Plant species richness elicits changes in the metabolome of grassland species via soil biotic legacy

Species‐rich plant communities can induce unique soil biotic legacy effects through changing the abundance and composition of soil biota. These soil legacy effects can cause feedbacks to influence plant performance. In addition, soil biota can induce (defensive) secondary metabolites in shoots and roots and thus affect plant–herbivore interactions. We hypothesize that plant diversity‐driven soil biotic legacy effects elicit changes in the shoot and root metabolome. We tested this hypothesis by establishing an experiment with four plant species. We grew plants in a sterile substrate inoculated with soil conditioned by different plant species communities: (a) monocultures of either of the four species, (b) the four species in a mixture, (c) an eight species mixture including all four species or (d) a sterile inoculum. After at least 8 weeks in the field, we estimated shoot herbivory. At the same time, we took root and shoot samples for metabolomics analyses by liquid chromatography quadrupole time‐of‐flight mass spectrometry. We found that shoot and root metabolomes of all plants grown in sterile soil differed significantly from those grown in living soil. The plant metabolomes in living soils differed by species and tissue. Across all species, shoots displayed a greater richness of secondary metabolites than roots. The richness of secondary metabolites differed by species and among living soils. The conditioning species richness significantly affected the Shannon diversity of secondary metabolites in Centaurea jacea. Shoot herbivory positively correlated with the richness and Shannon diversity of secondary metabolites in Leucanthemum vulgare. We detected multiple metabolites that together explained up to 88% of the variation in herbivory in the shoots of C. jacea and Plantago lanceolata. Synthesis. Our findings suggest that plant diversity‐driven shifts in soil biota elicit changes in the composition and diversity of shoot and root secondary metabolites. However, these plant responses and their effect on shoot herbivores are species‐specific. Tracking changes in plant secondary chemistry in response to soil biotic legacy effects will help to understand the mechanisms that govern species‐specific plant–plant and plant–herbivore interactions.


| INTRODUC TI ON
Each plant species harbours a unique rhizosphere community (Bezemer et al., 2010). In plant communities, each plant species thus contributes to the establishment of soil communities (Wardle et al., 2004). Relationships between plant community diversity and soil biota diversity have been reported to vary from neutral to positive over time in plant diversity experiments (Eisenhauer et al., 2010;Lange et al., 2015;Strecker, Macé, Scheu, & Eisenhauer, 2016). Such plant-induced changes in the soil community can remain operational over time and thus result in soil biotic legacy effects (Kardol, Cornips, van Kempen, Bakx-Schotman, & van der Putten, 2007).
Soil legacy effects can be either positive or negative, depending on whether the conditioned soil biota increase or reduce the performance of individual plants (Ehrenfeld, Ravit, & Elgersma, 2005;Kulmatiski, Beard, Stevens, & Cobbold, 2008;van der Putten et al., 2013). Negative intraspecific legacy effects can result from specialized soil pathogens. Soil pathogens can accumulate because of high and species-specific root exudation (Steinauer, Chatzinotas, & Eisenhauer, 2016;van de Voorde, van der Putten, & Bezemer, 2011). At the same time, the same root exudates may suppress root pathogens of neighbouring heterospecific plants, thus providing positive interspecific legacy effects (van de Voorde et al., 2011). Hence, chemical traits of individual plants can shape soil legacy effects through changing soil biota community composition and activity.
These altered soil communities, in turn, can also influence chemical traits of individual plants. For instance, the composition and concentration of defensive plant secondary metabolites, such as glucosinolates, iridoid glycosides or pyrrolizidine alkaloids, can change in response to soil microbial composition, nematodes and mycorrhizal fungi Hol et al., 2010;Kos, Tuijl, de Roo, Mulder, & Bezemer, 2015b;Wurst, Wagenaar, Biere, & Van der Putten, 2010).
These interactions with plant growth facilitators and plant antagonists are thus likely to influence the diversity of secondary metabolites that a plant produces. This additionally implies that soil legacy effects may also affect the entire plant metabolome, that is the entirety of all metabolites synthesized by an organism (Oliver, Winson, Kell, & Baganz, 1998).
Changes in the composition and concentration of plant metabolites are known to affect important ecological functions, such as the resistance to above-ground herbivory. Herbivory can induce metabolite synthesis locally (induced defences ) in the attacked plant tissue or systemically throughout the plant . Systemic induction can also elicit changes at the concentration of shoot metabolites as a consequence of interactions with soil biota. This systemic induction can then affect the resistance to above-ground herbivores (van Dam & Heil, 2011). Such a response in above-ground herbivores thus constitutes indirect soil legacy effects.
Indirect soil legacy effects can result in increased or reduced performance of specific plant species (Karban, Agrawal, Thaler, & Adler, 1999). This may ultimately affect the fitness of a species and their abundance in the plant community.
Here we analyse soil legacy effects as reflected in shifts in the individual plant metabolome at the end of a plant-soil feedback experiment (Dudenhöffer, Ebeling, Klein, & Wagg, 2018). We defined soil legacy effects on the metabolome as shifts in the composition or the diversity of secondary metabolites. Furthermore, we defined the strength of the legacy effects as the magnitude of the difference in the metabolite profile among different living soils. The plant-soil feedback experiment was designed to test how soils conditioned by different plant communities affect key plant life stages. In the context of the experiment, soil conditioning affected flower production and plant fitness with mostly neutral effects on plant biomass (Dudenhöffer et al., 2018).
Based on the previously reported effects, we hypothesized that (a) plants grown in living soil differ in their metabolome compared to plants grown in sterile soil, and that (b) plants grown in soil conditioned by plant communities that differ in their species richness display different shoot and root metabolomes. In addition, we hypothesized that (c) the individual diversity of secondary metabolites increases with an increasing species richness of the soil conditioning plant community, and that (d) the diversity of secondary metabolites correlates with shoot herbivory. In order to test these hypotheses, we grew four common grassland plant species in three living soils that differed in the diversity of the conditioning plant community. In addition, we added a control group grown in sterile soil. 4. Synthesis. Our findings suggest that plant diversity-driven shifts in soil biota elicit changes in the composition and diversity of shoot and root secondary metabolites.
However, these plant responses and their effect on shoot herbivores are speciesspecific. Tracking changes in plant secondary chemistry in response to soil biotic legacy effects will help to understand the mechanisms that govern species-specific plant-plant and plant-herbivore interactions.

K E Y W O R D S
above-ground-below-ground interactions, biodiversity-ecosystem function, chemical diversity, eco-metabolomics, herbivory, Jena Experiment, metabolite profile 2 | MATERIAL S AND ME THODS

| Experimental design
In summer 2014, we set up a soil legacy experiment with four common central European grassland herb species (Centaurea jacea L., Knautia arvensis (L.) Coult., Leucanthemum vulgare Lam., and Plantago lanceolata L.). We used 3 L pots filled with 2,700 g autoclaved (120°C for 20 min) sand-field soil mixture (50:50, v/v) taken from the 'Jena Experiment' (www.the-jena-exper iment.de) field site (Thuringia, Germany; 50°55′N, 11°35′E, 130 m a.s.l, Roscher et al., 2004). In order to remove roots and coarse stones, we sieved the field soil through a 5 cm mesh. We added 100 g living or sterile soil inoculum to each pot and thoroughly mixed the inoculum with the sterile sand-field soil substrate. Finally, we added 200 g of the sterile background substrate on top, thus minimizing cross contamination (Dudenhöffer et al., 2018). For each pot, the inoculum comprised only 3.33% (w/w) of the entire substrate, with 96.66% (w/w) being the same sterile standard background substrate.
We established three living soil and one sterile soil treatments for each plant species. The living soil inocula had been conditioned by different plant species compositions, that is monocultures of either of the four species (CR1 -conditioning species richness 1), the four species in mixture (CR4) or an eight species mixture (CR8) including all four plant species supplemented by the four common grass species Festuca rubra L., Helicotrichon pubescens (Huds.) Domort., Phleum pratense L. and Poa pratensis L. All plant communities had been sown in plots of 3.5 m × 3.5 m in summer 2010 and are part of the 'Trait Based Experiment' (for more details see Ebeling et al., 2014). More specifically, soil collection for this study was conducted 4 years after establishment of the plots. We collected multiple soil cores from the upper 10 cm of each plot along a transect throughout the length of the plot to account for within-plot variability. We sieved each living soil inoculum through a 1 cm mesh and subsequently stored all soil inocula at 4°C for 24 hr prior to the experimental setup. The sterile soil inoculum was a mixture of equal parts of all living soil inocula sterilized by autoclaving at 120°C for 20 min. This created a common baseline for the sterile soil inoculum treatment.
The full experimental design resulted in 128 pots (Table S1). Each combination of plant species and corresponding soil inoculum treatment was replicated eight times, arranged in eight blocks (Dudenhöffer et al., 2018). The plants of six pots died during the experiment and were thus not available for further analyses; mortality was not related to any experimental treatment (Table S1). Initially, 20 (non-sterilized) seeds per pot were sown (Rieger-Hofmann GmbH, Blaufelden-Raboldshausen, Germany) at a depth of 1 cm and subsequently covered with clear plastic cellophane in order to keep humidity in the pots high to encourage germination and seedling establishment. We transferred all pots to two climate chambers (four blocks per chamber) equipped with artificial light (four Osram Powerstar HQI-T 1000/D, E40, 1,000 W, 80,000 lm lamps per chamber) with a photoperiod of 16 hr in light at 20°C and 8 hr in darkness at 16°C. We removed the plastic cellophane once seedlings had established. In mid-November 2014, after 11 weeks, we reduced the number of plants per pot to three individuals by cutting the other plants just below the shoot meristems. The corresponding roots remained in the soil to decompose. Decomposing roots can elicit negative as well as positive effects on plant biomass production (Zhang, Van der Putten, & Veen, 2016). In the context of this experiment, however, we detected mostly neutral effects on plant biomass (Dudenhöffer et al., 2018). We then transferred all pots to an unheated glasshouse located at the Botanical Garden in Jena, Germany. There, the plants were confronted with a natural winter photoperiod and 8°C. In early May 2015, we moved all pots to an open area at the field site of the Jena Experiment maintaining the original eight blocks (for more details see Dudenhöffer et al., 2018).
During the field phase of the experiment, the lower half of each pot was covered in a bag that was closed to the bottom. This protected the pot against invasion of external soil biota and below-ground herbivores. At the same time, it allowed only the natural occurring above-ground herbivores and pollinators to interact with all plants. At the end of the experiment, we validated the sterile soil inoculum treatment by assessing the presence of mycorrhizal structures in roots (Dudenhöffer et al., 2018).
Only three samples of the sterile treatment displayed the presence of mycorrhizal structures in roots. These samples, however, did not affect any of our consecutive analyses/results and were thus included.

| Sampling and sample processing
We harvested the shoot and root biomass of one plant per pot at the end of the flowering period, which occurred in July 2015 for K. arvensis, L. vulgare and P. lanceolata, and in September 2015 for C. jacea. We separated the shoot and root biomass by cutting the plants with scissors and removed all flowers from the shoot samples. We counted the total number of leaves and the number of leaves with herbivore damage. Herbivore damage included signs of sucking, chewing and mining on leaves. We washed the roots twice in tap water to remove soil particles, and then dried the samples with paper towels. This process took roughly 30 s, and samples were then immediately stored in paper bags on dry ice to stop further metabolism. In the laboratory, samples were stored in a −80°C freezer, and subsequently, freeze-dried (LABCONCO FreeZone Plus 12 Liter, Kansas City, USA) for 72 hr. Dried samples were stored in zip-lock bags filled with silica gel at room temperature until further processing. We measured the dry weight in milligram and ground each sample to a fine homogenous powder using a ball mill (Retsch mixer mill MM 400; Haan, Germany).

| Metabolome extraction and analysis
We extracted 20 mg dried ground plant tissue of each sample in 1 ml of extraction buffer (methanol/50 mM acetate buffer, pH 4.8; 50/50 [v/v]). The samples were homogenized for 5 min at 30 Hz using a ball mill (Retsch mixer mill MM 400), and subsequently centrifuged (25,155 g, 10 min, 4°C). The supernatant was collected in a 2 ml Eppendorf tube. We repeated the extraction procedure with the remaining pellet and combined the supernatant with the first one. We centrifuged (25,155 g, 5 min, 4°C) all extracts, transferred 200 μl to an HPLC vial and added 800 μl extraction buffer, resulting in a 1:5 dilution.

| LC-MS data processing
We converted the LC-qToF-MS raw data to the mzXML format by using the CompassXport utility of the DataAnalysis vendor software.
Subsequently, we trimmed each data file by excluding the same non-informative regions at the beginning and end of each run using the msconvert function of ProteoWizard v3.0.10095 (Chambers et al., 2012). We performed peak picking, feature alignment and feature group collapse in r v3.3.3 (R Core Team, 2017) using the Bioconductor (Huber et al., 2015) packages 'xcms' (Benton, Want, & Ebbels, 2010;Smith, Want, O'Maille, Abagyan, & Siuzdak, 2006;Tautenhahn, Böttcher, & Neumann, 2008) and 'CAMERA' (Kuhl, Tautenhahn, Böttcher, Larson, & Neumann, 2012). We performed simulation experiments to analyse the best set of parameters prior to data processing (Table S2). These parameters included, among others, the signal-to-noise ratio which determines the proportion of low-intensity metabolites. Based on our tests, we chose a low signal-to-noise ratio, that is the inclusion of low-intensity metabolites. We used the following 'xcms' parameters: peak picking method 'centWave' (snthr = 10; ppm = 10; peakwidth = 4, 10); peak grouping method 'density' (minfrac = 0.7; bw = 3; mzwid = 0.005); retention time correction method 'symmetric'. We used 'CAMERA' to annotate adducts, fragments and isotope peaks with the following parameters: extended rule set (https ://gitlab.com/users/ stans trup/groups); perfwhm = 0.6; calcIso = TRUE; calcCaS = TRUE, graphMethod = lpc. Lastly, we collapsed each annotated feature group, hereafter referred to as 'metabolite' which is described by mass-to-charge ratio (m/z) and retention time (rt), using a maximum heuristic approach. This means in detail that the intensity values of the feature, which most often displayed the highest intensity across all samples represent the feature group. We performed pre-processing with 'xcms' and 'CAMERA' separately for each species and tissue type. We merged the four species-specific feature lists by m/z and rt values, allowing for a retention time window of 10 s and a mass deviation of 5 ppm. We tentatively identified metabolites through the comparison of LC-MS/MS data with literature references. We submitted high-resolution m/z values to the MassBank of North America (MoNA, http://mona.fiehn lab.ucdav is.edu/) spectral database for comparison using a mass tolerance of 0.5 D. In addition, we calculated low-resolution molecular weights, molecular formulae for putative molecular ions in neutral form, and particle weights for mass spectrometry generated fragments using ChemDraw Ultra 8.0 (www.cambr idges oft.com).
In order to test our hypotheses and to accommodate the experimental design, we analysed the data only within species and tissue.
We visualized the differences in metabolome composition between sterile soil and living soil and among the different living soil inoculums by performing Partial Least Squares -Discriminant Analyses (PLS-DAs). Differences in metabolome composition are not only based on the presence, absence, or identity of metabolites. The intensity of the corresponding signals in the mass spectrometer (which is proportional to the concentration of a particular metabolite) also contributes to the metabolome composition. Therefore, we ran pairwise multi-response permutation procedures (MRPP) on log + 1transformed metabolite intensity data to test for significant differences in the metabolite profile between our different treatments.
The MRPP dissimilarity matrix was Bray-Curtis and each analysis was permuted 10,000 times.
We calculated two metrics of metabolite diversity: (a) the number of metabolites within a plant individual (hereafter, richness of secondary metabolites) and (b) the abundance-weighed diversity of metabolites expressed as the Shannon-Weaver index (Hill, 1973) based on plant individual-level metabolite intensities (hereafter, Shannon diversity of secondary metabolites). We used Dunnett's test for single step comparison to compare the richness and Shannon diversity of secondary metabolites expressed by plants in either living soil against the expression in sterile soil. This analysis is similar to contrasts but corrects for the multiple comparison problem. In order to test if the richness of secondary metabolites increases with increasing conditioning species richness, we calculated a linear mixed effects model. We calculated a similar linear mixed effects model to test if the Shannon diversity of secondary metabolites increases with increasing conditioning species richness. In addition, we analysed if the richness and Shannon diversity of secondary metabolites correlates with above-ground herbivory using linear mixed effects models. The linear mixed effects models were based on restricted maximum likelihood estimation and Type I analyses of variance (ANOVA) with Satterthwaite approximation for degrees of freedom. In the first two cases, the richness or Shannon diversity of secondary metabolites was the dependent variable. As explanatory variables, we fitted tissue, the conditioning species richness (CR) of the living soil inocula and the interaction of both. In order to account for the spatial arrangement and non-random design, we applied 'block' as the random effect. In the last case, the dependent variable was shoot herbivory (expressed as relative number of damaged leaves in percent). The explanatory variables were either the richness or the Shannon diversity of secondary metabolites, and random effects were the CR of the living soil inocula (random slope) and 'block' (random intercept).
The LASSO algorithm assumes that the herbivory responses can be 'predicted' by a linear combination of metabolite intensities. LASSO estimates the coefficients of this linear combination by shrinking coefficients of predictors (here metabolites) using an l1 penalty in order to minimize the mean squared error in the herbivory. Some coefficients are penalized to zero and non-zero coefficients of predictors (metabolites) indicate that these are important 'features' for predicting herbivory with the least error. We used the 'cv.glmnet' function, including a leave one out cross validation, provided by the 'glmnet' (Friedman, Hastie, & Tibshirani, 2010) package to determine the sets of metabolites for each species and tissue type that could explain the herbivory pattern. The cross-validation process returns the most parsimonious model that has a cross-validated error within one standard deviation of the minimum.

| Soil biota effects on the composition and diversity of plant metabolomes
We compared the metabolomes of plants grown either in sterile or living soil, and observed significant differences across species, above-and below-ground ( Figure 1). The metabolomes of plants grown in living soil were more similar to each other than to the metabolomes of plants grown in sterile soil. Based on the results of Dunnett's test, we found significant differences in the richness and Shannon diversity of secondary metabolites (Table S3) compared to living soil in L. vulgare and P. lanceolata (with the exception of P. lanceolata plants grown in CR1 soil; Figure 2). Shoot herbivory did not significantly differ between plants grown in sterile soil and plants grown in living soil in either plant species (Table S3).

| Soil legacy effects on the composition of plant metabolomes
The three living soil inocula differed in their effect on shoot and root metabolomes across all plant species (Figure 3). Consecutive pairwise comparisons revealed that the differences in shoot metabolomes were more prevalent than in root metabolomes. The shoot metabolomes differed significantly between CR4 soil and CR8 soil across all species (Table 1). In P. lanceolata, the shoot metabolomes also differed significantly between all three living soil inocula. Root metabolomes differed between CR1 soil and CR4 soil in samples of C. jacea and L. vulgare (Table 1). In addition, root metabolomes differed between CR4 soil and CR8 soil in samples of C. jacea, L. vulgare and P. lanceolata. The root metabolome of K. arvensis was unaffected by the CR of the soil inoculum ( Figure 3f).

| Soil legacy effects on the diversity of plant metabolomes
We found significant differences in the richness of secondary metabolites between shoots and roots across all species, with a higher richness of secondary metabolites in shoots than in roots (Table 2). In addition, we found significant differences in the Shannon diversity of secondary metabolites between shoot and roots in K. arvensis, P. lanceolata and marginally significant differences in C. jacea (Table 2). In contrast, the Shannon diversity of secondary metabolites between shoot and roots in L. vulgare did not differ (Table 2).
The richness of secondary metabolites significantly increased with increasing CR in K. arvensis and P. lanceolata. In contrast, the richness of secondary metabolites significantly decreased with increasing CR in L. vulgare (Figure 2; Table 2). We observed a tissue-specific response of the richness of secondary metabolites to increasing CR in C. jacea (Table 2). An increase in CR increased the richness of secondary metabolites in C. jacea shoots, but reduced the richness in C. jacea roots (Figure 2).
The Shannon diversity of secondary metabolites significantly responded to an increase in CR in C. jacea, only (Table 2). However, we observed a tissue-specific response with an increase in the Shannon diversity of secondary metabolites with increasing CR in C. jacea shoots, but a decrease in C. jacea roots (Figure 2).

| Linking richness, Shannon diversity and identity of secondary metabolites to herbivory
We analysed if the amount of above-ground herbivory relates to the richness of shoot secondary metabolites and their Shannon diversity.
In addition, we related shoot herbivory to a combination of specific shoot secondary metabolites and found links between the identities of secondary metabolites and shoot herbivory. By using LASSO, we detected a combination of 15 metabolites in C. jacea shoot samples, which explained 88.8% of the total variation in shoot herbivory. Furthermore, we detected a combination of nine metabolites in P. lanceolata shoot samples that explained 86.1% of the total variation in shoot herbivory. In contrast, the LASSO regression found no congruent combination of metabolites in samples of K. arvensis and L. vulgare (Table 4). Because the full dataset contained on average 33.2% low-intensity metabolites, we repeated our analysis with a reduced dataset that only contained metabolites above median intensity. We did this as a sensitivity analysis of the results achieved for the full dataset. In the reduced

F I G U R E 3 Per species Partial Least Squares -Discriminant Analysis plots of the metabolites found in shoot (a-d) and root metabolomes (e-h). Plants grew in soils conditioned by plant communities differing in species richness. Ellipses represent the 95% confidence interval.
p-values are based on multi-response permutation procedures. The metabolite intensity matrix was log + 1 transformed for the purpose of data normalization. Abbreviations: CR 1 = conditioning species richness (CR) 1, that is monoculture (yellow circles, n = 7-8); CR 4 = 4-plant species mixture (orange squares, n = 7-8); CR 8 = 8-plant species mixture (red triangles, n = 7-8); expl. dataset, we detected a combination of 11 metabolites in C. jacea shoot samples that explained 69.2% of the total variation in shoot herbivory. From the full dataset to the reduced dataset, LASSO retained six metabolites that could best predict herbivory. On the contrary, the set of selected metabolites from the reduced dataset contained five additional metabolites that were not included in the selected metabolite combination from the full dataset. In P. lanceolata shoot samples, we detected a combination of seven metabolites in the reduced dataset. These seven metabolites explained 64.2% of the total variation in shoot herbivory (Table 4).
All seven metabolites that were part of the linear combination in the reduced dataset were part of the linear combination in the full dataset too.
We identified 10 putative metabolites in C. jacea samples and five putative metabolites in P. lanceolata samples (Table 5; Figure S1).
In addition, we tentatively identified seven metabolites in C. jacea and P. lanceolata samples that were not part of any linear combination (Table S4; Figure S1).

| D ISCUSS I ON
We demonstrated that both the root and shoot metabolome of four grassland species reacted to the presence of soil biota. Soils with a different legacy of plant species richness elicited shifts in the metabolite profiles. In addition, we detected combinations of metabolites that best explained the variation in shoot herbivory. Hence, our results point to soil legacy effects as a possible mechanism linking plant communities and above-ground herbivores through changes in secondary metabolites.  (Chialva et al., 2018;Joosten, Mulder, Klinkhamer, & van Veen, 2009). These differences were attributed to the absence of arbuscular mycorrhizal fungi (AMF) in the sterile soil (Chialva et al., 2018;Rivero, Gamir, Aroca, Pozo, & Flors, 2015). AMF colonization can alter the levels of secondary metabolites, such as alkaloids and flavonoids, as well as primary metabolites, such as amino acids and sugars, through compound-specific up-or downregulation (Rivero et al., 2015). Shoot Root 2015a) like we used in our experiment. Our experimental setup also meant that all plants grew in a mostly sterilized substrate (96.66% of the total soil per pot). Sterilization by autoclaving can result in a pulse of nutrients and toxins (Alphei & Scheu, 1993;Trevors, 1996). It may be possible that the plants in our experiment have shown species-specific responses to the pulse in nutrient and toxins. But to address this possibility, we performed all tests within a species, rather than between species. Our results show that the presence of soil biota had a metabolome-wide impact on four different plant species. Hence, our results have strong implications for results obtained in experiments that solely use sterile soil when it comes to their extrapolation to natural systems.

| Soil legacy effects on the composition of plant metabolomes
We observed that the CR levels affected the shoot and root metabolomes across all plant species. However, the response to soils with different CR levels was species-specific and tissue-specific. These species-specific and tissue-specific responses support our hypothesis (2) that plant metabolomes change according to the plant diversity-driven soil legacy they encounter. It proved difficult to compare our results to similar studies, because research on plant diversitydriven effects on plant metabolomes is scarce (Peters et al., 2018).
In one study, increasing plant diversity was linked to shifts in the above-ground metabolic profile of small-growing herbs but not of tall-growing herbs, with more than 100 detected metabolites that changed in concentration (Scherling, Roscher, Giavalisco, Schulze, & Weckwerth, 2010). In another case, metabolic fingerprinting revealed adaptation to monoculture or plant species mixture history (Zuppinger-Dingley, Flynn, Brandl, & Schmid, 2015). Here, the accumulation of soil pathogens in monocultures was suggested to drive shifts in certain metabolic groups. Our study now adds valuable insights by revealing that plant diversity-driven soil legacy effects induce shifts in the composition and diversity of secondary metabolites. Given the lack of further research on plant diversity-induced shifts in the metabolome, we compared our results to similar studies that focused on single species or single compound classes. These studies on single species or single compound classes confirm our interpretation of plant diversity-induced effects. For instance, bacterial or fungal root pathogens, and non-pathogenic soil bacteria, as well as mycorrhizal fungi caused changes in above-ground defence compounds Hol et al., 2010;van Dam & Heil, 2011). These changes in above-ground defence compounds range from a decrease to an increase depending on the plant species and below-ground interaction partner .
Although shoot and roots can differ in their response to soil biota, shifts in single compound classes were also reported. For instance, in the roots of P. lanceolata, nematodes had no effect on iridoid glycoside concentration, whereas soil micro-organisms increased iridoid glycoside levels (Wurst et al., 2010).
A particular in our study was that within the living soil treatments, the soil legacy of CR 8 soils led to distinct metabolomes expressed across all species and tissues. We attribute this effect in part TA B L E 3 Statistical parameters resulting from a Type 1 ANOVA on species-specific and tissue-specific shoot herbivory as a function of the richness and Shannon diversity of secondary metabolites variables. Significant differences (p < 0.05) are given in bold  F, F-value; p, p-value. to the higher conditioning species richness and to the presence of an additional plant functional group, that is grasses, in the conditioning phase. Our results contrast with those of Kos et al. (2015b). They found that the functional group of the conditioning plant species did not alter the concentration of pyrrolizidine alkaloids and amino acids in Jacobaea vulgaris. We believe that our approach of integrating a comprehensive part of the secondary metabolome in the analysis allowed for detection of plant functional group effects which went unnoticed so far.

| Soil legacy effects on the diversity of plant metabolomes
In addition to analysing the metabolite profile, we analysed the species-specific and tissue-specific richness and Shannon diver-  Macel et al., 2014). In addition, single one-dimensional variables, such as richness or Shannon diversity of secondary metabolites, might not be sufficient to describe the metabolome of a plant in plant-herbivore interactions.
Hence, we applied LASSO to detect linear combinations of metabolites that may explain variance in shoot herbivory. We applied LASSO to a full and a truncated dataset to test the robustness of our findings. In both cases, LASSO identified a single linear combination of metabolites that explained variation in shoot herbivory in C. jacea and P. lanceolata, only. Differences in the linear combination between both datasets in C. jacea indicate that low-intensity metabolites contributed to the plant-herbivore interaction. In P. lanceolata, seven to nine high-intensity metabolites explained shoot herbivory. We tentatively identified the metabolites of the linear combinations as quinic acid/quinic acid derivatives, chlorogenic acid derivatives, flavonoid glycosides, verbascosides and iridoid glycosides in C. jacea and P. lanceolata. These metabolites are known for their significant role in plant-herbivore interactions (Bowers & Puttick, 1988;Erb et al., 2009;Leiss, Maltese, Choi, Verpoorte, & Klinkhamer, 2009;Sutter & Müller, 2011;Treutter, 2006

| CON CLUS IONS
Plant diversity-driven soil legacy affects the performance of plants, as well as the composition and concentration of certain secondary metabolites. We show that soil biota in general induced changes in the root and shoot metabolome of common grassland plant species. In addition, we provide evidence that plant community properties, that is differences in plant species richness, can translate into responses in the composition and diversity of secondary metabolites. We suggest two different strategies of plants to deal with shoot herbivory, based on the species-specific composition of secondary metabolites. Based on our results, we conclude that plant diversity-driven soil legacy can affect plant-herbivore interactions through changes in secondary metabolites. This field of research may be a key to understand some of the mechanisms that govern species interactions including important ecosystem functions like herbivory.

ACK N OWLED G EM ENTS
We thank Silke Schroeck and Amelie Hauer for assistance at the biomass harvest. We further thank the technical staff and the gardeners of the Jena Experiment for their efforts in maintaining the experi-

DATA AVA I L A B I L I T Y S TAT E M E N T
The data are archived at MetaboLights: https ://www.ebi.ac.uk/ metab oligh ts/MTBLS544 .