Proteins in living organisms perform many functions by physically interacting with each other (Bork et al., 2004). Interactomes have been described as the genome-wide roadmaps of inferred protein–protein interactions (PPIs). Investigating the model of PPI can enhance our understanding of the cellular process and biological interactions within an organism. Interactomes of model organisms such as Arabidopsis thaliana (L.) and Saccharomyces cerevisiae (Meyen) were built using high-throughput experimental methodologies (Arabidopsis Interactome Mapping Consortium, 2011). Interactomes have also recently begun to expand to the non-model organisms by predictions based on orthology. Predicted interactomes in agronomically important organisms, such as Citrus sinensis, Oryza sativa (L.), Glycine max (L.) Merrill, and Zea mays (L.), have also provided valuable insight into disease resistance (Afzal et al., 2009; Musungu et al., 2015). In recent years, an abundance of PPI data has been developed through high-throughput technologies, including plant pathogens such as Fusarium graminearum (Zhao et al., 2009) that causes Fusarium head blight of both wheat and barley; Gibberella stalk rot of maize (Goswami and Kistler, 2004); and Magnaporthe grisea (He et al., 2008), cause of rice blast (Talbot, 2003; Dean et al., 2005).

Phomopsis longicolla T. W. Hobbs (syn. Diaporthe longicolla) is a fungal species of Ascomycota in the Diaporthaceae family. It is the primarily causal agent of Phomopsis seed decay (PSD) in soybean, G. max (L.) Merrill (Hobbs et al., 1985; Santos et al., 2011; Li et al., 2015a). This pathogen also causes stem lesion on velvet leaf plants (Li et al., 2001) and can live as an endophyte in the mangrove and Meliaceae plant species (Rhoden et al., 2012). P. longicolla has also been reported to produce a number of cytotoxic and antimicrobial secondary metabolites, such as dicerandrols and phomoxanthones (Isaka et al., 2001; Lin et al., 2010).

Soybean is one of the most important economic crops in the world with global production over 340 million metric tons (ASA, 2017). The soybean PSD disease results in poor seed quality and it is one of the most economically important seed diseases in soybean (Sinclair, 1993; Li, 2011). Management of PSD has been conducted using conventional tillage to reduce pathogen inoculum, rotation with non-host or non-legume crops, and early harvest (once soybeans have matured) to avoid late season wet weather. However, inconsistent reductions of PSD have been reported when those common agronomic practices were used. Fungicide treatments have also been an option to reduce PSD, but they were not always effective in controlling PSD (TeKrony et al., 1985; Wrather et al., 2004). Planting PSD-resistant cultivars is a cost-effective and long-term strategy to manage PSD. In past decades, research has been conducted to identify PSD-resistance sources by screening soybean germplasms, commercial cultivars, and breeding lines (Li et al., 2011, 2015c; Li and Chen, 2013; Li and Smith, 2016), investigating inheritance of resistance and genetic mapping of resistance to PSD (Zimmerman and Minor, 1993; Jackson et al., 2005, 2009; Smith et al., 2008), and breeding for resistant lines and cultivars (Minor et al., 1993; Pathan et al., 2009). However, information about mechanisms underlying the pathogenicity of P. longicolla on soybean is lacking. The prediction of PPI networks in P. longicolla has not been investigated and reported.

Although the genome of the P. longicolla isolate MSPL 10-6 has been sequenced (Li et al., 2015b, 2017), there are still many genes with unknown functions. Therefore, using the computational biology approach to analyze the interactome will help understand the different mechanisms underlying pathogenicity in PPI networks. Hence, the objectives of this study were to perform a genome-wide analysis of the predicted proteins interactome and to identify conserved global networks and pathogenicity subnetworks in P. longicolla causing PSD of soybean. This research will enhance our knowledge of the biology, pathogenicity, and protein interactions of P. longicolla and aid in developing improved disease management strategies for PSD.

Interologs were defined as a conserved interaction between a pair of proteins of a given organism which have interacting homologs in another organism (Yu et al., 2004). This method has been used to study and predict protein interactions successfully in other multiple organisms (Gu et al., 2011; Ho et al., 2012; Weßling et al., 2014), including but not limit to O. sativa, C. elegans, and S. pombe. Bioinformatic algorithms and programs, and corresponding parameters and weights used to produce P. longicolla interactomes in this study were similar to those used in A. thaliana (Geisler-Lee et al., 2007), O. sativa (Ho et al., 2012), and Z. mays (Musungu et al., 2015). In this study, a proteome-wide analysis of a predicted protein interactome using the interlog method was used to predict protein interactions in P. longicolla and developed to create resource for understanding the biology of the ascomycete P. longicolla. The PiPhom interactome can assist plant pathologists interested in possible gene-for-gene interactions and mycologists interested in possible industrial applications in agriculture. Additionally, the PiPhom may lead to a better understanding of this economically important soybean pathogen that causes seed decay. To date, analysis of the P. longicolla genome has been conducted to determine the genome features (Li et al., 2017) and comparative genome study with other soybean ascomycete pathogens (Li and Musungu, unpublished). The primary goal of the study was to predict PPIs and gain a functional understanding of proteins involved in the developmental processes, plant cell wall degrading enzymes (PCWDEs) and pathogenicity proteins which are important components of P. longicolla. In our previous study, PCWDEs encoded within the P. longicolla genome were determined (Li et al., 2017). In this study, a cysteine-rich secretory protein (g7307.t1) that had five interacting partners within the network was identified. The cysteine rich protein was an interesting conserved discovery in the network because many of these proteins have been shown to function as secretory proteins in fungi and oomycetes, such as Phytophthora cactorum, Leptosphaeria maculans, and F. oxysporum (Sperschneider et al., 2015). Using the Systems Biology approaches, networks among families of PCWDEs were identified (Li et al., 2017). Additionally, proteins that represented a graphical significance within the network through degree of connectivity were characterized.

The level of CV can be used as a filter to identify true hypotheses and reduce the false positives when the data is used to build networks. Many previously reported common hetero-interactions were often the most abundant interactions in PPIs (Musungu et al., 2015). However, that was not the case in the PiPhom interactome, in which many homo-interactions were identified. The networks of P. longicolla displayed many of the core regulatory machineries in the cell that were important in multiple eukaryotic systems and organisms. One of the expected pathways mined from the CV analysis of the network was the DNA repair machinery subnetworks, which were conserved throughout eukaryotes (Liu et al., 1999; Ohbayashi et al., 1999) and recovered from similar interactomes, such as S. cerevisiae (Yu et al., 2008), Z. mays (Musungu et al., 2015), and A. thaliana (Geisler-Lee et al., 2007) among others. The higher the CV, the greater the likelihood of the conserved interactions detected in PiPhom interactome.

Connectivity in computational biology has been demonstrated to aid in the generation of hypotheses for targeted pathogenicity analysis. This has been demonstrated in multiple interactome studies when working on systems with minimal biological information. In the PiPhom interactome, the highly connected proteins were ubiquitous partners and co-factors such as cullins, scaffolding proteins, and proteins involved in degradation pathways. This overlapped with the previous interactome studies, such as A. thaliana, Z. mays, and P. patens, which had highly conserved pathways represented by cullins. The highly connected interactions are likely to have a large degree of connectivity since it is evolutionarily conserved (Evlampiev and Isambert, 2008).

The enrichment of the PiPhom protein network led to identification of pathogenesis pathways, such as nitrogen metabolism, carbohydrate degrading enzymes and cell to cell signaling processes. It has already been demonstrated that mitogen activated protein kinases are important in the pathogenicity of multiple pathogens such as F. graminearum, F. solani, and M. grisea (Xu and Hamer, 1996; Di Pietro et al., 2001; Jenczmionka et al., 2003; Ramamoorthy et al., 2007). Moreover, a total of 477 pathogenicity-associated proteins were detected in the PiPhom interactome. The abundance of pathogenicity factors was likely due to yeast data as well as the addition of the pathogen C. albicans in which 80 of 478 proteins were inferred within pathogenicity networks. This is an abundant amount in comparison to previous studies where about 100 pathogenicity proteins were detected for the M. grisea PPI (He et al., 2008). Within the network, there were common pathogenicity genes, which are seen in other ascomycetes. For example, STE11 was associated with pathogenicity in Botrytis cinerea and F. graminearum (Izumitsu et al., 2009; Leroch et al., 2013; Gu et al., 2015). Connection between phosphorylation of serine-threonine/tyrosine-protein kinases and transcription factors was present within the interactome. Those pathogenic proteins have been used as targets for targeted mutagenesis (Caracuel et al., 2003; Shimizu et al., 2003; Ramamoorthy et al., 2007).

Analysis of orthology is a key concept during interactome analysis of protein–protein interactions. This is because interactions are likely to occur if there is conservation through different organisms which can be attributed to fitness of organism or random mutations (Lynch, 2007). These features of the interlog method have been demonstrated in previous interactomes such as A. thaliana, Z. mays, H. sapiens, and S. cerevisiae. Additionally, because our dataset is made up of interactions from S. cerevisiae, multiple developmental pathways in important life cycles of plant pathogens were identified. This was highlighted in the results with proteins such as FUS3 which has been implicated in pathogenesis in fungi such as A. alternata, F. oxysporum, and M. oryzae (Wilson and Talbot, 2009; Lin et al., 2010; Pareek and Rajam, 2017). Moreover, the HOG1 protein was identified, which was initially characterized and conserved in S. cerevisiae to be involved in osmotic signaling. It has also been demonstrated in multiple fungi to be involved in the regulation of pathogenesis. For example, in Zymoseptoria tritici, a hemibiotrophic pathogen, when HOG1 was targeted for knockout it displaced a loss in pathogenicity (Mehrabi et al., 2006). This was additionally seen in pathogens such as Penicillium digitatum, Magnaporthe oryzae and several others (Motoyama et al., 2008; Wang et al., 2014).

The PiPhom interactome was built using preexisting data from the biogrid and resembles other inferred networks built by the same methodology. The connectivity of PiPhom differed from previous interactomes like Zea mays, Physcomitrella patens, D. melanogaster, and O. sativa (Giot et al., 2003; Stark et al., 2006; Gu et al., 2011; Musungu et al., 2015; Schuette et al., 2015). The characteristic path length of PiPhom differed from the previously reported interactomes for organisms, such as A. thaliana (3.4), S. cerevisiae (2.6), Stegodyphus mimosarum (2.5) and H. sapiens (between 1 and 3; Gursoy et al., 2008; Taylor et al., 2009; Chen et al., 2012; Schuette et al., 2015; Wang and Jin, 2017). When contrasting the PiPhom with previous plant, animal, and fungal work, there was an abundance of signaling proteins in the conserved interactions, suggesting a difference in wiring vs. other fungal species. In contrast, the PiPhom conserved subnetwork modules were comprised of many of the pathways previously found in other interactomes, such as ubiquitination, methylation, pheromone signaling, developmental pathways and chromatin remodeling (Mosca et al., 2012). Thus the similarity between PiPhom and previous interactomes produced by similar methodology leads to increase in confidence in the novelties discovered (Geisler-Lee et al., 2007; Gu et al., 2011; Musungu et al., 2015). Furthermore, the other proteins that were found in the network will become targets for in-depth lab studies. Scientists who are interested in a particular PCWDE family would have the ability to mine the network. For example, the g5566t.1 gene that was identified to represent the highest degree of connectivity was initially identified as uncharacterized and containing cytochrome P450 domain information. While multiple genes have been identified to be involved with cytochrome P450 and pathogenicity for other pathogens, the protein sequence would have been initially missed because it was returned as uncharacterized during the initial BLAST analysis against F. graminearum. Utilizing the metadata in PiPhom, the yeast one-to-one ortholog NAB2 was inferred to be the closest ortholog of the g5566t.1 protein, partly because it had the largest degree of connectivity within the network. The protein was essential for cell viability in yeast (Anderson et al., 1993). Its primary function was inferred to be involved in RNA transport and confirmed primarily in yeast. However, mutations in A. oryzae have been able to show that knocking out of the gene altered the ability of the pathogen to grow compared to wild type (Yamada et al., 1999).

In addition, the interactions are informative for study of other biological processes due to the similarity between PhiBase proteins and other pathogens. From the analysis against the pathogen host interaction (PHI) database genes, multiple BLAST hits on virulence and pathogenicity factors were shared among ascomycetes. PiPhom will be used to select for genes of interest. For example, some of the orthologous proteins identified included CYP1, SGE1, PMK1 and many others from fungi and bacteria. The CYP1 and SGE1 have been characterized and previously shown to be involved in pathogenicity (Winnenburg et al., 2006). Another feature of the PiPhom interactome can be used to filter/identify the non-virulence and non-pathogenic hits that are stored within PHI database.

Overall, PiPhom showed several novel interactions specific to P. longicolla. This is likely due to the close ortholog overlap for P. longicolla proteins vs. yeast proteins within the network. The PiPhom interactome appeared different from previous studies of plant interactomes, such as maize (Musungu et al., 2015), in which the amount of interactions generated was limited by plant specific data.

The PiPhom interactome generated from this study provides a valuable resource for understanding the complexity of pathogenicity in P. longicolla. The orthologous proteins, such as FUS3, HOG, CYP1, SGE1, PCWDE, and the g5566t.1 gene identified in this study could play an important role in pathogenicity of P. longicolla. This research enhances our knowledge of the biology, pathogenicity, and protein interactions of P. longicolla and aids in developing improved strategies for managing PSD. Moreover, the PiPhom interactome can also lead to a better understanding of PPIs in soybean pathogens. Furthermore, the PPI may aid in targeting of genes and proteins for further studies of pathogenicity mechanisms.

SL conceived and led the project, interpreted results and wrote the manuscript. BM analyzed data and wrote the manuscript. DL and PJ provided suggestions for the project and edited the manuscript. All authors reviewed and approved the final version of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.