Effects of soil resource availability on patterns of plant functional traits across spatial scales

Abstract Identifying patterns and drivers of plant community assembly has long been a central issue in ecology. Many studies have explored the above questions using a trait‐based approach; however, there are still unknowns around how patterns of plant functional traits vary with environmental gradients. In this study, the responses of individual and multivariate trait dispersions of 134 species to soil resource availability were examined based on correlational analysis and torus‐translation tests across four spatial scales in a subtropical forest, China. Results indicated that different degrees of soil resource availability had different effects on trait dispersions. Specifically, limited resource (available phosphorus) showed negative relationships with trait dispersions, non‐limited resource (available potassium) showed positive relationships with trait dispersions, and saturated resource (available nitrogen) had no effect on trait dispersions. Moreover, compared with the stem (wood density) and architectural trait (maximum height), we found that leaf functional traits can well reflect the response of plants to nutrient gradients. Lastly, the spatial scale only affected the magnitude but not the direction of the correlations between trait dispersions and environmental gradients. Overall, the results highlight the importance of soil resource availability and spatial scale in understanding how plant functional traits respond to environmental gradients.


| INTRODUC TI ON
Identifying the patterns and drivers of plant community assembly has long been a central issue in ecology (Chapman & McEwan, 2018;Chesson, 2000). There is a growing consensus that the assembly of natural plant communities is governed by stochastic (random events related to dispersal, establish, mortality, etc.) and deterministic (Hubbell, 2001;Macarthur & Levins, 1967) processes. In particular, the two deterministic processes environmental filtering and biotic interactions have attracted more attention in community assembly research (Kuczynski & Grenouillet, 2018;Swenson & Enquist, 2009).
Environmental filtering increases species similarity through abiotic constraints, while biotic interactions lead to a limitation of the similarity of coexisting species (Gotzenberger et al., 2016;Ramm et al., 2018;Spasojevic & Suding, 2012). Plant functional traits describe ecological differences between different species; however, which plant functional traits could well reflect plant response to environmental gradients and how they respond remain questions (Costa et al., 2017;Wang et al., 2018).
Trait dispersion is a measure of the variation in functional traits within a community. Revealing how trait dispersions vary with environmental gradients is important to advance the predictive ability of functional ecology (Muscarella & Uriarte, 2016). A widely used explanatory proposition for the relationships between trait dispersions and environmental gradients is the stress-dominance hypothesis (SDH), which predicts that environmental filtering plays a major role in stressful environments, yielding a clustered pattern of trait dispersion, whereas biotic interactions determine community assembly in benign environments, favoring an overdispersed pattern of trait dispersion (Coyle et al., 2014;Swenson & Enquist, 2007;Weiher & Keddy, 1995). Therefore, an increasing trait dispersion pattern from stressful to benign environments should be expected (Costa et al., 2017;Spasojevic & Suding, 2012;Wang et al., 2018). Although many studies have tested the universality of the SDH in forest communities, empirical supports are still contradictory (Coyle et al., 2014;Lhotsky et al., 2016;Spasojevic & Suding, 2012;Wang et al., 2018).
Firstly, testing of the SDH depends on the proxy used to measure environmental gradients (Costa et al., 2017;Lhotsky et al., 2016;Wang et al., 2018). It is noteworthy that different studies often calculate standing biomass as proxies of environmental gradients (Costa et al., 2017;Lhotsky et al., 2016; and this method implicitly assumes that the environmental stress increases with decreasing standing biomass (Grime, 1977;Gross et al., 2010).
However, the variation in standing biomass could be driven by the topography, soil, or microhabitat properties in the research area (Gross et al., 2010;Michalet, 2006), which makes the comparison between different studies without basis and leads to conflicting empirical supports . Secondly, whether the SDH is true for different functional traits is still unknown, because different functional traits may respond differently to environmental gradients Wang et al., 2018).
Although soil resource availability is a major driving force in community trait structure (John et al., 2007;Katabuchi et al., 2012;Pinho et al., 2018), it has been rarely considered when analyzing how trait dispersions vary with environmental gradients. As the "big three" of crucial soil nutrients, the resource availability of nitrogen (N), phosphorus (P), and potassium (K) strongly affects competition between plant species, as species vary in their ability to cope with different nutrient resources (Koerselman & Meuleman, 1996;Mao et al., 2019).
In response to N limitation, legumes could fix atmospheric N in a symbiotic relationship with bacteria , and plant species could also coexist by varying in the form in which they preferentially absorb N (e.g., NO 3 − , NH 4 + , or organic N) in an N-limited community (Ehrenfeld et al., 2005). Moreover, plants can respond in two ways to overcome P limitation: increasing P-use efficiency aboveground, and/or adjusting their P-uptake strategies, including root morphological, physiological, and biotic adaptations (Ehrenfeld et al., 2005;Kitayama, 2013;Vitousek et al., 2010). Recent studies have revealed that plants that had the highest mycorrhizal dependency could be supplied 90% P by mycorrhizal fungi (van der Heijden et al., 2006(van der Heijden et al., , 2015. Plants could also absorb nutrients over their needs to prevent other species from taking over (Van Wijk et al., 2003). For instance, when there is plenty of K in the soil then plants secure it more than the normal amount and these reserves could be used to support growth when external nutrients are not available (Chapin, 1980).
Overall, plants evolve a variety of strategies to cope with the change in soil resource availability, which can ultimately be reflected in the distribution patterns of plant functional traits (Suding et al., 2005).
Here, we tested how trait dispersions varied with soil resource availability, which may contribute to reconciling the contrasting relationships between trait dispersions and environmental gradients. We hypothesized that different degrees of soil resource availability (limited resource, non-limited resource, and saturated resource) should have different effects on trait dispersion patterns. Firstly, trait dispersion should be highest in habitats with the lowest concentration of limited resources, where competition for those resources is strongest (Figure 1). It means that the less the limited resources, the more the intense competition between plants.
F I G U R E 1 Conceptual models of the effects of soil resource availability on trait dispersion. Limited resource has a significant negative effect on trait dispersion from more stressful to less stressful habitats; Non-limited resource has a significant positive effect on trait dispersion from less benign to more benign habitats; saturated resource has no significant effect on trait dispersion from less saturated to more saturated habitats Empirical studies have shown that plants in infertile soils produce more roots to increase the competitive ability for belowground nutrients (Chen et al., 2021;Schenk, 2006). Secondly, non-limited resources should have significant positive effects on trait dispersions from less benign to more benign habitats (Figure 1). In fact, plants can adjust energy allocation between shoots and roots in different environments (Campbell et al., 1991). Previous studies found that aboveground competition for light was most intense in benign habitats (Grime, 1977;Weiss et al., 2019;Wilson & Tilman, 1993) and nutrient addition (single nutrient or multiple nutrients in combination) ultimately shifted biotic interactions from belowground competition for nutrients toward aboveground competition for light (Harpole et al., 2016). Lastly, saturated resources should have no significant effects on trait dispersions because these resource supplies exceed the biological demands ( Figure 1).
It is undeniable that trait dispersion may depend on which functional traits are involved Wang et al., 2018).
As different functional traits are often related to different ecological strategies (Violle et al., 2007), testing our hypotheses based on single and multivariate trait dispersions may provide more insights about particular niche axes (Lhotsky et al., 2016;Spasojevic & Suding, 2012). Moreover, the detectability of trait dispersion is also scale-dependent (Cavender-Bares et al., 2009;Weiher & Keddy, 1995;Zhang et al., 2018). For example, trait overdispersion has been more often detected at small spatial scales where competitive adversity predominates Price et al., 2017). Additionally, previous studies also found that the relative importance of environmental variables to community assembly increased with increasing spatial scales (Chase, 2014;Legendre et al., 2009). Thus, multiple spatial scale analysis is helpful to evaluate the relationships between trait dispersions and environmental gradients, since a large number of quadrats represent a wide range of spatial variability in soil resource availability and species composition (Coyle et al., 2014).
The 20-ha subtropical forest plot in Dinghushan (DHS), China, is characterized by the following soil features: (1) limited available phosphorus (AP) due to low soil pH caused by nitrogen deposition and highly weathered soil feature of the old age of this forest, (2) non-limited available potassium (AK) and (3) saturated available nitrogen (AN) due to long-term high nitrogen deposition in this region Koerselman & Meuleman, 1996;Lin et al., 2013;Lu et al., 2010;Mo et al., 2006). All of these soil features provide an ideal background for examining how trait dispersions vary with soil resource availability. Here, we analyzed trait dispersion patterns of 134 species across six functional traits and four spatial scales while taking into account the edaphic data. We hypothesized that: (1)

| Functional trait measurement
For the 134 species that accounted for 99.0% of all individuals with DBH ≥ 1.0 cm, six plant functional traits including a stem trait (wood density, WD), an architectural trait (maximum height, Hmax), and four leaf traits: leaf area (LA), specific leaf area (SLA), leaf dry matter content (LDMC), and leaf thickness (LT), were measured Shen et al., 2016). Traits selected in this study are expected to be good predictors of the response of plant species to variation in resource availability Pinho et al., 2018;Poorter et al., 2008). LA (cm 2 ) is related to light acquisition (Hao et al., 2018). SLA (cm 2 g −1 ) is a good indicator of the potential relative growth rate of plants (Hao et al., 2018;Wright et al., 2004). LDMC (g g −1 ) is associated with leaf life span and correlated with leaf resistance (Cornelissen et al., 2003;Vaieretti et al., 2007). LT (mm) has important consequences for leaf water content and nutrient cycling (Afzal et al., 2017;Pérez-Harguindeguy et al., 2016). WD (g cm −3 ) represents a trade-off between low construction costs and high growth rates vs. high construction costs and low growth rates (Poorter et al., 2008). Hmax (m) is indicative of plant competitive vigor and strategy (Hao et al., 2018). Wood samples for each species followed Cornelissen et al. (2003), and WD was calculated as the ratio of dry mass to fresh volume. Specifically, 6-12 individuals were randomly selected for each species and 10 mature leaves for each individual were measured . Detailed sampling protocols for these six functional traits were described in previous studies Shen et al., 2016). We calculated the mean value of each trait for each species  to represent the characteristics of species because of the larger variations in functional traits between than within species (Garnier et al., 2001;Shipley, 2007).

| Soil data measurements
Because of the thin soil layer in the DHS plot, topsoil (0-10 cm depth) was collected using a 5.0-cm-diameter soil auger after removing the litters on the soil surface based on regular grids of 30 m × 30 m (Lin et al., 2013). Each of the 238 grid points was paired with two additional sample points at 2, 5, or 15 m in a random compass direction from the grid to capture fine-scale variation in soil properties (John et al., 2007). In total, 710 soil samples (four of 714 samples were not taken because they fell in creeks or on rocks) were collected and nine soil properties were measured (Lin et al., 2013), including soil moisture (SM), organic matter (OM), soil pH (pH), total phosphorus (TP), available phosphorus (AP), total potassium (TK), available potassium (AK), total nitrogen (TN), and available nitrogen (AN). Kriging methods were used to obtain the predicted values of the nine soil properties for each quadrat at 5 m × 5 m, 10 m × 10 m, 20 m × 20 m, and 50 m × 50 m spatial scale (Gallardo, 2003). Descriptive statistics of the edaphic variables across four spatial scales are given in Table 1.

| Statistical analysis
2.4.1 | Comparing the observed trait dispersions with those from null models Trait dispersion is quantified by a multidimensional functional diversity index: functional dispersion (FDis) (Laliberté & Legendre, 2010). FDis is defined as the mean distance of individual species to the centroid of all species in the multidimensional trait space within a given community (Laliberté & Legendre, 2010). Among several available metrics of functional diversity (Mouchet et al., 2010), the main reasons why FDis was selected were as follows: Firstly, FDis is independent of species richness ; furthermore, FDis can take relative abundances of the species into account ; finally, FDis can be used for single or multiple traits (Spasojevic & Suding, 2012). Moreover, FDis well represents the degree of trait dissimilarity among coexisting species, and thus, it is closely related to the strategies of resource utilization (Chiang et al., 2016;Hao et al., 2018).
Before the analysis, all of the six functional traits were rescaled to center on zero with a standard deviation of 1 to eliminate the effects of the magnitudes of the data on the calculation of FDis (Hao et al., 2018;Villeger et al., 2008). Besides, once the species-species Euclidean distance matrix is obtained from the species-trait matrix, a principal coordinates analysis (PCoA) is then performed on the distance matrix, and the resulting PCoA axes were used as the new "traits" together with a speciesabundance matrix to compute the FDis (Hao et al., 2018;Laliberté & Legendre, 2010). Detailed algorithms on how to perform multivariate dispersions in PCoA space from the species-species Euclidean distance matrix and how to correct for negative eigenvalues can refer to Anderson (2006). FDis was calculated by the R package "FD" (Laliberté et al., 2014). To test whether any observed trait dispersion is a random distribution or shows trait clustering or trait overdispersion within a quadrat, we generated 999 random assemblages for each quadrat, keeping the same number of species abundances and occurrence frequency in the DHS plot and only randomly shuffling taxon names (Yang et al., 2014). The null models were run by the R package "picante" (Kembel et al., 2010). Based on the 999 random assemblages, a standardized effect size of FDis (ZFDis) for each quadrat following Gurevitch et al. (1992)

| Principal components analysis
Because the axis scores of PCA based on soil factors were usually calculated as indicators of environmental gradients (Costa et al., 2017;Coyle et al., 2014;John et al., 2007), we also conducted the PCA from the R package "vegan" (Oksanen et al., 2017) for the nine measured soil variables. To determine how many PCA axes should be retained, Horn's parallel analysis was also conducted and components with adjusted eigenvalues greater than 1 are retained (Dinno, 2018). Finally, the retained PCA axes were selected as one kind of composite indicator of the environmental gradients.

| Correlational analysis and torustranslation test
Pearson's correlation coefficient (r) was calculated between trait dispersion and each of the five environmental variables (PC1, PC2, AN, AP, and AK) at four spatial scales. We tested whether r between trait dispersion and each environmental variable at each spatial scale was significant using torus-translation tests, which take into account the inherent spatial autocorrelation in both trait dispersions and environmental variables (Harms et al., 2001). calculated based on the number of times that the observed r was higher or lower than the predicted r values (Comita et al., 2007). If the observed r value was lower than 5.0% or higher than 95.0% of the expected r values in a given spatial scale, then we could infer that the environmental variable has a significant negative or positive effect on trait dispersion at a significance level of 0.05. Overall, a total of 32,000, 8000, and 320 translated maps were generated at 5 m × 5 m, 10 m × 10 m, and 50 m × 50 m spatial scale, respectively.
For each environmental variable, r values were plotted against the spatial scales (quadrat area), and their trends were tested by fitting linear regression models. All analyses were conducted in R 3.5.1 (R Core Team, 2018).  (Figure 2c). This set of edaphic factors was also described by the PCA. According to Horn's parallel analysis, 2 axes should be retained for all spatial scales considered except for 50 m × 50 m spatial scale ( Figure S1). To ensure the consistency of analysis, the first two axes of PCA for each spatial scale were selected for further analysis. Specifically, they explained 77.9%, 78.0%, 77.9%, and 79.1% of the variance across four spatial scales, respectively (Table 3). In all cases, PC1 showed a soil fertility gradient from infertile to fertile habitats (Table 3). It should be addressed that AP, pH, and SM were negatively correlated with the PC1 (Table 3). PC2

| Environmental gradients
represented a gradient from stressful (low TK and SM) to benign (high TK and SM) conditions (Table 3).

| Trait dispersion patterns along with spatial scales
We quantified ZFDis for a multivariate trait that considered all traits in combination. Overall, trait dispersions in our plot showed both trait clustering and overdispersion patterns across four spatial scales  (Table 4). Overall, a wider range of spatial variability in patterns of plant functional traits was detected with the increasing spatial scale.

| The effects of soil resource availability on trait dispersions
Different degrees of soil resource availability showed distinct relationships with the multivariate trait dispersions across spatial scales ( Figure 4). Specifically, PC2 and non-limited AK showed significant positive relationships with multivariate trait dispersions (Figure 4b,d) and limited AP had significant negative relationships with trait dispersions across spatial scales (Figure 4c). However, neither PC1 nor saturated AN exhibited significant correlations with trait dispersions across spatial scales (Figure 4a,e).
Moreover, the effects of soil resource availability on most patterns of individual leaf functional traits showed similar results with the multivariate trait dispersion (Figures 5-7). However, the architectural trait (Hmax) and stem trait (WD) mainly showed no responses to environmental gradients (Figures 5-7). It should be pointed out that trait dispersions of Hmax showed positive correlations with AK across spatial scales ( Figure 6). Overall, the magnitudes of correlation between environmental variables and trait dispersions increased from small to large spatial scales, while directions of correlation between environmental variables and trait dispersions did not change from small to large spatial scales (Figures 5-7).

| DISCUSS ION
Many studies have stressed the importance of abiotic environments to patterns of plant functional traits (Chapman & McEwan, 2018;Liu et al., 2013;Uriarte et al., 2004), but how trait dispersions respond to soil resource availability remains elusive. Our results found that the relationships between trait dispersion and environmental gradient depended on soil resource availability and spatial scale.  Yu et al., 2018). Here, we expected that the less the soil resource availability, the more the intense belowground competition would be (Chen et al., 2021;Tilman, 1982;Zemunik et al., 2015), leading to an increasing trait dispersion pattern. Consistent with our hypothesis, limited soil resource (AP) showed negative relationships with the multivariate trait dispersions across spatial scales in this subtropical forest (Figure 4)

F I G U R E 3
Distribution patterns of the multivariate trait dispersion (ZFDis, the standardized effect size of trait dispersion) at four spatial scales. The color bar on the right of each map indicates the ZFDis values. "ZFDis > 0" indicates that trait dispersion is overdispersed, and "ZFDis < 0" indicates that trait dispersion is clustered. The lines in each map represent the contour lines at 20-m intervals (1993) also found that belowground competition was most intense in the lowest N availability plots and decreased significantly with N addition in an N-limited sand plain.
P is primarily derived from bedrock weathering, and its availability declines with the increase in soil age (Wright et al., 2011;Yu et al., 2018). Foliar N:P ratio has been proposed as an effective indicator of P limitation (Güsewell, 2004;Koerselman & Meuleman, 1996). Liu et al. (2013) reported the mean value of foliar N:P ratios in this forest was 29.6, which far exceeded the critical threshold of 12.5 to 26.3 (Tessier & Raynal, 2003). Moreover, the average content of AP is 1.81 mg kg −1 (Figure 2a), indicating severe P limitation in the DHS plot.
Here, we also explored how individual trait responded to P limitation. Compared with the stem (WD) and architectural traits (Hmax), leaf traits are more sensitive to P limitation and these trait dispersions were highest in the most P-deficient habitats. Except for fulfilling a structural role, P plays an integral role in photosynthesis, plant growth, and resistance to pathogens and abiotic stresses (Kitayama, 2013;Thuynsma et al., 2016). Overall, the results showed that leaf functional traits could well reflect the adaptive strategies of plants to survive in low P habitats (Roy-Bolduc & Hijri, 2011;Thuynsma et al., 2016).

| The effect of non-limited soil resource on patterns of plant functional traits
Plants can store more than normal amounts of K to support growth, and this phenomenon is called luxury consumption (Chapin, 1980). K is considered seldom limited plant growth in the natural community (Koerselman & Meuleman, 1996). The average content of AK is 55.03 mg kg −1 (Figure 2b). The moderate supply of AK (Sun, 2005) indicates K is a non-limited soil resource in the DHS plot. Thus, we predicted trait dispersions increased with the content of the non-limited AK from less benign to more benign habitats. In line with our hypothesis, significant positive relationships between AK and the multivariate trait dispersions indicate that the intensity of aboveground competition for light may be greater in more benign habitats (Weiss et al., 2019;Wilson & Tilman, 1993).  The results based on the individual leaf and architectural traits also support our hypothesis. K plays essential roles in plant processes, including regulation of plants' responses to light and stress resistance (Traenkner et al., 2018;Wang et al., 2013). For the leaf traits, LA and SLA are associated with photosynthesis (Hao et al., 2018;Wright et al., 2004), and LDMC and LT are related to plant resistance to physical hazards (Afzal et al., 2017;Vaieretti et al., 2007).
Thus, we can infer that competing for more AK not only improves photosynthesis but also increases resistance to biotic and abiotic stress. Moreover, trait dispersion of Hmax also showed a positive relationship with the increasing content of AK, indicating the important role of Hmax in determining the coexistence of species at different vertical layers .
It should be noted that there is still no clear consensus on which functional traits are more related to which ecological processes (Lhotsky et al., 2016;Yang et al., 2018). Because different functional traits may represent different niche axes Violle et al., 2007), how these functional traits respond to specific environmental gradients remains to be further explored (Spasojevic & Suding, 2012  tively (Lu et al., 2018). Fang et al. (2006) also pointed out that the DHS plot can be considered as an N-saturated ecosystem, resulting in increased leaching of various forms of N.
Although N deposition was thought to play a major role in biodiversity loss due to its effects on soil acidification, aluminum mobility, nutrient base cations, and the ratios of N versus other elements in plant tissue (Gress et al., 2007;Tian et al., 2018), we failed to detect the influence of increasing N availability on trait dispersions across spatial scales. There may be two possible explanations. Firstly, the mature forest in this plot is a regional climax forest type and has been protected for more than 400 years, which has become N-saturated from both the old age of this forest and chronic highlevel N deposition in this region Mo et al., 2006).
Presumably, the species in this forest may be adapted to this kind of high-N conditions and their competition for N might be minimal (Lu et al., 2010). This could be supported by the fact that the N contents in both leaves and roots did not significantly increase with chronic N deposition in this plot (Liu, Zhou, et al., 2010). Secondly, in a field experiment that carried out near this plot, Lu et al. (2010) found that N additions only decreased the abundance of understory ferns, moss, and seedlings, but had no significant effects on shrubs and canopy trees. Besides, the effect of increasing N availability on community structure and composition should be time-dependent (Güsewell, 2005), and detecting generalizable patterns of trait dispersions at different temporal scales should be rewarding.

| The effect of PCA axis on patterns of plant functional traits
A surprising finding was that the PC1 that demonstrated soil fertility gradient from infertile to fertile soils had no significant effects on trait dispersions across spatial scales (Figure 4a)

| The effect of spatial scale on relationships between trait dispersion and soil resource availability
We also found that spatial scale only affected the magnitude but not the direction of the correlations between trait dispersions and environmental gradients.  (Cavender-Bares et al., 2009;Li et al., 2019). Moreover, community assembly in natural communities is governed by stochastic and deterministic processes (Hubbell, 2001;Macarthur & Levins, 1967).
As spatial scale declines to encompass fewer individuals and less environmental heterogeneity, the relative importance of those stochastic events to the community assembly increases (Chase, 2014), resulting in a weak correlation between environmental gradients and trait dispersions at smaller spatial scales. Previous studies also found that the relative importance of environmental variables to community assembly increased with increasing spatial scales (Chase, 2014;Legendre et al., 2009), leading to a stronger correlation between environmental gradients and trait dispersions at larger spatial scales. Thus, multiple spatial scale analysis is helpful to evaluate the relationship between trait dispersions and environmental gradients.  Violle et al., 2007), we also found that leaf functional traits can well reflect the response of plants to nutrient gradients. Lastly, we point out that spatial scale only affects the magnitude but not the direction of the correlations between trait dispersions and environmental gradients. Overall, these findings are essential for a better understanding of the forces that determine the structure and dynamics of natural communities and to advance the predictive theory of functional ecology.

ACK N OWLED G M ENTS
Many constructive comments on the manuscript were received in the ForestGEO's 2019 Annual Analytical Workshop at the Nanyang Technological University in Singapore. We are grateful to Drs.
Nathan G. Swenson, Marco D. Visser, Jie Yang, Xiaoyang Song, Qinhong Ma, and Marielle Smith for their useful suggestions.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
The tree census data for the 20-ha Dinghushan plot in 2005 are available through the online portal at: http://www.fores tgeo.si.edu.
The data of plant functional traits are openly available in (Shen et al., 2016)