Plot diversity differentially affects the chemical composition of leaves, roots and root exudates in four subtropical tree species

Plants produce thousands of compounds, collectively called the metabolome, which mediate interactions with other organisms. The metabolome of an individual plant may change according to the number and nature of these interactions. We tested the hypothesis that tree diversity level affects the metabolome of four subtropical tree species in a biodiversity ecosystem-functioning experiment, BEF-China. We postulated that the chemical diversity of leaves, roots and root exudates increases with tree diversity. We expected the strength of this diversity effect to differ among leaf, root and root exudates samples. Considering their role in plant competition, we expected to find the strongest effects in root exudates. In an ecometabolomics approach, roots, root exudates and leaves of four tree species (Cinnamomum camphora, Cyclobalanopsis glauca, Daphniphyllum oldhamii, Schima superba) were sampled from selected plots in BEF-China. Samples were extracted and analysed using Liquid Chromatography-Time of Flight-Mass Spectrometry. The exudate metabolomes were normalized over their non-purgeable organic carbon level. Multivariate analyses were applied to identify the effect of both neighbouring (local) trees and plot diversity on tree metabolomes. The species and sample specific metabolites were assigned to major compound classes using the ClassyFire tool, whereas m/z features related to diversity effects were annotated manually. Individual tree species showed distinct leaf, root and root exudate metabolomes. The main compound class in leaves were the flavonoids, whereas carboxylic acids, prenol lipids and specific alkaloids were most prominent in root exudates and roots. Overall plot diversity had a stronger effect on metabolome profiles than the diversity of local, directly neighbouring trees. Leaf metabolomes responded more often to tree diversity level than exudates, whereas root metabolomes varied the least. We found not overall correlation between metabolite richness or diversity and tree diversity. Synthesis: Classification of metabolites supported initial ecological interpretation of differences among species and organs. Particularly the metabolomes of leaves and root exudates respond to differences in tree diversity. These responses were neither linear nor uniform and individual metabolites showed different dynamics. More controlled interaction experiments are needed to dissect the causes and consequences of the observed shifts in plant metabolomes.

root exudates increase mineral nutrient bioavailability through rhizosphere priming effects (Dijkstra et 151 al., 2013) and mediate mutualistic interactions with beneficial microorganisms, such as mycorrhizal fungi 152 (Ferlian et al., 2018). In addition, they deter or resist pathogenic microbes, invertebrate herbivores and 153 parasitic plants, as well as suppress competing plant species (Baetz andMartinoia, 2014, Zeng, 2014). In 154 consequence, root exudates might play a key role in direct and indirect tree-tree interactions. We 155 therefore expect that both root, root exudate and leaf metabolomes respond to tree diversity levels in 156 their environment. 157 Root exudates, root and leaves of the four selected tree species were sampled in plots with different 158 levels of diversity, ranging from monocultures to 16 and 24 species plots. We also recorded the local 159 species diversity among the trees directly surrounding the target trees. This allowed us to test the effect 160 of surrounding tree diversity levels (local diversity) and total plot diversity on the metabolomes of 161 leaves, roots and root exudates. We hypothesized that roots, root exudates and leaves showed species-162 specific metabolomic profiles, and that within species, root and root metabolomes would be most 163 similar. We also expected that local neighbour diversity would have a larger effect on plant 164 metabolomes than plot level diversity, and that metabolite diversity would increase with tree diversity 165 level. Finally, we postulated that the metabolomes of root exudates respond stronger to plot diversity 166 than roots or leaves, because of the positive effects of diversity on the production of fine roots (Sun et 167 al., 2017), which might also enhance root exudation. 168

Field location 171
This study was carried out in the Biodiversity-Ecosystem Functioning Experiment China (BEF-China), 172 which has been set-up and managed since 2008 (Bruelheide et al., 2014). It is located in Jiangxi Province, 173 China (29°08′-29°11′N, 117°90′-117°93′E) on two sites, A and B, which were planted in 2009 and 2010, 174 respectively. A former timber forest was replanted with local tree species in monoculture and mixed 175 stands using a "broken stick" design. Using a pool of 40 tree species, extinction scenarios were simulated 176 with tree richness levels of 1, 2, 4, 8, 16 and 24 species on a total of 566 plots of 25.8 m x 25.8 m and 177 400 trees each. The trees are planted on a rectangular grid with 1.29 m distance. Thus, every tree has 178 eight potential neighbours at a distance of 1.29 m or 1.82 m (diagonal). 179 This study was carried out on site B, which ranges in altitude from 113 to 182 m and with slopes from 180 15-43 degrees. The site has a subtropical climate. From 1971-2000 the mean annual temperature was 181 16.7 °C and the mean annual precipitation 1800 mm. This has increased to 17.9 °C and 2076 mm, 182 respectively, during 2013-2017. January is the coldest month with a mean temperature of 0.4 °C and 183 July is the hottest with a mean temperature of 34.2 °C. The (natural) vegetation is characterized by 184 subtropical forest with a mixture of evergreen and deciduous species. 185

Sampling design 212
The sampling took place from 19 th of October until 28 th of October 2019. We took our samples in the 13 213 plots in the BEF China experiment that contained our target species. We sampled root exudates, roots 214 and leaves of the above-mentioned four tree species in plots with species richness levels 1, 2, 4, 8, or 215 16(24) species. In monoculture plots (1 per species), we sampled three randomly selected trees. In 2-216 species plots, 4-8 replicates were sampled for each species. In plots with 4, 8, 16/24 species, we sampled 217 2-7 trees per species. The detailed sampled species combinations and sampling replicate numbers were 218 shown in Fig S1. In total, we sampled 84 trees, yielding 84 samples for root exudates, roots and leaves 219 each. 220 After scrutiny of the metabolomes, it became apparent that 9 of the 84 root (exudate) samples were not 221 from the target tree. Pre-experiments showed that under field conditions, the root excavated for 222 exudate collection is not always originating from the target tree, even when taking all possible care. 223 Eight of the root and root exudate samples could be reassigned to one of the other three species based 224 on their metabolome. One sample could not be reassigned, thus, for the roots, 83 samples remained for 225 statistical analyses, for the exudates 81 (2 were lost during sample preparation) and for the leaves the 226 complete 84. 227 228 229

Leaves, roots, and root exudate collection 230
For the collection of root exudates, we followed the protocol of (Phillips et al., 2008), with slight 231 modifications. The root exudates were collected from the middle of two trees under each combination 232 (Fig. S1). We carefully excavated the roots from the upper 10 cm of mineral soil, starting from the base 233 of the tree, to identify terminal fine root strands (< 2 mm). Adhering soil particles were carefully 234 removed with demineralized water. Roots were gently dried using paper tissue and placed into a 30 ml 235 glass syringe, which was then filled with new, unused glass beads (500-750 μm, acid washed, ACROS 236 Organics, ThermoFisher, New Jersey, USA/Geel, Belgium) The outlet of the glass syringe was attached to 237 a plastic tube with a plastic 50 ML syringe attached to it. The glass syringe with the roots was filled with 238 nutrient solution (0.50 mM NH 4 NO 3 , 0.10 mM KH 2 PO 4 , 0.20 mM K 2 SO 4 , 0.15 mM MgSO 4 , 0.30 mM CaCl 2 ) 239 from the top. The solution was sucked through the plastic syringe and discarded. The root was left to 240 equilibrate for 20 min, after which fresh nutrient solution was added and the procedure was repeated. 241 After this second washing step, the glass syringe was closed with cotton wool and sealed with parafilm. 242 Everything was wrapped in aluminium foil and left for 48 h. After 48 h, the roots were cut off and the 243 syringes, which contained the root and glass beads, were placed in plastic bags and brought into the lab. 244 In addition, randomly chosen leaves were sampled for each tree from the shaded part of the crowns. 245 Each target tree's position and neighbourhood species were recorded at the time of sampling. These 246 data were used to calculate neighbourhood diversity i.e. the number of species that were directly 247 neighbouring the target tree. 248 To recover the root exudates, a 0.22 μm sterile filter was placed on a vacuum pump manifold over a 60 249 mL glass/plastic vial. The glass syringe with the roots and the glass beads was mounted on top of the 250 manifold and filled with 60 ML nutrient solution. Then, the vacuum pump was started and the exudate 251 solution was collected into the vial. Five ml of the exudate solution of each sample was aliquoted for 252 Non-Purgeable Organic Carbon (NPOC) analysis (see below). The remaining root exudate solution was 253 frozen at -20 °C in 60 ml HDPE cryo-vials until freeze-drying but thawed for shipment. After the 254 collection of exudate solution, the roots were cleaned with water to remove residual glass beads. Root 255 and leaf samples were put in envelopes, oven dried for 30 min under 105 °C, and then oven dried for 24 256 h under 60 °C. 257

Sample preparation for LC-MS 258
Dried roots and leaves were extracted for LC-MS according to a standard protocol: Per 20 mg powdered 259 material, 1 ml of extraction buffer (75% v/v methanol, HPLC grade, 25% v/v acetate buffer (2.3 mL acetic 260 acid and 3.41g ammonium acetate in 1L 18MΩ water, pH set to 4.8) plus 50µl 100mM IAA-Valin as 261 internal standard) was added and shaken with ceramic beads in a tissue homogenizer (Retch MM400, 262 Retch GmbH, Haan, Germany) for 5 min at 30Hz. Samples were centrifuged for 15 min at RT at 15.000g. 263 The pellet was reextracted with another 1 ml of extraction buffer per 20 mg starting material, 264 centrifuged again and both supernatants were unified. Samples were diluted 1:5 with the extraction 265 buffer, kept at -20°C overnight, centrifuged at 15.000 g for 10 min and transferred to HPLC vials. 266 Exudates were freeze-dried after shipment, redissolved in 1 ml water and the bottles washed with 1 ml 267 methanol. The samples were transferred to a new 2 ml reaction tube, completely dried (vacuum 268 centrifuge 27°C) and we tried to redissolve them in a smaller volume of the extraction buffer. However, 269 this was not possible, so 1 ml of water was added, dissolution was aided by vortexing, ultrasonic bath 270 and heating to 40°C, the samples centrifuged, the remaining pellet again dissolved in another 1 ml of 271 water, centrifuged again and both supernatants were unified. After that, the samples were 272 concentrated, but not to complete dryness, in a vacuum centrifuge. 273

Metabolome composition of exudates, roots and leaves is defined by species identity 407
Principal component analysis (PCA) showed that the metabolomes of the four tree species can be 408 distinguished in all three tissue types sampled (Fig. 1 A-C). In general, the metabolome of exudates 409 showed a larger variation indicated by wider confidence bands and larger score distances within the 410 species (Fig 1A) than leaf and root metabolomes (Fig 1B-C). The total number of mass features (25,145) in 411 root exudates was more than twice as high as in root (12,205 in Fig. 1 D-F) (Table S2 and S3; Fig. 2B and 2C, third 417 column). Interestingly, the leaf metabolomes of D. oldhamii contained relatively few alkaloids (Fig 2D,  418 third column), the discriminating compounds were glycosides, most likely with a flavonoid backbone, a 419 disaccharide, a putative coumarin and a steroidal compound (Table S4). The other three species mainly 420 separated on PC2 (Fig 1 A-C). Whereas the exudate metabolomes (Fig. 1A:) of C. camphora, Cy. glauca 421 and S. superba showed some overlap, their leaf and root metabolomes clearly separated (Figs 1 B-C;). 422 The presence of laurolitsine-like alkaloids separated both C. camphora exudates and root metabolomes 423 from the other species (Table S2 and S3). Several terpenoids and a putative phenylpropanoid in C. 424 camphora exudates (high loadings on PC2; Supplemental Table S2); and a procyanidin, a putative 425 reticuline, a sesquiterpene and an unknown metabolite in C. camphora roots were the most responsible 426 for separating this species from the other two. The C. camphora leaves were typified by specific 427 terpenoids. In S. superba roots and root exudates, saponines had the highest influence in separating S. 428 superba from the rest (Table S2 and S3). A coumarin (in exudates) and a putative phenylpropanoid (roots) 429 contributed to the separation. For Cy. glauca leaves, quinic acid, polyphenols and terpenoids were the 430 metabolites with the highest loading and therefore discriminating these metabolomes from the other 431 species. We could not identify discriminating features for roots and root exudates of Cy. glauca, because 432 there was too much overlap with the metabolomic profiles of C. cinnamonum and D. oldhamii in PC1 vs. 433 PC2 (Fig 1A). 434 When comparing samples within species, it became apparent that in all four species the exudates had the 435 highest number of unique mass features, followed by leaves and roots ( Fig. 2A). The number of root 436 specific features was generally less than 4%, with the exception of D. oldhamii (8.4%). In D. oldhamii, C. 437 camphora and S. superba, the numbers of features shared between roots and exudates were greater than 438 those shared between roots and leaves ( Fig. 2A) 439 In other to better characterize the metabolomes of exudates, roots, and leaves per species we used an 440 automated approach to classify our tentatively annotated features in a chemical class by using the 441 ClassyFire tool ( (Feunang et al., 2016); see methods section). This tool uses structural information (e.g. 442 SMILES, InChIs or IUPAC names) to categorize chemical entities into hierarchical chemical classes. Thus, 443 we were able to categorize between 22% (roots Cy. glauca) and 46% (leaves D. oldhamii) of the m/z 444 features into main compound classes (Table S7). The resulting plots reveal that carboxylic acids are a major 445 common compound class in root exudates (Fig 2 B). In roots, specific alkaloids, flavonoids and prenol lipids 446 are among major compound classes we could categorize. The leaf metabolomes of all four species are 447 dominated by phenylpropanoids with flavonoids as the most abundant chemical glass, followed by 448 organoheterocyclic compounds (e.g. imidazopyrimidines, indoles, azaspirodecanes) or organooxygen 449 compounds (e.g. alcohols and polyols, carbohydrates and conjugates, carbonyl compounds) (Fig. 2 D). 450 Prenol lipids were omnipresent, but most prominently present in roots and exudates of S. superba as well 451 as in exudates and leaves of C. camphora (Fig. 2 B-C). Aporphines, a class of alkaloids, and the structurally 452 related isoquinolines, were most prominent in C. camphora roots, but less in the exudates, whereas the 453 classified metabolome of both D. oldhamii roots and root exudates consisted for a large part of the typical 454 Daphnipyllum alkaloids (Fig. 2 B-C, 3 rd column). 455 Interestingly, neither in C. camphora nor in D. oldhamii leaves, alkaloids were as pronouncedly present as 456 they were in their exudates or roots. The plots also show that on the level of compound class, the chemical 457 profiles of roots and root exudates are more similar to each other than to leaf metabolomes (Fig. 2B-D). 458 This is also in line with the high number of overlapping features for roots and exudates ( Fig. 2A). Finally, 459 we found coumarins to be present in C. camphora and S. superba exudates, as well as in D. oldhamii leaf 460 metabolomes. Sphingolipids formed a notable part of D. oldhamii roots and root exudates, whereas the 461 roots of Cy. glauca showed tannins and flavonoids as major classes (Fig. 2D). 462 463

Effects of plot and local tree diversity on metabolome composition 464
Because the metabolomes differed substantially among species and tissues types, we assessed the effect 465 of plot and local tree diversity by species and sample. Partial least squares discriminant analysis (PLS-DA) 466 showed that plot diversity affected the metabolome composition dependent on the sample type as well 467 as on species identity. The exudate metabolomes of three of the four tree species (Cy. glauca, D. oldhamii 468 and S. superba) were significantly affected by plot diversity level ( Fig. 3; MRPP P < 0.05; Table S5). 469 In the roots, only the metabolomes of D. oldhamii varied significantly with plot diversity level, whereas 470 the leaf metabolomes of C. camphora, Cy. glauca and S. superba all varied significantly due to plot diversity 471 The effect of local diversity was less pronounced; only S. superba exudates showed a significant difference 472 (Fig. 4), while none of the root metabolomes was affected significantly. The leaf metabolomes of the same 473 three species (C. camphora, Cy. glauca and S. superba) responding to plot diversity, also showed a 474 significant response to local diversity (Fig. 4). Neither for plot nor for local diversity, any of the pairwise 475 comparisons in the MRPP analyses showed statistically significant differences (Table S5 and S6). 476 For all species and samples that showed a significant overall response to plot or local diversity, we picked 477 the features with the highest VIP (variable importance in projection) value to analyse their response to 478 tree diversity in more detail (Table S8, S9, and S10). Several of these features were annotated to 479 compound classes that also constituted a large fraction of the respective metabolomes. In the exudates 480 and roots of D. oldhamiii, two of the typical Daphniphyllum alkaloids were responding to plot diversity, 481 and in the exudates of S. superba the levels of saponins and phenylpropanoids varied with plot and local 482 diversity (Fig. 5 B and D). In the leaves of C. camphora, Cy. glauca and S. superba, flavonoids (e.g. 483 kaempferols, rutin and other quercetins) and terpenes were among the compounds responding to 484 variation in diversity level. 485 Despite the presence of features responding to biodiversity level, there was no overall uniform response 486 on species, sample or compound level. For example, in S. superba leaves, quinic acid decreased in intensity 487 with plot and local diversity, while in the same species rutin was increasing with increasing local or plot 488 diversity (Fig 6 C and F). Daphniphyllum alkaloids either showed a hump-shaped response or rather 489 decreased at intermediate plot diversity levels in D. oldhamii exudates (Fig 5 B). In roots, they either 490 peaked at diversity level 8, or increased with diversity (Fig 5 E). Within species and samples, we identified 491 some common patterns. Saponin levels in S. superba commonly decreased (Fig. 5 C and D), whereas a 492 farnesene-like sesquiterpene in C. camphora consistently increased with local and plot diversity (Fig. 6 A  493 and D). In general, the peak intensity of the single features varied substantially across species and sample 494 types, likely due to sampling issues and low replication rates. 495 496

Effects of plant diversity on species chemical diversity 497
Finally, we correlated chemical richness, i.e. the number of mass features with intensity > 1000, or the 498 chemical diversity, calculated as the Shannon diversity of the mass features, with either local diversity or 499 plot diversity. Spearman rank-correlations revealed that metabolite richness was not significantly affected 500 by plot diversity in any sampling type or tree species. The leaf metabolite richness of Cy. glauca (P = 0.053) 501 and D. oldhamii (P= 0.071) showed a marginally significant response, but in opposite directions. 502 Metabolite richness increased with plot diversity in Cy. glauca leaves, and decreased with plot diversity in 503 D. oldhamii (Fig. S3; third column). Local diversity only affected leaf metabolomes of Cy. glauca (P= 0.049) 504 and marginally so, the exudates of S. superba (P= 0.056). In both cases, local diversity positively correlated 505 with metabolite richness (Fig. S4). In none of the samples we found a significant correlation between 506 Shannon diversity of mass features and local or plot diversity level (Figs S5 and S6). 507 508 DISCUSSION 509 Overall, exudates, roots and leaves each showed clearly species-specific metabolomic profiles. Leaf and 510 root metabolomes were more distinctive than root exudates. The prominent presence of specific 511 alkaloids separated the C. camphora and D. oldhamii root and root exudate metabolomes from each 512 other as well as from the other two tree species. As postulated, root and root exudate metabolomes 513 were more similar to each other than to leaf metabolomes. Carboxylic acids formed a large part of all 514 exudates, whereas phenylpropanoids, in particularly flavonoids, dominated the classifiable subset of all 515 leaf metabolites. All four tree species showed a metabolomic response to tree diversity in at least one of 516 the metabolomes sampled. Both root exudates and leaf metabolomes responded more often to 517 differences in tree diversity levels than roots. Plot diversity level had an overall larger effect on tree 518 metabolomes than local diversity. Important features driving the differences in the metabolomes were 519 saponines in S. superba exudates , D. oldhamii specific alkaloids in roots, kaempferols, quercetins and 520 quinic acid in S. superba leaves, and sesquiterpenes in C. camphora. profiles. Compared to leaves and roots, however, exudates had higher numbers of specific features with 526 a higher intra-specific variance. Moreover, the profiles showed more overlap, despite the fact that we 527 normalized peak areas over NPOC to mitigate differences in exudation rates. Likely this is due to the fact 528 that exudates were sampled in situ, which means that these samples inevitably contained compounds 529 from microbes and organic matter in the rhizosphere. Additionally (local) differences in nutrient status 530 among plots unrelated to plot diversity may have caused differences in exudation patterns (Meier 2020). 531 Commonly, much less than 10% of the mass features in metabolomic experiments can be assigned, even 532 in model systems (da Silva et al., 2015). This limitation prohibits the ecological interpretation of 533 differences in metabolomic profiles (Peters et al., 2018). By using bioinformatics tools to compare m/z 534 features to online metabolite databases, we could broadly classify up to 45% of the features to 535 compound classes (Figure 2). This did not only allow us to derive broad ecological functions, but also 536 created a starting point for more targeted analyses combined with biological experiments to assess the 537 ecological relevance of particular metabolites in more detail. 538 Carboxylic acids were one of the most abundant metabolite classes in the root exudates (van Dam and 539 Bouwmeester, 2016). Other carboxylic acids (mono-, di-, and tricarboxylic acids) are found in the root 540 exudates of birch and spruce (Sandnes et al., 2005). It is known that carboxylic acids, like citrate or 541 malate, play a role in P mobilization (Inderjit and Weston, 2003). In D. oldhamii the second largest (in 542 exudates) and largest group (in roots) of annotated features were alkaloids. Daphniphyllum alkaloids are 543 structurally diverse group of metabolites, which were isolated from the bark of Daphniphyllum spp. and 544 were also reported in leaves and fruits (Wu et al., 2013). Therefore, it is surprising that we found 545 Daphniphyllum alkaloids in roots and root exudates, but were not or only scarcely present in leaves. We 546 identified one study showing that these alkaloids may have insecticidal activities (Li et al., 2009). Considering that alkaloids in general serve as anti-herbivore defences (Mithofer and Boland, 2012), it is 553 likely that the alkaloids in D. oldhamii and C. camphora roots and exudates serve as defences against 554 root feeders and pathogens. Further experiments, preferably under controlled conditions, are needed to 555 falsify this hypothesis. Prenol lipids made up a substantial fraction of annotated features in the 556 exudates, roots and leaves of C. camphora and S. superba. This class contains molecules consisting of 557 one or several isoprene (C5) units (Fahy et al., 2005). It includes mono (C10) -and sequiterpenoids 558 (C15), which are known for their roles in direct and indirect defence against herbivores and pathogens. 559 In addition, also carotene (C40) which plays a role in photosynthesis (Fahy et al., 2005) and saponines 560 which have been shown to modulate soil microbial communities (Fujimatsu et al., 2020), belong to this 561 group. The roots of Cy. glauca contained tannins as a major class. Tannins are typical for the oak family, 562 including Cy. glauca (Wakamatsu et al., 2020), which are known as defences to a broad range of 563 herbivores (Barbehenn and Constabel, 2011) . As mentioned earlier, our metabolomic analyses of these 564 four chemically poorly described tree species is a starting point for testing their chemical ecology in 565 more detail. 566 Root exudate and root metabolomes showed a high overlap in the number of features detected as well 567 as in the chemical classes that could be annotated (Fig 2). In particular, D. oldhamii and S. superba 568 exhibited a high similarity in root exudate and root metabolites. On the other hand, C. camphora shows 569 a high overlap in the number of features, but the composition of root exudates and roots are very 570 different. This might be due to the different ways in which metabolites are exuded by plants (Oburger 571 and Jones, 2018) or the interspecific interactions with trees in the local neighbourhood (Xia et al., 2016). 572 The metabolomes of root exudates shared also a number of features with the leaf metabolomes. This 573 was not reflected in the composition of the annotated metabolome. The shared features might belong 574 to the group that could not be annotated and therefore makes an interpretation more difficult. Leaf 575 metabolites could be introduced into the soil via leachate from leaf litter and show up in our exudate 576 samples, despite cleaning and washing of roots prior to sampling. 577

578
In addition to the species-specific metabolomic profiles, we also found that the metabolomes of the 579 different species and organs responded to differences in tree diversity levels. Especially exudate and leaf 580 metabolomes varied with plot diversity level, whereas the fine roots we sampled had rather constant 581 profiles. The latter is in line with a recent study, showing that the glucosinolate profiles of fine roots of 582 Brassica spp. did not change in response to local or systemic herbivory (Tsunoda et al., 2018). Fine roots 583 of trees have high turnover rates, and therefore trees may invest less in defending their fine roots 584 (Bouma et al., 2001, Yanai andEissenstat, 2002). This is supported by the fact that we found fewer 585 organ-specific metabolites in roots than in leaves. 586 Other than expected, plot diversity had an overall stronger effect on exudate and leaf metabolome 587 profiles than local diversity. Our hypothesis was based on the fact that direct neighbouring trees would 588 have a stronger effect on our target trees than more remote trees in the plot. The BEF China experiment 589 was deliberately planned to have a high tree density (0.6 tree m -2 ) (Bruelheide et al., 2014), which means 590 that ten years after planting the roots may have grown sufficiently to contact also more remote trees. 591 Even in grasslands, where plants are growing in much closer proximity, the local neighbourhood also 592 determined a small part of the variance in exudate composition (Dietz et al., 2019). In addition, trees 593 may be connected via widespread mycorrhizal networks, thus influencing each other beyond the local 594 neighbourhood (Courty et al., 2010). This may cause that overall plot diversity has a larger effect on root 595 exudates than previously expected. 596 Only for Cy. glauca leaves we found a consistent and statistically significant increase of chemical 597 diversity with local and plot tree diversity. Also, the number of exudate metabolites in S. superba 598 showed a positive trend (P = 0.056) with local diversity, whereas the number of metabolites in D. 599 oldhamii leaves tended to decrease with plot diversity. Whereas such variable relationships between 600 diversity level and chemical richness or diversity is in line with studies analysing grassland species (Ristok 601 et al., 2019), the high variation among samples may also have limited our ability to detect general trends 602 Partly, this is due to the low replication level of species combinations, which is a common issue in BEF 603 experiments manipulating plant diversity (Uthe et al., 2020). The experimental plots and as well our 604 samples are designed to vary species diversity, whereby plots of the same diversity will differ in the 605 composition of species (Bruelheide et al., 2014). This caused that each focal tree within diversity level 606 was exposed to different interacting trees. This variation among plots of the same diversity level could 607 dilute specific responses, leading to a less pronounced response in the metabolome. Additional variation 608 can have emerged due to the vastness of the experiment and distances among (replicate) plots 609 (Bruelheide et al., 2014). The vast spatial scale may cause differences in slope, humidity, soil 610 characteristics et cetera, which may have affected the metabolomes. For example, substrate properties 611 like acidity and the type of topsoil have been shown to influence the exudation in beech (Meier et al., 612 2020). Soil acidity is associated with nutrient availability, because a decrease in soil pH will reduce 613 availability of N and P and also change the microbial composition in the soil (Rousk et al., 2009). 614

CONCLUSION: 615
We showed that plant diversity affects aboveground and belowground metabolomes of four subtropical 616 tree species. Overall, plot diversity as well as the local neighbourhood has an impact on the metabolome 617 of root exudates and leaves. Studies on tropical trees have linked metabolomic diversity to differences 618 in insect community composition (Richards et al., 2015, Zu et al., 2020. Our observation that tree 619 diversity affects leaf metabolomes thus may impact herbivore communities on trees of the same 620 species, but growing in plots with different diversity levels. Similarly, differences in root exudate 621 composition may shape the microbial diversity in the rhizosphere (Haichar et al., 2008). We also showed 622 that the response of specific metabolites to plot diversity is dynamic and not linearly correlated to 623 diversity gradients. This suggests that in fact not species diversity, but species composition might be the 624 important driver of changes in plant chemical diversity. In future studies, the results of our field analyses 625 may be used to manipulate single factors at a higher replication level to disentangle the individual 626 contributions of species identity and neighbouring trees on tree metabolomes. These future studies may 627 benefit from our ecometabolomics workflow to identify chemical classes of metabolites that could be 628 investigated with more targeted approaches. showing the number of mass features unique to a single species (light yellow) or shared between two 989 (light orange), three (dark orange) or all species (red). The area of single sections corresponds to the 990 proportion of the number of features compared to the total number. 991 Figure 5 Boxplots of the top 5 VIPs (Variable Importance in Projection) of PLS-DA that showed a 1025 significant global difference in the metabolome (MRPP P < 0.05) in Fig. 4 ("Box": 25 th to 75 th percentile 1026 and median, "whiskers": 10 th and 90 th percentile). 1027