A metabolomic study of Gomphrena agrestis in Brazilian Cerrado suggests drought-adaptive strategies on metabolism

Drought is the main factor that limits the distribution and productivity of plant species. In the Brazilian Cerrado, the vegetation is adapted to a seasonal climate with long- and short-term periods of drought. To analyze the metabolic strategies under such conditions, a metabolomic approach was used to characterize Gomphrena agrestis Mart. (Amaranthaceae) a native species that grows under natural conditions, in a rock-field area. Roots and leaves material from native specimens were sampled along different seasons of the year and LC–MS and GC–MS analyzed for multiple chemical constituents. The datasets derived from the different measurements were combined and evaluated using multivariate analysis. Principal component analysis was used to obtain an overview of the samples and identify outliers. Later, the data was analyzed with orthogonal projection to latent structures discriminant analysis to obtain valid models that could explain the metabolite variations in the different seasons. Two hundred and eighty metabolites were annotated, generating a unique database to characterize metabolic strategies used to cope with the effects of drought. The accumulation of fructans in the thickened roots is consistent with the storage of carbons during the rainy season to support the energy demand during a long period of drought. The accumulation of Abscisic acid, sugars and sugar alcohols, phenolics, and pigment in the leaves suggests physiological adaptations. To cope with long-term drought, the data suggests that tissue water status and storage of reserves are important to support plant survival and regrowth. However, during short-term drought, osmoregulation and oxidative protection seems to be essential, probably to support the maintenance of active photosynthesis.


Results
Water status of the soil and plants. The Table 1. From September to November (2013) and February to July (2014), great variations in rainfall were observed in the region, with the highest index in April (75 mm of rain). In February, even in middle of the rainy period, a reduction in rainfall rates was detected (13 mm in February). The past data for the region have shown lower precipitation rates in February (INMET, 2017), leading to a short dry period during the rainy season (this period is popularly called "Veranico"). In July, which is a part of the dry season, no precipitation was recorded.
The soil moisture was positive correlated (p < 0.05) with the rainfall recorded for the period and ranged from 0.4 to 17.3% (Fig. 1). Interestingly, during the rainy season (from November to April), a decrease in the percentage of soil moisture, in February, matches with the Veranico period. In the months that comprise the dry periods, soil moisture was 0.4% in early September 2013 and 1.05% in July 2014.
Phenology. The predominant phenological stage annotated for the species was compiled and represented in Fig. 2. At the end of the dry period, in early September, the plants were characterized in a dormant state, showing senescent floral branches and older leaves. In the DR period in October, the plants entered the sprouting state, with emission of new branches and leaves. In the Veranico period in February, presence of inflorescences with reddish floral parts was observed. In the rainy period in April, the fruiting phase was characterized by the presence of inflorescences with senescent and paleaceous floral parts. The fruiting phase lasted until May in the DR period, and the senescence of the floral branches was remarkable. In July, at the beginning of the dry period, the plants entered dormancy. In the dormant state, the younger leaves on the branches of the plants were still green.
Abscisic acid content. In Fig. 3, the relative water content (RWC), abscisic acid (ABA) content, and Sm are summarized as an average of the samples of the D, dry period; RD, transition from dry to rainy period; V, veranico period; R, rainy period; and RD, transition from rainy to dry period. No correlation was observed between Sm and leaf and root RWC values of the plants. Nevertheless, in the Veranico period, RWC of the leaves was slightly lower when there was a decrease in Sm (Fig. 3A).
Abscisic acid (ABA) content in the roots and leaves was measured because of its important role in plant responses to drought and other abiotic stresses (Fig. 3B). The general pattern of ABA accumulation was similar in the leaves and roots; however, ABA content was significantly higher in the leaves (Tukey's test, p < 0.05) in the V period than in the DR and R periods.
Metabolic profiling. To obtain information on the metabolic phenotypes of G. agrestis grown under natural conditions in a rock-field area of the Cerrado in the dry and rainy seasons, we performed a multi-metabolomic analysis of the leaves and roots. The datasets derived from different measurements were combined and evaluated using multivariate analysis. Principal component analysis (PCA) was used to obtain an overview of the samples www.nature.com/scientificreports/ and identify outliers (data not shown). Later, the data was analyzed with orthogonal projection to latent structures discriminant analysis (OPLS-DA) to obtain valid models that could explain the metabolite variations in the different seasons. From the generated loading plots, the metabolites were filtered by variable importance in the projection (VIP) (cutoff ≥ 1) and listed. On the basis of the selected VIP metabolites, OPLS-DA score plots of the leaves and roots are shown in Fig. 4. To validate the strategy, several pair-wise OPLS-DA models were calculated between the groups by using the selected metabolites (Supplementary Table 2). Because all comparisons resulted in valid models, the distinct metabolites were clustered, and the heatmaps (Fig. 5A,B) show a global metabolite profile of the leaves and roots of G. agrestis in different seasons. The metabolites derived from the untargeted approach with GC-MS and LC-MS analyses were systematically annotated (Fig. 5A,B, as described in the "Material and methods"). The combination of targeted and untargeted approaches allowed us to have a significant coverage of all major metabolite classes that could characterize the plant adaptation to the environmental changes, resulting in the annotation of 280 metabolites (Supplementary  Tables 3, 4, 5, 6, 7 and 8); 215 were detected in the leaves and 195 in the roots.
For better interpretation, the annotated metabolites were clustered using a non-hierarchical analysis according to their metabolite class (Fig. 6). As expected, different metabolite patterns were observed in the leaves and roots of G. agrestis. The analysis of the leaves showed a unique pattern for Veranico samples, and the accumulation of Design of sampling and sample grouping for the metabolomic analysis. The groups are as follows: D, dry period (12 samples, red color); DR, transition from dry to rainy period (18 samples, blue); V, veranico, a short dry period during the rainy period (6 samples, yellow); R, rainy period (12 samples, green); and RD, transition from rainy to dry period (18 samples, brown). Below, months of the year and the number of the sample in the corresponding month. Soil moisture was used to select samples for metabolomics. www.nature.com/scientificreports/ organic acids, sugars, galactolipids, phenolics, and chlorophyll degradation products were observed. Interestingly, contrasting accumulation of some sugars, phosphocholine (PC), triacylglycerolipids (TAG), and galactolipids (MGDG and DGDG) were verified between Veranico and RD. Accumulation of amino acids and pigments (chlorophylls and ketocarotenoids) was observed in the leaves during the transition from drought to rainy season. In the roots, a distinct fructan pattern was observed: plants grown during Veranico and rainy accumulated fructans containing less than 10 units of fructose, and plants grown during dry and DR accumulated more complex fructans. In general, higher levels of amino acids were observed in the DR and R seasons.
Drought-discriminating metabolites. The impact of water availability in the soil on the leaves and roots of G. agrestis grown in the rainy, dry, and Veranico seasons is shown in Figs. 7 and 8 and Supplementary Tables 9  and 10. In general, changes in the metabolism of sugar, specially fructans, lipids, amino acids, and phenolics  , and TG (53:3) was pronounced during the dry season, which is in contrast to the reduced levels of fructans containing small units of fructose and amino acids GABA, phenylalanine, tyrosine, and valine during the rainy (R) season. However, during Veranico, accumulation of the sugars lactose, sorbitol, and mannitol, fructans 2_DP and 3_DP, lipid TG (55:4), and the phenolic 3,4-dihydroxybenzoic acid was observed, and a decrease in the amino acids valine, phenylalanine, tryptophan, and glycine betaine was observed. The metabolism of sugars, lipids, phenolics, and pigments www.nature.com/scientificreports/ was affected in the leaves in the dry, Veranico, and rainy seasons ( Fig. 8), with pronounced accumulation of galactolipids, xanthophylls, chlorophyll intermediates, and several classes of phenolics such as phenylpropanoids, benzoic acid, and flavonoids in Veranico. Although the dry and Veranico seasons were characterized by low water availability in the soil, the metabolism of G. agrestis responded differently.

Discussion
In the Brazilian Cerrado, the vegetation is adapted to a seasonal climate with short and long periods of drought. This condition causes the local species to experience low water availability during specific periods of the year. In the present study, the water status in the soil of a rock-field area in the Cerrado changed from low to high moisture, which directly impacted the water status of plants collected from the same area during different months of the year (Figs. 1 and 3). Endemic species, like Gomphrena, in the Brazilian Cerrado are adapted to such water variations and can survive via different and perhaps unknown physiological strategies. In general, plant adaptation can include changes in the developmental stage and phenology of plants obtained during different seasons (Fig. 2). The global analytical approach used to analyze G. agrestis allowed us to obtain a unique metabolite fingerprint of plants growing under natural conditions in Cerrado's dry (D: July and September), dry-rainy (DR: end of September and October), Veranico (V: February), rainy (R: March and April), and rainy-dry (RD: May and June) seasons. The results provide several hints about how this endemic specie can tolerate such drastic changes in the soil water availability throughout the year. The drought stress on the plants occur when either the water supply to their roots becomes limited or when the transpiration rate becomes too intense and causes a system imbalance 1 . In both situations, the stress starts to affect the water functions, culminating in reduced growth. Here, the leaves and roots of G. agrestis showed reduced RWC during Veranico, resulting in an increase in ABA content (Fig. 3B). The results suggest that G. agrestis controls the stomatal movement as a strategy to keep the tissues hydrated or maintain the CO 2 supply for photosynthesis, as observed in many other species [32][33][34][35][36] . The maintenance of open stomata can lead to a greater loss of water and consequently dehydration. However, if the soil is sufficiently moist or if the moisture is recovered soon, it is easy for the plant to replenish water and maintain both photosynthesis www.nature.com/scientificreports/ and the growth rate. Because of the physical characteristics of the rock-field soil (shallow and sandy) and lower precipitation rates during the Veranico period, the soil moisture decreases quickly and probably limits the water availability for the plants (Fig. 3C). ABA has an important role under stress conditions, especially during drought 6,37,38 . In the present study, we found a pronounced increase in ABA levels in the leaves during Veranico. Generally, the root system is affected to the greatest extent when there is water scarcity or the availability of water is inconsistent 39 . The fact that the ABA levels were not significant in the roots suggest that Gomphrena plants have other alternatives to compensate for the drought. Fructans in the tissues may act as osmotic solutes to maintain the water status of tissues 28 . Our results support their suggestion because we found increased levels of fructans containing up to 8 units of fructose during the Veranico season (Fig. 7). Similar results were obtained by 40 , who reported no changes in the RWC in the roots of Gomphrena marginata (also growing in a rock field) during the dry period, suggesting that accumulation of fructans could result in osmoregulation. Similar results were found in Vernonia herbacea, another local species present in the rock fields of Cerrado 29,41 . The dry season (during which the soil moisture levels were lower; Fig. 3C) was characterized by the accumulation of more complex fructans (Fig. 7). In general, complex fructans (containing up to 22 fructose units) were higher during the dry and DR periods. Fructans with lower DP were predominant during the Veranico and rainy periods (Figs. 6B and 7). These contrasting patterns are consistent with the involvement of fructans in drought strategies: complex fructans represent a carbon source that supports initial growth or regrowth during the beginning of the rainy season 26,40 . Simple fructans in the rainy and Veranico periods can be explained by the water-favorable condition for synthesis and turnover of the energy metabolism in the rainy season or a strategy to support osmoregulation during the Veranico.
The accumulation of sugar alcohols such as arabitol and ribitol in the leaves during the dry season (Fig. 8) suggests their involvement in the non-photochemical quenching during drought 42 . However, other simple sugars (e.g., glucose, fructose) may also play a role in drought tolerance in plants by reducing the effects of osmotic stress, maintaining turgor, stabilizing cell membranes, and protecting plants from damage 43 . The level of xylose was also high in the dry season, and it is a component of cell wall metabolism and suggested to be involved in drought stress through cell wall modification 35 . www.nature.com/scientificreports/ Accumulation of phenolic metabolites during the dry and Veranico seasons was observed. The phenylpropanoid pathway was more pronounced during Veranico, resulting in the accumulation of sinapinic acid, sinapyl alcohol, 1-O-(4-coumaroyl)-B-d-glucoside, feruloyl-glucoside, and caffeoylshikimate as well as different flavonoids, which is in contrast to the accumulation of benzoic acid derivatives like vanillic acid and 1-galloyl b-d-glucose during the dry season (Fig. 8). Such accumulation of phenolic metabolites in both seasons might be related to plant protection against oxidative damage that occurs during these two periods of low water availability 21,45 . The production of reactive oxygen species may be the most important secondary effect of drought and can result in chloroplast membrane damage. We observed the accumulation of several galactolipids during Veranico. These lipids are major components of the photosynthetic apparatus 46 .
Pigments like carotenes can act as antioxidants and energy quenchers 47 . Carotenes and xanthophylls, as well as chlorophyll metabolites like chlorophyllide, pheophorbide and pheophytin, were accumulated in the plants during Veranico (Fig. 8). Pheophytin is involved in the process of electron transfer in PS II, working as a bridge of electrons between the chlorophyll P680 and plastoquinone 48,49 . Previous studies have investigated how this mechanism works and the function of pheophytin 50,51 . In this study, the increased levels of pheophytin may be due to induced chlorophyll degradation 52 , the stress, or a response that benefits the plant. However, no chlorophyll changes were observed in the plants collected during Veranico. Therefore, pheophytin accumulation may be a mechanism that helps in photosynthesis efficiency by either acting in the flux of electrons or protecting the system from damage 50,53 . The role of pheophytin in plants is not well understood and needs to be studied under stress conditions.
It is important for a plant to adapt to yearly (long-term) and short-term changes in drought and other environmental factor to coordinate growth and stress-related responses. Stomatal closure acts by reducing the loss of water and maintaining the hydration state in the tissues, but it also limits CO 2 influx for photosynthesis. G. agrestis is a C4 plant 54 and therefore exhibits efficient photosynthetic metabolism under drought conditions. Probably the plant showed high photosynthesis rates in the Veranico period, even with stomatal control. This strategy is important under mild or short-term water stress conditions because the plant can sustain growth, which is primarily affected. The adaptions of metabolism, as observed in the present study, suggests strategies to maintain photosynthesis during the Veranico period. This is of great importance to G. agrestis because this period coincides with the flowering time (Fig. 2), which requires high quantities of assimilated carbon. The minimum hydration necessary for survival, cell enlargement, or maintenance of metabolic activity is provided by regulated stomata control and other associated strategies, such as osmoregulation.

Conclusions
In this study, we used a metabolomic approach to understand and describe the metabolic adaption of a native species to seasonal changes in drought. We showed that fructans are accumulated in the thickened roots, suggesting a metabolic pattern that is consistent with the storage of carbons during the water-favorable season to support energy demand during the long period of drought and regrowth as well as metabolic adjustments for osmoregulation. In the leaves, ABA, simple sugars, sugar alcohols, phenolics, and pigment metabolism indicate the importance of metabolic responses that should act together to modulate general physiological adaptations such as stomatal control, photosynthesis, and oxidative stress. The metabolic pattern in the Veranico period suggests that during short-term drought, the maintenance of active photosynthesis seems to be more important, and stomatal control, osmoregulation, and protection from oxidative damage may be the strategies used by the species.

Methods
Geographical location. The study was conducted in the Environmental Preservation Area "Serra do Resplandecente Encantado", a public area in the municipality of Itacambira, north of Minas Gerais State (16° 59′ 47″ S, 43° 20′ 01″ W), Brazil. In this area, which is part of the Espinhaço mountain range, rock-field formations are predominant.
Plant material and sampling. The study was conducted in accordance with relevant guidelindes and brazilian legislation 55

Relative water content.
To characterize the water status of the plants, 10 leaves and root fragments were collected from each plant and weighed for determining the fresh weight (FW). Then, they were immersed in distilled water for 6 h to determine the turgid weight (TW), followed by drying at 70 °C to determine the dry weight (DW). The relative water content (RWC) was estimated using the following equation Soil moisture. Soil moisture (Sm, %) was measured using the gravimetric method 57 . In each field expedition, six soil samples were collected between 0 and 20 cm, which corresponds to the effective root depth. The soil samples were weighed to determine FW and then dried at 70 °C to measure DW. Sm (%) was determined using the following formula: Sm (%) = (FW − DW/DW) × 100. , and the flow-rate was 0.5 mL/min. The compounds were eluted with a linear gradient consisting of 0.1-10% B over 2 min, B was increased to 99% over 5 min and held at 99% for 2 min; B was decreased to 0.1% for 0.3 min and the flow-rate was increased to 0.8 mL/min for 0.5 min; these conditions were held for 0.9 min, after which the flow-rate was reduced to 0.5 mL/min for 0.1 min before the next injection. The compounds were detected with an Agilent 6550 Q-TOF mass spectrometer equipped with a jet stream electrospray ion source operating in positive or negative ion mode. The settings were kept identical between the modes, with exception of the capillary voltage. A reference interface was connected for accurate mass measurements; the reference ions purine (4 µM) and HP-0921 (Hexakis (1H, 1H, 3H-tetrafluoropropoxy phosphazine) (1 µM) were infused directly into the MS at a flow rate of 0.05 mL/min for internal calibration, and the moni- www.nature.com/scientificreports/ nebulizer pressure 35 psig. The sheath gas temp was set to 350 °C and the sheath gas flow 11 L/min. The capillary voltage was set to 4000 V in positive ion mode, and to 4000 V in negative ion mode. The nozzle voltage was 300 V. The fragmentor voltage was 380 V, the skimmer 45 V and the OCT 1 RF Vpp 750 V. The collision energy was set to 0 V. The m/z range was 70-1700, and data was collected in centroid mode with an acquisition rate of 4 scans/s (1977 transients/spectrum). For metabolite annotation autoMSMS was performed on pooled QC-samples at 3 different collision energies, 10, 20 and 40 eV. The fructans compounds were detected with an Agilent 6550 QTOF mass spectrometer equipped with a Jet Stream electrospray ion source operating in positive and negative ion mode 60 . The MS/MS spectra were obtained under the same conditions, with the collision energy from 10 to 40 V.
Lipidomic analysis with LC-QTOF MS. The lipid analysis was performed in the positive ion mode 61 . In brief, lipid extracts based on chloroform/methanol extraction was chromatographic separation was performed on an Acquity UPLC CSH, 2.1 × 50 mm, 1.7 µm C18 column in combination with a 2.1 mm × 5 mm, 1.7 µm VanGuard precolumn (Waters Corporation, Milford, MA, USA) held at 60 °C. The gradient elution buffers were A (60:40 acetonitrile:water, 10 mM ammonium formate, 0.1% formic acid) and B (89.1:10.5:0.4 2-propanol:acetonitrile:water, 10 mM ammonium formate, 0.1% formic acid), and the flow-rate was 0.5 mL/ min. The compounds were detected with an Agilent 6550 Q-TOF mass spectrometer equipped with a jet stream electrospray ion source operating in positive ion mode. All mass spectrometer settings as for untargeted LC-MS analysis. All generated files were processed using Profinder B.08.00 (Agilent Technologies).
Metabolite annotation. The metabolites were annotated by manual interpretation of the high mass accuracy of the fragments produced by MS/MS experiments and/or comparison with public (Kegg and PlantCyc) and in house database. Additional MS/MS networking (Global Natural products social molecular networking 62 ) was performed as a quality control to detect adduct masses that somehow were not excluded during the processing data. For annotation of fructans, the degree of polymerization, which means the number of fructose units in the molecule structure was used. Glycerolipids annotation was performed by comparison with in house lipid spectral databases. The lipid classes were differentiated by the presence of diagnostic fragments m/z 184.0733 (PC), m/z 243.0945 (MGDG-Na + ) or neutral losses of 162.0528 (DGDG), 161.0450 (PI) and 141.0191 (PE). Spectral information of phenolics and lipids are presented in Supplementary Tables 5 and 7.
Amino acid analysis with LC-QqQ MS. The extracts were derivatized with the Waters AccQ•Tag method, in accordance with the manufacturer's protocol. The analysis was performed using a 1290 Infinitely UHPLC system from Agilent Technologies (Waldbronn, Germany) with G4220A binary pump, G1316C thermostated column compartment, and G4226A autosampler with G1330B autosampler thermostat coupled to an Agilent 6490 triple quadrupole mass spectrometer equipped with a jet stream electrospray source operating in the positive ion mode 63 . The amino acid multiple-reaction-monitoring (MRM) transitions were optimized using MassHunter MS Optimizer software (Agilent Technologies Inc., Santa Clara, CA, USA), and the data were quantified using MassHunter Quantitation software B07.01 (Agilent Technologies); the amount of each amino acid was calculated on the basis of the calibration curves.
ABA analysis with LC-QqQ MS. For the ABA analysis 64 the analytes were separated using a 1290 UHPLC system from Agilent Technologies (Waldbronn, Germany), with a G4220A binary pump, G1316C thermostated column compartment, and G4226A autosampler with thermostat. A 2 µL aliquot of the sample was injected onto a Waters column (TSS3, C18; 2.1 × 50 mm, 1.7 µm) at 40 °C in a column oven. The analysis was performed in multiple-reaction-monitoring (MRM) mode, in which the fragmentation conditions for the analyses were optimized using MassHunter MS Optimizer software (Agilent Technologies Inc., Santa Clara, CA, USA). MRM scan was performed monitoring m/z 263 → 153 for ABA and m/z 269 → 159 for d6-ABA as quantifiers. Transitions m/z 263 → 219 for ABA and m/z 269 → 225 for d6-ABA were used as qualifier ions. The data were quantified using MassHunter Quantitation software B07.01 (Agilent Technologies); the amount of ABA was calculated on the basis of the calibration curve done with d6-ABA (1 pg/µL) and ABA standards (from 0 to 10 pg). Statistical analysis. The generated datasets from the different analyses were checked using statistical multivariate analysis in SIMCA-P 13 software (Umetrics AB, Umeå, Sweden). The samples were compared using PCA and OPLS-DA analysis. Before the analysis, the missing data were set to the mean value of each variable and were mean-centered and scaled to the unit variance. The samples were grouped according to the environment characterization into five groups: 1, dry (D: 12 samples); 2, transition between dry to rainy (DR: 18 samples); 3, "Veranico", a short dry period during the rainy season (V: 6 samples); 4, rainy (R: 12 samples); and 5, transition between rainy to dry (RD: 18 samples). To identify the most important metabolites in the OPLS-DA models, the VIP was used, and variables showing VIP values greater than 1 were considered of high importance 65 . The OPLS-DA models were validated using the goodness of fit (R 2 ) and prediction (Q2) parameters. Further statistical analysis and visualization (ANOVA, Tukey's test, t-test, Benjamini and Hochberg correction, and heatmaps) were performed using R-software version 3.4.1 66 .