Friend or foe: differential responses of rice to invasion by mutualistic or pathogenic fungi revealed by RNAseq and metabolite profiling

The rice endophyte Harpophora oryzae shares a common pathogenic ancestor with the rice blast fungus Magnaporthe oryzae. Direct comparison of the interactions between a single plant species and two closely-related (1) pathogenic and (2) mutualistic fungi species can improve our understanding of the evolution of the interactions between plants and fungi that lead to either mutualistic or pathogenic interactions. Differences in the metabolome and transcriptome of rice in response to challenge by H. or M. oryzae were investigated with GC-MS, RNA-seq, and qRT-PCR. Levels of metabolites of the shikimate and lignin biosynthesis pathways increased continuously in the M. oryzae-challenged rice roots (Mo-roots); these pathways were initially induced, but then suppressed, in the H. oryzae-challenged rice roots (Ho-roots). Compared to control samples, concentrations of sucrose and maltose were reduced in the Ho-roots and Mo-roots. The expression of most genes encoding enzymes involved in glycolysis and the TCA cycle were suppressed in the Ho-roots, but enhanced in the Mo-roots. The suppressed glycolysis in Ho-roots would result in the accumulation of glucose and fructose which was not detected in the Mo-roots. A novel co-evolution pattern of fungi-host interaction is proposed which highlights the importance of plant host in the evolution of fungal symbioses.

Interactions between plants and fungi span a broad continuum from pathogenic to mutualistic 1 . Whereas asymptomatic fungal endophytes exemplify the mutualistic or commensalistic region of this spectrum, other fungi are virulent pathogens that kill their host plant; still others strongly reduce plant performance and fitness [2][3][4] . As sessile organisms, plants are unable to escape attack by parasites, so strong defense mechanisms are needed for them to effectively respond to and manage pathogens 5 . Plants also engage in mutualistic interactions with beneficial microorganisms such as root-associated fungi to extend access to nutrients 6 . Plants can tune their physiological responses to prevent detrimental interactions or to support advantageous interactions.
The commonalities and differences between pathogenic and symbiotic colonization strategies of various fungi have been investigated in many studies [7][8][9][10] . It has been reported that both pathogenic and mutualistic interactions follow similar developmental programs, progressing from host identification through

Results
Changes in the rice root metabolome. GC-MS analysis was used to evaluate Ho-roots at 2, 6, and 20 days after inoculation (DAI), Mo-roots at DAI2 and DAI6, and control roots (Control-roots). By comparing the mass spectra of analyte peaks with those of commercial reference standard compounds, a total of 58 sample metabolites were identified (Supplementary Fig. 1; Supplementary Table S1; Supplementary  Table S2). In our principal component analysis (PCA) of all samples, the first two PCs accounted for 64.7% of the total variance of the data, and differentiated among developmental stages (PC1) as well as between fungi (both H. and M. oryzae) and the control samples (PC2, Fig. 1), suggesting that the majority of the variance in the data resulted from the treatments. Two days after the initial inoculation, there was some degree of overlap in the clustering of the Ho-root and Mo-root samples in the PCA scores plot (Fig. 1). However, samples within each treatment clustered tightly, and different treatments were discriminated at subsequent stages (DAI6, Fig. 1), clearly showing the effect of the different species of fungi on the roots. The profiles indicated a bigger discrepancy in metabolic activity between the Ho-roots and Control-roots than between the Mo-roots and Control-roots ( Fig. 1), at both the 2 and 6 day time points.
In order to further understand the differences between the roots challenged by the different fungi, orthogonal partial least-squares-discriminant analysis (OPLS-DA) models were generated. The OPLS-DA scores plots showed a significant clustering of the rice roots that received different treatments (Control-roots, Ho-roots, and Mo-roots) (Fig. 2). In the OPLS-DA models of the fungi-challenged roots at DAI2, DAI6, and DAI20, the first two components described 90.6%, 90.5%, and 98.5% of the variation and predicted 79.4%, 84.3%, and 96.9% for each infection stage, respectively, according to cross-validation. Furthermore, permutation tests were performed with PLS-DA models to validate each  OPLS-DA model, and these tests showed that all of the Q2 and R2 values were higher in the permutation tests than in the OPLS-DA models (Fig. 2). These results demonstrated both high goodness of fit and high predictive capability for the OPLS-DA models. Therefore, we concluded that using the first two components to examine the rice root metabolite profiling results was sufficient.
In OPLS-DA modeling, the contribution of each metabolite to OPLS1 and OPLS2 is computed, and each metabolite is given a loading (weighted value) for both OPLS1 and OPLS2. By combining the VIP (variable importance in the projection) values in the loadings plot ( Supplementary Fig. S2), metabolites with VIP > 1 were selected as differentially accumulated metabolites. Twenty-two such compounds were thusly identified (Fig. 3), and Student's t-tests showed that these metabolites were significantly different at least between one treatment and the corresponding control samples (p < 0.05). 6 metabolites were differentially accumulated by H. and M. oryzae at DAI2; and similarly, 11 differentially accumulated metabolites were identified at DAI6 (Fig. 3). The greater number of differentially accumulated metabolites at DAI6 compared to DAI2 was consistent with an observed overlap of the clusters for the Ho-root and Mo-root samples at DAI2 in the PCA analysis. The most significant metabolites influencing the separation in the OPLS-DA models of rice roots of different treatments are listed. Transcript fold-changes (log2) of Ho-roots vs. Control-roots samples at DAI2, 6 and 20 are indicated. Green, increase in the abundance; red, decrease in the abundance. Asterisks refer to the highly accumulated (white) and reduced (yellow) metabolites with statistical significant changes compared with control determined using a Student's t-test (P< 0.05). I, amino acids; II, central carbon metabolism; III, TCA/glyoxylate cycle; IV, shikimate pathway and aromatic amino acids; V, others. Ho-roots-DAI2, Ho-roots-DAI6, and Ho-roots-DAI20 refer to DEGs of the H. oryzae-challenged roots at 2, 6, and 20 days after inoculation, respectively. Mo-roots-DAI2 and Mo-roots-DAI6 refer to DEGs of the M. oryzae-challenged roots at 2 and 6 days after inoculation, respectively. Three of the twenty-two differentially accumulated metabolites were the disaccharides sucrose, maltose, and trehalose (Fig. 3). In addition, four intermediates of glycolysis and the TCA cycle including succinate, malic acid, fructose, and glucose were detected as differentially accumulated metabolites (Fig. 3). The aromatic amino acids phenylalanine, tyrosine, and tryptophan, together with quinate and shikimate, two precursor compounds in aromatic amino acid biosynthesis, were identified as differentially accumulated metabolites (Fig. 3). In general, the differentially accumulated metabolites were associated with a small number of distinct areas of metabolism, with carbohydrate metabolism and aromatic amino acid metabolism predominating.

Identification of differently expressed genes (DEGs).
It has been shown that the infection process of rice roots by H. and M. oryzae proceed with a similar infection strategy at an early stage (DAI2) but clearly differentiate at the middle stage (DAI6) 9,22,23 . Furthermore, metabolic profiling showed that rice roots responded more similarly to H. and M. oryzae infection at DAI2 than at DAI6. Thus, in this study, we focused on the differences in the response of rice roots to H. and M. oryzae infection at the middle and later stages as compared to an early stage, in order to identify gene expression patterns that changed between these infection stages.
Consequently, RNA isolated from the roots of Ho-roots-DAI2, Ho-roots-DAI6, Ho-roots-DAI20, Mo-roots-DAI2, and Mo-roots-DAI6 (most of the roots infected by M. oryzae were dead at DAI20) were evaluated with RNA-seq analysis. Three biological replicates were prepared and analyzed for each sample. In total, 121.71, 109.81, and 102.93 million reads were uniquely mapped to the rice genome sequence (O. sativa Japonica Group cultivar Nipponbare; ENSEMBL release 6.12) for the Ho-roots-DAI2, Ho-roots-DAI6, and Ho-roots-DAI20 samples, respectively (Supplementary Table S3), with a mean coverage of 37 × per biological replicate. As compared to Ho-roots-DAI2, there were 1085 genes in Ho-roots-DAI6 that were expressed at significantly different levels (343 up-regulated, 742 down-regulated) (Fig. 4a). Between the Ho-roots-DAI2 and the Ho-roots-DAI20 samples, 1482 genes were expressed at significantly different levels (634 up-regulated, 864 down-regulated) (Fig. 4a). The total number of uniquely mapped reads for Mo-roots-DAI2 and Mo-roots-DAI6 were 115.83 and 87.76 million, respectively, with a mean coverage of 33 × per biological replicate (Supplementary Table S3). 99 significantly up-regulated genes and 172 significantly down-regulated genes were identified in the Mo-roots-DAI6-DAI2 comparison (Fig. 4a), indicating that a much lower number of DEGs in Mo-roots were detected in the DAI6-DAI2 comparison than that in the Ho-root DAI6-DAI2 comparison. Furthermore, the expression of seven housekeeping genes in Ho-roots were similar to those of Mo-roots (Supplementary Table S4), suggesting that the basic level of gene expression was consistent in rice roots infected with either M. or H. oryzae.
One of the most striking features of the GO terms specifically enriched in the Ho-roots-DAI6-DAI2 comparison, as compared to the Mo-roots-DAI6-DAI2 comparison, was the relatively high number of GO terms involved in carbohydrate metabolism (Supplementary Table S5). This finding was consistent with the results from the metabolome analysis. These GO terms included "polysaccharide catabolic process" (GO:0000272), "cellular polysaccharide biosynthetic process" (GO:0033692), "monosaccharide catabolic process" (GO:0046365), "glycolysis" (GO:0006096), and "lignin catabolic process" (GO:0046274).
To comprehensively assess the biological functions of the DEGs, all DEGs were mapped to KEGG database terms with the goal of identifying significantly enriched metabolic or signal transduction pathways (Supplementary Table S6). Consistent with the GO enrichment analysis, pathways involved in carbohydrate metabolism, such as "starch and sucrose metabolism", "fructose and mannose metabolism", "glycolysis/gluconeogenesis", and "lignin biosynthesis" were identified in the Ho-roots-DAI6-DAI2 comparison, while no obviously enriched metabolic pathways were detected in the Mo-roots-DAI6-DAI2 comparison, with the exception of "lignin biosynthesis".
Validation of the RNA-seq data. The expression levels of 26 genes were assessed using qRT-PCR to validate the RNA-seq results (Supplementary Fig. S4; Supplementary Table S7). Among these 26 genes, 4 genes were housekeeping genes, 14 genes were involved in the starch degradation, glycolysis, and TCA pathways, 3 genes were involved in lignin biosynthesis, and 5 genes were involved in responses to biotic or abiotic stress in the M. and H. oryzae-challenged roots. Log2 transcript fold changes (DAI6 vs. DAI2, DAI20 vs. DAI2) measured using RNA-seq and qRT-PCR were used to assess the correlations between gene expression profiles obtained using the two experimental approaches. High linear regressions were observed for the DAI6-DAI2 comparison: for H. oryzae, R 2 = 0.886, and for M. oryzae, R 2 = 0.860. For the qRT-PCR DAI20-DAI6 comparison, only one gene was differently expressed, which was consistent with the results of the RNA-seq analysis. Although the fold change values for the expression of some genes measured by RNA-seq or qRT-PCR were slightly different, trends of all the examined genes at the various time points were the same. These results indicate highly consistent results from both techniques. The similar consistence between the RNA-seq and qRT-PCR results were also detected for the gene expressions of H. and M. oryzae in the same RNA-seq data sets used in this study 9 , suggesting the successful application of RNA-seq to analyze the gene expressions of the interacted fungi and plants simultaneously.
Divergent carbohydrate metabolism in rice roots in response to H. or M. oryzae colonization. As compared to the Control-roots, the changes in the concentrations of most carbohydrates were similar in the Ho-roots and Mo-roots, and included reductions of sucrose and maltose and increases of mannitol and trehalose (Fig. 5a). The concentration of glucose was significantly increased in the Ho-roots as compared to the Control-roots at all stages measured. However, a similar increase of glucose was not detected in the Mo-roots. The concentration of fructose increased dramatically at DAI6 in the Ho-roots as compared to both the Control-roots-DAI6 and the Mo-roots-DAI6 samples (Fig. 5a).
samples though the glucose concentration in the Mo-roots sample did not increase (Fig. 5a). This phenomenon can potentially be explained by the different expression patterns of genes coding for enzymes in glucose and fructose metabolism in the Ho-root and Mo-root samples. The expression of most genes encoding enzymes involved in glycolysis and the TCA cycle, including phosphoglucomutase, fructokinase, aldolase, glyceraldehyde-3-phosphate, phosphoglycerate kinase, enolase, dehydrogenase E, and malate dehydrogenase, were suppressed in the Ho-root samples, but enhanced (or not different than the controls) in the Mo-root samples (Fig. 5c). Suppressed glycolysis activity would theoretically result in the accumulation of glucose and fructose. Consistent with the suppressed expression of malate dehydrogenase in the Ho-root samples, there was an accumulation of malic acid in Ho-root samples during the colonization process; this was not detected in the Mo-root samples (Fig. 5a). These results suggest that differential gene expression programs were adopted by Ho-roots and Mo-roots to maintain different carbohydrate concentrations. These concentrations may be important in how rice roots respond differentially to the presence of a mutualist or a pathogen.
The shikimate pathway and defensive lignification in H. or M. oryzae-challenged roots. Sustained increases in both quinate and shikimate were observed in the Mo-root samples, indicating apparently enhanced flux through the shikimate pathway (Fig. 6a). In contrast, the concentrations of quinate and shikimate increased at DAI2 but then decreased to normal levels at DAI6 and DAI20 in the Ho-root samples, representing a modulation of the shikimate pathway different from that observed in the Mo-root samples (Fig. 6a). Consistent with the reduced shikimate pathway activation in the Ho-root samples, the enzymes catalyzing lignin biosynthesis, including phenylalanine ammonia-lyase, cinnamoyl-CoA reductase, peroxidase, and laccase, were all down regulated at DAI6 compared to DAI2 (Fig. 6c). However, enhanced (or unchanged) expression of these genes was detected in the Mo-root samples (Fig. 6c). Phenylalanine levels increased dramatically at DAI6 in the Mo-root samples compared to the controls; no such difference was detected between the control and the Ho-root samples (Fig. 6a). Besides, as a typical phenylpropanoid metabolite, the content of ferulic acid, which involved in resistance to pathogenic fungi 24 , was decreased at DAI20 in Ho-root samples compared to the controls (Fig. 6a). These results suggested that defensive phenylpropanoid metabolism was suppressed at the onset of the endophytic interaction between H. oryzae and rice, but that it was enhanced or sustained in the interaction between rice and M. oryzae. Consistently, the elevated tryptophan contents in Ho-root samples at DAI6 decreased to the same level of controls at DAI20 (Fig. 6a), the pathway of which also involved in the defense responses of rice against pathogenic infection 25 .

Discussion
Driven by the devastating impact of diseases in agriculture, extensive plant pathology research has been conducted that has resulted in substantial knowledge about how plants defend themselves against pathogens. There has also been intense research into how plants engage in mutualistic root symbioses [26][27][28][29] . In pathogenic interactions, nutrients flow from the host to pathogens; in mutalistic interactions, there is nutrient trafficking in both directions 30 . The interaction between plant roots and arbuscular mycorrhizal (AM) is a well-studied example of mutualism 31 . AM mutualism is an ancient plant-microbe interaction that is considered to have been crucial for the successful colonization of land by plants 32,33 . A study showed that 40% of the genes in rice roots whose expression was induced by AM were also induced when roots were infected with pathogenic fungi, revealing an ancient pattern of response to fungal infection 34 . Different from the AM symbiosis, benifical endophytism is evolutionarily transient and has originated multiple times 12,13 . It has been shown that some endophytes have evolved recently from pathogenic ancestors 9 . Empirical comparisons of the responses of a single host plant species to these mutualistic and pathogenic fungi with the same pathogenic ancestor can help to uncover the underling mechanism for the emergence of novel mutualistic interactions that have evolved recently.
It has been shown that H. and M. oryzae evolved from a common pathogenic ancestor 9 . Morphological observations suggested that similar root colonization strategies are used by H. and M. oryzae at the early stages of infection 9,23 . This assertion is bolstered here with evidence from the host plant side of the interaction showing a greater degree of divergence between the metabolite profiles of the H. and M. oryzae-treated rice roots at the middle stage (DAI6) as compared to the early stage (DAI2). Further, while the unique effects of the different fungi colonizations were evident at DAI6, there was some degree of overlap of the clusters for the Ho-root and Mo-root samples at DAI2. Even though rice roots respond similarly when challenged by these two fungi at the early stages of such interactions, our metabolic profiling analyses indicated a larger discrepancy of metabolic activity between Ho-roots and Control-roots than between Mo-roots and Control-roots at both DAI2 and DAI6, which indicates that rice roots challenged by M. and H. oryzae differentiate between day 2 and day 6 of fungal challenge. One possible explanation for this could be that the H. oryzae-rice interactions originate from the interaction between rice and a phytopathogenic ancestor. This more complex set of physiological responses observed in rice roots colonized by H. oryzae is reflected in the much larger number of DEGs that were detected in the Ho-root samples as compared to the Mo-root samples.
Many studies have revealed that sophisticated reprogramming mechanisms in a host's metabolism and physiology are required for successful colonization and/or establishment of pathogenicity [35][36][37] . In the establishment of AM symbiosis, a consistent set of genes is expressed in plant roots, including genes Scientific RepoRts | 5:13624 | DOi: 10.1038/srep13624 involved in nutrient transport, responses to biotic and abiotic stimuli, and the regulation of developmental processes 26,27,38,39 . For example, transcriptional changes triggered in Solanum lycopersicum roots and shoots by Glomus mosseae were detected by microarray and were mainly involved in primary and secondary metabolism, defense, and transcriptional regulation 27 . A similar study showed that genes induced during interactions between Medicago truncatula and AM had functions related to sugar transport, secondary metabolism, proteolysis, and changes in signaling 26 . These results suggest that a profound molecular reprogramming occurs in the roots during AM mutualism. Most of these studies of AM mutualism are based on cDNA microarray data and are thus limited by the fact that probe sets can only be deisgned based on known transcripts; many genes may be missed 27,39 . To better explore the intersection between endophytic and pathogenic interaction in rice roots, we used RNA-Seq to profile transcription changes. A large number of genes displayed differential expression between the Ho-root and Mo-root samples, though few such genes were shared in common in both sample types. This result highlights the differences in the reprogramming triggered by the establishment of either a mutualistic or a pathogenic interaction. GO enrichment analysis of these DEGs identified many enriched terms, most of which related to carbohydrate metabolism and/or responses to biotic and abiotic stress. Consistently, metabolic profile analysis showed that many of the differentially accumulated metabolites were part of carbohydrate or aromatic amino acid metabolism.
It has been shown that the reinforcement of cell walls by phenolics is part of a defense reaction against both pathogens and endophytes 35,36,40,41 . Early defensive activation of the phenylpropanoid pathway leads to production of lignin precursors for cell wall strengthening in resistance responses 42,43 . Phenylpropanoids are produced from precursors derived from the shikimic acid pathway, a pathway known to be transcriptionally activated by pathogen attack [44][45][46] . In our study, a transcriptional activation of lignin biosynthetic genes accompanied by elevated phenyalanine and tyrosine accumulation levels was evident in the Mo-roots. Consistently, the content of shikimate, a metabolite upstream of phenylalanine, was also observed to increase. Activation of the shikimate pathway leads to the biosynthesis of the lignin precursor compounds that are used for cell-wall strengthening in resistance responses 43 . The enhancement of the shikimate pathway and the transcriptional activation of lignin biosynthesis genes observed here indicate the presence of induced defenses in rice in response to M. oryzae. This has also been reported in other host-pathogen interactions 35,36 . Similarly, the shikimate pathway was induced in the Ho-root samples at DAI2, indicating induced plant defense responses in response to H. oryzae. However, at DAI6 and DAI20, the activities of shikimate pathway and lignin biosynthesis in the Ho-root samples were at the same levels as in the control samples. Similarly, defense-and stress-response genes were also induced in M. truncatula roots during AM mutualism 38 . Different from what occurs in AM mutualism, the rice defense responses observed in our study were activated but subsequently suppressed in the endophytic interaction with H. oryzae. DEGs involved in responses to biotic and abiotic stress were up-regulated in the Mo-root samples but down-regulated in the Ho-root samples. These results reveal the distinct responses of rice to infection by different fungi; showing that the defense responses in the interaction of rice with H. oryzae are suppressed, unlike the situation with rice and M. oryzae. These observations are consistent with the knowledge that the H. oryzae genome encodes more elements to suppress plant defense responses than does the M. oryzae genome 9 , including LysM-bearing proteins that are secreted and interact with chitin and a larger number of small secreted cysteine-rich proteins (SSCRP). Further, it is known that some virulence-associated gene orthologs are down regulated in H. oryzae but up regulated in M. oryzae; none of the SSCRP orthologs in H. and M. oryzae were regulated identically between the two species 9 . These different gene expression patterns in H. oryzae are also related to the suppression of plant defense responses.
It is known that in susceptible host-pathogen interactions in plants, invertase activity increases after infection 21,35,47 . This is accompanied by a progressive accumulation of hexoses and sucrose, and a reduction in photosynthesis in host leaves 47 . For example, in Brachypodium distachyon infected with M. oryzae, a rapid elevation of sucrose, glucose, and fructose content in leaves is evident when visible lesions first appear, and this is mirrored by a large increase in mannitol content 21 . Glucose and fructose levels in the rice roots of this study were not significantly affected by M. oryzae, though the levels of sucrose and maltose decreased, a result that may reflect the tissue-adapted invasion strategy of M. oryzae 22 . Unlike pathogens, mutualistic partners can improve the growth and fitness of the host plant by 'paying' their hosts with nutrients that are difficult to access by the plant or through indirect 'payment' such as the production of plant protective anti-herbivory alkaloids 48 . Improved plant fitness as a result of this interaction, however, has a price. For example, up to 20% of photosynthetically fixed carbon of host plants is translocated to the AM fungi which are obligate symbionts that entirely depend on carbon and energy supplied to them by their autotrophic hosts 49,50 . Together with their fungal partner, the root systems of ectomycorrhizal plants can receive about half of the plant's photosynthetically fixed carbon 51 . In contrast to the Mo-root samples, the glucose and fructose levels increased dramatically in the Ho-root samples. For AM, carbon produced by the host plant is taken up in the form of glucose or hexose 52,53 . The increase in free hexose content found in the Ho-root samples is typical for a sink tissue, and the free hexoses within rice roots could be used by H. oryzae as an easily accessible carbon source. In addition to its role as a carbon source, hexose has been shown to play other important roles in plant-fungi interactions, such as inhibiting the fungal expression of plant cell wall degrading enzymes 9 . It has been shown that the hexose content in rice is correlated to the nutritional preferences of H. or M. oryzae, and play important roles in the evolution of the interactions between rice and H. or M. oryzae leading to either mutualism or disease. However, the cause of the different contents of glucose and fructose in rice roots with different fungi-plant interactions is still unclear. The metabolite profiling and comparative transcriptomics analysis in this study imply that the accumulated glucose and fructose content in the Ho-root samples may relate to the apparently suppressed flux of glucose through glycolysis and the TCA cycle that we deduced from the suppressed expression of genes encoding enzymes involved in glycolysis and the TCA cycle. Sucrose degradation can also contribute to pools of fructose, the precursor of fructose phosphate. With Scientific RepoRts | 5:13624 | DOi: 10.1038/srep13624 the suppressed expression of fructokinase in the Ho-root samples, less fructose would theoretically be available for glycolysis, which could lead to an increase in fructose content. The rapid increase of glucose and fructose in H. oryzae-challenged roots was accompanied by a very significant increase in mannitol, especially at DAI6. Mannitol has been previously proposed to be an important carbohydrate for fungal growth 54 . In the interactions between fungal Neotyphodiumlolii endophytes and Lolium perenne, it has been shown that mannitol levels are significantly positively correlated with fungal biomass 36 . As mannitol is membrane-impermeable, the conversion of imported hexoses to mannitol might maintain a gradient for the continued uptake and sequestration of carbohydrates 21 .
It is now clear that clavicipitaceous endophytes arose from insect-parasitic ancestors 14 . However, this evolutionary pattern does not fit the often-expected transition from phytopathogens to endophytes 6 . We addressed this transition by suggesting a novel evolutionary model of the beneficial endophyte H. oryzae, which originated from a phytopathogenic ancestor 9 . Although M. oryzae preferentially infects aerial parts of rice with appressoria, Tucker 55 proposed that hyphopodia represent primitive appressoria and that the ability to develop appressorium-mediated penetration is acquired later with the incorporation of novel genes or functions involved in surface recognition (pth11 and mpg1), peg formation (gas1, gas2 and pls1), turgor formation (cpkA) and melanin synthesis. M. poae and G. graminis, the closely related fungi of H. and M. oryzae, are both pathogenic and obligated root-infection 9 . Similar to M. oryzae, both M. poae and G. graminis share the same pathogenic ancestor with H. oryzae 9 . Thus, the root-infection by these fungi is a more ancient character than the aerial part-infection, which is preserved by both H. oryzae and its closely related pathogenic fungi, indicating that the comparative analysis of M. and H. oryzae on rice roots can illuminate well the novel co-evolution pattern of fungi-host interaction. In the present study, our findings highlight the importance of the host in the evolutionary processes of fungi. The development of these diverse interaction strategies may have been caused by competition between H. oryzae and related species such as M. oryzae to occupy different ecological niches. It has been shown that H. oryzae colonization can restrict root infection by M. oryzae 23 , suggesting that competition between these closely related species does indeed occur. To make full use of a given carbon source, the ecological niches for their common pathogenic ancestors might have changed. This is revealed by the fact that H. oryzae has lost the ability to infect the aerial parts of host plant, a capacity that is retained by M. oryzae 9 . We have developed an experimental system that allows us to use the same plant to study the colonization biology of two different types of plant-fungi interactions with closely related fungi. Our results with this experimental system contribute significantly to the knowledge and genomics resources of the study of plant-fungi interactions and should prove useful in attempts to employ friendly microbes in crop production systems to control devastating plant diseases.

Methods
Biological materials and culture conditions. The fungal endophyte H. oryzae strain R5-6-1, the rice blast fungus M. oryzae strain Guy11, and the blast-susceptible rice cultivar CO-39 (O. sativa) were used in these experiments. Rice seeds were surface sterilized as described previously 23 . To promote root growth, the sterilized seeds were planted in half-strength Murashige & Skoog (1/2 MS) solid medium and cultured in the dark at 30 °C for 4 days 22 . The seedlings were then transferred to sterilized square culture dishes (13 × 13 cm), with 5 plants per plate, and grown vertically for an additional 6 days with a 16-h light/8-h dark photoperiod at 28/24 °C. H. oryzae strain R5-6-1 was cultured on complete medium (CM) 56 at 25 °C in the dark; M. oryzae strain Guy11 was cultured on the same medium, but with a 16-h light/8-h dark photoperiod.
Inoculation, co-culture, and harvest of rice roots. Conidia of M. oryzae were harvested from 10-day-old cultures grown on solid CM. Germinating phialidic conidia of H. oryzae were harvested from 4-day-old potato dextrose broth (with 5 g glucose/L), as conidia harvested from cultures grown on solid CM are not able to germinate 9 . Inoculations and co-culturing were performed as described previously 9 . Based on a previous study 9 , we chose to sample rice roots infected by H. oryzae at 2, 6, and 20 days after inoculation (DAI) and to sample rice roots infected by M. oryzae at 2 and 6 DAI (at DAI20, most of the roots infected by M. oryzae were dead). These particular sample tissues and time points represent contexts during which the main differences of the respective colonization strategies employed by H. and M. oryzae should be detectable. All samples were harvested and immediately flash frozen in liquid nitrogen. The control samples were prepared using sterile water instead of a conidia suspension for inoculation.

Sample Preparation for GC-MS analysis.
Grouped roots from 10 independent rice plants were grouped as a single experimental replicate, and 5-6 such biological replicates were prepared and analyzed for each treatment. Briefly, the root samples (10 mg dry weight per sample) were freeze-dried for 48 h and ground to a powder. Quality control (QC) samples were obtained by thoroughly blending the same amount of each ground sample 57 . The sample preparation for the GC-MS analysis was performed as previously described 9 . Chromatography was performed on an Agilent 7890/5975C-GC-MSD using an Agilent DB-5MS UI column (30 m × 0.25 mm × 0.25 mm) and helium as a carrier gas.
Metabolite assays by GC-MS. Samples  was injected with a split ratio of 10:1. The parameters and the procedures for the GC-MS analysis were performed as previously described 9 . Components of the total ion chromatogram were extracted using the Automatic Mass Spectral De-convolution and Identification System (AMDIS) (NIST, Gaithersburg, MD, USA). Metabolites were identified by comparing their mass spectra with those of commercial standards [57][58][59] . The peak areas were analyzed with an in-house automatic integration method established in Agilent MSD ChemStation. Manual corrections were performed to guarantee the accuracy of the integrations. The relative intensities of the various metabolites were obtained by normalizing the intensity of individual ion traces, which are indicative of the respective compounds, to the response of an internal reference compound. The internal standard and any known artificial peaks were removed from the results matrix 57 .
Statistical and multivariate analysis. The data matrix was imported into SIMCA-P 13.0 (Umetrics, Umeå, Sweden) for multivariate data analysis, and Pareto-scaled to minimize the influence of baseline deviations and noise 60 . Principal component analysis (PCA) was used initially to obtain an understanding of the relationships among the samples and metabolites. An orthogonal partial least-squares discriminate analysis (OPLS-DA) was performed to further differentiate the contributions of particular metabolites to the separations of the different sample groups. The efficiency and reliability of the OPLS-DA models were verified by percent variation of the x and y variables explained by the model (R2X, R2Y) and the predictive performance of the model (Q2) 57 . In addition, permutation tests were performed to validate the OPLS-DA models 61 .
Transcriptome analysis. The technique of RNA-seq was used in this study for the advantages of analyzing the gene expressions of the interacted fungi and plants simultaneously without the indiscrimination. The RNA-seq data of Mo-roots and Ho-roots used in the transcriptome analysis were same to the data sets used in our previous report 9 which only focused on the fungi in the rice-fungi interaction. As the genome sequences of rice cultivar CO-39, M. and H. oryzae are available 9,62,63 , all the clean reads were mapped to the genome sequences using TopHat v 2.0.9 64 , and an expression profile was created using Cufflinks v2.0.2 65 . The abundances of gene expression were reported as normalized fragments per kb of transcript per million mapped reads. Transcripts with a significant p-value (< 0.05) and a greater than two-fold change in transcript level between the Ho-roots and Mo-roots were considered as 'differentially expressed' . All the p-values were corrected for false discoveries resulting from multiple hypothesis testing using the Benjamini-Hochberg procedure. GO enrichment analysis was performed using a hypergeometric test with TopGO 66 , and p-values were adjusted with Bonferroni correction for multiple testing. The FDR criterion was selected as FDR < 0.05. Heatmaps of gene expression profiles were generated using R (www.R-project.org) based on significant expression changes (log2 fold change).
Quantitative RT-PCR. Total RNA of Ho-roots or Mo-roots was extracted as described above. Three independent biological replicates were performed for each sample. Quantitative RT-PCR was performed as described previously 9 . Four commonly used house-keeping genes were used (Supplementary Table  S7). One of the house-keeping genes encoding elongation factor 1-alpha was used as an endogenous control, and the stability of expression of the three other house-keeping genes was assessed. The relative gene expression levels of target genes were calculated using the 2 -∆∆Ct method 67 . The sequences of the primers used in the qRT-PCR analysis are listed in Supplementary Table S7.