Trait–performance relationships of grassland plant species differ between common garden and field conditions

Abstract The way functional traits affect growth of plant species may be highly context‐specific. We asked which combinations of trait values are advantageous under field conditions in managed grasslands as compared to conditions without competition and land‐use. In a two‐year field experiment, we recorded the performance of 93 species transplanted into German grassland communities differing in land‐use intensity and into a common garden, where species grew unaffected by land‐use under favorable conditions regarding soil, water, and space. The plants’ performance was characterized by two independent dimensions (relative growth rates (RGR) of height and leaf length vs. aboveground biomass and survival) that were differently related to the eight focal key traits in our study (leaf dry matter content (LDMC), specific leaf area (SLA), height, leaf anatomy, leaf persistence, leaf distribution, vegetative reproduction, and physical defense). We applied multivariate procrustes analyses to test for the correspondence of the optimal trait–performance relationships between field and common garden conditions. RGRs were species‐specific and species ranks of RGRs in the field, and the common garden were significantly correlated. Different traits explained the performance in the field and the common garden; for example, leaf anatomy traits explained species performance only in the field, whereas plant height was found to be only important in the common garden. The ability to reproduce vegetatively, having leaves that are summer‐persistent and with high leaf dry matter content (LDMC) were traits of major importance under both settings, albeit the magnitude of their influence differed slightly between the field and the common garden experiment. All optimal models included interactions between traits, pointing out the necessity to analyze traits in combination. The differences between field and common garden clearly demonstrate context dependency of trait‐based growth models, which results in limited transferability of favorable trait combinations between different environmental settings.


| INTRODUC TI ON
Plant functional traits are connected with species differences in productivity and performance (Comas, Becker, Cruz, Byrne, & Dierig, 2013;Enquist et al., 2007;Poorter & Bongers, 2006). Moreover, it has been shown that traits strongly depend on the environment (Bruelheide et al., 2018;Díaz, Cabido, & Casanoves, 1998). Therefore, different combinations of traits may be advantageous under different environmental conditions. For example, under field conditions in managed grasslands other traits may be important as compared to conditions without land-use and competition. However, it is still poorly understood how trait-performance relationships of different plant species vary under different environmental settings.
Differences in relative growth rates (RGRs) reflect speciesspecific adaptations to abiotic factors such as climate, water, nutrient, and light availability as well as biotic factors such as competition, pathogens, or herbivory (Bultynck, Fiorani, & Lambers, 1999;Poorter, 1989). Stress-tolerant species are considered to have low potential RGRs suitable for environments with low nutrient supply, whereas species with high potential RGRs are superior in highly productive habitats (Grime & Hunt, 1975;Lambers & Poorter, 1992). However, RGR of the same species varies with environment.
Under common garden conditions, species can be grown under favorable conditions regarding soil and water and growth is unaffected by mowing, grazing, or negative species interactions such as competition. RGRs under these conditions can be expected to be higher and approach the species' potential growth rates, compared to natural field conditions. Under controlled conditions, species RGRs are expected to be mainly correlated with leaf traits (Grime et al., 1997). High leaf dry matter content (LDMC) of plant species indicates low productive species, whereas high specific leaf area (SLA) is considered characteristic of competitive species (Suter & Edwards, 2013). For example, in a greenhouse experiment high potential growth rates were found to be correlated with high SLA Poorter & Remkes, 1990), which, however, is not universally true, as was demonstrated for woody species (Böhnke & Bruelheide, 2013;Paine et al., 2015). This contrast between high LDMC and high SLA that distinguishes slow from fast-growing species is known as the leaf economics spectrum (Wright et al., 2004).
In contrast, under realistic field conditions in managed grasslands abiotic and biotic factors can be expected to reduce species' growth rates. Disturbances caused by land-use have strong influences on species growth (Deng, Sweeney, & Shangguan, 2014;Herz et al., 2017). Furthermore, resources have to be shared with other species or defended against herbivores, which both can limit growth rates (Lind et al., 2013). However, plants were also found to increase growth rates to compensate for biomass loss from grazing (Zheng et al., 2014). Under realistic field conditions, other traits may be advantageous than in interaction-free environments. For example, under high grazing pressure plants may grow smaller (Lienin & Kleyer, 2012) and/or have chemical and physical defense traits to avoid or reduce herbivory (Hanley, Lamont, Fairbanks, & Rafferty, 2007). Disturbances by trampling may benefit plants with clonal growth organs as they are able to invade much faster into disturbed areas (Bullock et al., 2001;Klimešová, Latzel, Bello, & Groenendael, 2008). Competition between plant species may also favor species with certain traits, for example, it has been reported that under competition for nutrients plants grow longer roots or under competition for light plants grow taller (Craine & Dybzinski, 2013). Roscher et al. (2011) found that a combination of monoculture biomass, plant growth rates, and resource-use traits associated with nutrient and light acquisition best explained non-legume species performance in a grassland biodiversity experiment. concerning soil, water, and competition regimes. In this context, we tested two main hypotheses: 1. We expected that growth rates are highly species-specific and that abiotic and biotic factors under field conditions reduce relative growth rates (RGRs), but result in similar overall patterns compared to RGRs in the common garden. In particular, we hypothesized that the species' RGRs observed in the field experiment correspond to the RGRs when grown under common garden conditions.
2. Secondly, we hypothesized that the magnitude of growth and performance are correlated with different plant traits in field and common garden. In particular, we expected strong correlations with traits of the leaf economics spectrum (LES) (Wright et al., 2004) in the common garden experiment, whereas the traits vegetative reproduction and physical defense should be more relevant in the field experiment.

| Field experiment
Different grassland species were planted into managed grassland communities, making use of the network of experimental plots in the German Biodiversity Exploratories (Fischer et al., 2010). In each of the three study regions (Schwäbische Alb, South Germany; Hainich, Central Germany and Schorfheide, Northeast Germany), 18 grassland plots were selected, each six of them representing the three main land-use types (meadow, pasture, and mown pasture). The plots differed in land-use intensity, species richness, and functional diversity. Each plot was divided into eight subplots of 1 × 1 m which each were planted with six phytometers of six different species, selected from a total pool of 130 species. The six species planted in every subplot were selected specifically based on every plot's species composition, and thus differed among plots. Detailed information of the experimental design and the different planting scenarios is reported in Breitschwerdt, Jandt, and Bruelheide (2015).

| Data analysis
Relative growth rates (RGR) were calculated for every individual plant according to formula (1) (Hoffmann & Poorter, 2002;, where M is any growth variable and t is the time span in weeks between the two monitoring dates 1 and 2.  Table S2. For the 93 species, we compiled a full trait matrix of eight traits (SLA, LDMC, height, leaf anatomy (succulent, scleromorphic, mesomorphic, hygromorphic, helomorphic, and hydromorphic), leaf persistence (in spring, summer or overwinter green or evergreen), leaf distribution (evenly spread leaves, rosettes, or semi-rosettes), physical defense, and vegetative reproduction). Trait values were measured in 2011 (Breitschwerdt et al., 2015) and complemented from the databases LEDA (Kleyer et al., 2008), BIOPOP (Poschlod, Kleyer, Jackel, Dannemann, & Tackenberg, 2003), BIOLFLOR (Klotz, Kühn, & Durka, 2002), and Rothmaler (Jäger & Werner, 2001). The species-mean values of all traits are provided in Supporting Information Table S1. As none of the 93 species had hydromorphic leaves or leaves that are persistent over winter, these trait states were excluded from calculations. Furthermore, we excluded leaf persistence evergreen and leaf distribution evenly spread leaves from analyses to avoid redundant information.

| PCA and procrustes analyses
We performed principal component analyses (PCAs) with the vegan package of R (Oksanen et al., 2015). We based the calculations on all species-mean performance variables (mean RGR of height, plant projection area, leaf length, and number of leaves, biomass and (1) survival rates), and carried out PCAs separately for the field experiment and the common garden experiment.
These performance PCAs of field and common garden observations were compared to a PCA based on all species traits, using procrustes rotation (procrustes function in vegan). Applied to the pair of performance and trait PCAs, the procedure rotates and scales the PC scores of the second PCA to maximally fit those of the first target PCA, minimizing the sum of squared differences. To test whether different traits were relevant in the field and the common garden, we then searched for an optimized corresponding trait PCA that best explained performance, using a forward selection of traits, also including two-way interaction of traits. This optimization procedure was carried out separately for the field and common garden PCA. We developed a stepwise forward selection by adding that trait or trait interaction to the trait PCA that resulted in the best correlation in a symmetric procrustes rotation between the performance and trait matrix. The correlation coefficient was obtained by the protest command of the procrustes analyses in vegan package of R (Oksanen et al., 2015), using 9,999 permutations. We run this forward selection until further addition of traits or trait interactions did no longer improve the procrustes correlation coefficient between each of the two performance PCAs and the corresponding trait PCA. We also considered forward selection of predictors using redundancy analyses (RDAs), but in comparison with the procrustes approach found RDA to be too greedy, resulting in much longer final trait lists. Furthermore, the automated forward procedure (ordistep or ordiR2step in vegan) could not be used, because the trait states of the same trait had to enter the model as a group (e.g., leaf anatomy, leaf persistence, and leaf distribution), also interactions with the different trait states had be handled as group and once discarded traits had be considered in subsequent steps. Therefore, we considered procrustes analyses the most appropriate way to select the best combination of predictor traits.

| Univariate analyses
In addition, we employed ordinary linear regression models, relating the final traits to all performance variables of the field and the common garden experiment. Furthermore, the species' ranks of each performance variable in both experiments (field and common garden) were compared using a Spearman correlation test.

| RE SULTS
All performance variables except RGR plant projection area showed a strong correlation between the mean values of the 93 study species obtained in the field compared to the common garden experiment according to Spearman's rank correlation coefficient (see Supporting Information Table S4). The best match was encountered for biomass production (r s = 0.42), followed by RGR leaf length and survival (each 0.32), RGR height (0.28), and RGR number of leaves Comparing the performance PCAs of the field and the common garden experiment with procrustes analyses resulted in a correlation of 0.31 (p = 0.0003, Table 1). The PCA of all traits showed that the dimensions of 14 traits were not well captured by only one or two axes, which explained 17% and 13% of variation in trait values (Supporting Information Figure S1). Thus, we used the whole ordination of traits as predictor for performance. In both PCAs of performance in the field and in the common garden, the procrustes correlation with the PCA based on all traits was insignificant (Table 1) (Figure 2b)). The procrustes correlation coefficients between the performance PCAs for field or common garden data, and the corresponding trait PCA based on the optimized set of traits were 31% and 37%, respectively (p = 0.028 and 0.0001, Table 1). However, the reciprocal application of trait PCA optimized for the field performance PCA to the common garden performance PCA and vice versa resulted in insignificant correlations (Table 1).
The trait-wise analyses of all performance variables in separate linear regression models (Table 2) (Table 1), and the significant univariate relationships with performance variables (Table 2) (Table 2). Furthermore, RGRs of plant projection area and survival were positively correlated with mesomorphic anatomy.
RGR of height was positively correlated both with leaf persistence in spring and the interaction between LDMC and leaf persistence in summer ( Note. Traits in the field experiment were LDMC, leaf anatomy (succulent, scleromorphic, mesomorphic, hygromorphic, and helomoprhic), leaf persistence (green in spring or summer), vegetative reproduction, and the three interaction traits between LDMC with leaf anatomy succulent, leaf persistence green in summer and vegetative reproduction. Traits in the common garden experiment were LDMC, height, leaf persistence (green in spring green or in summer), vegetative reproduction, and the interaction between LDMC and vegetative reproduction.
TA B L E 1 Results of procrustes analyses based on the principal component analyses (PCAs) of all species' performance variables (RGR of height, plant projection area, leaf length and number of leaves, biomass, and survival) and traits Biomass and RGR number of leaves correlated negatively with succulent anatomy and the interaction of LDMC with succulent anatomy (Table 2).
In the common garden experiment, the finally selected traits displayed significant linear relationships with almost all performance variables except for RGR plant projection area (Table 2).
Highest correlations were found between biomass and the plant height, showing that under unconstrained conditions biomass increased with potential plant height (Figure 3c). Similarly, RGR of leaf length was also positively correlated with plant height (Table 2). Negative correlations were found between RGR height and the interaction of LDMC and vegetative reproduction, showing that species with vegetative reproduction decreased in RGR height with increasing LDMC (Figure 3d). The single predictors LDMC and vegetative reproduction both had a negative impact on RGR height, while survival and RGR number of leaves increased with increasing LDMC and for species with vegetative reproduction (Table 2).

| D ISCUSS I ON
As expected, growth rates in the field were much smaller than under common garden conditions. Most aspects of growth as well as survival were species-specific to some degree, as revealed by the significant rank correlations between field and common garden vari- out being grazed or mown. There were even traits with opposing effects on growth. While the ability to reproduce vegetatively characterized slow-growing species with respect to RGR leaf F I G U R E 2 Procrustes analyses of PCAs: (a) PCA of traits in the field rotated to match PCA of performance in the field; (b) PCA traits in the common garden rotated to match PCA performance in the common garden. For the field experiment, the optimized remaining traits were as follows: LDMC, leaf anatomy (succulent, scleromorphic, mesomorphic, hygromorphic, and helomorphic), leaf persistence (persistent in spring and persistent in summer), vegetative reproduction, and the three interaction traits between LDMC with leaf anatomy succulent, persistence summer, and vegetative reproduction. For the common garden experiment, the optimized remaining traits were as follows: LDMC, height, leaf persistence (persistent in spring and persistent in summer), vegetative reproduction, and the interaction between LDMC and vegetative reproduction. Arrows show procrustes errors (longer arrows = higher errors) calculated by rotating species in 9,999 permutations and comparing species positions of two PCA until finding positions with least differences. For abbreviations of species names see Supporting Information Table S1. Only species with highest scores on axes (above the 95th percentile or below the 5th percentile) are shown length in the common garden (e.g., Festuca ovina, F. guestfalica, and Silene flos-cuculi), this trait was characteristic of fast-growing species in the field (e.g., Galium mollugo and Scirpus sylvaticus). In the field experiment, species with vegetative reproduction dis- Nevertheless, there were also traits that played important roles under both fields and common garden conditions. LDMC and the interaction between LDMC and vegetative reproduction as well as leaf persistence were relevant under both conditions. The key role of these traits was also reported by Gross, Suding, and Lavorel (2007) who found that LDMC and lateral spread were suitable predictors of growth under different nutrient, shade, and clipping intensities.
The importance of LDMC supported our expectation that LES traits are key predictors for performance. However, we found LDMC to be a better predictor for plant biomass production than SLA, as was pointed out also in previous studies (Kröber et al., 2015;Smart et al., 2017). More generally, leaf traits seem to be better predictors when based on mass rather than area (Lloyd, Bloomfield, Domingues, & Farquhar, 2013;Osnas, Lichstein, Reich, & Pacala, 2013). Overall, LES traits became only meaningful in combination with other traits.
The final trait models in our study all included the ability to reproduce vegetatively, confirming previous findings that LES traits alone are poor predictors for plant growth (Paine et al., 2015). Similarly, LES traits were found to have a subordinate role in community assembly as response to land-use. For example, Dirks, Dumbur, Lienin, TA B L E 2 Correlations between optimal traits found for each experiment (field and common garden) and the respective performance variables (biomass, survival, RGR of height, plant projection area, leaf length, and number of leaves) of each experiment Note. Final traits were correlated in lm models in R with performance variables of field and common garden. Values are Pearson correlations coefficients. Significances are indicated with *. Significance levels are as following: from 0 to 0.001 = ***, from 0.001 to 0.01 = **, from 0.01 to 0.05 = *. Correlations in bold fonts are shown in Figure 3.
Kleyer, and Grünzweig (2017) found that size and reproduction traits rather than leaf economic traits drove the composition of Mediterranean annual vegetation along a land-use intensity gradient. This emphasizes the general importance of traits concerning clonal growth and vegetative reproduction for plant performance (Klimešová, Tackenberg, & Herben, 2016 conductance, leading to a higher photosynthetic capacity (Tomás et al., 2013). However, scleromorphic leaves were not advantageous in the field because species with mesomorphic leaves regrow more easily after mowing and grazing under real-world land-use conditions. As the field experiment was conducted in three different regions of Germany and contained grassland plots F I G U R E 3 Correlations of (a) RGR leaf length in the field with vegetative reproduction, (b) RGR plant projection area in the field with the vegetative reproduction, (c) biomass in the common garden with the trait height, and (d) RGR height in the common garden with the interaction trait LDMC-vegetative reproduction. Final traits were correlated in lm models with performance variables of field and common garden, respectively. The graphs show predictor variables with high correlation coefficients (for significance levels see Table 2) of different management regimes of grazing and mowing, future studies should aim at analyzing trait differences for particular land-use types.
In conclusion, our study showed that species-specific traits were capable to predict different dimensions of plant performance, characterized by relative growth rates and survival both under field and controlled common garden conditions. We found a prominent role of vegetative reproduction for plant performance, albeit with opposing effects under common garden and field conditions, and of LDMC. Zeuner for assistance during the field work. The work has been funded by the DFG Priority Program 1374 "Infrastructure-Biodiversity-Exploratories" (Project BE RICH, BR 1698/11-2).
Field work permits were issued by the responsible state environmental offices of Baden-Württemberg, Thüringen, and Brandenburg. We acknowledge the financial support of the Open Access Publication Fund of the Martin Luther University Halle-Wittenberg.

CO N FLI C T O F I NTE R E S T
The authors declare no competing financial interests.

AUTH O R CO NTR I B UTI O N S
H.B. and U.J. developed the conceptual and methodological foundation of this study and designed the experiment. E.B. coordinated and collected field data. E.B. and H.B. conducted the statistical analyses, wrote the first draft of the manuscript, and prepared the figures.

DATA ACCE SS I B I LIT Y
All data used in this paper is included in the Supporting Information (Tables S1-S3).