Metabolic Profiling and Metabolite Correlation Network Analysis Reveal That Fusarium solani Induces Differential Metabolic Responses in Lotus japonicus and Lotus tenuis against Severe Phosphate Starvation

Root fungal endophytes are essential mediators of plant nutrition under mild stress conditions. However, variations in the rhizosphere environment, such as nutrient depletion, could result in a stressful situation for both partners, shifting mutualistic to nonconvenient interactions. Mycorrhizal fungi and dark septate endophytes (DSEs) have demonstrated their ability to facilitate phosphate (Pi) acquisition. However, few studies have investigated other plant–fungal interactions that take place in the root environment with regard to phosphate nutrition. In the present research work, we aimed to analyze the effect of extreme Pi starvation and the fungal endophyte Fusarium solani on the model Lotus japonicus and the crop L. tenuis. We conducted metabolomics analysis based on gas chromatography-mass spectrometry (GC-MS) on plant tissues under optimal conditions, severe Pi starvation and F.solani presence. By combining statistical and correlation network analysis strategies, we demonstrated the differential outcomes of the two plant species against the combination of treatments. The combination of nutritional stress and Fusarium presence activated significant modifications in the metabolism of L. japonicus affecting the levels of sugars, polyols and some amino acids. Our results display potential markers for further inspection of the factors related to plant nutrition and plant–fungal interactions.


Introduction
Lotus spp. belong to the Fabaceae family, which contains approximately 130 species. Lotus legumes have the ability to tolerate different biotic and abiotic stresses, and some of them are forage crops in constrained areas that have limitations for traditional agriculture use due to the restrictive soil conditions [1]. Lotus tenuis is a legume used as a forage resource in the Flooding Pampa region (Buenos Aires, Argentina). This region is mainly characterized by flooding, high salinity and high pH values, which lead to the low availability of nutrients such as phosphorus and iron in the soil. L. tenuis can grow and fully develop in these restrictive conditions; thus, it has been successfully naturalized in the Flooding Pampa region [1][2][3]. Research performed on L. tenuis has been accompanied by research on the model L. japonicus [4]. Both species have been used to characterize the fungal endophyte F. solani in Lotus species. To assess the effect of F. solani, we inoculated roots of the model species L. japonicus and of the forage crop L. tenuis with the endophyte and explored interactions with Pi nutrition in root and shoot tissue by comparing the Pi sufficiency to Pi starvation in inoculated and noninoculated plants. We profiled and analyzed the primary metabolism by conventional statistical approaches and took a step beyond by including a comparative correlation network approach as a strategy to detect changes in interactions between central metabolites in response to our experimentally defined conditions.

Plant Material, Inoculation, Growth Conditions and Sampling
Seeds of L. japonicus (ecotype Gifu B-129) and L. tenuis (cv. "Nahuel") were scarified with sulfuric acid for 3 min and washed 10 times with distilled water. Furthermore, the sample seeds were surface disinfected with 5% sodium hypochlorite and washed with sterile distilled water. The scarified-disinfected seeds were incubated overnight in sterile distilled water until imbibition. The seeds were placed in Petri dishes containing an agarwater medium (0.5% w/v) and cultivated in growth chamber under a 16/8 h photoperiod at 24/19 • C (day/night), a light intensity of 240 mmol −2 s −1 and 60% humidity for 10 days. The fungal endophyte F. solani strain 142L52B (in the following Fsol), which was isolated from roots of L. tenuis growing in constrained soils [14], was used to inoculate roots of L. japonicus and L. tenuis under experimental conditions.
The fungal strain was cultivated in a Sabouraud agar medium (SAM) [33] for 12 days at 28 • C in dark conditions. Solid SAM (approximately 1 cm −2 ) plugs containing Fsol mycelia were used as inoculants. The inoculation of Lotus spp. roots with Fsol was carried out during the seedling transplant. Seedlings were transferred to pots containing a quartzsand:perlite substrate (proportion 3:1). Each seedling was inoculated with SAM plugs containing Fsol mycelia (FUS+). Control inoculation was performed with sterile plugs of fresh media (noninoculated). The inoculum was placed at 1 cm depth in each pot. Simultaneously, sufficient Pi and Pi starvation treatments were performed using modified Evans solution [34] containing 0.1 mM KH 2 PO 4 and 0.1 mM K 2 HPO 4 as Pi sources or an Evans solution without the Pi fraction. Each combination between species and inoculation was performed under optimal Pi conditions (P+) or Pi starvation (P-) using the aboveindicated nutrient solutions.
The final set of experimental treatments of L. japonicus and L. tenuis included the following combinations: Control (noninoculated, optimal Pi), FUS+ (inoculated, optimal Pi), P-(noninoculated, Pi starvation) and FUS+P-(inoculated, Pi starvation). The experiment was carried out with 10 biological replicates per treatment combination. Plants were harvested 32 days postinoculation and following Pi treatment. Shoots and roots were separated, frozen, ground in liquid nitrogen and stored at −80 • C until analysis.

In Vitro Cultivation of Fusarium solani 142L52B and Sampling of Mycelia and Secreted Compounds
To determine the compounds secreted by Fsol and the compounds accumulated by the mycelial cells, we analyzed polar metabolites from mycelia and the supernatant medium produced by Fsol under in vitro conditions. Cultivation of Fsol was performed by inoculating one plug of approximately 1 cm −2 in 50 mL of Sabouraud broth medium (SAB) incubated on a shaker for 7 days at 28 • C. Fresh SAB was used as the control. The experiment was conducted in five biological replicates. After 7 days, mycelia were separated from the supernatant medium by filtration and ground to a fine powder in liquid nitrogen. The mycelia and supernatant were stored at −80 • C until metabolite extraction.

Metabolite Extraction and Profiling of Plant Tissues, Mycelia and Fungal Supernatant
The extraction of polar metabolites from Lotus tissues was conducted using methanol:chloroform:water liquid-phase extraction [35]. The general extraction procedures were carried out as reported by Erban et al. [36]. An extraction solution containing 300 µL of methanol and 30 µL of a 13 C 6 -sorbitol stock solution (0.2 mg/mL in distilled water) was added to 200 mg of finely ground fresh material. The mixture was heated at 70 • C for 15 min, followed by the addition of 200 µL of chloroform and incubation at 37 • C for an additional 5 min. Polar phase separation was induced by adding 400 µL of distilled water and centrifugation at 20,000× g for 5 min. An aliquot of 160 µL in the polar phase was dried for further chemical derivatization. It consisted of methoxyamination and the subsequent silylation, namely, by incubation with 40 µL of 40 mg/mL methoxyaminehydrochloride in pyridine for 90 min at 37 • C and an 80 µL silylation mixture containing 70 µL N,O-bis-(trimethylsilyl)-trifluoroacetamide (BSTFA) and 10 µL alkanes in pyridine for 30 min at 30 • C.
Gas chromatography (GC) was performed with a phenyl/polysiloxane (5:95%) coated 40 m capillary column with an inner diameter of 0.25 mm. The column included 10 m of a nonseparating precolumn. One µL of the final sample volume was injected. Mass spectrometry (MS) was performed at nominal mass resolution using an electron impact ionization time of flight mass spectrometer [36].
Data mining was conducted using TagFinder Software [37] after chromatogramspecific alkane-based retention index calculations were performed. The chromatograms were baseline corrected, and mass-fragment abundances were retrieved as chromatographic peak heights. Annotation was manually supervised using the Golm Metabolome Database Mass-spectral Collection (GMD, http://gmd.mpimp-golm.mpg.de/, accessed on 15 June 2018) [38]. At least three mass fragments per analyte were used within the TagFinder software for linearity and specificity checks and for relative quantification. Abundance data, i.e., the peak heights of these mass features were normalized to the internal standard for 13 C 6 -sorbitol and the sample fresh weight. In the case of fungal supernatants, the abundance data were normalized to 300 µL instead of fresh weight.

Statistical Analysis of Metabolite Data
The normalized abundance data of the shoots, roots, fungal mycelia and fungal supernatant were separately analyzed. Missing value (NA) replacement was achieved using the input.knn function of the impute package [39] for the R environment. The normalized abundance of mass features representing each metabolite was divided by the averaged normalized abundance of the control treatment of each plant species to obtain the relative abundance ratios, i.e., fold changes relative to the noninoculatedoptimal Pi control conditions. These ratios were subsequently log 2 -transformed. The log 2 -transformed ratios were used to compare the metabolic responses to each treatment combination. To determine the interaction between the "Pi status" (optimal/starvation) and "Fsol presence" (control/inoculated), experimental factors, two-way analysis of variance (two-way ANOVA) was applied to the combination of these treatments within each of the Lotus species (p < 0.05). Pairwise comparisons were assessed using the Student's Test (p < 0.05). Only the mass features of the metabolites that were detected in at least 60% (before the NA replacement) of the experimental replications of each combination of the treatments were considered for the statistical analysis. The average values obtained from non-sample SAB medium samples were subtracted to estimate the quantity of secreted fungal metabolites in the cultivation supernatants. Principal component analysis (PCA) was performed after missing value substitution using the packages FactoMiner [40] and Factoextra [41] for the R environment.

Network Analysis of the Metabolite Data
To assess the patterns of correlation and the respective Hub metabolites in correlation networks, we conducted network analysis based on correlation matrices of Log 2transformed fold changes of normalized metabolite abundances. Instead of noninformative correlation across all combinations of experimental conditions, we analyzed the subsets of each treatment, namely, Control, FUS+, FUS+P-and P-of the shoot and root organs, and the two Lotus species separately. Pairwise Pearson s correlation coefficients were calculated among all the metabolites of each of the subsets. To avoid the selection of an arbitrary numerical correlation threshold, we applied the context likelihood of relatedness algorithm (CLR) to define or reject a correlation between metabolites [42]. The CLR was applied to the complete correlation matrix to remove spurious correlations [43]. The calculation of correlations and CLR transformation were performed using the functions cor and clr of the stats and parmigene R-packages, respectively [44]. We constructed weighted permuted networks based on the resulting adjacency matrices of metabolites using the functions graph_from_adjacency_matrix (weighted = TRUE, mode = "undirected", diag = FALSE) and permute (sample (vcount)) from the igraph R package [45]. This procedure resulted in the treatment-specific individual networks of each experimental condition, organ and species. We selected the degree of node (in the following degree) as a network topology parameter, which allowed us to select the nodes with the most interactions in the network. The degree of a node is defined by the number of edges linking it to other nodes within the network [46]. If the topology of a network contains few highly connected nodes, they are defined as Hubs [47]. In our current study, the most interactive nodes, i.e., the nodes with the highest degree values, represented the Hub metabolites of each experimental condition. We assigned relevance values to the Hub metabolites following previous studies that proposed and explored this concept [43,47]. As an additional parameter to characterize metabolites according to their position relative to the other metabolites of the network, we included betweenness centrality (in the following betweenness) calculations for each node. The betweenness value of a node is determined by the number of geodesic distances between the other two nodes in the network [46]. This parameter captures, in layman's terms, the extent that a given node is "in-between" other nodes on paths across a network and thereby indicates the importance of a given node for connecting any two other nodes in a network. Betweenness is considered an indicator of the importance of a node for the potential flux of information between nodes of a network [48].

Interaction between the Fusarium solani Endophyte and Lotus spp. under Extreme Pi Starvation
To assess the involvement of the Pi status in the response of both Lotus species to inoculation with the endophyte Fsol, we analyzed the biomass after a growth period of 32 days, according to the time of the most evident differences between control and Pi starvation conditions, and before starting the flowering period. Pi starvation treatment significantly affected both Lotus species. Pi starvation reduced the biomass of shoots and roots, independent of the presence of Fsol ( Figure 1A,B).
To assess the impact on central metabolism, we analyzed a metabolite fraction enriched for primary metabolites that contained 160 polar metabolites in the shoot analyses and 158 in roots [35,49,50]. In agreement with the effect on biomass, Pi starvation affected the primary metabolism as a dominant experimental factor, with only minor contributions of Fsol inoculation to the total variance of our data set ( Figure 1B,C). In addition to the effects of experimental interventions, samples grouped according to the plant species were the second most important factor affecting primary metabolism in Lotus species, similar to earlier comparative metabolite profiling studies on abiotic stresses [35,49]. Multivariate analysis by PCA of shoot data revealed that PC1 explained 32.8% of the variance. PC1 grouped Pisufficient and Pi-starved samples irrespective of the presence of the endophyte ( Figure 1C). The species difference and presence of species-specific responses to Pi starvation became apparent through PC2, with 11.4% of the variance explained. The PCA results of the root data displayed similar patterns to those detected for shoots ( Figure 1D). In this case, the grouping induced by Pi starvation was more evident in PC2, with 14.6% of the variance explained, and the metabolic responses of the roots from the two species overlapped. After statistical analysis, detailed differences between all treatments and the control condition became apparent, including FUS+ inoculation and the combination FUS+P-. The results of the statistical testing of all metabolites detected in the shoots and roots are reported in Supplementary Tables S1 and S2, respectively. In addition to statistically significant differences, we searched for metabolites that became nondetectable after Fsol treatment and/or due to Pi starvation. Conversely, we found metabolites that were absent in the control conditions and became detectable only in the P-and/or FUS+P-treatment conditions (Supplementary Tables S1 and S2). These metabolites were considered relevant without statistical testing and were scored as present or absent.
In addition, Fsol inoculation of L. tenuis under optimal Pi supply caused a significant decrease in the concentration of phosphoric acid ( Figure 2; Supplementary Table S2), while L. japonicus plants, under the influence of Fsol, maintained phosphoric acid levels similar to the control conditions. Two-way ANOVA (p < 0.05) applied to analyze further phosphorylated compounds detected in the roots was not significant (Supplementary Table S2).
Despite the decrease in Pi-related compounds in the roots, the concentration of phosphorylated metabolites in the shoots remained mostly unaffected according to pairwise statistical testing ( Figure 2). However, the results obtained from two-way ANOVA revealed interactions between the Pi status (optimal/starvation) and Fsol presence (control/inoculated) experimental factors. Significant interactions were detected for the myoinositol-P in L. japonicus and L. tenuis shoots and for ribulose 1,5-diphosphate (Ribulose 1,5-2P) in the L. tenuis shoots ( Figure 2). The differences observed in both Lotus species for myo-inositol-P exhibited a statistically significant increase in the shoots of L. japonicus for the FUS+P-treatment, while the opposite effect was observed in L. tenuis.

Metabolites Representing Potential Carbon Sources
The partitioning of carbon resources is a key factor in plant-fungal interactions. We emphasize carbohydrate partition analysis by examining the potential carbon mobilization within the plants and the pathways involved in this process. Our results displayed differ-ential responses regarding the sugar quantities after Pi starvation and Fsol inoculation. The concentration of sugars detected in the shoots and that detected in the roots are shown in Figure 3. . Log 2 -transformed relative concentration of sugars detected in the shoots and roots of L. japonicus and L. tenuis after 32 days of treatment: Control (noninoculated-optimal Pi), FUS+ (inoculated-optimal Pi), P-(noninoculated-Pi starvation) and FUS+P-(inoculated-Pi starvation). Statistical differences between the Control (noninoculated-optimal Pi) and each treatment (FUS+, P-and FUS+P-) were assessed using Student s test (mean ± standard deviation, n = 10), * p < 0.05, ** p < 0.01, *** p < 0.001, nd: not detected. Two-way ANOVA results of the interaction FUS+P-; # p < 0.05, ## p < 0.01 and ### p < 0.001 for L. tenuis.
The results revealed changes in the sucrose levels. The amount of sucrose decreased significantly in the roots of L. tenuis for all the treatments evaluated. In contrast, L. japonicus responded with a significant decrease only after Pi starvation. Furthermore, fructose increased in the shoots of L. japonicus after Pi starvation, while in L. tenuis, fructose and glucose decreased slightly in response to the Pi starvation treatment. Moreover, Fsol elicited the reduction of fructose in L. tenuis under Pi-sufficient conditions. The shoots of L. tenuis exhibited significant interactions of the Pi status and Fsol experimental factors for four of the six sugars detected in our analysis. This interaction had an additive effect on the sugar levels ( Figure 3; Supplementary Table S1).
We detected an increase in maltose levels in the treatments involving Pi starvation. Both plant species showed the same tendency of a major increase in maltose content in shoots for the FUS+P-treatment combination. This effect was strongest in L. japonicus and significant in the two-way ANOVA of L. tenuis (Figure 3).
The sugar conjugate galactinol, which is a biosynthesis intermediate of the raffinose family of sugars, was also detected by our analysis. The levels of galactinol decreased exclusively in the shoots of L. tenuis under all the treatment combinations ( Figure 3). In addition, differences in the levels of raffinose were detected in both Lotus species. In L. japonicus, statistically significant differences were observed in the P-treatment when the levels of raffinose were increased, while in L. tenuis, the reduction of this sugar was observed in the FUS+ treatment ( Figure 3).
In the shoots, α-α-trehalose was reduced in all the treatments of L. tenuis. In contrast, this sugar was detected in roots at increasing levels when inoculated with Fsol. The amount of α-α-trehalose was significantly higher in L. japonicus following inoculation with Fsol. In the case of L. tenuis, the increase in this sugar in the roots was only significant in the FUS+P-treatment (Supplementary Table S2).
Polyols (also called sugar alcohols) exhibited a specific differential response according to the plant species and treatments. Arabitol, erythritol and glycerol accumulated in the shoots of both plant species under Pi starvation. On the other hand, the ononitol and myo-inositol levels decreased in both species under the same conditions. Ononitol and pinitol significantly decreased in the shoots of L. tenuis under the Fsol treatment. However, pinitol accumulation was detected in the roots of both plant species (Figure 4).

Organic Acids and Polyhydroxy Acids
The organic acid content belonging to the tricarboxylic acid pathway (TCA) can be modified under stress conditions, such as waterlogging [51], alkalinity [6] and salinity [49]. Modifications in the levels of these acids were detected in the treatments involving Pi starvation in both plant species.
The concentration of citric acid increased in the shoots and roots of L. japonicus under Pi starvation conditions, while in L. tenuis, this effect only occurred in roots. The 2methyl malic acid responses were similar to that of citric acid ( Figure 4). Succinic acid increased significantly in the shoots of L. japonicus under Pi starvation, while in L. tenuis, this compound increased only in the roots.
The glyceric acid and erythronic acid levels decreased under the FUS+P-and Ptreatments in both Lotus species. The same effect was observed in L. tenuis for threonic acid. The rest of the detected polyhydroxy acids increased significantly under Pi starvation, e.g., saccharic acid, which is an intermediate compound in the TCA cycle ( Figure 4; Supplementary Table S1).

Amino Acids and Other Nitrogen-Containing Compounds
Overall, the detected amino acids increased in both Lotus species and organs after treatment with Pi starvation (Figure 4; Supplementary Tables S1 and S2). Asparagine and tryptophan displayed the most relevant increases in the roots. These amino acids also showed interactions between our experimental factors, as detected by two-way ANOVA ( Figure 4; Supplementary Table S2). Among other nitrogen-containing compounds, Nacetyl-glucosamine, 6-amino-1-methyl-uracil and 3-hydroxy-pyridine exhibited a statistically significant increase in the shoots of both Lotus species under the treatments involving Pi starvation. In the roots of L. japonicus, nitrogen-containing compounds were not detected or were not present in statistically significant amounts, except 3-hydroxypyridine. In the roots of L. tenuis, significant decreases in ethanolamine were detected for all the treatment combinations. A significant increase in 6-amino-1-methyl-uracil was detected in the roots of L. tenuis for the treatments involving Pi starvation (Figure 4, Supplementary Table S2). The results of two-way ANOVA applied to shoot measurements displayed a statistically significant interaction between factors for putrescine detected in the shoots of both Lotus species (Supplementary Table S1).

Metabolites of Mycelia and Secreted into the Medium by Fusarium solani
To detect the compounds that can be potentially secreted by the endophyte, we characterized the liquid supernatant medium after growing Fsol under in vitro conditions. The characterization of the primary metabolism of mycelial cells allowed the prediction of the compounds necessary for the normal growth of Fsol under optimal conditions. The results obtained in this analysis of metabolites produced by Fsol demonstrate that phosphorylated compounds were present mostly as intracellular components of the mycelia rather than secreted into the surrounding medium ( Figure 5).  Glycerol-3P, glucose-6P and mannose-6-phosphate (mannose-6P) were the phosphorylated compounds with the highest apparent abundance detected in our analysis. These compounds were mainly detected in mycelium cells. Phosphoric acid and phosphoric acid monomethyl ester were present in mycelia at higher relative abundances than in the supernatant. In addition to other compounds, such as amino acids and polyols, myo-inositol and arabitol were detected in higher amounts in mycelium cells. As expected, the mycelia contained a higher amount of sugar than the supernatant medium ( Figure 5). α,α-Trehalose was detected in high relative abundance in mycelia compared to the supernatant and blank medium. The relative abundance observed for this compound exceeded the values detected for other sugars in mycelia, such as glucose and isomaltose.

Correlation Network Analysis
To gain more insights into the changes in the metabolic network covered by our profiling analyses, we generated correlation networks of metabolite abundances across all replicate measurements of each of the 16 experimental conditions, namely, Control, FUS+, FUS+P-and P-treatments of shoots and roots from L. japonicus and L. tenuis, respectively. The resulting 16 correlation networks of metabolite nodes connected by edges representing CLR positive abundance correlations, generated as outlined in our "Material and Methods" section, were highly interconnected, as can be expected from metabolite profiling analyses enriched for primary metabolites (Supplementary Figure S1 and Table S3). The networks had similar overall shapes. Differences among the networks were difficult to detect and to recognize by visual inspection. Therefore, we performed an in-depth analysis of the network properties of degree and betweenness (refer to "Material and Methods" for definitions). We specifically identified Hub metabolites, i.e., those nodes in our networks that were most interconnected compared to average or typical nodes and thereby characterized the differences among the network topologies that were related to the investigated experimental conditions.
The degree distribution analysis of our correlation networks ( Figure 6A,B) revealed degree ranges of~20-60 or in part extending to >70 of the correlation networks of shoots ( Figure 6A,C). In agreement with the smaller number of detected metabolites of the corresponding correlation networks of roots, the degrees ranged from~15 to~50 ( Figure 6B).
The degree values did not follow normal distributions but were, in most cases, multimodal. Following visual inspection, multimodality was not found to be associated with our experimental conditions. Therefore, we neglected the multimodality aspect and focused on the clear shift of the degree distributions towards the higher values that we observed in the shoots of L. japonicus, specifically under the FUS+P-and P-treatments. In contrast to the other experimental conditions, multiple metabolites that we will term in the following as Hub metabolites passed an arbitrarily chosen degree threshold, i.e., >50, in the FUS+P-or P-L. japonicus shoot samples. (Figure 6A). None of the other degree distributions exhibited such clear shifts towards higher degree values except to a minor extent for the FUS+Ptreatment of L. tenuis shoots. In this case, a small number of metabolites exceeded 50 degrees. Other than the correlation networks of shoot metabolites, no significant shifts of degree distributions that were associated with either treatments or species differences were observed for the root correlation networks ( Figure 6B). Control (non-inoculated-optimal Pi), FUS+ (inoculated-optimal Pi), P-(non-inoculated-Pi starvation) and FUS+P-(inoculated-Pi starvation), Fum: fumaric acid, Azel: azelaic acid, Ben: benzoic acid, Succ: succinic acid, Citric: citric acid, Lac: lactic acid, Tart: tartaric acid, Glutar: glutaric acid, DihydroMal: Di-hydroxy-malonic acid, 2Me.Malic: 2-methyl-malic acid, 2I.Malic.acid: 2-isopropyl-malic acid, 6A. Cap: 6-amino-caproic acid, Asn: asparagine, β-Ala: β-alanine, Ala: alanine, Asp: aspartic acid, His: histidine, Ile: isoleucine, Trp: tryptophan, Gln: glutamine, Lys: lysine, Val: valine, Orn: ornithine, Thr: threonine, Leu: leucine, 3H. Pyr: 3-hydroxy-pyridine, 6AMe. Uracil: 6-amino-1-methyl-uracil, N-ac-glu: N-Acetylglucosamine, Me.P.acid: phosphoric acid monomethyl ester, P.acid: phosphoric acid, RuBP: ribulose 1,5-2P, GPGlycerol: glycerophosphoglycerol, G6P: glucose-6P, MyoInP: myo-inositol-phosphate, Gal: galactonic acid, Galac: galactaric acid, Ribo: Ribonic acid, Gly: glyceric acid,Ery: erythronic acid, 4L.thr.ac: threonic acid-1,4-lactone, thr.ac: threonic acid, αα-tre: α-α-trehalose, "/" indicates "and". Motivated by the obviously altered network degree distributions associated with the treatment of Pi starvation in the absence of Fsol inoculation or in combination, we performed an in-depth analysis of the degree values of shoot metabolites. We highlighted global metabolite classes in this analysis, e.g., amino acids, and compared them across all eight metabolic correlation networks of the shoot samples ( Figure 6C). Hub metabolites were present in all chosen metabolite classes, e.g., diverse organic or amino acids, nitrogencontaining compounds, phenylpropanoids, and diverse carbohydrates ( Figure 6C). In general, and as expected from the degree distribution analysis, most Hub metabolites were present in the FUS+P-L. japonicus shoot network. Fewer Hub metabolites were present in the L. tenuis FUS+P-network or in the P-L. japonicus network. We found three main classes of metabolite observations; the first was the Hub metabolites (1) that had high degree values specifically in the FUS+P-L. japonicus shoot network. The second class was Hub metabolites (2) that were common to and highly ranked within the FUS+P-and P-L. japonicus networks. The FUS+P-L. japonicus shoot network yielded more Hub metabolites (Supplementary  Table S3) and, in many cases, higher degrees than the respective P-network. The third class of Hub metabolites (3) had high degrees in the combinatorial stress FUS+P-of both the L. japonicus and L. tenuis shoot networks.
Among class (1), the specific Hub metabolites of the FUS+P-L. japonicus shoot samples were sugars and polyols. Glucose, sucrose and raffinose had high degree values in the L. japonicus FUS+P-network, whereas α-α-trehalose ranked highest in the respective control and in most of the other networks ( Figure 6C). This observation was accompanied by high degree values for pinitol, ononitol, and myo-inositol in the L. japonicus FUS+P-shoot network. The pinitol degree value of 75 from this network was the largest overall observed degree value and was far higher than that in the respective L. tenuis FUS+P-network. Other FUS+P-L. japonicus-specific observations included phenylpropanoids, transferulic acid and 4-hydroxycinnamic acid, fumaric acid or, to a lesser degree, succinic acid.
Concerning class (2), i.e., the commonly high rankings in FUS+P-and P-L. japonicus Hub metabolites, we found several nitrogen-containing compounds and amino acids, for example, Asp and β-Ala. These compounds had the highest degree values in the L. japonicus FUS+P-and P-networks compared to 6-aminocaproic acid, which ranked highest in most of the other networks ( Figure 6C). This class contained diverse metabolites, such as catechin, benzoic acid and several polyhydroxy acids, e.g., erythronic or galactonic acid.
Observation class (3) contained a set of biotic or abiotic stress-related compounds that had high degree values in the combinatorial FUS+P-stress networks. Salicylic acid glucopyranoside had the highest degrees in the L. japonicus and L. tenuis FUS+P-networks ( Figure 6C). Azelaic acid ranked highest in the L. tenuis FUS+P-network and, similar to 2-methyl malic acid, was a Hub metabolite in the L. japonicus FUS+P-, L. tenuis FUS+, and L. tenuis Control networks. Catechin was a common Hub metabolite of the L. japonicus and L tenuis FUS+P-networks and overlapped with class (2) featured in the L. japonicus Pnetwork. Catechin shared this pattern with, for example, N-acetyl-glucosamine, galactonic acid, threonic acid or phosphoric acid mono-methyl-ester ( Figure 6C).
To further explore and specify the multiple observed Hub metabolites in our correlation networks of shoot samples, we calculated the betweenness centrality of a node as a second topology parameter. The betweenness of metabolite nodes was independent of the degree (Supplementary Figure S2). As specified in our "Materials and Methods" Section 2, betweenness can be considered an alternative indicator of the importance of a node for a potential flux of information between distant nodes within a network [48]. We assigned high relevance to metabolites with a combination of high degree and betweenness values and metabolites with high betweenness and low to medium degrees. Guided by the previous analyses, we focused on the combinatorial FUS+P-stress of the L. japonicus (Supplementary Figure S2A) shoot samples compared to the L. tenuis shoot samples (Supplementary Figure S2B).
For the correlation network of the combinatorial FUS+P-treatment of L. japonicus, 10 of 13 metabolites yielded high betweenness values, which were arbitrarily set as larger than the 95% quantile (Supplementary Figure S2A). This global result underlined the difference in the metabolic combinatorial stress state compared to single stress conditions. A set of Hub metabolites reported earlier for the combinatorial stress state had high betweenness values, namely, β-Ala, benzoic acid, myo-inositol, arabitol, ononitol galactaric acid, phosphoric acid and sucrose (Supplementary Figure S2A).
Applying the same betweenness threshold to the L. tenuis correlation networks, we discovered three metabolites with high betweenness, i.e., salicylic acid glucopyranoside, β-Ala, and ribonic acid (Supplementary Figure S2B). In agreement with its link to biotic stress responses, salicylic acid glucopyranoside had high betweenness in the L. tenuis FUS+ and FUS+P-networks but a high degree only under combinatorial stress. β-Ala had high betweenness and degree values only under FUS+ single stress conditions, and ribonic acid had high degree values under all stress conditions but increased degree values from FUS+ to P-and FUS+P-stresses (Supplementary Figure S2B).

Pi Starvation Masks the Effect of the Endophyte Fusarium solani in Both Lotus Species
The results obtained in this study demonstrated that extreme Pi starvation in Lotus spp. constitute a stronger stress than the effect caused by the interaction with Fsol. The measurement of biomass performed in the experiment exhibited a morphological response to the combination of both treatments (FUS+ and P-) and displayed the same effect as the response produced by the effect of Pi starvation itself (P-). According to this result, the hypothesis that Pi constitutes a determinant factor in the interaction between Fsol and Lotus species is not supported. Although for biomass no differences were observed between the P-and FUS+P-treatments, particular variations in the levels of metabolites exposed differences in the responses by the two Lotus species to nutritional stress mediated by the presence of Fsol in combination with Pi starvation. The multivariate approach based on PCA indicated that roots were affected by Pi starvation, independent of plant species and Fsol presence. On the other hand, shoots revealed a differential response to Pi starvation according to the Lotus species. The Pi deprivation effect has been found to be related to a complex reprogramming of metabolism [52]. This assumption, in addition to the observations obtained in this work, could indicate that particular outcomes in response to Pi starvation cannot be fully explained by observing only biomass.
Roots were certainly the first organ affected by the nutritional stress evaluated. In this trend, the information obtained from the metabolomics analysis shows that Pi depletion is evident in the root system. However, shoots conserve a certain concentration of phosphorylated metabolites to maintain metabolic functions.
The biomass production response suggests that extreme Pi starvation might be a minor aspect of the differential behavior of the interactions between Lotus spp. and Fsol observed in our previous study. Moreover, different patterns observed in the phosphorylated compounds in the correlation network indicate the possibility that both Lotus species have different mechanisms to deal with Pi starvation, independent of the presence of Fsol.

Modulation of Carbohydrate Availability by Lotus spp.-Fusarium solani Interactions under Pi Starvation Varies According to the Lotus Specie
In addition to the differential response observed in phosphorylated compounds, we also detected differential behaviors in the response related to the sugar content in L. japonicus and L. tenuis under Pi starvation. Our results show an increase in sugar compounds as a consequence of Pi deprivation. Nevertheless, this effect was only observed in L. japonicus, while L. tenuis exhibits decreases in the levels of glucose and fructose under the same stress conditions.
The induced carbohydrate shortage extends to the invertase-cleavage products of sucrose. Pi starvation induces a reduction in glucose/fructose that is highly similar in L. tenuis compared to L. japonicus roots. However, L. japonicus accumulated glucose in shoots in response to the Pi limitation. This result clearly demonstrated the differential carbohydrate availability in L. japonicus compared to L. tenuis. Taken together with sucrose observations, L. tenuis appears to become limited in soluble carbohydrates after inoculation with Fsol and Pi starvation. Furthermore, L. tenuis may not be able to control and limit carbohydrate/carbon withdrawal by Fsol. Thus, Fsol might be on the verge of being a mild pathogen of L. tenuis. This result is consistent with our previous study of the interaction of Fsol-Lotus spp., which demonstrated the differences in the carbohydrate contents of the two Lotus species after infection by Fsol under optimal growth conditions [14].
The raffinose pathway may support the observation of Fsol-induced sucrose and hexose shortages in L. tenuis. This pathway is also linked to galactose (conjugated from galactinol) and myo-inositol. Myo-inositol approximately mirrors the sucrose and raffinose concentrations in L. tenuis roots and the Pi levels in shoots. Indeed, raffinose was reduced in L. tenuis after Fsol inoculation but, like glucose, increased in L. japonicus as part of the Pi limitation stress response. This observation confirms the Fsol-induced sucrose shortage in L. tenuis. The galactinol levels also mirror sucrose reduction and may indicate a limitation of the galactinol precursors after Fsol inoculation of L. tenuis. In fact, galactinol reduction is more significant than raffinose reduction and part of the Pi limitation response. In summary, the effects observed in the raffinose biosynthesis pathway support the hypothesis that Fsol induces sucrose shortage in shoots and roots and glucose/fructose shortage predominantly in shoots. Both sucrose shortage and reduced hexose availability appear to limit galactinol/raffinose biosynthesis.
Maltose levels increased under Pi starvation. This result could be related to the transitory starch breakdown to supply the energetic requirements induced by the stress condition [53]. This effect was more evident when Fsol was present in combination with Pi starvation and was significant for the FUS+P-interaction in L. tenuis. Moreover, the levels of maltose in L. tenuis increase as a consequence of the additive effect of Fsol and Pi starvation. This result might reinforce the assumption regarding the hexoses requirement in the interaction between Fsol and L. tenuis.
The levels of α-α-trehalose decreased significantly in L. tenuis under the treatments involving Fsol, Pi starvation and the combination of both conditions. This compound has previously been described as a signaling molecule in abiotic stresses and plant microbial interactions [54]. Rhizobia-legume interactions analyzed in Phaseolus vulgaris demonstrated that an exogenous application of trehalose induces sucrose synthase and alkaline invertase activities [55]. This evidence reinforces the hypothesis of a sucrose shortage in L. tenuis in interaction with Fsol. In addition, this sugar was detected at a high level as a mycelium component.
It is relevant to note that the increase in the levels of α-α-trehalose in the roots of Lotus spp. is likely due to the presence of the endophyte Fsol. These results suggest that α-α-trehalose could be considered a potential indicator of Fsol presence in the root system. In addition, this sugar is highly related to important fungal functions. Trehalose has been described as a reserve carbohydrate and stress-related metabolite in yeast and filamentous fungi [56]. Moreover, the accumulation of trehalose by yeast under Pi starvation has also been described [57]. It also plays important roles in relevant biological traits, such as sporulation and infectivity of certain fungal pathogens [58], including Fusarium [59].
Sugars constitute more than photosynthesis-derived products. Different studies reported to date have demonstrated the role of sugars as signaling compounds [60][61][62][63]. From a nutritional point of view, it has been demonstrated that the Pi levels influence the sugar response [29]. The concentration of sugars increases under low Pi levels and enhances the expression of genes involved in the Pi starvation response [64]. Here, we observed an increase in sugar content only in L. japonicus. According to our results, L. japonicus may exhibit better performance than L. tenuis against Pi depletion. However, L. tenuis grows and successfully develops in restricted environments that have diverse abiotic stresses, including Pi limitation. Taken together, the evidence, in addition to our results, elucidates the occurrence of differential responses of both Lotus species to the same stress. In this trend, if sugars are able to act as signaling compounds in L. japonicus, it might not be the same mechanism used by L. tenuis to achieve the same response to the nutritional stress.

Nitrogen-Related Compounds Responded Differentially to Pi Starvation
In legumes, Pi starvation involves a cost of carbon and nitrogen in the root systems [65]. Pi starvation in legumes affects the utilization of organic nitrogen sources and the availability of organic carbon required for the synthesis of amino acids in nodules [66]. Despite the absence of nodulation under our experimental conditions, we detected differences in the content of amino acids provoked by Pi starvation.
We detected a significant increase in asparagine in the roots of both Lotus species under the effect of extreme Pi starvation. This response was also observed previously by Almeida et al. [67] in the roots of Trifolium repens under severe Pi starvation. Previous studies have stated that the synthesis of this amino acid can be induced by mineral deficiencies [66]. Similar behavior was observed in the arginine levels. The enhancement in the biosynthesis of arginine has also been described in legumes under nutritional stress involving Pi [68].
We observed an increase in tryptophan under Pi starvation. The concentration of tryptophan in the roots and shoots of both plant species might lead to the production of indole acetic acid (IAA). This effect has been correlated with the promotion of root growth, which triggers modification of the root architecture to improve nutrient acquisition [7,8,[69][70][71]. Moreover, auxin and peptides have been described as regulatory components in signaling to regulate Pi acquisition. The pathways involving the synthesis of tryptophan have been associated with the production of the secondary metabolites and proteins necessary to address the shortage of amino acids [72]. Nevertheless, our results demonstrated an increase in the amount of tryptophan, but instead of shortage, we detected an increase in the levels of the other amino acids. However, only the involvement of asparagine with the rest of the metabolites was evident in the correlation network analysis. This result emphasizes the role of asparagine in the response to extreme Pi starvation and underlines the differential response between the model (L. japonicus) and the crop (L. tenuis).
The accumulation of polyamines has been correlated with the tolerance of L. tenuis and L. japonicus to abiotic stress [73][74][75][76]. However, the response of polyamines to different stress conditions in Lotus spp. seems to be dependent on the type of stress. Despite the relationship between polyamines and stress tolerance, it has also been demonstrated that the putrescine content is not in agreement with the saline stress in L. tenuis [9]. In our analysis, the content of putrescine, instead of accumulating, decreased in the roots of L. tenuis and shoots of L. japonicus under the Pi starvation treatment. Nevertheless, we observed an increase in putrescine in the shoots of L. japonicus under the combination of Pi starvation and the Fsol effect. Similar behavior was observed in the symbiosis between L. tenuis and mycorrhizal fungi under salinity stress. Echeverria et al. [76] stated that polyamine content is affected by the symbiosis between L. tenuis and Rhizophaguis irregularis (ex. Glomus intraradices) only when salt stress is present. This result added to the evidence observed in previous studies aimed at determining the multiple roles of polyamines in different biotic and abiotic stress conditions.

Organic Acids Might Play an Important Role in the Response to Pi Starvation
Elevated concentrations of organic acids belonging to the TCA cycle were exhibited in the treatments involving Pi starvation. An increase in succinic acid under Pi starvation was previously observed in the common bean by Hernandez et al. [77]. Nevertheless, the results obtained in the common bean differ from our results regarding other acids. We observed an increase in malic acid and citric acid in L. japonicus mediated by Pi starvation. This effect has been described before in Brassica napus [78,79], rice [80] and legumes such as Virgilia divaricata [81], chickpea [82] and lupine [83]. Moreover, Touhami et al. [84] demonstrated that forage legumes showed higher concentrations of the organic anions used to improve the acquisition of Pi in the root system than nonlegume grasses. The increase in the concentration of organic acids in roots has been related to the secretion of root exudates containing anions to solubilize phosphoric resources [85,86].
The metabolite correlation network analysis showed that fumaric acid had the highest number of correlations with other metabolites in L. japonicus for the treatment combining Fsol presence and Pi starvation. This result suggests fumaric acid as a potential Hub metabolite in the interaction between Fsol and L. japonicus under the effect of Pi starvation, revealing the possible role of fumaric acid in the regulation of the TCA cycle under these experimental conditions. Additionally, the role of this acid as a source of carbon transport could be considered [87].

Role of Polyols in the Pi Starvation Response
Our results showed an increase in the amounts of some polyols as a consequence of Pi starvation, such as arabitol, erythritol and glycerol. The accumulation of ononitol has been related to other abiotic stresses in Lotus spp. [49]. Nevertheless, in our study, the amount of this polyol was significantly reduced under Pi starvation conditions, with a further effect on the accumulation of pinitol. In this trend, it seems unlikely that ononitol is a metabolite representative of all abiotic stresses.
The multiple roles of polyol compounds have been previously reported as osmoprotectant molecules. However, their importance in the stress response has also been related to their ability to react as a quencher of reactive oxygen species [88]. Moreover, the physiochemical properties of polyols influence photosynthesis, respiration and developmental processes because they are a sink of photosynthates [89]. Polyols are an important part of photosynthesis-derived compounds, and they are transported as storage molecules via phloem [89]. Our results demonstrated the accumulation of some sugar alcohols under Pi starvation, while the concentrations of ononitol and pinitol decreased under the same conditions. This effect could be explained by their synthetic pathway, which is dependent on the concentration of Glu6P as the main precursor and is strongly affected by Pi starvation.
Polyols also accumulate in fungi for multiple functions, such as carbohydrate storage, osmoregulation, translocation of compounds and enzyme regulation [90]. These compounds were detected as mycelium components of Fsol. This observation supports the role of these compounds in fungal metabolism. In this trend, the variability in the number of polyols detected in roots could be partially explained by the presence of endophytes.

Phosphorylated Compounds Are Accumulated Instead of Excreted by Fusarium solani
The results obtained in the metabolic analysis of mycelia and supernatant medium after the growth of Fsol demonstrated that phosphorylated compounds are constitutive metabolites of the mycelia rather than excreted in the medium. Because phosphorus is a limiting nutrient in the production of secondary metabolites by fungi [91], its function in the metabolism of Fsol is relevant. In addition, the metabolic profile obtained from Fsol showed the accumulation of phosphoric acid in the mycelia in values corresponding to a ratio of 0.5 of glucose. This assumption might suggest competition between Lotus plants and Fsol for Pi resources under extreme Pi starvation. However, our results demonstrate neither amelioration nor declining Pi stress under the effect of Fsol. This premise is relevant considering that Fsol was isolated from the roots of L. tenuis growing in soils with low Pi availability [14].

Correlation Network Analysis Remarks
The degree patterns in the control conditions displayed lower degree values than those obtained in the different treatments involving Fsol and Pi starvation in both Lotus species. This effect could be interpreted as an equilibrium under control conditions, mediated primarily by the behavior of pathways under optimal conditions. On the other hand, an increase in the degree values for the different treatments evaluated indicates specific metabolites that vary similarly to the stress conditions. This effect was also observed in the rise or impairment of specific metabolites absent in control conditions. The detection values were observed only in the FUS+, P-or FUS+P-conditions. Network analysis showed the relevance of pinitol in the involvement with the other metabolites analyzed. This effect, observed only in L. japonicus for the combination of Fsol and Pi starvation, denotes the differential mechanisms of response to the Pi stress combination for both Lotus species. Our results also exposed the presence of myo-inositol and other alcohol sugars as metabolites that can be determinant in response to the combination Pi starvation Fsol. In addition, the involvement of sugars and sugar conjugates in the metabolic response to FUS+P-was confirmed in the analysis of topology parameters derived from the correlation network analysis. The presence of indicators of the involvement of polyols and sugars resulting from the observations obtained from the degree and betweenness centrality analyses reinforces the hypothesis regarding the sugar shortage in L. tenuis mediated by Fsol.
The relationship between pinitol and myo-inositol displays the involvement of polyols and sugar pathways in the interaction between L. japonicus-Fsol-Pi starvation. These compounds may be the origin of new hypotheses and experiments to perform a deep characterization of the pathways involved in the synthesis, storage and translocation of carbon substrates. In addition, the involvement of fumaric acid in this system reveals potential alternative pathways to address the energy demand produced by the combination of Pi starvation and Fsol presence.

Conclusions
F. solani cannot modify the effect of extreme nutritional stress. However, the effects observed in carbon mobilization depend on the Lotus species and might be facilitated by the interaction with Fsol. The presence of several Hub metabolites involved in the combination of Fsol and Pi starvation exposed the complex outcome caused by F. solani on the metabolism of L. japonicus under this condition. Among our results, sugars and sugarrelated compounds exhibited relevant alterations in the metabolic response to the FUS+Pcombination. A deeper focus on the mechanisms underlying these particular interactions would provide information that allows the performance of plant-fungal interactions in constrained environments to be predicted.

Supplementary Materials:
The following are available online at https://www.mdpi.com/article/ 10.3390/jof7090765/s1, Figure S1: Network plots based on the pairwise Pearson-CLR-corrected correlations of Log 2 -fold changes of metabolite relative concentrations, Figure S2: Scatter plot of the degree and betweenness centrality parameters obtained from the networks based on pairwise Pearson-CLR-corrected correlations of Log 2 -fold changes of metabolite relative concentrations for shoots (A) and roots (B), Table S1: Results of two-way ANOVA and Student s test applied to metabolites detected in shoots, Table S2: Results of two-way ANOVA and Student s test applied to metabolites detected in roots, Table S3: Topological parameters of each correlation network.

Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.