Differential Communications between Fungi and Host Plants Revealed by Secretome Analysis of Phylogenetically Related Endophytic and Pathogenic Fungi

During infection, both phytopathogenic and endophytic fungi form intimate contact with living plant cells, and need to resist or disable host defences and modify host metabolism to adapt to their host. Fungi can achieve these changes by secreting proteins and enzymes. A comprehensive comparison of the secretomes of both endophytic and pathogenic fungi can improve our understanding of the interactions between plants and fungi. Although Magnaporthe oryzae, Gaeumannomyces graminis, and M. poae are economically important fungal pathogens, and the related species Harpophora oryzae is an endophyte, they evolved from a common pathogenic ancestor. We used a pipeline analysis to predict the H. oryzae, M. oryzae, G. graminis, and M. poae secretomes and identified 1142, 1370, 1001, and 974 proteins, respectively. Orthologue gene analyses demonstrated that the M. oryzae secretome evolved more rapidly than those of the other three related species, resulting in many species-specific secreted protein-encoding genes, such as avirulence genes. Functional analyses highlighted the abundance of proteins involved in the breakdown of host plant cell walls and oxidation-reduction processes. We identified three novel motifs in the H. and M. oryzae secretomes, which may play key roles in the interaction between rice and H. oryzae. Furthermore, we found that expression of the H. oryzae secretome involved in plant cell wall degradation was downregulated, but the M. oryzae secretome was upregulated with many more upregulated genes involved in oxidation-reduction processes. The divergent in planta expression patterns of the H. and M. oryzae secretomes reveal differences that are associated with mutualistic and pathogenic interactions, respectively.


Functional annotation of secretomes
Blast2GO analysis revealed that 400, 556, 338, and 328 secreted proteins had no functional annotation, whereas 742, 814, 663, and 646 secreted proteins, in the H. oryzae, M. oryzae, G. graminis, and M. poae secretomes, respectively, had some form of annotation such as InterPro, SUPERFAMILY, or PFAM entries (Fig 1; S1 Table). Among the four unannotated secretomes, 197 (49%), 388 (70%), 154 (46%), and 165 (50%) were singletons, which were not included in any orthologue groups. The general enzymatic activities (EC classification) and Gene Ontology (GO) biological process annotations of the annotated secretomes of the four species were involved in some form of catalytic activity, represented primarily by hydrolyases and oxidoreductases (Fig 3; S3 Table). The major targets of the hydrolyases were lipids, carbohydrates, and proteins (Fig 3). To identify the metabolic pathways involved, enzymes from the four secretomes were mapped onto the Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic pathways (Table 1), showing the high representation of enzymes related to phenylpropanoid biosynthesis, amino/ nucleotide sugar metabolism, starch/sucrose metabolism, and aminobenzoate degradation.

Comparative PFAM analysis of secretomes
Of the H. oryzae, M. oryzae, G. graminis, and M. poae annotated protein sequences, 576, 601, 535, and 497, respectively, contained at least one PFAM domain (S4 Table) (Table 2). In addition, Peptidase_S8 (PF00082) was also abundant in the four secretomes (Table 2). For carbohydrate-binding modules, WSC (PF01822) and LysM (PF01476) domains were more abundant in the H. oryzae, G. graminis, and M. poae secretomes, whereas the CBM_1 (PF00734) domain was more abundant in the M. oryzae secretome. Another abundant PFAM domain was a type B carboxylesterase (COesterase, PF00135) implicated in lipid degradation (Table 2). These results suggested the most abundant PFAM domains were hydrolytic enzymes involved in the breakdown of host plant cell walls.

Identification of motifs in unannotated secretomes
Conserved motifs were detected in the unannotated H. and M. oryzae secretomes. Three constrained motifs were identified, ranging in abundance from 12 to 81 sites (Fig 4). Compared to M. oryzae, the unannotated H. oryzae secretome contained far more sites of motif 1 (48 sites) and motif 2 (53 sites), with 27 proteins containing both motifs. Motif 1 contained L, A, and R residues which were conserved and organised in an LXAR pattern. This motif was identified in 23 orthologue groups. Motif 2 contained R and D residues in an RD pattern; 28 orthologue groups contained proteins with motif 2, and 10 groups shared motif 1. Finally, motif 3 was the least abundant motif, identified in only 12 proteins distributed in 4 groups, but none of the groups overlapped with those of motifs 1 or 2. Motif 4 contained a conserved V/TCC pattern that was preceded by a conserved C residue, located at an average position of 36 amino acids (Fig 4).

Divergent in planta expression patterns of H. and M. oryzae secretomes
Of 1142 predicted members of the H. oryzae secretome, 918 genes encoding secreted proteins (80%) were expressed at 2 or 6 days after infection (HDAI2 or HDAI6) in rice roots, including 631 annotated and 287 unannotated genes (S6 Table). In total, 967 genes (71%) out of 1370 M. oryzae secretome genes were expressed at 2 or 6 days after infection (MDAI2 or MDAI6) in rice roots, with 635 annotated and 282 unannotated genes (S6 Table). Although 394 (43%) and 477 (49%) genes showed similar expression levels for DAI2 and DAI6 (fold change < 2), the expression patterns differed between the H. and M. oryzae secretomes during the infection of rice roots (Fig 5; S6 Table). Furthermore, 149 proteins of the H. oryzae secretome exhibited time-dependent expression (103 for HDAI2 and 46 for HDAI6), a much larger number than for M. oryzae (22 for MDAI2 and 11 for MDAI2) (Fig 5; S6 Table). Sixty one genes in the annotated H. oryzae secretome were upregulated during the infection process (DAI6/DAI2 > 2), a markedly smaller number than for M. oryzae (162 genes) (Fig 5A; S6 Table). At the same time, 201 annotated genes of the H. oryzae secretome were downregulated (DAI6/DAI2 < 0.5), many more than for the M. oryzae secretome (105 genes). However, no divergence in expression patterns was found between the unannotated secretomes ( Fig 5B; S6 Table). We further analysed the functions of the secretome genes that exhibited differential expression levels (Fig 6). Most of the H. oryzae genes involved in plant cell wall degradation, such as proteins containing Glyco_hydro, peptidase, lipase, tyrosinase, cutinase, cellulase, CBM, and LysM domains, were downregulated during infection. However, most of the genes with the same function were upregulated during the M. oryzae infection process. In addition, far more genes of the M. oryzae secretome involved in oxidation-reduction processes (including proteins with FAD_binding, GMC_oxred, and Cu-oxidase domains) were upregulated compared with H. oryzae (Fig 6).
For the H. oryzae secretome, we detected expression of 63 genes encoding proteins containing motif 1, 2, or 3, including 26 genes with both motifs 1 and 2. Among these, 24 genes were downregulated, whereas two were upregulated. Fewer genes encoding proteins with motifs 1, 2,  Table).

Discussion
Promotion and regulation of the infection process is essential for both endophytic and pathogenic fungi. Analysis of the secretome is a powerful tool to investigate how fungi manage the infection process [5,22]. For example, analysis of the secretomes of Puccinia graminis and Melampsora larici-populina using an in silico comprehensive analysis pipeline identified eight families of candidate effectors with high value for functional characterisation [22]. Investigation of the arsenal of proteins secreted by Sclerotinia sclerotiorum and Botrytis cinerea, identified refined secretomes with 432 and 499 proteins, respectively, providing a greater understanding of the necrotrophic infection process [5]. In our previous study of the H. oryzae genome, we focused on the characterisation of 167 small secreted cysteine-rich proteins (SSCRPs), which contained more than 4% cysteine residues and were at most 200 amino acids in length [16]. The prediction of these secreted proteins used a less stringent selection criteria and was based on a limited interspecies comparison, while the expression patterns were almost ignored. Therefore, in this study, the detailed analysis of the secretomes of four closely related fungi, namely H. oryzae, M. oryzae, G. graminis, and M. poae (including both an endophyte and pathogens) provides us with much more information about how secretomes affect infection by fungi with different lifestyles, and allow more informed choices to be made when choosing secreted proteins to be verified for function during infection.
We predicted full and refined H. oryzae, M. oryzae, G. graminis, and M. poae secretomes containing 1142, 1370, 1001, and 974 proteins, respectively, thereby significantly expanding the analysis of SSCRPs reported previously [16]. Some secreted protein-encoding genes such as the Avr-Pita1 genes (MGG_15370, MGG_14981, and MGG_15212), which were clustered in an M. oryzae-specific group, and the Avr-piz gene MGG_18041, which was present as a singleton, may function as avirulence genes in M. oryzae. Frequent presence/absence of polymorphisms in Avr-genes and rapid evolution of avirulence genes have been detected in M. oryzae and diverse repeated sequences are closely associated with most Avr-genes [23]. Approximately 75% of proteins from the H. oryzae, G. graminis, and M. poae secretomes were consistently present in orthologue groups, more than those of M. oryzae (56%). In addition, the number of proteins with paralogues and species-specific groups in M. oryzae was greater than for the other three species. Furthermore, more singletons were present in the M. oryzae secretome. These results suggest that the M. oryzae secretome may have evolved more rapidly than those of the three related species, resulting in many species-specific secreted proteins. This process would have also led to the retention of fewer secreted proteins of the common ancestor in M. oryzae than in the three other species.
PFAM analysis of the four secretomes revealed the abundance of enzymes involved in the breakdown of host plant cell walls, including hydrolysing enzymes that degrade various plant host substrates. Evaluation of the enzymatic activities, GO biological processes, and KEGG metabolic pathways assigned to the secreted proteins confirmed the high representation of polysaccharide catabolic and proteolytic processes among the four secretomes. GH61, the most abundant PFAM domain, is important for degrading lignocellulose in conjunction with cellulases [24]. Although it is not truly a glycosidase, this family is listed as a glycosyl hydrolase in the CAZy classification [25]. In addition, Peptidase_S8, a family of serine proteases, may contribute to plant cell wall degradation by targeting structural proteins, such as hydroxyproline-rich glycoproteins [26,27]. LysM domains could bind to chitin, a major constituent of fungal cell walls, which can be recognised by host cell surface receptors and trigger an immune response [28]. To overcome immunity, both endophytes and pathogens secrete effector molecules involved in fungal cell wall modification [29,30]. Four secreted LysM-containing proteins of H. and M. oryzae are expressed consistently during the infection of rice roots.
Characterisation of the role of reactive oxygen species (ROS) has shown that the redox state of the host substrate is extremely important in disease establishment [31,32]. At the same time, ROS originating from a mutualistic endophyte are required to inactivate plant defence responses and maintain mutualism [33]. It has been shown that virulent pathogens, such as M. oryzae, have developed mechanisms for scavenging ROS [34], whereas H. oryzae possesses melanin as an adaptive countermeasure to tolerate oxidative stress [20]. Consistent with these findings, our functional analysis of the four secretomes revealed a large number of proteins predicted to be involved in oxidation-reduction interactions. Enzymes that use FAD as a cofactor are primarily oxygen-dependent oxidoreductases, whereas GMC oxidoreductases constitute a large protein family with oxidoreductase activity which catalyses the transfer of electrons between molecules. In fungi, these enzymes are involved in many processes, such as defence and virulence mechanisms, browning and pigmentation, and melanin production [35,36]. In addition, tyrosinases, which are abundant in H. oryzae, may also be involved in multiple biological processes, including melanin production [35,36].
Using a de novo motif search, we detected only three conserved patterns in the combined unannotated secretomes of H. and M. oryzae, which is much less than was described in a similar study [22]. One possible reason for fewer conserved patterns is the rapid evolution of the M. oryzae secretome. This explanation is also supported by the fact that many more motif sites were detected in the H. oryzae secretome than in the M. oryzae secretome. Furthermore, compared to M. oryzae, more motif-containing proteins were expressed during the infection process in H. oryzae. These results suggest that the motif-containing genes may play more important roles for infection by H. oryzae than M. oryzae.
The induction of host defences has been reported in the symbiotic interaction of plants and fungi [21,37]. However, divergent responses of rice to infection by H. oryzae and M. oryzae have been identified, including suppressed defence responses in the interaction with H. oryzae, in contrast with the persistent defence responses induced by M. oryzae [37]. The expression levels of the H. oryzae secretome genes involved in plant cell wall degradation were consistently downregulated, but were upregulated in M. oryzae. The upregulated expression of the M. oryzae secretome genes involved in oxidation-reduction processes may have also resulted from the continuous defence responses of rice roots given that ROS play a vital role in stress responses, programmed cell death (PCD), and plant defences [32]. It has been shown that the host plays important roles in the evolutionary processes of fungi [16]. Thus, the different secretome expression patterns of H. and M. oryzae may represent differences in the feedback of these two kinds of fungi when they interact with the host plant, suggesting communication between fungi and host plants. These results also highlight how secretion and regulation of the expression of proteins and enzymes are important means of fungi-plant communication.

Conclusions
In this study, we characterised the full and refined secretomes of the endophytic fungus H. oryzae and the related phytopathogenic fungi M. oryzae, G. garminis, and M. poae. The secretomes were abundant in proteins involved in the breakdown of host plant cell walls and in oxidationreduction processes, and exhibited divergent expression patterns during infection of the same host plant by endophytes or pathogens. The next step is to determine whether the observed differences hold for pairs of related mutualists and endophytes in additional host species across the plant kingdom.

Secretome prediction
The proteomes of M. oryzae (version 8), G. graminis (version 1), and M. poae (version 1) used in this study were downloaded from the Broad Institute (https://www.broadinstitute.org/) and Genbank (accession number: PRJNA13840, PRJNA37931, and PRJNA13840 respectively). The H. oryzae proteome was the same as that used in our previous report (accession number: PRJNA252809, version 1) [16]. Predictions of signal peptides and cleavage sites were performed using SignalP v4.1 software [38] and sub-cellular targeting was analysed using TargetP v1.1 software [39] with default parameters. Only proteins predicted as being secreted by both methods were considered for further analyses. The proteins were scanned for transmembrane (TM) spanning regions using TMHMM v2.0 software [40] to exclude proteins likely to be retained in the plasma membrane. GPI-anchor proteins were predicted using PreGPI software [41] with default parameters. The final numbers of proteins belonging to the H. oryzae, M. oryzae, G. graminis, and M. poae secretomes were 1142, 1370, 1001, and 974, respectively.

Functional annotation
To characterise putative functions of predicted secretomes, we used BLAST2GO software with automated annotation [42]. Protein sequences were compared with the non-redundant sequence database using the BLASTP algorithm with an e-value less than 1.0e-05 recording max. 20 hits. Proteins with significant hits were classified into GO categories by 'Mapping' . Data from both analyses were merged into the annotation. PFAM domains were mapped onto proteins of the predicted secretomes using the PFAM batch search server [43] with default parameters. Orthologue groups were identified using OrthoMCL v2.0 software [44] with default parameters. First, orthologue, co-orthologue, and inparalogue pairs were identified with OrthoMCL. The pairs and their weights were used to construct an OrthoMCL graph for clustering with the MCL algorithm [45]. A de novo protein motif search was performed on the combined unannotated H. and M. oryzae secretomes using MEME software [46]. The program was set to report the 10 most robust motifs of 4-12 amino acids, occurring zero or one time per sequence with an e-value less than 1.0e-02.

Expression analysis
H. and M. oryzae RNA-seq data were the same as sets used in our previous study [16], which described the transcriptomes of the two fungi at 2 and 6 days after infection (DAI) of rice roots. Briefly, the blast-susceptible rice cultivar CO-39 (Oryzae sativa), the M. oryzae strain Guy11 and H. oryzae strain R5-6-1 were used in the experiment. Rice seeds were surface sterilized as previously described [20] and planted in half-strength Murashige & Skoog solid medium at 30°C in the dark for 4 days and then grown vertically under a 16-h light/8-h dark photoperiod at 28/24°C for an additional 6 days. M. oryzae were cultured under a 16-h light/ 8-h dark photoperiod at 25°C, while H. oryzae were cultured in the dark at 25°C. Conidia of M. oryzae were harvested from 10-day-old cultures grown on solid complete medium, and germinating phialidic conidia of H. oryzae were harvested from 4-day-old potato dextrose broth. Inoculations and co-culturing were performed as described previously [16]. All samples were harvested and immediately flash frozen in liquid nitrogen. Total RNA was extracted and then sequenced using the Illumina HiSeq 2000 based on 100bp paired-end read sequencing [16]. All the clean reads were mapped to the genome sequences using TopHat v 2.0.9 [47], and an expression profile was created using Cufflinks v2.0.2 [48]. The RNA-seq data were also validated by quantitative real time (qRT)-PCR, the results of which showed high consistency [16]. Gene expression levels were quantified by fragments per kilobase of exon per million fragments (FPKM) analysis and genes with FPKM values greater than one were considered to be expressed. Heat maps of gene expression profiles were generated using R (www.R-project.org) based on expression changes (log2-fold changes). All data were checked to avoid problems related to Excel converting gene identifiers [49].
Supporting Information S1