Metabolomic effects of the colonization of Medicago truncatula by the facultative endophyte Arthrobacter agilis UMCV2 in a foliar inoculation system

Biofertilizer production and application for sustainable agriculture is already a reality. The methods for biofertilizers delivery in crop fields are diverse. Although foliar spray is gaining wide acceptance, little is known about the influence that the biochemical features of leaves have on the microbial colonization. Arthrobacter agilis UMCV2 is a rhizospheric and endophytic bacteria that promotes plant growth and health. In this study, we determined the capacity of the UMCV2 strain to colonize different leaves from Medicago truncatula in a foliar inoculation system. By using two powerful analytical methods based on mass spectrometry, we determined the chemical profile of the leaves in 15-d old plants. The metabolic signatures between the unifoliate leaf (m1) and the metameric units developing above (m2 and m3) were different, and interestingly, the highest colony forming units (CFU) was found in m1. The occurrence of the endophyte strongly affects the sugar composition in m1 and m2 leaves. Our results suggest that A. agilis UMCV2 colonize the leaves under a foliar inoculation system independently of the phenological age of the leaf and it is capable of modulating the carbohydrate metabolism without affecting the rest of the metabolome.


Results
Chemical signature of M. truncatula leaves. Control leaves were characterized using metabolic phenotyping by DLI-ESI-MS. This tool provided the broad molecular weight range of compounds in the leaves. The mass spectra obtained showed 1121 ions, but the ion intensities in the unifoliate leaf (m1) were different from those obtained in the first (m2) and second trifoliate leaves (m3) (Fig. 1a,b). The ordering analysis clearly separated the m1 from the m2 and m3 samples with a high significance (p < 0.001), contrasted using PERMANOVA (Fig. 1c). To determine the most important ions that distinguish m1 from m2 and m3 leaves, we constructed a Random Forest (RF) model that delivered the 30 most important ions (Fig. 2a). The PCA from the selected ions displayed a group belonging to m1 with an error of less than 1% (Fig. 2b), thus, the separation of m1 from m2 and m3 was even greater after the ion selection (Figs. 1c and 2b). In addition, the RF model did not distinguish between m2 and m3, with an error of up to 40%. These results suggest that the physiological condition of the m1 leaf has a clearly distinct metabolic identity compared to the m2 and m3 leaves, which presented the same chemical profile at the time of harvest.
The DLI-ESI-MS results of the 30 ions were compared with the SpiderMass database. Only 13 ions were putatively identified ( Table 1) Table 1).
The relative signal magnitude of the 182.07 m/z ion that matched to indole acetaldehyde in the database was higher in m2 and m3 leaves, whereas the magnitudes of the remaining ions were all higher in m1 leaves. Thus, the 182.07 m/z ion could be established as a metabolite marker for the phenological age of M. truncatula leaves (Fig. 2b).
Detection and quantification of A. agilis UMCV2 in internal leaf tissues. The anatomical structure and chemical features of leaves are presumed to affect the internalization, survival, and growth of the endophytes 7,25 . In regards to the chemical landscape from different metameric units of M. truncatula, we determined that the chemical profile of the m1 leaf was dissimilar to those of m2 and m3; nevertheless, this difference might not influence the endophytic colonization of A. agilis UMCV2. The combination of FISH ( Fig. 3a-c), qPCR technique, and total viable account determination (Fig. 3d) showed that the endophyte is able to colonize the m1, m2, and m3 leaves after foliar inoculation. Furthermore, m1 showed the highest colonization, while m2 and m3 showed no differences in bacterial DNA content (p = 0.1687) (Fig. 3d). In addition, we measured the bacterial population sizes in m1 leaves at different post-inoculation times (24,72, and 120 h); again, no significant differences were observed in the CFUs. The inoculation experiment was repeated with the half of the cells in the inoculum but no significant differences were observed in this experiment either ( Supplementary Fig. S1), showing that the size of the initial inoculation concentration did not modify the final colonization obtained, tending to stabilize to values lower than 1 × 10 6 CFU/leaf.

Metabolite profiling and sugar content in leaves during plant-microbe interaction.
To determine if the metabolite profile changed after the inoculation, we analyzed the leaves by DLI-ESI-MS at day 5 of the interaction. As the ions obtained were the same for both the inoculated and uninoculated leaves, a heat map was constructed with the ions selected by the RF model (Fig. 4). The hierarchical clusters grouped the samples according to the leaf development and inoculation. The control m1 leaf samples clustered separately from the m2 and m3 samples, which were indistinguishable. Furthermore, it was not possible to associate the global chemical changes obtained by DLI-ESI-MS with the occurrence of the endophyte. As DLI-ESI-MS did not cover the complete metabolome, we performed a GC-MS analysis as a complementary method for the identification and quantification of monosaccharides and disaccharides in the leaves. A factorial analysis performed with the results obtained by the chromatographic determinations showed differences in sugar content between the control and inoculated leaves (p = 1.12 × 10 −5 ) ( Table 2). Fructose and ethyl-D-glucopyranoside accumulated more in inoculated m1 leaves compared to uninoculated controls; in contrast, glucose and myo-inositol were more www.nature.com/scientificreports www.nature.com/scientificreports/ abundantly detected in inoculated m2 leaves. Sucrose also was identified, but no differences in levels were found between treatments. Furthermore, no statistical differences in individual or total sugar content (glucose, fructose, myo-inositol, sucrose, and ethyl-D-glucopyranoside) were found between control and inoculated m3 leaves. However, total sugars showed the highest values in the m1 leaves, which corresponded to the largest bacterial colonization.

Discussion
Plant growth promoting rhizobacteria (PGPR) are microorganisms that greatly influence plant growth and productivity and exert an optimal effect on plant health 26 ; therefore, in crop fields, they are either irrigated in the rhizosphere or applied as foliar spray 13,27,28 . Some PGPR can colonize plants and thrive inside the tissues, gaining access to the host plants via openings in the root or through stomata, wounds, and hydathodes in the shoot when they fall on the foliage either by artificial inoculation or because of the wind 11 .
The rhizobacterium A. agilis UMCV2 used in this study is a facultative endophyte, and has been shown to promote the growth and development of M. truncatula under iron-sufficient and -deficient growth conditions 21,22,29,30 . This study was performed to determine if A. agilis UMCV2 was able to colonize the different metameric units of M. truncatula seedlings under a foliar inoculation system. In addition, taking advantage of this form of bacterial inoculation, we performed a metabolomic profiling to determine whether the chemical composition linked to the phenological age of the leaves influences endophyte growth and colonization. First, we analyzed the metabolite profile of M. truncatula leaves to discern the chemical differences between them. At the time of the harvest, the seedlings had one unifoliate leaf (m1), and two trifoliate leaves (m2 and m3). Using DLI-ESI-MS and statistical models for data analysis, we determined that the metabolite profile of m1 was different from that of m2 or m3. Although the ion signals were the same, the intensities of many of them displayed higher relative abundances in m1 leaves. The m1 leaf is the first true leaf produced by the epicotyl; the meristematic region of the epicotyl then continues to grow, adding trifoliate leaves on alternate sides of the developing stem 31 . Thus, the differences in ion intensities among the leaves may correspond to the leaf phenological age [32][33][34][35] . Some of the compounds putatively identified are part of the membranes (e.g., 1, 2-dilinolenoyl-digalactosyldiacylglycerol).  www.nature.com/scientificreports www.nature.com/scientificreports/ Monogalactosyldiacylglycerol (MGDG) and digalactosyldiacylglycerol (DGDG) constitute 80% of the total membrane lipids of the thylakoid membranes and they act directly in several important plastid roles, particularly during photosynthesis 36,37 . In contrast, isochorismate is involved in plant defense responses as a precursor of salicylic acid (SA), a phenolic phytohormone widely studied for its role in activating defense responses during phytopathogen attacks 38 . Indeed, SA accumulation in older leaves has been positively correlated with increased disease resistance compared to the youngest leaves [39][40][41] . Similarly, saponins are also involved in plant defenses against herbivores and pathogens, in addition to having roles as allelopathic agents 42 .
Brassinosteroids are steroids that regulate many aspects of plant growth and development, especially in stress adaptation 16 . They act both independently and in conjunction with other phytohormones, controlling different plant metabolic activities. The brassinosteroid that we found in the m1 leaf is hydroxylated at C-26, leading the inactive form; 43 it should be noted that the most bioactive form named brassinolide was previously identified in M. truncatula by using a targeted analysis by GC-MS 16 . Additionally, we found increased levels of the auxin-related compound indole-3-acetaldehyde (182.07 m/z) in the youngest leaves (m2 and m3). This compound www.nature.com/scientificreports www.nature.com/scientificreports/ acts as storage compound for the production of indole-3-acetic acid 44 . Auxins induce G1 to S (G1/S) phase transition of the cell cycle, promoting proliferation, cell wall laxity, and cell expansion 45,46 . Thus, the presence of indole-3-acetaldehyde in m2 and m3 leaves is a biomarker of the leaf phenological age in M. truncatula seedlings.
With a complementary study using GC-MS, we monitored quantitative differences in soluble sugar content in the leaves (e.g., D-glucose, D-fructose, myo-inositol, ethyl D-glucopyranoside, and sucrose). These small sugars are critical for plant growth, development, and stress responses 47,48 . Measurement of the total sugar concentration in each leaf indicated that m1 leaves accumulate more sugars than m2 and m3. According to the literature, the unifoliate leaf is a source tissue that supports the growth and development of sink tissues displaying active growth through sucrose translocation 49 .
At 5-d of foliar inoculation and using a confocal microscope, we observed bacterial microcolonies in the m1, m2, and m3 leaves, indicating that A. agilis UMCV2 is able to internalize and colonize all the tissues independently of the physicochemical features of the leaves. The pattern of colonization followed by endophytes is typically not homogenous and it is specific of the microbial taxon and strain type and also they present affinity towards their hosts 25,[50][51][52] . Therefore, physiological parameters such as plant genotype, developmental stage, nutritional status, type of organs (leaf vs. root), leaf age, among others, are involved in the colonization process 6,7,10 . The present study shows that A. agilis UMCV2 colonizes the leaves of M. truncatula, and the plant maintains the colonization independently of the metabolic differences between them. A heavy inoculation does not produces significate differences in the colonization of m1 leaves over time. Previous studies have indicated that endophytes must pass the first line of defense of the plant immune system and determine how to protect themselves from it, but at the same time, the plants have different defense strategies to keep microbial growth in control 53,54 . Thus, The red, yellow, and blue marks correspond to m1, m2, and m3 leaves, respectively. The black and grey marks correspond to control and inoculated leaves, respectively. The control m1 leaf clearly clusters outside the m2 and m3 leaves. No differences were found between control and inoculated leaves. The heat map demonstrates the high repeatability of the measurements.

Sugars (µg/g FW) Control
A  Table 2. Sugar content in control and inoculated leaves of Medicago truncatula. Data shown are means ± SE for samples from 7 replicates. Control leaves were collected from 15-day-old plants. Plants were inoculated at 10 days of age, and the interaction lasted for 5 days. The same letter represents no significant differences between means by a two factors ANOVA followed by a Tukey´s post-test (p ≤ 0.05).
an exchange of numerous chemical signals between the plant and endophyte then allows successful colonization by the endophyte, suggesting that survival of the endophyte in the intracellular environment is likely to be a specific adaptation 1,2,15 . For example, in the symbiotic interaction between M. truncatula and rhizobia, the NAD1 (Nodules with Activated Defense 1) gene is expressed in the later stage of nodulation to maintain the rhizobial endosymbiosis in the nodule 55 , further studies are needed to determine the molecular mechanisms by which the plant allows A. agilis UMCV2 to thrive inside the plant. The classification models of the metabolite profiles obtained by DLI-ESI-MS failed to discriminate between inoculated and uninoculated leaves at 5 d of the post-inoculation time; not even in m1 leaves in which a higher colonization was observed compared to m2 and m3 leaves. This result was not completely unexpected since one characteristic of commensal and mutualistic endophytes is that they do not cause any harm to the host plant, but under specific biotic or abiotic stresses, they may induce metabolic changes in the plant. Therefore, we conducted a finer study using GC-MS and observed an increased content of sugars in the inoculated m1 and m2 leaves compared to non-inoculated controls (p = 1.12 × 10 −5 ). Sugar metabolism is essential for plant growth and development, but also it works as sensing elements to environmental stimuli, so alterations in the sugar content in plants have been reported as a response to biotic and abiotic signals [56][57][58][59] . It has been established that endophytes modulate the photosynthetic capacity of a plant which results in production of sugars that provide the basic carbon scaffolding necessary to form both plant and microbial structures 60 . We also noted that the sugar content did not increase in inoculated m3 leaves. The m3 leaf is the youngest leaf and is a sink tissue; thus, the presence of the endophyte would constitute an additional sugar demand.
In summary, our study showed that rhizobacteria A. agilis UMCV2 is able to colonize the inner leaf tissues of M. truncatula when it is inoculated in the foliage, but it is likely to be more easily established in the unifoliate leaves, so the hypothesis that the interconnected phenological age and chemical composition of the leaves affects the endophytic colonization is not discarded. These findings provide a first step in understanding the complexity of the metabolic environment that microorganisms must endure in the leaves following foliar fertilization. It will be very interesting to conduct a subsequent metabolomic studies to determine if A. agilis UMCV2 impacts or not in the broad range of compounds formed at early stages of the colonization or during the presence of biotic or abiotic stimulus, or furthermore, if A. agilis UMCV2 as a facultative endophyte contributes to the production of metabolites, that help the plant to adapt to, or cope with, environmental stimuli 61 .

Materials and Methods
Biological material and growth conditions. Medicago truncatula seeds (ecotype Jemalong A17) were chemically scarified 62 with 1 mL of concentrated sulfuric acid; the seeds were immersed in a vial containing the acid, and then agitated for 8 min. The acid was decanted, and the seeds were rinsed with five washes of sterile deionized water. For sterilization, seeds were immersed in 12% sodium hypochlorite for 2 min and rinsed with abundant sterile deionized water. Seeds were placed on vertically oriented Petri dishes containing 1% agar and 0.6% sucrose and vernalized at 4 °C for 3 days. Seeds were germinated until the main root reached a length of 2 to 3 cm. One seedling was subsequently transferred into individual glass jars containing 20 mL of Hoagland medium and 1% agar. Later, the seedlings were returned to the Percival growth chamber with a photoperiod of 16 h a day at a light intensity of 200 μmol m −2 s −1 at 22 °C for 15 days.
The M. truncatula leaves were named according to a standardized numerical coding system 31 to define the phenotypic development of this legume plant. The metamer associated with the unifoliate leaf was designed as metamer 1 (m1), and the units developing above as m2 (the first trifoliate leaf), and m3. Here, the 15-day-old plants collected for chemical and molecular analysis had the unifoliate leaf (m1) and the first and second trifoliate leaves (m2 and m3, respectively; Fig. 1a). The m1 emerged on the fourth day after seed germination; at harvest, therefore, m1 was at 11 days of development, m2 at 8 days, and m3 at 6 days.
The root-associated bacteria A. agilis UMCV2 was isolated from lightly acid soil, as previously described 29 . The bacterium was grown on nutrient agar (NA) at 26 °C. The inoculation was made in one leaf per plant with a natural hair brush 3 mm in diameter. Two brushstrokes were made (approximately 2.2 × 10 6 colony-forming units, CFU, per brushstroke), one in the upper surface and the other under the leaf, ensuring that none of the bacterial suspension fell into the growth medium. Plants were placed in three groups of twenty-four. In the first group of plants, only the m1 leaves were inoculated; in the second group, only the m2 were inoculated; and in the third, only the m3 were inoculated. The inoculated plants were returned to the Percival growth chamber, and the interaction was allowed to proceed for 5 days, until the plants were 15 days old. Then, the leaves were harvested for mass spectrometry analyses and qPCR bacterial DNA quantification.

Inoculation experiments.
To determine if a smaller bacterial concentration in the inoculum lead to a greater endophytic colonization, we performed an additional inoculation experiment, in which the m1 leaves were inoculated with a bacterial solution prepared at 0.1 DO 590 .
Mass spectrometry analysis of leaves. Leaves  www.nature.com/scientificreports www.nature.com/scientificreports/ Each milled leaf was analyzed by direct liquid injection-electrospray ionization-mass spectrometry (DLI-ESI-MS) 63 using a ZQ-detector 2 Waters ® mass spectrometer in scan mode with a range of 50 to 2000 m/z, programmed with a cone temperature of 135 °C, capillary voltage of 3 V, and taking 1 scan s −1 , until 1 minute of data acquisition. The aqueous solution containing 75% methanol acidified with 0.1% formic acid (blank sample) was injected at the beginning and at the end of each round of samples for washing purposes, and three technical samples were included at different days to corroborate reproducibility over time. The experiments were repeated three times.
Statistical models for data analysis. For mass spectra processing, the MALDIquant package in R language was used 64 . The intensity of each ion signal (m/z) was normalized using the Statistics-sensitive Non-linear Iterative Peak-clipping (SNIP) algorithm to correct the baseline; and for ion detection, we used the absolute mean deviation (MAD) algorithm as the noise estimator with a signal-to-noise ratio of six. Finally, alignment of the detected signals was performed using a tolerance of ±0.35 m/z. We used a principal component analysis (PCA) to compare the DLI-ESI-MS metabolic fingerprint data of the leaves.
To determine the contribution of every ion, we used the rattle package (version 5.1.0) in R to construct a Random Forest (RF) model 65 . Implementation of the RF algorithm using m/z features, allows the creation of a large number of decision trees that classify the experimental units as belonging to any one treatment with high predictability and the characteristics that best predict it. For the model, 70% of the matrix was used to train the model, 15% for the validation step, and 15% for testing. Five hundred decision trees were built to finally obtain a predictive model that could classify samples depending on leaf age or the presence of the UMCV2 strain. Subsequently, the 30 most important ions for leaf classification were identified by the Mean Decrease Gini criterion; ions with higher values play the most important role in predicting the correct leaf samples classification. Analysis of ordination by PCA was performed again using the "classifier" ions. To validate the separation of the data obtained between treatments, a heat map was constructed in combination with a hierarchical cluster analysis using the Euclidean distance between experimental units and Ward's algorithm for classification.
Putative metabolite identification. The most important ions were putatively identified according to the RF algorithm, using SpiderMass software 66 and a M. truncatula metabolite database (PlantCyc database, https:// www.plantcyc.org/). Ion identification was based on the mass-to-charge fit with a tolerance of ±0.35 m/z. Analysis of soluble sugars. Sugars were extracted from 50 mg of controls and inoculated m1, m2, and m3 leaves with 1 mL of 80% ethanol at room temperature with continuous agitation for 12 h. Each treatment consisted of twenty composite samples of three plants. The sample was centrifuged at 15 000 rpm for 15 min and the extract was evaporated to complete dryness at 40 °C and then derivatized and analyzed using the GC-MS method according to a previous report 67 . The GC-MS analysis was performed in a gas chromatograph (Agilent 6850 Series II; Agilent, Foster City, CA, USA), equipped with an Agilent MS detector, model 5973, and an HP-5 MS capillary column (5% [w/v] phenyl methyl polysiloxane, 30 m × 0.25 mm I.D., film thickness of 0.25 μm). Glucose, fructose, myo-inositol, and sucrose were identified using a combination of a NIST 2.0 mass spectra database search and a comparison with the pure standard (Sigma-Aldrich) retention times and mass spectra. For quantification, the standards were mixed (1 μg each), derivatized, and 1 μL was injected into the GC. The peak area of each standard within the mixture was correlated with the peak area of the eluted compound in the sample. The analysis was repeated three times. Data were analyzed using a two factor ANOVA (inoculum and leaf) followed by a Tukey´s post-test (p ≤ 0.05).
In situ hybridization assay. To corroborate that bacterial DNA detected by qPCR originated from bacteria exhibiting endophytic growth, inoculated leaves were randomly collected and images of the inner planes of the clarified tissues were obtained using a confocal microscope (FV1200, Olympus Corporation, Tokyo, Japan) as previously reported 20 . The inoculated bacteria were localized using fluorescence in situ hybridization, according to the FISH protocol described previously 68 . Actinobacteria 5 'Rhodamine Green TM-X (NHS Ester) TAT AGT TAC CAC CGC CGT was used 69 , which hybridizes with the 23 S ribosomal RNA of A. agilis and is visualized as green fluorescence (excitation and emission wavelengths: 502 and 527 nm, respectively).
Bacterial quantification. Leaves (m1, m2, and m3) from control and inoculated plants (n = 8) were superficially disinfected with 20% sodium hypochlorite for 3 min, washed three times with abundant sterile deionized water, and ground in liquid nitrogen. The method for bacterial DNA quantification by qPCR was as previously described 20 . DNA was extracted from powdered leaves employing the methodology reported previously 70 ; the 2CV2F and 2CV2R primers designed to specifically amplify an amplicon in the A. agilis UMCV2 16S-23S internal transcribed spacer (ITS) were used for absolute quantification of bacterial DNA in a StepOne thermocycler (Applied Biosystems, Foster City, CA, USA). A calibration curve was used from 100 ng to 20 fg with a coefficient of determination (R 2 ) of 0.998 and a maximum dissociation peak at 83.78 °C. In each assay, 50 ng of total DNA was used as template. Three independent analysis were done. To estimate the number of CFUs in the inoculated leaves, we performed a regression curve by using the total viable account obtained from 100 μL of five serial bacterial dilutions (0.5, 0.4, 0.3, 0.2, and 0.1 at 590 nm) and the DNA content in each solution. Serial dilutions of each bacterial suspension were plated in nutrient agar and after 5 d of incubation at 23 °C, the colonies produced were counted to calculate the CFU at each bacterial suspension. The extraction of the total bacterial DNA was made from 1 mL of each bacterial suspension as previously described 70 . The amount of DNA was quantified with a Thermo Fisher NanoDrop 2000. The regressions curve produced with the data of total viable account and DNA amount of each bacterial suspension had a R 2 value of 0.9448.