Funct Integr Genomics DOI 10.1007/s10142-016-0481-4
ORIGINAL ARTICLE
Topological characteristics of target genes regulated by abiotic-stress-responsible miRNAs in a rice interactome network Linzhong Zhang 1 & Hongdong Xuan 2 & Yongchun Zuo 3 & Gaojian Xu 2 & Ping Wang 1 & Youhong Song 4 & Shihua Zhang 1,5
Received: 6 August 2015 / Revised: 18 January 2016 / Accepted: 20 January 2016 # Springer-Verlag Berlin Heidelberg 2016
Abstract A great number of microRNAs (miRNAs) have been identified in responding and acting in gene regulatory networks associated with plant tolerance to abiotic stress conditions, such as drought, salinity, and high temperature. The topological exploration of target genes regulated by abioticstress-responsible miRNAs (ASRmiRs) in a network facilitates to discover the molecular basis of plant abiotic stress response. This study was based on the staple food rice (Oryza sativa) in which ASRmiRs were manually curated. After having compared the topological properties of target genes (stress-miR-targets) with those (non-stress-miR-targets) not regulated by ASRmiRs in a rice interactome network, we found that stress-miR-targets exhibited distinguishable topological properties. The interaction probability analysis and kcore decomposition showed that stress-miR-targets preferentially interacted with non-stress-miR-targets and located at the Linzhong Zhang and Hongdong Xuan contributed equally to this work. Electronic supplementary material The online version of this article (doi:10.1007/s10142-016-0481-4) contains supplementary material, which is available to authorized users. * Shihua Zhang [email protected]
1
School of Science, Anhui Agricultural University, Hefei 230036, China
2
College of Information and Computer Science, Anhui Agricultural University, Hefei 230036, China
3
College of Life Sciences, Inner Mongolia University, Hohhot 010021, China
4
School of Agronomy, Anhui Agricultural University, Hefei 230036, China
5
State Key Laboratory of Tea Plant Biology and Utilization, Anhui Agricultural University, Hefei 230036, China
peripheral positions in the network. Our results indicated an obvious topological distinction between the two types of genes, reflecting the specific mechanisms of action of stressmiR-targets in rice abiotic stress response. Also, the results may provide valuable clues to elucidate molecular mechanisms of crop response to abiotic stress. Keywords Rice . Abiotic stress . miRNA . Target gene . Topological analysis
Introduction Generally, plant growth and productivity are greatly affected by abiotic stresses, such as drought, salinity, and high temperature (Budak et al. 2015a). MicroRNAs (miRNAs) are a class of endogenous non-coding small RNAs that posttranscriptionally control gene expression via either translational repression or mRNA degradation (Bartel 2004). Initially identified as regulators of developmental timing in Caenorhabditis elegans, the investigations of the functional mechanisms of these small RNAs have been an attractive territory (Shah and Chen 2014; Rogers and Chen 2013). In the kingdom of plant species, many studies have shown that miRNAs play important roles in regulating many biological processes, including leaf morphogenesis, vascular development, floral organ identity, and abiotic/biotic stress response (Budak et al. 2015b; Xie et al. 2015). During abiotic stress conditions, miRNAs act post-transcriptionally in gene regulatory networks pertaining to plant stress response (Sunkar et al. 2012). Under these circumstances, the action of target genes of abiotic-stress-responsible miRNAs (ASRmiRs) can be elaborately modified in a network fashion to facilitate stress tolerance at physiological, cellular, biochemical, and molecular levels (Urano et al. 2010; Cabello et al. 2014), shedding
Funct Integr Genomics
light in the disclosure of plant adaption mechanisms in view of target genes of ASRmiRs. To our knowledge, cellular systems featured with extremely complicated interacting networks of nucleotides, proteins, metabolites, and other small molecules (Barabasi and Oltvai 2004). The topological exploration of these networks is of great value in understanding the mechanisms of cellular action (Jeong et al. 2001; Kuchaiev et al. 2010). In the diverse interacting networks, protein-protein interactions (PPIs) are crucial for almost all levels of cellular functions, including genetic regulation, metabolism, and signal transduction (Morsy et al. 2008). In the past decade, miRNAs have emerged as primary regulators of gene expression controlling plant adaptation to stress conditions (Mallory and Vaucheret 2006). Therefore, the topological investigation of miRNA targets facilitates to unravel the structural and dynamical mechanisms of the interactome network in the control of ASRmiRs. Rice (Oryza sativa) is one of the staple food crops with rich genomic and proteomic data are available. Therefore, based on the rice species, we manually curated a complete list of ASRmiRs and computationally predicted their target genes. The topological properties of target genes (stress-miR-targets) regulated by ASRmiRs and those (non-stress-miR-targets) not regulated by ASRmiRs were compared in a predicted rice interactome network, showing that stress-miR-targets exhibited distinguishable topological properties from non-stressmiR-targets. In the rice interactome network, based on the interaction probability analysis, stress-miR-targets were shown to preferentially interact with non-stress-miR-targets. In addition, the k-core decomposition of the network revealed that stress-miR-targets might locate at the peripheral positions. All these findings indicated an evident topological distinction between the two types of genes, revealing the underlying mechanisms of actions of stress-miR-targets in rice abiotic stress response.
Materials and methods Prediction of a rice interactome network Several high-throughput experimental technologies, such as yeast two-hybrid system (Y2H) (Bartel and Fields 1997) and bimolecular fluorescence complementation (BiFC) (Wong and O’Bryan 2011), were usually used to chart the interactome (protein-protein interaction (PPI)) networks of model species, including C. elegans, Drosophila melanogaster, and Saccharomyces cerevisiae (Uetz et al. 2000; Zhong and Sternberg 2006; Giot et al. 2003). However, these largescale experiment strategies are usually cost-intensive and time-consuming. To overcome such limitations, a number of computational methods have been developed for the prediction of interactome network, such as phylogenetic profile,
correlated mutation, and machine learning (Cheng and Perocchi 2015; Pazos et al. 1997; Shen et al. 2007). Among these, we applied a widely used interolog method to predict the rice interactome network (Matthews et al. 2001). This approach is based on a simple principle: if two proteins (A and B) in rice are orthologous with two proteins (A’ and B’), respectively, in another model species, and the interaction between A’ and B’ has been experimentally validated; in such case, A and B can be predicted to interact with each other. In this study, as illustrated in Fig. 1, we chose seven model species including Homo sapiens, Arabidopsis thaliana, C. elegans, D. melanogaster, Escherichia coli K12, Mus musculus, and S. cerevisiae, and their experimentally validated PPI data were retrieved from publically available databases: BioGrid, MINT, DIP, and IntAct (Stark et al. 2006; Chatr-Aryamontri et al. 2007; Salwinski et al. 2004; Aranda et al. 2010). The orthologous relations between rice and the seven model species were decided using the reciprocal best-hit approach proposed by Brown and Jurisica (2005). The highquality PPI data is undoubtedly critical for the identification of basic protein functions and mechanisms of action. Therefore, we used two indexes of subcellular co-localization and coexpression, described by our colleagues, respectively (Brandao et al. 2009; De Bodt et al. 2009), as the quality control for the predicted PPI data. In principle, interacting protein pairs sharing the same subcellular compartment or having the co-expression correlation coefficient score greater than 0.5 were retained for further use. Definition of interaction probability A wide range of biological functions are determined by crosstalking proteins rather than isolated individuals, and different types of protein interactions in the network constitute the basic driving force in cellular operations (Pawson 2004). Therefore, the analysis of interaction probability among different types of proteins is beneficial for the dissection of protein functions. In the rice interactome network, three kinds of interactions among the two types of proteins (or genes, as stress-miR-targets and nonstress-miR-targets) were illustrated as interaction between the same type of genes (stress-miR-targets interact with stress-miRtargets, non-stress-miR-targets interact with non-stress-miR-targets) and interaction between different types of genes (stressmiR-targets interact with non-stress-miR-targets). Based on this, we defined the interaction probability among stress-miR-targets and non-stress-miR-targets as follows: 8 Nij > > < 2 Cn Pði; jÞ ¼ Nij > > : ni n j
if i and j belong to the same type of genes ðcase 1Þ if i and j belong to different types of genes ðcase 2Þ
In this formula, the two indexes i and j denote one of the two gene types (stress-miR-targets or non-stress-miR-targets).
Funct Integr Genomics
Fig. 1 The flowchart that illustrated the prediction of the rice interactome network using an interolog method
Nij represents the actual interaction number among the same type of genes (case 1) or different types of genes (case 2) in the Þ rice interactome network. C2n equals to nðn−1 2 , denoting all the possible interaction number in one of the two gene types, where n is the total gene number of this gene type. If i and j belong to different gene types (case 2), ni × nj is all the possible interaction number between the two types of genes, where, ni and nj denote the total gene number of stress-miR-targets and non-stress-miR-targets, respectively.
Network measurements describing the topological properties The exploration of topological properties for stress-miR-targets in the rice interactome network enables the dissection of rice
stress response mechanisms. For this purpose, two types of network measurements were predefined for the description of their topological properties in the interactome network. The first type describes the topological feature of individual gene in the local interactome network, such as degree, clustering coefficient (CC), betweenness (BC), and topological coefficient (TC). The second type describes the relationship between an individual gene and the other particular gene group in the global interactome network, such as 1N index, the shortest distance to the group of stress-miR-targets (SDT), and the average distance to the group of stress-miR-targets (ADT). The definitions of these measurements are detailed in Table 1. Degree of a node in the network can be determined by the number of edges directly linking to the node. Current efforts have shown the correlation between degree and essentiality (act as Bhubs^) of a certain node, and degree can be regarded
Funct Integr Genomics
as a predictor of essentiality in the network (Yu et al. 2007). The measurement of CC is used to characterize the modular property of a network, and it can be calculated as the fraction of the actual edges between the neighbors of a node against the possible maximum connections. BC is measured as the proportion of shortest paths between paired nodes in the network that a given node locates on. The centrality of an interested node can be characterized by the BC measurement, and a bottleneck node in the network can be considered if the node has a high BC. TC is a conventional measurement that describes the extent to which a node shares neighbor nodes with other individual nodes in the network. k-core decomposition analysis We used the k-core decomposition method, proposed by our colleagues and successfully performed in certain species (Wuchty and Almaas 2005), to disclose the inherent hierarchical structure of the rice interactome network. This is useful in dissecting the topological importance of stress-miR-targets in the global network. The k-core algorithm aims to decompose a network into k-cores (k-subgraphs) by iteratively removing all the nodes with the degree less than k, resulting in a spectrum of subgraphs (k-cores) that gradually display the backbone of the network. For each of the k-iterations, the remaining subgraph with every node having a degree of no less than k is called the kcore of the network. In these k-cores, we respectively calculated the proportion of stress-miR-targets to all the nodes in each kcore. This serial of proportion values can be used to describe the enrichment extent to which stress-miR-targets are represented in subgraphs (k-cores) of the rice interactome network. Network visualization and analysis We used the Cytoscape software (Smoot et al. 2011) with the force-directed layout to visualize the interaction relationships among stress-miR-targets and non-stress-miR-targets in the rice interactome network. The first type of network measurements, including degree, CC, BC, and TC, were calculated using the Cytoscape plugin NetworkAnalyzer (Assenov et al. 2008). We calculated the second type of network measurements (1N index, SDT, and ADT) using our in-house Matlab scripts. In addition, we used the Cytoscape plugin MCODE (Bader and Hogue 2003) to identify gene modules in the rice interactome network.
Results
literature mining approach (Zhang et al. 2013). In this study, we used a similar approach for the rice species. To obtain a complete list of ASRmiRs in rice, we conducted an extensive literature mining as that of the PASmiR project to discern published articles that give experimental evidence for rice miRNAs in response to certain abiotic stress. We first searched in Google Scholar, Scopus, and PubMed (Henderson 2005; Burnham 2006; Wheeler et al. 2007) in retrieving relevant articles using keyword Brice AND microRNA AND abiotic stress.^ Then, we manually reviewed the full text if the two sections of title and abstract of an article provided the evidence that the article described the role of rice miRNAs involved in abiotic stress response. Finally, we collected the relevant publications that assessed the up-/down-regulated expression pattern of rice miRNAs during certain abiotic stress conditions. As a result, 184 rice ASRmiRs were retrieved from the original publications, and 29 (15.76 %) miRNAs are involved in an average of 3–5 abiotic stress conditions (Supplementary File 1). Prediction of target genes regulated by rice ASRmiRs Using the interolog method, we computationally predicted the rice interactome network, which represented 22,933 nonredundant interacting pairs covering 2975 unique proteins under the two indexes of quality control. In this network, genes were divided into two main types: genes (stress-miR-targets) regulated by ASRmiRs and those (non-stress-miR-targets) not regulated by ASRmiRs. To this end, cDNA sequences of these 2975 genes were retrieved from the website of the international rice genome sequencing project (IRGSP) (Kawahara et al. 2013). We downloaded mature sequences of the 184 rice ASRmiRs from miRBase (Griffiths-Jones et al. 2006) and used the web tool psRNATarget (Dai and Zhao 2011) for the miRNA target identification. The psRNATarget tool was designed as a successful plant small RNA target analysis server, which can be specifically used for miRNA target prediction in the plant community. Of the total 2975 genes in the rice interactome network, 776 (26.1 %, namely stress-miR-targets) were predicted to be targeted by ASRmiRs (Supplementary File 2). We used the Cytoscape software to visualize the rice interactome network where a circular node represented a gene and an edge represented the interacting relationship between two different genes (Fig. 2a). To demonstrate the network more clearly, the two types of genes of stress-miR-targets and non-stressmiR-targets were marked with different colors in the rice interactome network. Interaction probability among stress-miR-targets and non-stress-miR-targets
Curation of rice ASRmiRs We have previously completed a PASmiR project by having collected abiotic-stress-responsible miRNAs in plants using a
We compared the probability of the three kinds of interactions, and the results showed that stress-miR-targets had the lower tendency to interact among themselves in the network
Funct Integr Genomics Table 1
Definitions of the seven network measurements
Typea
Network measurements
Function
Brief description
Type 1
Degree Clustering coefficient (CC)
ki
The number of links to node i. ki is the number of neighbors of node i, and ni is the number of connected pairs between all the neighbors of node i. σkj denotes shortest paths between node pairs k and j, σkij denotes that pass through node i. J(i, j) is the number of neighbors shared between the node i and node j, plus one if there is a direct link between i and j. avg(J(i, j)) is the average value of J(i, j). ki is degree of node i. kpi is the number of links between node i and stress-miR-targets.
2ni k i ðk i −1Þ
Betweenness centrality (BC)
∑ k≠ j≠i∈V
Topological coefficient (TC)
Type 2
1N index Shortest distance to the group of stress-miR-targets (SDT) Average distance to the group of stress-miR-targets (ADT)
σ ki j σk< j
avg ð J ði; jÞÞ ki
k pi ki
di
∑
j∈M
di j jM j
The shortest distance from node i to stress-miR-targets. M denotes the set of indices corresponding to stress-miR-targets; dij denotes length of the shortest path between node i and node j.
a
Network measurements can be divided in the following two types: type 1 describes the topological feature of individual gene in the local network, and type 2 describes the relationship between individual gene and the other particular gene group in the global network
(Fig. 2b). It is noted that interacting gene partners tend to have similar cellular functions and locate closer to form a functional module in the network (Bu et al. 2003). Thus, we can further infer that stress-miR-targets were likely to participate in
distinct functional modules associated with a certain abiotic stress response. It is noted that the probability of stress-miRtargets interacting with stress-miR-targets was the lowest. Therefore, we performed the chi-square test to validate the
Fig. 2 The rice interactome network. a The visualization of a giant connected component and the isolated ones for the whole rice interactome network: brown nodes represented stress-miR-targets, and green nodes represented non-stress-miR-targets. b The stress-miR-target LOC_Os01g34200.1 (targeted by osa-miR1432-3p) was involved in a subnetwork where LOC_Os01g34200.1 had few links to non-stress-miR-targets, but non-stress-miR-targets had more links among themselves. The
annotated function of LOC_Os01g34200.1 was indicated as Belemental activities^ in the gene ontology (GO) analysis, and the annotated function of the other genes (non-stress-miR-targets) in the subnetwork was mainly indicated as Bresponse to abiotic stimulus,^ Bcellular homeostasis,^ and Btransporter activity.^ c A subnetwork represented three cross-connected modules where stress-miR-targets tended to be on peripheral positions in these different modules
Funct Integr Genomics Fig. 3 The comparison among three kinds of interaction probabilities in the rice interactome network. Three kinds of interactions were indicated as different colors. In the plot, the Y axis represented −ln (interaction probability), and the red star symbol indicated the significant difference with the chi-square test (P value p < 0.01)
significant differences of the probabilities between the interaction of stress-miR-targets with stress-miR-targets and the other two kinds of interactions, stress-miR-targets with nonstress-miR-targets and non-stress-miR-targets with nonstress-miR-targets, indicating that the probability of interactions within stress-miR-targets was significantly lower than those of the other two kinds of interactions (as shown in Fig. 3, with a red star symbol indicating the significant difference, chi-square test with P value
ORIGINAL ARTICLE
Topological characteristics of target genes regulated by abiotic-stress-responsible miRNAs in a rice interactome network Linzhong Zhang 1 & Hongdong Xuan 2 & Yongchun Zuo 3 & Gaojian Xu 2 & Ping Wang 1 & Youhong Song 4 & Shihua Zhang 1,5
Received: 6 August 2015 / Revised: 18 January 2016 / Accepted: 20 January 2016 # Springer-Verlag Berlin Heidelberg 2016
Abstract A great number of microRNAs (miRNAs) have been identified in responding and acting in gene regulatory networks associated with plant tolerance to abiotic stress conditions, such as drought, salinity, and high temperature. The topological exploration of target genes regulated by abioticstress-responsible miRNAs (ASRmiRs) in a network facilitates to discover the molecular basis of plant abiotic stress response. This study was based on the staple food rice (Oryza sativa) in which ASRmiRs were manually curated. After having compared the topological properties of target genes (stress-miR-targets) with those (non-stress-miR-targets) not regulated by ASRmiRs in a rice interactome network, we found that stress-miR-targets exhibited distinguishable topological properties. The interaction probability analysis and kcore decomposition showed that stress-miR-targets preferentially interacted with non-stress-miR-targets and located at the Linzhong Zhang and Hongdong Xuan contributed equally to this work. Electronic supplementary material The online version of this article (doi:10.1007/s10142-016-0481-4) contains supplementary material, which is available to authorized users. * Shihua Zhang [email protected]
1
School of Science, Anhui Agricultural University, Hefei 230036, China
2
College of Information and Computer Science, Anhui Agricultural University, Hefei 230036, China
3
College of Life Sciences, Inner Mongolia University, Hohhot 010021, China
4
School of Agronomy, Anhui Agricultural University, Hefei 230036, China
5
State Key Laboratory of Tea Plant Biology and Utilization, Anhui Agricultural University, Hefei 230036, China
peripheral positions in the network. Our results indicated an obvious topological distinction between the two types of genes, reflecting the specific mechanisms of action of stressmiR-targets in rice abiotic stress response. Also, the results may provide valuable clues to elucidate molecular mechanisms of crop response to abiotic stress. Keywords Rice . Abiotic stress . miRNA . Target gene . Topological analysis
Introduction Generally, plant growth and productivity are greatly affected by abiotic stresses, such as drought, salinity, and high temperature (Budak et al. 2015a). MicroRNAs (miRNAs) are a class of endogenous non-coding small RNAs that posttranscriptionally control gene expression via either translational repression or mRNA degradation (Bartel 2004). Initially identified as regulators of developmental timing in Caenorhabditis elegans, the investigations of the functional mechanisms of these small RNAs have been an attractive territory (Shah and Chen 2014; Rogers and Chen 2013). In the kingdom of plant species, many studies have shown that miRNAs play important roles in regulating many biological processes, including leaf morphogenesis, vascular development, floral organ identity, and abiotic/biotic stress response (Budak et al. 2015b; Xie et al. 2015). During abiotic stress conditions, miRNAs act post-transcriptionally in gene regulatory networks pertaining to plant stress response (Sunkar et al. 2012). Under these circumstances, the action of target genes of abiotic-stress-responsible miRNAs (ASRmiRs) can be elaborately modified in a network fashion to facilitate stress tolerance at physiological, cellular, biochemical, and molecular levels (Urano et al. 2010; Cabello et al. 2014), shedding
Funct Integr Genomics
light in the disclosure of plant adaption mechanisms in view of target genes of ASRmiRs. To our knowledge, cellular systems featured with extremely complicated interacting networks of nucleotides, proteins, metabolites, and other small molecules (Barabasi and Oltvai 2004). The topological exploration of these networks is of great value in understanding the mechanisms of cellular action (Jeong et al. 2001; Kuchaiev et al. 2010). In the diverse interacting networks, protein-protein interactions (PPIs) are crucial for almost all levels of cellular functions, including genetic regulation, metabolism, and signal transduction (Morsy et al. 2008). In the past decade, miRNAs have emerged as primary regulators of gene expression controlling plant adaptation to stress conditions (Mallory and Vaucheret 2006). Therefore, the topological investigation of miRNA targets facilitates to unravel the structural and dynamical mechanisms of the interactome network in the control of ASRmiRs. Rice (Oryza sativa) is one of the staple food crops with rich genomic and proteomic data are available. Therefore, based on the rice species, we manually curated a complete list of ASRmiRs and computationally predicted their target genes. The topological properties of target genes (stress-miR-targets) regulated by ASRmiRs and those (non-stress-miR-targets) not regulated by ASRmiRs were compared in a predicted rice interactome network, showing that stress-miR-targets exhibited distinguishable topological properties from non-stressmiR-targets. In the rice interactome network, based on the interaction probability analysis, stress-miR-targets were shown to preferentially interact with non-stress-miR-targets. In addition, the k-core decomposition of the network revealed that stress-miR-targets might locate at the peripheral positions. All these findings indicated an evident topological distinction between the two types of genes, revealing the underlying mechanisms of actions of stress-miR-targets in rice abiotic stress response.
Materials and methods Prediction of a rice interactome network Several high-throughput experimental technologies, such as yeast two-hybrid system (Y2H) (Bartel and Fields 1997) and bimolecular fluorescence complementation (BiFC) (Wong and O’Bryan 2011), were usually used to chart the interactome (protein-protein interaction (PPI)) networks of model species, including C. elegans, Drosophila melanogaster, and Saccharomyces cerevisiae (Uetz et al. 2000; Zhong and Sternberg 2006; Giot et al. 2003). However, these largescale experiment strategies are usually cost-intensive and time-consuming. To overcome such limitations, a number of computational methods have been developed for the prediction of interactome network, such as phylogenetic profile,
correlated mutation, and machine learning (Cheng and Perocchi 2015; Pazos et al. 1997; Shen et al. 2007). Among these, we applied a widely used interolog method to predict the rice interactome network (Matthews et al. 2001). This approach is based on a simple principle: if two proteins (A and B) in rice are orthologous with two proteins (A’ and B’), respectively, in another model species, and the interaction between A’ and B’ has been experimentally validated; in such case, A and B can be predicted to interact with each other. In this study, as illustrated in Fig. 1, we chose seven model species including Homo sapiens, Arabidopsis thaliana, C. elegans, D. melanogaster, Escherichia coli K12, Mus musculus, and S. cerevisiae, and their experimentally validated PPI data were retrieved from publically available databases: BioGrid, MINT, DIP, and IntAct (Stark et al. 2006; Chatr-Aryamontri et al. 2007; Salwinski et al. 2004; Aranda et al. 2010). The orthologous relations between rice and the seven model species were decided using the reciprocal best-hit approach proposed by Brown and Jurisica (2005). The highquality PPI data is undoubtedly critical for the identification of basic protein functions and mechanisms of action. Therefore, we used two indexes of subcellular co-localization and coexpression, described by our colleagues, respectively (Brandao et al. 2009; De Bodt et al. 2009), as the quality control for the predicted PPI data. In principle, interacting protein pairs sharing the same subcellular compartment or having the co-expression correlation coefficient score greater than 0.5 were retained for further use. Definition of interaction probability A wide range of biological functions are determined by crosstalking proteins rather than isolated individuals, and different types of protein interactions in the network constitute the basic driving force in cellular operations (Pawson 2004). Therefore, the analysis of interaction probability among different types of proteins is beneficial for the dissection of protein functions. In the rice interactome network, three kinds of interactions among the two types of proteins (or genes, as stress-miR-targets and nonstress-miR-targets) were illustrated as interaction between the same type of genes (stress-miR-targets interact with stress-miRtargets, non-stress-miR-targets interact with non-stress-miR-targets) and interaction between different types of genes (stressmiR-targets interact with non-stress-miR-targets). Based on this, we defined the interaction probability among stress-miR-targets and non-stress-miR-targets as follows: 8 Nij > > < 2 Cn Pði; jÞ ¼ Nij > > : ni n j
if i and j belong to the same type of genes ðcase 1Þ if i and j belong to different types of genes ðcase 2Þ
In this formula, the two indexes i and j denote one of the two gene types (stress-miR-targets or non-stress-miR-targets).
Funct Integr Genomics
Fig. 1 The flowchart that illustrated the prediction of the rice interactome network using an interolog method
Nij represents the actual interaction number among the same type of genes (case 1) or different types of genes (case 2) in the Þ rice interactome network. C2n equals to nðn−1 2 , denoting all the possible interaction number in one of the two gene types, where n is the total gene number of this gene type. If i and j belong to different gene types (case 2), ni × nj is all the possible interaction number between the two types of genes, where, ni and nj denote the total gene number of stress-miR-targets and non-stress-miR-targets, respectively.
Network measurements describing the topological properties The exploration of topological properties for stress-miR-targets in the rice interactome network enables the dissection of rice
stress response mechanisms. For this purpose, two types of network measurements were predefined for the description of their topological properties in the interactome network. The first type describes the topological feature of individual gene in the local interactome network, such as degree, clustering coefficient (CC), betweenness (BC), and topological coefficient (TC). The second type describes the relationship between an individual gene and the other particular gene group in the global interactome network, such as 1N index, the shortest distance to the group of stress-miR-targets (SDT), and the average distance to the group of stress-miR-targets (ADT). The definitions of these measurements are detailed in Table 1. Degree of a node in the network can be determined by the number of edges directly linking to the node. Current efforts have shown the correlation between degree and essentiality (act as Bhubs^) of a certain node, and degree can be regarded
Funct Integr Genomics
as a predictor of essentiality in the network (Yu et al. 2007). The measurement of CC is used to characterize the modular property of a network, and it can be calculated as the fraction of the actual edges between the neighbors of a node against the possible maximum connections. BC is measured as the proportion of shortest paths between paired nodes in the network that a given node locates on. The centrality of an interested node can be characterized by the BC measurement, and a bottleneck node in the network can be considered if the node has a high BC. TC is a conventional measurement that describes the extent to which a node shares neighbor nodes with other individual nodes in the network. k-core decomposition analysis We used the k-core decomposition method, proposed by our colleagues and successfully performed in certain species (Wuchty and Almaas 2005), to disclose the inherent hierarchical structure of the rice interactome network. This is useful in dissecting the topological importance of stress-miR-targets in the global network. The k-core algorithm aims to decompose a network into k-cores (k-subgraphs) by iteratively removing all the nodes with the degree less than k, resulting in a spectrum of subgraphs (k-cores) that gradually display the backbone of the network. For each of the k-iterations, the remaining subgraph with every node having a degree of no less than k is called the kcore of the network. In these k-cores, we respectively calculated the proportion of stress-miR-targets to all the nodes in each kcore. This serial of proportion values can be used to describe the enrichment extent to which stress-miR-targets are represented in subgraphs (k-cores) of the rice interactome network. Network visualization and analysis We used the Cytoscape software (Smoot et al. 2011) with the force-directed layout to visualize the interaction relationships among stress-miR-targets and non-stress-miR-targets in the rice interactome network. The first type of network measurements, including degree, CC, BC, and TC, were calculated using the Cytoscape plugin NetworkAnalyzer (Assenov et al. 2008). We calculated the second type of network measurements (1N index, SDT, and ADT) using our in-house Matlab scripts. In addition, we used the Cytoscape plugin MCODE (Bader and Hogue 2003) to identify gene modules in the rice interactome network.
Results
literature mining approach (Zhang et al. 2013). In this study, we used a similar approach for the rice species. To obtain a complete list of ASRmiRs in rice, we conducted an extensive literature mining as that of the PASmiR project to discern published articles that give experimental evidence for rice miRNAs in response to certain abiotic stress. We first searched in Google Scholar, Scopus, and PubMed (Henderson 2005; Burnham 2006; Wheeler et al. 2007) in retrieving relevant articles using keyword Brice AND microRNA AND abiotic stress.^ Then, we manually reviewed the full text if the two sections of title and abstract of an article provided the evidence that the article described the role of rice miRNAs involved in abiotic stress response. Finally, we collected the relevant publications that assessed the up-/down-regulated expression pattern of rice miRNAs during certain abiotic stress conditions. As a result, 184 rice ASRmiRs were retrieved from the original publications, and 29 (15.76 %) miRNAs are involved in an average of 3–5 abiotic stress conditions (Supplementary File 1). Prediction of target genes regulated by rice ASRmiRs Using the interolog method, we computationally predicted the rice interactome network, which represented 22,933 nonredundant interacting pairs covering 2975 unique proteins under the two indexes of quality control. In this network, genes were divided into two main types: genes (stress-miR-targets) regulated by ASRmiRs and those (non-stress-miR-targets) not regulated by ASRmiRs. To this end, cDNA sequences of these 2975 genes were retrieved from the website of the international rice genome sequencing project (IRGSP) (Kawahara et al. 2013). We downloaded mature sequences of the 184 rice ASRmiRs from miRBase (Griffiths-Jones et al. 2006) and used the web tool psRNATarget (Dai and Zhao 2011) for the miRNA target identification. The psRNATarget tool was designed as a successful plant small RNA target analysis server, which can be specifically used for miRNA target prediction in the plant community. Of the total 2975 genes in the rice interactome network, 776 (26.1 %, namely stress-miR-targets) were predicted to be targeted by ASRmiRs (Supplementary File 2). We used the Cytoscape software to visualize the rice interactome network where a circular node represented a gene and an edge represented the interacting relationship between two different genes (Fig. 2a). To demonstrate the network more clearly, the two types of genes of stress-miR-targets and non-stressmiR-targets were marked with different colors in the rice interactome network. Interaction probability among stress-miR-targets and non-stress-miR-targets
Curation of rice ASRmiRs We have previously completed a PASmiR project by having collected abiotic-stress-responsible miRNAs in plants using a
We compared the probability of the three kinds of interactions, and the results showed that stress-miR-targets had the lower tendency to interact among themselves in the network
Funct Integr Genomics Table 1
Definitions of the seven network measurements
Typea
Network measurements
Function
Brief description
Type 1
Degree Clustering coefficient (CC)
ki
The number of links to node i. ki is the number of neighbors of node i, and ni is the number of connected pairs between all the neighbors of node i. σkj denotes shortest paths between node pairs k and j, σkij denotes that pass through node i. J(i, j) is the number of neighbors shared between the node i and node j, plus one if there is a direct link between i and j. avg(J(i, j)) is the average value of J(i, j). ki is degree of node i. kpi is the number of links between node i and stress-miR-targets.
2ni k i ðk i −1Þ
Betweenness centrality (BC)
∑ k≠ j≠i∈V
Topological coefficient (TC)
Type 2
1N index Shortest distance to the group of stress-miR-targets (SDT) Average distance to the group of stress-miR-targets (ADT)
σ ki j σk< j
avg ð J ði; jÞÞ ki
k pi ki
di
∑
j∈M
di j jM j
The shortest distance from node i to stress-miR-targets. M denotes the set of indices corresponding to stress-miR-targets; dij denotes length of the shortest path between node i and node j.
a
Network measurements can be divided in the following two types: type 1 describes the topological feature of individual gene in the local network, and type 2 describes the relationship between individual gene and the other particular gene group in the global network
(Fig. 2b). It is noted that interacting gene partners tend to have similar cellular functions and locate closer to form a functional module in the network (Bu et al. 2003). Thus, we can further infer that stress-miR-targets were likely to participate in
distinct functional modules associated with a certain abiotic stress response. It is noted that the probability of stress-miRtargets interacting with stress-miR-targets was the lowest. Therefore, we performed the chi-square test to validate the
Fig. 2 The rice interactome network. a The visualization of a giant connected component and the isolated ones for the whole rice interactome network: brown nodes represented stress-miR-targets, and green nodes represented non-stress-miR-targets. b The stress-miR-target LOC_Os01g34200.1 (targeted by osa-miR1432-3p) was involved in a subnetwork where LOC_Os01g34200.1 had few links to non-stress-miR-targets, but non-stress-miR-targets had more links among themselves. The
annotated function of LOC_Os01g34200.1 was indicated as Belemental activities^ in the gene ontology (GO) analysis, and the annotated function of the other genes (non-stress-miR-targets) in the subnetwork was mainly indicated as Bresponse to abiotic stimulus,^ Bcellular homeostasis,^ and Btransporter activity.^ c A subnetwork represented three cross-connected modules where stress-miR-targets tended to be on peripheral positions in these different modules
Funct Integr Genomics Fig. 3 The comparison among three kinds of interaction probabilities in the rice interactome network. Three kinds of interactions were indicated as different colors. In the plot, the Y axis represented −ln (interaction probability), and the red star symbol indicated the significant difference with the chi-square test (P value p < 0.01)
significant differences of the probabilities between the interaction of stress-miR-targets with stress-miR-targets and the other two kinds of interactions, stress-miR-targets with nonstress-miR-targets and non-stress-miR-targets with nonstress-miR-targets, indicating that the probability of interactions within stress-miR-targets was significantly lower than those of the other two kinds of interactions (as shown in Fig. 3, with a red star symbol indicating the significant difference, chi-square test with P value
