS, Smith AJ, et al.   The Database of Interacting Proteins: 2004 update. Nucleic Acids Res  2004; 32: D449– 51. Google Scholar CrossRef Search ADS PubMed  40 Poole RL. The TAIR database. Methods Mol Biol  2007; 406: 179– 212. Google Scholar PubMed  41 Kerrien S, Aranda B, Breuza L, et al.   The IntAct molecular interaction database in 2012. Nucleic Acids Res  2012; 40: D841– 6. Google Scholar CrossRef Search ADS PubMed  42 Kumar R, Nanduri B. HPIDB–a unified resource for host-pathogen interactions. BMC Bioinformatics  2010; 11 (Suppl 6): S16. Google Scholar CrossRef Search ADS PubMed  43 Krogh A, Larsson B, von Heijne G, et al.   Predicting transmembrane protein topology with a hidden Markov model: application to complete genomes. J Mol Biol  2001; 305: 567– 80. Google Scholar CrossRef Search ADS PubMed  44 Kawahara Y, Oono Y, Kanamori H, et al.   Simultaneous RNA-seq analysis of a mixed transcriptome of rice and blast fungus interaction. PLoS One  2012; 7: e49423. Google Scholar CrossRef Search ADS PubMed  45 Bendtsen JD, Nielsen H, von Heijne G, et al.   Improved prediction of signal peptides: SignalP 3.0. J Mol Biol  2004; 340: 783– 95. Google Scholar CrossRef Search ADS PubMed  46 Horton P, Park KJ, Obayashi T, et al.   WoLF PSORT: protein localization predictor. Nucleic Acids Res  2007; 35: W585– 7. Google Scholar CrossRef Search ADS PubMed  47 Smoot ME, Ono K, Ruscheinski J, et al.   Cytoscape 2.8: new features for data integration and network visualization. Bioinformatics  2011; 27: 431– 2. Google Scholar CrossRef Search ADS PubMed  48 Fawcett T. An introduction to ROC analysis. Pattern Recogn Lett  2006; 27: 861– 74. Google Scholar CrossRef Search ADS   49 Liu B, Wang S, Long R, et al.   iRSpot-EL: identify recombination spots with an ensemble learning approach. Bioinformatics  2017; 33: 35– 41. Google Scholar CrossRef Search ADS PubMed  50 Chou KC, Shen HB. Recent progress in protein subcellular location prediction. Anal Biochem  2007; 370: 1– 16. Google Scholar CrossRef Search ADS PubMed  51 Fan GL, Zhang XY, Liu YL, et al.   DSPMP: discriminating secretory proteins of malaria parasite by hybridizing different descriptors of Chou's pseudo amino acid patterns. J Comput Chem  2015; 36: 2317– 27. Google Scholar CrossRef Search ADS PubMed  52 Urban M, Cuzick A, Rutherford K, et al.   PHI-base: a new interface and further additions for the multi-species pathogen-host interactions database. Nucleic Acids Res  2017; 45: D604– 10. Google Scholar CrossRef Search ADS PubMed  53 Assenov Y, Ramirez F, Schelhorn SE, et al.   Computing topological parameters of biological networks. Bioinformatics  2008; 24: 282– 4. Google Scholar CrossRef Search ADS PubMed  54 Mi H, Muruganujan A, Thomas PD. PANTHER in 2013: modeling the evolution of gene function, and other gene attributes, in the context of phylogenetic trees. Nucleic Acids Res  2013; 41: D377– 86. Google Scholar CrossRef Search ADS PubMed  55 Dobay MP, Stertz S, Delorenzi M. Context-based retrieval of functional modules in protein-protein interaction networks. Brief Bioinform  2017. doi: 10.1093/bib/bbx029. 56 Gu H, Zhu P, Jiao Y, et al.   PRIN: a predicted rice interactome network. BMC Bioinformatics  2011; 12: 161. Google Scholar CrossRef Search ADS PubMed  57 Kwon C, Bednarek P, Schulze-Lefert P. Secretory pathways in plant immune responses. Plant Physiol  2008; 147: 1575– 83. Google Scholar CrossRef Search ADS PubMed  58 Chen X, Ronald PC. Innate immunity in rice. Trends Plant Sci  2011; 16: 451– 9. Google Scholar CrossRef Search ADS PubMed  59 Morano KA. New tricks for an old dog: the evolving world of Hsp70. Ann N Y Acad Sci  2007; 1113: 1– 14. Google Scholar CrossRef Search ADS PubMed  60 Chen X, Lin S, Liu Q, et al.   Expression and interaction of small heat shock proteins (sHsps) in rice in response to heat stress. Biochim Biophys Acta  2014; 1844: 818– 28. Google Scholar CrossRef Search ADS PubMed  61 Collins NC, Thordal-Christensen H, Lipka V, et al.   SNARE-protein-mediated disease resistance at the plant cell wall. Nature  2003; 425: 973– 7. Google Scholar CrossRef Search ADS PubMed  62 Duplan V, Rivas S. E3 ubiquitin-ligases and their target proteins during the regulation of plant innate immunity. Front Plant Sci  2014; 5: 42. Google Scholar CrossRef Search ADS PubMed  63 Craig A, Ewan R, Mesmar J, et al.   E3 ubiquitin ligases and plant innate immunity. J Exp Bot  2009; 60: 1123– 32. Google Scholar CrossRef Search ADS PubMed  64 Yu H, Kim PM, Sprecher E, et al.   The importance of bottlenecks in protein networks: correlation with gene essentiality and expression dynamics. PLoS Comput Biol  2007; 3: e59. Google Scholar CrossRef Search ADS PubMed  © The Author 2017. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Briefings in Bioinformatics Oxford University Press

Prediction of protein–protein interactions between fungus (Magnaporthe grisea) and rice (Oryza sativa L.)

Loading next page...
 
/lp/ou_press/prediction-of-protein-protein-interactions-between-fungus-magnaporthe-ctidNpixEp
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
1467-5463
eISSN
1477-4054
D.O.I.
10.1093/bib/bbx132
Publisher site
See Article on Publisher Site

Abstract

Abstract Rice blast disease caused by the fungus Magnaporthe grisea (M. grisea) is one of the most serious diseases for the cultivated rice Oryza sativa (O. sativa). A key factor causing rice blast disease and defense might be protein–protein interactions (PPIs) between rice and fungus. In this research, we have developed a computational pipeline to predict PPIs between blast fungus and rice. After cross-prediction by interolog-based and domain-based method, we achieved 532 potential PPIs between 27 fungus proteins and 236 rice proteins. Accuracy of jackknife test, 10-fold cross-validation test and independent test for these PPIs were 90.43, 93.85 and 84.67%, respectively, by using support vector machine classification method. Meanwhile, the pathogenic genes of blast fungus were enriched in the predicted PPIs network when compared with 1000 random interaction networks. The rice regulatory network was downloaded and divided into 228 subnetworks with over six nodes, and the top seven subnetworks affected by blast fungus through PPIs were investigated. The results indicated that 34 upregulated and 12 downregulated master regulators in rice interacting with the fungus proteins in response to the infection of blast fungus. The common master regulators in rice in response to the infection of M. grisea, Xanthomonas oryzae pv.oryzae and rice stripe virus were analyzed. The ubiquitin proteasome pathway was the common pathway in rice regulated by these three pathogens, while apoptosis signaling pathway was induced by fungus and bacteria. In summary, the results in this article provide insight into the process of blast fungus infection. rice (Oryza sativa L.), blast fungus (M. grisea), interspecies interaction, master regulators Introduction Rice blast, bacterial blight and stripe, which are caused by Magnaporthe grisea, Xanthomonas oryzae pv.oryzae (Xoo) and rice stripe virus (RSV), respectively, are three of the most widespread and devastating rice diseases, especially rice blast. The rice plant is susceptible to the blast fungus in all of its growth period, with infection affecting leaves, nodes, panicles and roots [1]. Annual yield losses of rice destroyed by blast are sufficient to feed 60 million people [2]. In addition, the blast fungus is also harmful to wheat and other small grains and leads to a large reduction in yield [3]. Strategies such as fungicides, resistant cultivars, agronomical practices and biotechnological methods have been developed to overcome this disastrous disease [4]. However, use of plant host-resistant ability is the most effective and economic way to control the diseases [5]. Thus, it is essential to explore the pathogenicity of blast fungus and the resistant ability of rice. The infection of M. grisea is considered as a hemibiotrophic process, and it must go through the important biotrophic phase in its early stages [6, 7]. There have been minutely cytological interpretations of occurring events during rice blast infection [8–10]. During the lengthy warfare between pathogens and hosts, plants have evolved elaborate defense systems, including pathogen-associated molecular patterns (PAMPs), pathogen-triggered immunity (PTI) and effector-triggered immunity (ETI) system [11]. The molecular mechanisms of rice PTI and ETI to resist the infection of blast fungus have been partially illustrated. The genes underlying rice plant defenses comprise a substantial portion of the host genome through the direct interactions with pathogen proteins or the initiation of rice plant defense responses to the infection [12]. It is critical to understand the protein–protein interaction (PPI) network (i.e. interactome) between plant and pathogen for studying the molecular basis of pathogenesis [13, 14]. In the past decade, the genomes of rice (Oryza sativa) and blast fungus (M. grisea) have been sequenced and well annotated [1, 15]. This enables the pathosystem between rice and blast fungus to be a model of the study of plant–microbe interactions. Although many biochemical techniques have been developed to verify PPIs, the experimental identification of PPIs between plant and pathogen is still a time-consuming and challenging work [16]. Till now, few pairs of PPIs between rice and blast fungus have been identified experimentally, which is insufficient to elucidate the molecular mechanism of pathogenicity [17–19]. Experimental methods are the ideal tools for validating PPIs, but these should be more efficient for large-scale PPI prediction [20]. With the massive accumulation of genomic data and known PPIs, computational methods to predict PPIs as an alternative to biochemical techniques have been improved with high performance recently. These include predictors based on genomic data [21, 22], protein structure [23], domain information [24, 25], protein sequence [26] and Gene Ontology (GO) annotation semantic similarity [27]. Most of the predictors follow the common principle that text mining is used to extract the information from the known PPIs in the biomedical literatures [28]. The method based on interolog, a PPI predictor based on protein sequence similarity, has been used widely to predict PPIs on different scales. He et al. [13] used the interolog-based method to predict PPIs in blast fungus genome, while Wang et al. [29] constructed the interacting network of heat shock protein 70 s (Hsp70s) in rice genome. Meanwhile, PPI predictors based on protein domain information and support vector machine (SVM) predictors based on protein structural information have been used alone or complementary to sequence similarity predictors to predict intraspecies PPIs [30, 31] and achieve high performance [32]. To the best of our knowledge, computational efforts to predict interspecies PPIs have been hardly reported. Li et al. [30] successfully used the method based on interolog and a domain-base method to construct PPI network between Ralstonia solanacearum and Arabidopsis thaliana. The regulatory network in host plants affected by pathogen infection through the PPIs information could be developed to understand the influence of pathogen proteins to the host [33]. The ClusterViz program in Cytoscape was introduced to analyze the regulatory network [34]. Master regulators, genes at the top of a regulatory hierarchy, usually regulate lots of downstream genes [35]. Lefebvre et al. [36] developed a tool, Viper, to analyze master regulators in the regulatory network. In this research work, PPIs between blast fungus and rice were crosswise predicted using the method based on interolog and the domain-based method. Then, the potential PPIs were further confirmed by using SVM and the enrichment of pathogenic proteins. After that, we analyzed the regulatory network in rice in response to the infection of blast fungus through the predicted PPI network. Materials and methods Data sets A total of 11 054 protein sequences in blast fungus genome were downloaded from the M. grisea database [1] (http://www.broadinstitute.org/annotation/genome/magnaporthe_grisea/MultiHome.html), while 66 338 protein sequences in rice genome were obtained from MSU Rice Genome Annotation Project Database [15] (ftp://ftp.plantbiology.msu.edu/pub/data/Eukaryotic_Projects/o_sativa/annotation_dbs/pseudomolecules/version_7.0/all.dir/). To perform the method based on interolog, 635 008 experimentally verified PPIs were firstly downloaded from five public databases, including Biomolecular Interaction Network Database [37] (BIND), Molecular INTeraction database [38] (MINT), Database of Interacting Proteins [39] (DIP), The Arabidopsis Information Resource [40] (TAIR) and IntAct [41]. Overall, 23 845 host–pathogen PPIs were obtained from Host–Pathogen Interaction Database [42] (HPIDB; http://agbase.msstate.edu/hpi/main.html). The microarray data of rice in response to three biotic stresses, including M. grisea, Xoo and RSV infection, were downloaded from Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/). The GEO Accession for rice under M. grisea, Xoo and RSV infection was GSE7256, GSE19844 and GSE11025, respectively. RiceNet data under the biotic stress, including every nodes and links, were obtained from http://www.functionalnet.org/ricenet. Identification of secreted and membrane proteins The proteins in M. grisea were inferred to be transmembrane when the number of predicted transmembrane helices was >1 by using TMHMM [43]. According to the methods introduced by Kawahara et al. [44], SignalP [45] and WoLFPSORT [46], predictors were used to identify secretory proteins with the default criterion. The proteins containing signal peptide predicted by SignalP and locating extracellular predicted by WoLFPSORT were deduced to be secreted proteins. Prediction of PPIs between blast fungus and rice using the method based on interolog The potential PPIs between blast fungus and rice were predicted using the method based on interolog (Figure 1). Briefly, each protein in M. grisea and O. sativa was first searched against BIND, MINT, DIP, TAIR and IntAct database to identify homologs with 10−5E-value, 30% sequence identify and 40% aligned sequence length coverage cutoff. Proteins in M. grisea and O. sativa were BLASTed against pathogenic proteins and host proteins in HPIDB, respectively, to identify their homologs with the same criterion. If the corresponding homologs of one protein pair from M. grisea and O. sativa had at least one interaction in the above databases, the protein pair is predicted to interact. Figure 1. View largeDownload slide Prediction pipeline of the potential PPIs between blast fungus and rice. Figure 1. View largeDownload slide Prediction pipeline of the potential PPIs between blast fungus and rice. Verification of PPIs between blast fungus and rice using domain-based method The potential PPIs between blast fungus and rice were first verified by using the domain-based method (Figure 1). Each protein in the potential PPIs was searched against the Pfam database (www.pfam.com) with 10−5E-value and 90% sequence identify. When an interacting Pfam domain was found in the interacting protein pair, the potential PPI between blast fungus and rice was further identified. After that, the PPIs predicted by the method based on interolog and verified by the domain-based method were intersected into a PPI network between blast fungus and rice by using Cytoscape [47]. Those PPIs with blast fungus proteins predicted to be non-membrane or nonsecreted ones were removed from the established PPI network. Three different methods based on SVMs and the enrichment of pathogenic proteins were used to confirm the above predicted PPIs in the network. Following Shen et al. [32], the sequence information of every predicted PPI was extracted to assess the prediction effectiveness using 10-fold crossover test, independent test [48] and jackknife test [49]. For 10-fold crossover test, we randomly selected proteins in blast fungus and rice genome to construct a negative data set with the same size of the predicted PPIs. Then, the negative set and the predicted PPIs data were mixed together and divided into 10 sub-data sets. One of these sub-data sets and the other nine sub-data sets were used as test and training data set, respectively, to detect the mean index with 10 repetitions. In the independent test, the size of negative data set was 20 times the size of predicted PPIs. The negative data set and the predicted PPI pairs were also merged together and divided into three subsets, two of which was used as training data set and the other one was used as test data set. Because the jackknife test was regarded as the most objective and effective approach to evaluate the accuracy (ACC) of various predictors [50, 51], the predictive performance was further examined by the jackknife test. Moreover, fungus proteins in the predicted PPIs should match the pathogenic proteins in Phi-base [52]. Simultaneously, we built 1000 random PPI networks containing the above predicted 523 PPIs and calculated the P-value and significance of comparison by the following formula:   P=nnp≥iN (1) where np is the number of M. grisea pathogenic proteins found in every random PPI network, and i is the number of M. grise pathogenic proteins in the predicted interactome; nnp≥i, denotes the number of random PPI networks containing more pathogenic proteins than in the predicted interactome; N is the total number of random PPI networks. Analysis on the PPI network topology Two topological parameters (i.e. degree and betweenness) for each protein in the PPI network were calculated using NetworkAnalyzer [53]. In a PPI network, each protein is represented as a node. The degree of a node is simply defined as the number of interactions that a node has. The betweenness is a centrality measure of a node in a network. The betweenness centrality of a node n can be calculated as the formula.   Cb(n)=∑s≠t≠n(σst(n)/σst) (2) where s and t are nodes in the network that are different from n, σstdenotes the number of shortest paths from s to t and σst( n) is the number of shortest paths from s to t that pass through the node n. Analysis of the impact of fungus proteins on rice regulatory network The regulatory networks in rice in response to biotic stresses were downloaded from http://www.functionalnet.org/ricenet. The ClusterViz program in Cytoscape, which is based on the FAG-EC algorithm [34], was used to cluster rice regulatory networks into different functional modules. The first 10 largest networks named Clusters 1–10 were selected to analyze the impact of pathogenic genes in blast fungus on the functional modules in rice. The 236 rice proteins involved in the above predicted PPIs were first searched against these 10 clusters. If a match is not found then, the downstream proteins of the 236 rice proteins in the predicted PPI networks were searched. After that, GO enrichment of each cluster was determined using the Fisher exact test followed by the false discovery rate (FDR) correction with P < 0.05. Analysis of master regulators in rice in response to blast fungus infection First, the original microarray data were normalized using Robust Multi-Array in oneChannelGUI running by R program and the ID of probe was transformed into gene ID in MSU Rice Genome Annotation Project Database by BioMart. When a gene corresponded to more than one probe, the expression of this gene was represented by a probe which had a largest coefficient of variation (V) calculated as the formula:   V=σE (3) where σ is the SD of the expression of a probe in all samples, and E is the average of the expression of a probe in all samples. The rowTtest in viper was used to analyze the differentially expressed genes (DEGs) in the microarrays under three biotic stresses with P < 0.05 corrected by Bonferroni. DEGs, and RiceNet data were submitted to viper, and msviper was used to search master regulators with P < 0.05 corrected by Bonferroni. The degree of a regulator was set as at least 20, and there would be some false-positive regulators, which were defined as shadow regulators [36]. To remove them, the function of shadow in viper was used with the default criterion. The synergy of master regulators was obtained by using msviperSynergy with the default criterion. Analysis of GO enrichment and metabolic pathway The PANTHER database was used to analyze the GO enrichment of all the subnetworks with P < 0.05 corrected by Bonferroni [54]. PANTHER GO slim was applied in GO annotations, while metabolic pathway of every gene was obtained from PANTHER database directly. Results Prediction of PPIs between M. grisea and O. sativa After predicting PPIs using the method based on interolog and verifying by using domain-based method, we achieved 523 potential PPIs between M. grisea and O. sativa, which is shown in Figure 2 and Supplementary Table S1. The methods based on SVM and enrichment of pathogenic proteins were used to confirm the above PPIs between M. grisea and O. sativa. Compared with the random 1000 PPI networks, the pathogenic proteins were significantly enriched in the predicted PPIs (empirical P-value = 0). Additionally, ACC of jackknife test, 10-fold cross test and independent test based on SVM were 90.43, 93.85 and 84.67%, respectively (Table 1). Table 1. Classification results of PPIs in jackknife test, 10-fold crossover test and independent test Test  ACC (%)  Sp (%)  Sn (%)  AUC  Jackknife test  90.43  87.19  93.68  0.904  10-folds crossover test  93.85  95.67  85.41  0.955  Independent test  84.67  84.57  94.26  0.935  Test  ACC (%)  Sp (%)  Sn (%)  AUC  Jackknife test  90.43  87.19  93.68  0.904  10-folds crossover test  93.85  95.67  85.41  0.955  Independent test  84.67  84.57  94.26  0.935  Sp, Sn and AUC denote specificity, sensitivity and area under curve, respectively. Table 1. Classification results of PPIs in jackknife test, 10-fold crossover test and independent test Test  ACC (%)  Sp (%)  Sn (%)  AUC  Jackknife test  90.43  87.19  93.68  0.904  10-folds crossover test  93.85  95.67  85.41  0.955  Independent test  84.67  84.57  94.26  0.935  Test  ACC (%)  Sp (%)  Sn (%)  AUC  Jackknife test  90.43  87.19  93.68  0.904  10-folds crossover test  93.85  95.67  85.41  0.955  Independent test  84.67  84.57  94.26  0.935  Sp, Sn and AUC denote specificity, sensitivity and area under curve, respectively. Figure 2. View largeDownload slide PPIs between M. grisea and O. sativa. Each node represents a protein, and each edge denotes an interaction. Red and blue nodes are M. grisea and O. sativa proteins, respectively. Figure 2. View largeDownload slide PPIs between M. grisea and O. sativa. Each node represents a protein, and each edge denotes an interaction. Red and blue nodes are M. grisea and O. sativa proteins, respectively. As shown in Figure 2, 27 fungus proteins and 236 rice proteins were involved in the PPIs between M. grisea and O. sativa. This result indicated that one fungus protein averagely had 19 interacting partners from rice, and one rice protein interacted approximately with two pathogen proteins. In total, 137 of 236 O. sativa proteins that interacted with pathogen proteins could be found in the regulatory networks in rice. The average degree and betweenness of proteins in rice regulatory network and the pathogen-targeted proteins in rice regulatory network were 64.02, 1.39E-04 and 164.15, 1.43E-04, respectively. This meant that the potential pathogen-targeted proteins had a higher degree as well as a larger betweenness than other proteins in the network (Wilcoxon rank-sum test, P = 3.431e-14 and 3.158e-14). It was worth to notice here that we focused on the topology of proteins inside rice regulatory network, which were target by blast fungus through PPI, not the PPIs between blast fungus and rice [55]. To determine the functional module of rice proteins interacting with pathogen proteins, GO enrichment of these proteins was analyzed. The results showed that GO terms proteolysis (GO:0006508), vesicle-mediated transport (GO:0016192), exocytosis (GO:0006887), protein transport (GO:0015031), protein kinase activity (GO:0004672) and kinase activity (GO:0016301) were enriched significantly in these 236 rice proteins (Supplementary Table S2). Furthermore, the results in the analysis of GO functional classification implied that these potential pathogen-targeted rice proteins mainly participated in metabolic process, ligand–receptor binding, response to stimulus and function of membrane (Supplementary Figure S1). Modular analysis of regulatory networks in rice in response to biotic stresses By using the FAG-EC algorithm, the regulatory networks in rice in response to biotic stresses were clustered into 228 subnetworks, each of which contained more than six nodes. The subnetworks with maximum and minimum nodes was named Clusters 1 and 228, respectively. We found that Cluster 1 contained 45 potential pathogen-targeted proteins, while Cluster 2, Cluster 3, Cluster 5, Cluster 6, Cluster 7, Cluster 9 and Cluster 10 connect with the predicted PPIs network through one rice protein (Figure 3A). The results also indicated that GO annotations were significantly enriched in the above seven clusters except for Cluster 10 by using the Fisher exact test followed by the FDR correction with P < 0.05 (Supplementary Table S2). Figure 3. View largeDownload slide The sub-regulatory networks in rice affected by blast fungus proteins. (A) Seven main clusters. (B) Rice genes connected to multiple functional subnetworks. Red nodes denote fungus proteins, pink nodes denote fungus interactors and blue nodes denote rice proteins. Figure 3. View largeDownload slide The sub-regulatory networks in rice affected by blast fungus proteins. (A) Seven main clusters. (B) Rice genes connected to multiple functional subnetworks. Red nodes denote fungus proteins, pink nodes denote fungus interactors and blue nodes denote rice proteins. Cluster 1 was the main functional module of rice regulatory networks. There were 3684 genes in the cluster, which was much more than the others. According to the GO enrichment and PANTHER metabolic pathway analysis, we found that 143 GO terms (Supplementary Table S2) and several metabolic pathway were enriched in Cluster 1 (Table 2). The first three enriched pathways were ubiquitin proteasome pathway, apoptosis signaling pathway and general transcription regulation. In addition, those 45 potential pathogen-targeted proteins in Cluster 1 had a higher degree as well as a larger betweenness than other proteins in the network (Wilcoxon rank-sum test, P = 2.156e-10 and 3.221e-10), indicating that it should play an important role in the rice regulatory network. Table 2. PANTHER pathway enrichment analysis in Cluster 1 PANTHER pathway  P-value  Ubiquitin proteasome pathway  2.20E-17  Apoptosis signaling pathway  5.24E-14  General transcription regulation  7.12E-14  Transcription regulation by bZIP transcription factor  1.03E-12  Cell cycle  2.95E-09  Huntington disease  6.06E-09  Inflammation mediated by chemokine and cytokine signaling pathway  1.98E-08  DNA replication  6.98E-08  Tryptophan biosynthesis  1.93E-07  Cytoskeletal regulation by Rho GTPase  2.29E-05  Adenosine triphosphate synthesis  8.81E-05  De novo purine biosynthesis  5.12E-04  Tricarboxylic acid cycle  1.24E-02  Lysine biosynthesis  4.31E-02  De novo pyrimidine deoxyribonucleotide biosynthesis  3.27E-20  PANTHER pathway  P-value  Ubiquitin proteasome pathway  2.20E-17  Apoptosis signaling pathway  5.24E-14  General transcription regulation  7.12E-14  Transcription regulation by bZIP transcription factor  1.03E-12  Cell cycle  2.95E-09  Huntington disease  6.06E-09  Inflammation mediated by chemokine and cytokine signaling pathway  1.98E-08  DNA replication  6.98E-08  Tryptophan biosynthesis  1.93E-07  Cytoskeletal regulation by Rho GTPase  2.29E-05  Adenosine triphosphate synthesis  8.81E-05  De novo purine biosynthesis  5.12E-04  Tricarboxylic acid cycle  1.24E-02  Lysine biosynthesis  4.31E-02  De novo pyrimidine deoxyribonucleotide biosynthesis  3.27E-20  Table 2. PANTHER pathway enrichment analysis in Cluster 1 PANTHER pathway  P-value  Ubiquitin proteasome pathway  2.20E-17  Apoptosis signaling pathway  5.24E-14  General transcription regulation  7.12E-14  Transcription regulation by bZIP transcription factor  1.03E-12  Cell cycle  2.95E-09  Huntington disease  6.06E-09  Inflammation mediated by chemokine and cytokine signaling pathway  1.98E-08  DNA replication  6.98E-08  Tryptophan biosynthesis  1.93E-07  Cytoskeletal regulation by Rho GTPase  2.29E-05  Adenosine triphosphate synthesis  8.81E-05  De novo purine biosynthesis  5.12E-04  Tricarboxylic acid cycle  1.24E-02  Lysine biosynthesis  4.31E-02  De novo pyrimidine deoxyribonucleotide biosynthesis  3.27E-20  PANTHER pathway  P-value  Ubiquitin proteasome pathway  2.20E-17  Apoptosis signaling pathway  5.24E-14  General transcription regulation  7.12E-14  Transcription regulation by bZIP transcription factor  1.03E-12  Cell cycle  2.95E-09  Huntington disease  6.06E-09  Inflammation mediated by chemokine and cytokine signaling pathway  1.98E-08  DNA replication  6.98E-08  Tryptophan biosynthesis  1.93E-07  Cytoskeletal regulation by Rho GTPase  2.29E-05  Adenosine triphosphate synthesis  8.81E-05  De novo purine biosynthesis  5.12E-04  Tricarboxylic acid cycle  1.24E-02  Lysine biosynthesis  4.31E-02  De novo pyrimidine deoxyribonucleotide biosynthesis  3.27E-20  In Cluster 2, 21 GO terms were significantly enriched and 78 genes accounting for 60.93% of all genes were related to metabolic processes (Supplementary Table S2). Blast fungus showed impact on the metabolic process in rice through interacting with Cluster 2, such as hydrolase activity (P = 1.99E-23) and transport (P = 2.84E-18). Cluster 3 showed enrichment in regulatory activity, including regulation of catalytic activity, regulation of molecular function and kinase regulator activity. We note that Cluster 3 is a typical scale-free network (Figure 3A). The central regulatory gene, LOC_Os05g34770, had a high degree and was regulated by pathogenic genes indirectly through four rice proteins. Meanwhile, we found that Clusters 5, 6 and 7 mainly overrepresented peptidase activity, ubiquitin-protein ligase activity and oxidoreductase activity, respectively (Supplementary Table S2). In Cluster 9, receptor-mediated endocytosis, endocytosis and vesicle-mediated transport were enriched, indicating that rice plants could take advantage of those functions to resist pathogens (Supplementary Table S2). From the above seven clusters, 10 genes exist in more than one cluster, and these can be considered as the bottleneck of the network, which might be involved in the multiple cellular processes (Figure 3B). These 10 genes were LOC_Os02g06640, LOC_Os06g40560, LOC_Os01g62500, LOC_Os02g48290, LOC_Os02g10640, LOC_Os01g73530, LOC_Os05g44310, LOC_Os02g54340, LOC_Os06g09290 and LOC_Os03g16920. Master regulators in rice in response to pathogen infection Viper was used to search the master regulators in the regulatory network in rice under the infection of blast fungus. Based on the microarray data and the regulatory networks, we found 1676 (935 upregulated and 741 downregulated) master regulators in rice in response to the infection of M. grisea, among which 34 upregulated and 12 downregulated genes were also those predicted to be pathogen-targeted proteins (Supplementary Table S3). The results in PANTHER pathway analysis indicated that the upregulated master regulators mainly participated in ubiquitin proteasome pathway and apoptosis signaling pathway, while seven proteins were classified as Hsp70s family chaperone. Although there were no enriched PANTHER pathways in downregulated master regulators, Hsp70s family chaperone and hydrolase were also involved. Furthermore, synergistic effect analysis manifested that eight pairs of synergistic genes in the rice regulatory network respond to the infection of blast fungus, and they could regulate the expression of downstream genes (Table 3). Table 3. Synergistic genes in rice in response to the infection of blast fungus Synergistic genes  Predicted score  LOC_Os01g49330-LOC_Os01g70580  1.000000000  LOC_Os01g74190-LOC_Os02g39410  0.001138238  LOC_Os03g13170-LOC_Os03g51690  0.011223978  LOC_Os03g13170-LOC_Os03g55540  0.101105938  LOC_Os03g13170-LOC_Os04g01590  0.012648366  LOC_Os03g51690-LOC_Os04g01590  0.206007994  LOC_Os03g63420-LOC_Os04g42930  0.681000000  LOC_Os01g70580-LOC_Os01g74190- LOC_Os02g39410  0.007490552  Synergistic genes  Predicted score  LOC_Os01g49330-LOC_Os01g70580  1.000000000  LOC_Os01g74190-LOC_Os02g39410  0.001138238  LOC_Os03g13170-LOC_Os03g51690  0.011223978  LOC_Os03g13170-LOC_Os03g55540  0.101105938  LOC_Os03g13170-LOC_Os04g01590  0.012648366  LOC_Os03g51690-LOC_Os04g01590  0.206007994  LOC_Os03g63420-LOC_Os04g42930  0.681000000  LOC_Os01g70580-LOC_Os01g74190- LOC_Os02g39410  0.007490552  Table 3. Synergistic genes in rice in response to the infection of blast fungus Synergistic genes  Predicted score  LOC_Os01g49330-LOC_Os01g70580  1.000000000  LOC_Os01g74190-LOC_Os02g39410  0.001138238  LOC_Os03g13170-LOC_Os03g51690  0.011223978  LOC_Os03g13170-LOC_Os03g55540  0.101105938  LOC_Os03g13170-LOC_Os04g01590  0.012648366  LOC_Os03g51690-LOC_Os04g01590  0.206007994  LOC_Os03g63420-LOC_Os04g42930  0.681000000  LOC_Os01g70580-LOC_Os01g74190- LOC_Os02g39410  0.007490552  Synergistic genes  Predicted score  LOC_Os01g49330-LOC_Os01g70580  1.000000000  LOC_Os01g74190-LOC_Os02g39410  0.001138238  LOC_Os03g13170-LOC_Os03g51690  0.011223978  LOC_Os03g13170-LOC_Os03g55540  0.101105938  LOC_Os03g13170-LOC_Os04g01590  0.012648366  LOC_Os03g51690-LOC_Os04g01590  0.206007994  LOC_Os03g63420-LOC_Os04g42930  0.681000000  LOC_Os01g70580-LOC_Os01g74190- LOC_Os02g39410  0.007490552  Finally, we compared the different master regulators in rice in response to the infection of M. grisea, Xoo and RSV. There were 1053 and 715 master regulators in rice under the infection of Xoo and RSV, respectively. Taken into account of 1676 master regulators in rice under the infection of M. grisea, total 150 master regulators were common in rice in response to these three kinds of infection, 341 common in M. grisea and RSV, 587 in M. grisea and Xoo, while 235 in RSV and Xoo (Figure 4). Figure 4. View largeDownload slide The Venn diagram of the master regulators in rice in response to the infection of three pathogens. MGG, XOO and RSV denote M. grisea, Xoo and RSV, respectively. Figure 4. View largeDownload slide The Venn diagram of the master regulators in rice in response to the infection of three pathogens. MGG, XOO and RSV denote M. grisea, Xoo and RSV, respectively. Discussion Considering the critical role of PPIs in the blast fungus infection process, a protocol pipeline including three methods was used sequentially to predict and verify the PPIs between M. grisea and O. sativa in this article. First, the potential PPIs between M. grisea and O. sativa were predicted by using the method based on interolog. This method has been widely used in biological research work, such as prediction of PPIs in rice genome [56] and interacting network of Hsp70s in rice [29]. However, most previous researches predicted PPIs in intraspecies genome by using the experimentally identified PPIs as templates. In this research, experimentally identified PPIs including 63 PPIs between host and pathogen were downloaded and used as templates to predict PPIs between interspecies, M. grisea and O. sativa. Second, the domain-based method was applied as a complementary method to confirm the predicted PPIs between M. grisea and O. sativa. Recently, PPIs between R. solanacearum and A. thaliana were predicted by using the above two methods [30]. Third, SignalP and WolFPSORT were subsequently used to identify the secretory proteins in M. grisea, which were involved in the above predicted PPIs because the secretory proteins in pathogens were considered to interact with receptors in plants [30]. Only the cross-verified PPIs in these three methods were selected for the downstream analysis. Finally, we achieved 523 PPIs between M. grisea and O. sativa, among which 470 PPIs were predicted by the 63 PPIs templates between host and pathogen. The 523 PPIs between M. grisea and O. sativa were further confirmed. Although some methods including the similarity of GO annotation [56], subcellular localization and gene co-expression [29] had been used to evaluate the PPIs in intraspecies, they might not be adapted for PPIs between interspecies. In this study, we used the two methods that were introduced by Shen et al. and He et al., respectively [13, 32], to confirm the PPIs between M. grisea and O. sativa. The method of Shen et al. [32] was applied to investigate whether the protein sequences in PPI pairs could be statistically classified, while the method of He et al. [13] was used to figure out the enrichment of pathogenic proteins in the PPIs. The results in these two methods all supported the above predicted PPIs between M. grisea and O. sativa. In summary, 523 PPIs between M. grisea and O. sativa achieved in this research work were cross-verified by using three methods and were confirmed again by two strategies. The results in GO annotations analysis on the rice proteins involved in 523 PPIs between M. grisea and O. sativa showed that the most enrichment of GO annotation of these proteins was related to the function of vesicle-mediated transport and exocytosis, which could drive focal and/or nondirectional secretion of antimicrobial cocktails to resist microbial pathogens infection [57]. Alternatively, pathogenic proteins in fungus interacted with these vesicle-mediated transport and exocytosis-related proteins intercept the secretion machinery by blocking vesicle formation from intracellular membranes [57]. This might be one of the models showing the balance between immune responses of rice plants and counter defense of blast fungus. Receptor kinase function was enriched in the rice proteins involved in the 523 PPIs between blast fungus and rice. Receptor kinases in rice plants could recognize the conservative pathogenic molecules and subsequently active intracellular signaling pathways [58]. In summary, during the infection process of blast fungus, the kinase receptor-mediated recognition of rice plants detected the presence of blast fungus and triggered immune responses including activating the vesicle-associated exocytosis pathways. Analysis using microarrays on the regular network of rice in response to blast fungus infection showed that 935 master regulators were upregulated by blast fungus infection, among which 34 proteins interacted with blast fungus proteins. In total, 7 of those 34 master regulators belonged to the Hsp70s family, which could help hosts to resist environmental stresses [59]. This was consistent with previous research work where Hsp70s was upregulated in rice under high temperature [60], and was involved in macromolecular translocation, carbohydrate metabolism, innate immunity, photosystem II repair and regulation of kinase activities [29]. Meanwhile, another 2 of the 34 master regulators, LOC_Os01g08400.1 and LOC_Os01g73300.1, were annotated as SNARE proteins, which mediated the exocytosis in plant cell and the fusion between intracellular vesicles and cytomembrane [61]. Our results suggest that Hsp70s and SNARE proteins in rice were upregulated by the infection of blast fungus and then induce the downstream immune system in rice. In this research work, GO and PANTHER were used to explore the biological functions of subnetworks in rice, which were affected by blast fungus through the PPIs. In Cluster 1, there were several immune-related metabolic pathways, including the protein enzyme and apoptosis signaling pathways. Meanwhile, we found that the ubiquitin proteasome pathway was enriched, respectively, in rice under the infection of M. grisea, Xoo or RSV, further illustrating that ubiquitin proteasome pathway was the common response of rice plants to the infection of fungi, bacteria and viruses. It was reported that ubiquitin proteasome pathway was the key step in the process of plant immune response [62], while apoptosis signaling pathway was the final presentation forms of plant ETI pathway [11]. The results in this study implied that the pathogens inhibit the rice plant immune process through PPIs by disrupting hosts’ ubiquitination [63]. We found 10 bottleneck proteins related to multiple subnetworks, which might contribute to the regulation of metabolic process [64]. In total, 4 of 10 bottleneck proteins including LOC_Os06g4056, LOC_Os06g09290, LOC_Os02g54340 and LOC_Os02g10640, which were participating in ubiquitin protease pathway, were all connected to Cluster 5. The Cluster 5 subnetwork was related to proteolysis, peptidase and hydrolase activity. Furthermore, LOC_Os02g54340 and LOC_Os02g10640 were simultaneously connected to Cluster 6, which accounted for ubiquitin-protein ligase. These indicated that the interactions between fungus proteins and the bottleneck proteins showed impact on hydrolase and peptidase activity of rice plants, which might also be connected to immune metabolism [58]. Therefore, these 10 bottleneck proteins in rice might be the infecting target of blast fungus. Conclusion Using two well-known PPI prediction methods, 523 potential PPIs between M. grisea and O. sativa were predicted. We used jackknife test, 10-fold crossover test and independent test based on SVM to further confirm the PPIs with ACC of 90.34, 93.85 and 84.67%, respectively. Then, the regulatory network of rice, RiceNet data, was clustered into 228 subnetworks with over six nodes and the top seven clusters affected by blast fungus through PPIs was analyzed. In addition, 46 master regulators interacting with the fungus proteins, including 34 upregulated and 12 downregulated regulators, were observed under the infection of M. grisea. Our results uncover the interacting mechanism between M. grisea and O. sativa, and provide researchers a framework of validation for future experimental work. Key Points According to the classical gene-for-gene system, the key factor causing rice blast disease and defense might be PPIs between rice and fungus. After cross-verified by using three methods and confirmed again by two strategies, 523 PPIs between blast fungus and rice were achieved in this research work. The regulatory networks in rice in response to biotic stresses were divided into 228 subnetworks with over six nodes and the top seven subnetworks affected by blast fungus through PPIs were investigated. The result indicated that 34 upregulated and 12 downregulated master regulators in rice interacting with the fungus proteins in response to the infection of blast fungus. Supplementary Data Supplementary data are available online at http://bib.oxfordjournals.org/. Shiwei Ma is a PhD student at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Qi Song is an MSc student at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Huan Tao is a Lecturer at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Andrew Harrison is a Senior Lecturer at Department of Mathematical Sciences, University of Essex, UK. Shaobo Wang is an MSc student at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Wei Liu is an Associate Professor at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Shoukai Lin is a Lecturer at Fujian Provincial Key Laboratory of Ecology-toxicological Effects and Control for Emerging Contaminants, Putian University, China Ziding Zhang is a Professor of bioinformatics at the College of Biological Sciences, China Agriculture University, China. Shoukai Lin is a Lecturer at Fujian Provincial Key Laboratory of Ecology-toxicological Effects and Control for Emerging Contaminants, Putian University, China Yufang Ai is an Associate Professor at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Huaqin He is a Professor of bioinformatics at the College of Life Sciences, Fujian Agriculture and Forestry University, China. Acknowledgement The authors thank the anonymous referees whose constructive comments were helpful in improving the quality of this work. Funding Natural Science Foundation of China and Fujian (grant numbers 31270454, 81502091 and 2013J01077, 2014N5006), Innovative Foundation of FAFU (grant numbers CXZX2017132 and CXZX2017303) and Fujian-Taiwan Joint Innovative Centre for Germplasm Resources and Cultivation of Crop (grant number 2015-75. FJ 2011 Program). References 1 Dean RA, Talbot NJ, Ebbole DJ, et al.   The genome sequence of the rice blast fungus Magnaporthe grisea. Nature  2005; 434: 980– 6. Google Scholar CrossRef Search ADS PubMed  2 Parker D, Beckmann M, Enot DP, et al.   Rice blast infection of Brachypodium distachyon as a model system to study dynamic host/pathogen interactions. Nat Protoc  2008; 3: 435– 45. Google Scholar CrossRef Search ADS PubMed  3 Xue MF, Yang J, Li ZG, et al.   Comparative analysis of the genomes of two field isolates of the rice blast fungus Magnaporthe oryzae. PLoS Genet  2012; 8: e1002869. Google Scholar CrossRef Search ADS PubMed  4 Ribot C, Hirsch J, Balzergue S, et al.   Susceptibility of rice to the blast fungus, Magnaporthe grisea. J Plant Physiol  2008; 165: 114– 24. Google Scholar CrossRef Search ADS PubMed  5 Hulbert SH, Webb CA, Smith SM, et al.   Resistance gene complexes: evolution and utilization. Annu Rev Phytopathol  2001; 39: 285– 312. Google Scholar CrossRef Search ADS PubMed  6 Tao Y, Xie Z, Chen W, et al.   Quantitative nature of Arabidopsis responses during compatible and incompatible interactions with the bacterial pathogen Pseudomonas syringae. Plant Cell  2003; 15: 317– 30. Google Scholar CrossRef Search ADS PubMed  7 Kankanala P, Czymmek K, Valent B. Roles for rice membrane dynamics and plasmodesmata during biotrophic invasion by the blast fungus. Plant Cell  2007; 19: 706– 24. Google Scholar CrossRef Search ADS PubMed  8 Talbot NJ, Ebbole DJ, Hamer JE. Identification and characterization of MPG1, a gene involved in pathogenicity from the rice blast fungus Magnaporthe grisea. Plant Cell  1993; 5: 1575– 90. Google Scholar CrossRef Search ADS PubMed  9 Xu JR, Zhao X, Dean RA. From genes to genomes: a new paradigm for studying fungal pathogenesis in Magnaporthe oryzae. Adv Genet  2007; 57: 175– 218. Google Scholar PubMed  10 Tucker SL, Talbot NJ. Surface attachment and pre-penetration stage development by plant pathogenic fungi. Annu Rev Phytopathol  2001; 39: 385– 417. Google Scholar CrossRef Search ADS PubMed  11 Jones JD, Dangl JL. The plant immune system. Nature  2006; 444: 323– 9. Google Scholar CrossRef Search ADS PubMed  12 Bishop JG, Dean AM, Mitchell-Olds T. Rapid evolution in plant chitinases: molecular targets of selection in plant-pathogen coevolution. Proc Natl Acad Sci USA  2000; 97: 5322– 7. Google Scholar CrossRef Search ADS PubMed  13 He F, Zhang Y, Chen H, et al.   The prediction of protein-protein interaction networks in rice blast fungus. BMC Genomics  2008; 9: 519. Google Scholar CrossRef Search ADS PubMed  14 Kim JG, Park D, Kim BC, et al.   Predicting the interactome of Xanthomonas oryzae pathovar oryzae for target selection and DB service. BMC Bioinformatics  2008; 9: 41. Google Scholar CrossRef Search ADS PubMed  15 Kawahara Y, de la Bastide M, Hamilton JP, et al.   Improvement of the Oryza sativa Nipponbare reference genome using next generation sequence and optical map data. Rice  2013; 6: 4. Google Scholar CrossRef Search ADS PubMed  16 Bogdanove AJ. Protein-protein interactions in pathogen recognition by plants. Plant Mol Biol  2002; 50: 981– 9. Google Scholar CrossRef Search ADS PubMed  17 Jia Y, McAdams SA, Bryan GT, et al.   Direct interaction of resistance gene and avirulence gene products confers rice blast resistance. Embo J  2000; 19: 4004– 14. Google Scholar CrossRef Search ADS PubMed  18 Kanzaki H, Yoshida K, Saitoh H, et al.   Arms race co-evolution of Magnaporthe oryzae AVR-Pik and rice Pik genes driven by their physical interactions. Plant J  2012; 72: 894– 907. Google Scholar CrossRef Search ADS PubMed  19 Cesari S, Thilliez G, Ribot C, et al.   The rice resistance protein pair RGA4/RGA5 recognizes the Magnaporthe oryzae effectors AVR-Pia and AVR1-CO39 by direct binding. Plant Cell  2013; 25: 1463– 81. Google Scholar CrossRef Search ADS PubMed  20 Coelho ED, Arrais JP, Matos S, et al.   Computational prediction of the human-microbial oral interactome. BMC Syst Biol  2014; 8: 24. Google Scholar CrossRef Search ADS PubMed  21 Barker D, Pagel M. Predicting functional gene links from phylogenetic-statistical analyses of whole genomes. PLoS Comput Biol  2005; 1: e3. Google Scholar CrossRef Search ADS PubMed  22 Najafabadi HS, Salavati R. Sequence-based prediction of protein-protein interactions by means of codon usage. Genome Biol  2008; 9: R87. Google Scholar CrossRef Search ADS PubMed  23 Aloy P, Russell RB. Interrogating protein interaction networks through structural biology. Proc Natl Acad Sci USA  2002; 99: 5896– 901. Google Scholar CrossRef Search ADS PubMed  24 Huang C, Morcos F, Kanaan SP, et al.   Predicting protein-protein interactions from protein domains using a set cover approach. IEEE/ACM Trans Comput Biol Bioinform  2007; 4: 78– 87. Google Scholar CrossRef Search ADS PubMed  25 Wang RS, Wang Y, Wu LY, et al.   Analysis on multi-domain cooperation for predicting protein-protein interactions. BMC Bioinformatics  2007; 8: 391. Google Scholar CrossRef Search ADS PubMed  26 Zhang YQ, Zhang DL, Mi G, et al.   Using ensemble methods to deal with imbalanced data in predicting protein-protein interactions. Comput Biol Chem  2012; 36: 36– 41. Google Scholar CrossRef Search ADS PubMed  27 Maetschke SR, Simonsen M, Davis MJ, et al.   Gene Ontology-driven inference of protein-protein interactions using inducers. Bioinformatics  2012; 28: 69– 75. Google Scholar CrossRef Search ADS PubMed  28 Jaeger S, Gaudan S, Leser U, et al.   Integrating protein-protein interactions and text mining for protein function prediction. BMC Bioinformatics  2008; 9 (Suppl 8): S2. Google Scholar CrossRef Search ADS PubMed  29 Wang Y, Lin S, Song Q, et al.   Genome-wide identification of heat shock proteins (Hsps) and Hsp interactors in rice: Hsp70s as a case study. BMC Genomics  2014; 15: 344. Google Scholar CrossRef Search ADS PubMed  30 Li ZG, He F, Zhang ZD, et al.   Prediction of protein-protein interactions between Ralstonia solanacearum and Arabidopsis thaliana. Amino Acids  2012; 42: 2363– 71. Google Scholar CrossRef Search ADS PubMed  31 Ng SK, Zhang Z, Tan SH. Integrative approach for computationally inferring protein domain interactions. Bioinformatics  2003; 19: 923– 9. Google Scholar CrossRef Search ADS PubMed  32 Shen J, Zhang J, Luo X, et al.   Predicting protein-protein interactions based only on sequences information. Proc Natl Acad Sci USA  2007; 104: 4337– 41. Google Scholar CrossRef Search ADS PubMed  33 Tully JP, Hill AE, Ahmed HM, et al.   Expression-based network biology identifies immune-related functional modules involved in plant defense. BMC Genomics  2014; 15: 421. Google Scholar CrossRef Search ADS PubMed  34 Wang J, Li M, Deng Y, et al.   Recent advances in clustering methods for protein interaction networks. BMC Genomics  2010; 11 (Suppl 3): S10. Google Scholar PubMed  35 Chan SS, Kyba M. What is a master regulator? J Stem Cell Res Ther  2013; 3: 114. Google Scholar PubMed  36 Lefebvre C, Rajbhandari P, Alvarez MJ, et al.   A human B-cell interactome identifies MYB and FOXM1 as master regulators of proliferation in germinal centers. Mol Syst Biol  2010; 6: 377. Google Scholar CrossRef Search ADS PubMed  37 Liu T, Lin Y, Wen X, et al.   BindingDB: a web-accessible database of experimentally determined protein-ligand binding affinities. Nucleic Acids Res  2007; 35: D198– 201. Google Scholar CrossRef Search ADS PubMed  38 Licata L, Briganti L, Peluso D, et al.   MINT, the molecular interaction database: 2012 update. Nucleic Acids Res  2012; 40: D857– 61. Google Scholar CrossRef Search ADS PubMed  39 Salwinski L, Miller CS, Smith AJ, et al.   The Database of Interacting Proteins: 2004 update. Nucleic Acids Res  2004; 32: D449– 51. Google Scholar CrossRef Search ADS PubMed  40 Poole RL. The TAIR database. Methods Mol Biol  2007; 406: 179– 212. Google Scholar PubMed  41 Kerrien S, Aranda B, Breuza L, et al.   The IntAct molecular interaction database in 2012. Nucleic Acids Res  2012; 40: D841– 6. Google Scholar CrossRef Search ADS PubMed  42 Kumar R, Nanduri B. HPIDB–a unified resource for host-pathogen interactions. BMC Bioinformatics  2010; 11 (Suppl 6): S16. Google Scholar CrossRef Search ADS PubMed  43 Krogh A, Larsson B, von Heijne G, et al.   Predicting transmembrane protein topology with a hidden Markov model: application to complete genomes. J Mol Biol  2001; 305: 567– 80. Google Scholar CrossRef Search ADS PubMed  44 Kawahara Y, Oono Y, Kanamori H, et al.   Simultaneous RNA-seq analysis of a mixed transcriptome of rice and blast fungus interaction. PLoS One  2012; 7: e49423. Google Scholar CrossRef Search ADS PubMed  45 Bendtsen JD, Nielsen H, von Heijne G, et al.   Improved prediction of signal peptides: SignalP 3.0. J Mol Biol  2004; 340: 783– 95. Google Scholar CrossRef Search ADS PubMed  46 Horton P, Park KJ, Obayashi T, et al.   WoLF PSORT: protein localization predictor. Nucleic Acids Res  2007; 35: W585– 7. Google Scholar CrossRef Search ADS PubMed  47 Smoot ME, Ono K, Ruscheinski J, et al.   Cytoscape 2.8: new features for data integration and network visualization. Bioinformatics  2011; 27: 431– 2. Google Scholar CrossRef Search ADS PubMed  48 Fawcett T. An introduction to ROC analysis. Pattern Recogn Lett  2006; 27: 861– 74. Google Scholar CrossRef Search ADS   49 Liu B, Wang S, Long R, et al.   iRSpot-EL: identify recombination spots with an ensemble learning approach. Bioinformatics  2017; 33: 35– 41. Google Scholar CrossRef Search ADS PubMed  50 Chou KC, Shen HB. Recent progress in protein subcellular location prediction. Anal Biochem  2007; 370: 1– 16. Google Scholar CrossRef Search ADS PubMed  51 Fan GL, Zhang XY, Liu YL, et al.   DSPMP: discriminating secretory proteins of malaria parasite by hybridizing different descriptors of Chou's pseudo amino acid patterns. J Comput Chem  2015; 36: 2317– 27. Google Scholar CrossRef Search ADS PubMed  52 Urban M, Cuzick A, Rutherford K, et al.   PHI-base: a new interface and further additions for the multi-species pathogen-host interactions database. Nucleic Acids Res  2017; 45: D604– 10. Google Scholar CrossRef Search ADS PubMed  53 Assenov Y, Ramirez F, Schelhorn SE, et al.   Computing topological parameters of biological networks. Bioinformatics  2008; 24: 282– 4. Google Scholar CrossRef Search ADS PubMed  54 Mi H, Muruganujan A, Thomas PD. PANTHER in 2013: modeling the evolution of gene function, and other gene attributes, in the context of phylogenetic trees. Nucleic Acids Res  2013; 41: D377– 86. Google Scholar CrossRef Search ADS PubMed  55 Dobay MP, Stertz S, Delorenzi M. Context-based retrieval of functional modules in protein-protein interaction networks. Brief Bioinform  2017. doi: 10.1093/bib/bbx029. 56 Gu H, Zhu P, Jiao Y, et al.   PRIN: a predicted rice interactome network. BMC Bioinformatics  2011; 12: 161. Google Scholar CrossRef Search ADS PubMed  57 Kwon C, Bednarek P, Schulze-Lefert P. Secretory pathways in plant immune responses. Plant Physiol  2008; 147: 1575– 83. Google Scholar CrossRef Search ADS PubMed  58 Chen X, Ronald PC. Innate immunity in rice. Trends Plant Sci  2011; 16: 451– 9. Google Scholar CrossRef Search ADS PubMed  59 Morano KA. New tricks for an old dog: the evolving world of Hsp70. Ann N Y Acad Sci  2007; 1113: 1– 14. Google Scholar CrossRef Search ADS PubMed  60 Chen X, Lin S, Liu Q, et al.   Expression and interaction of small heat shock proteins (sHsps) in rice in response to heat stress. Biochim Biophys Acta  2014; 1844: 818– 28. Google Scholar CrossRef Search ADS PubMed  61 Collins NC, Thordal-Christensen H, Lipka V, et al.   SNARE-protein-mediated disease resistance at the plant cell wall. Nature  2003; 425: 973– 7. Google Scholar CrossRef Search ADS PubMed  62 Duplan V, Rivas S. E3 ubiquitin-ligases and their target proteins during the regulation of plant innate immunity. Front Plant Sci  2014; 5: 42. Google Scholar CrossRef Search ADS PubMed  63 Craig A, Ewan R, Mesmar J, et al.   E3 ubiquitin ligases and plant innate immunity. J Exp Bot  2009; 60: 1123– 32. Google Scholar CrossRef Search ADS PubMed  64 Yu H, Kim PM, Sprecher E, et al.   The importance of bottlenecks in protein networks: correlation with gene essentiality and expression dynamics. PLoS Comput Biol  2007; 3: e59. Google Scholar CrossRef Search ADS PubMed  © The Author 2017. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com

Journal

Briefings in BioinformaticsOxford University Press

Published: Oct 11, 2017

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off