Computational Prediction of Pathogenic Network Modules in Fusarium verticillioides

Abstract

Fusarium verticillioides is a fungal pathogen that triggers stalk rots and ear rots in maize. In this study, we performed a comparative analysis of wild type and loss-of-virulence mutant F. verticillioides co-expression networks to identify subnetwork modules that are associated with its pathogenicity. We constructed the F. verticillioides co-expression networks from RNA-Seq data and searched through these networks to identify subnetwork modules that are differentially activated between the wild type and mutant F. verticillioides, which considerably differ in terms of pathogenic potentials. A greedy seed-and-extend approach was utilized in our search, where we also used an efficient branch-out technique for reliable prediction of functional subnetwork modules in the fungus. Through our analysis, we identified four potential pathogenicity-associated subnetwork modules, each of which consists of interacting genes with coordinated expression patterns, but whose activation level is significantly different in the wild type and the mutant. The predicted modules were comprised of functionally coherent genes and topologically cohesive. Furthermore, they contained several orthologs of known pathogenic genes in other fungi, which may play important roles in the fungal pathogenesis.

Characterizing co-expression networks underpinning maize stalk rot virulence inFusarium verticillioidesthrough computational subnetwork module analyses

“…We developed a computational workflow that allows us to build co-expression networks from F. verticillioides NGS datasets 15 . We first inferred the co-expression networks for the wild type as well as the fsr1 mutant utilizing the preprocessed gene expression data by using the partial correlation 21 (Supplementary Information).…”

“…Our NGS study was designed to capture dynamic changes in gene expression during maize stalk colonization by F. verticillioides wild type and fsr1 mutant. To capture dynamic changes in transcriptome, samples were harvested from three distinct phases of stalk pathogenesis: establishment of fungal infection, colonization and movement into the vascular bundles, and host destruction and collapse 15 . A total of six independent biological replications were prepared and analyzed for each sample, since increasing the number of replicates was important for us to implement our computational analysis for identifying subnetwork modules that show strong differential expression.…”

“…As described in our previous work 15 , our strategy is to first construct the co-expression network of F. verticillioides using partial correlation, and search through these networks to detect subnetwork modules that are differentially expressed in the two F. verticillioides strains. Subsequently, we use the probabilistic pathway activity inference scheme 16 to predict the activity level of potential subnetworks, followed by applying a computationally efficient branch-out technique to find the subnetworks that display the largest differential expression.…”

FvLcp1, a type-D fungal LysM protein with Chitin-binding domains, is a secreted protein involved in host recognition and fumonisin production in Fusarium verticillioides - maize kernel interaction

“…By following our previously proposed analysis approach (Kim et al, 2018a), we started 148 extending subnetwork modules having member genes up to two from the seed genes ( Fig. 1).…”

“…We first preformed preprocessing the RNA-seq datasets by alignment using TopHat2 132 To search for genes encoding secreted protein, we assigned genes with signal peptide that were 138 7 significantly differentially expressed between the two maize kernels as seed genes by measuring 139 t-test statistics scores as well as F scores of ANOVA across all three PGEMs, thereby preparing 140 ten seed genes for each kernel (twenty in total). For co-expression network construction, we built 141 five different networks at five different levels (i.e., five different threshold cut-offs) as previously 142 applied (Kim et al, 2015;Kim et al, 2018a;Kim et al, 2018b), where the smallest size included 143 roughly 400,000 edges and the largest size contained around 2,000,000 edges. Additional detail 144 is provided in Supplementary Method A.…”