Niche differentiation along multiple functional‐trait dimensions contributes to high local diversity of Euphorbiaceae in a tropical tree assemblage

Understanding the mechanisms that drive community assembly in species‐rich tropical forest remains a fundamental challenge in ecology. Here, we integrated multivariate functional trait dimensions, phylogeny and metabolomics to test fundamental predictions concerning the role of differentiation with respect to abiotic and biotic niche axes in the maintenance of high local diversity of woody plants in the Euphorbiaceae. We measured 40 functional traits related to resource acquisition, photosynthetic capacity, hydraulic efficiency and secondary‐metabolite profiles generated using untargeted metabolomics in all 26 Euphorbiaceae species in a 20‐ha forest dynamics plot in tropical southwestern China. We examined the correlation structure of 40 traits using a trait networking approach. We coupled these traits with variation in soil nutrients, light environment, soil water content and herbivore pressure within the plot to assess niche differentiation in space. We compared phylogenetic signal among multivariate trait dimensions and secondary metabolites to assess niche differentiation in evolutionary time. Network analysis revealed that a small number of traits with high network centrality reflected variation in ecological strategy among the Euphorbiaceae. Using these high‐centrality traits, we observed significant functional turnover along environmental gradients defined by light, soil moisture, soil nutrients and leaf herbivory, respectively. Most resource utilization traits showed significant phylogenetic signal, whereas almost all defensive traits lacked phylogenetic signal, including species similarity with respect to plant secondary metabolites. Synthesis. Our results suggest that resource‐utilization traits and the habitat associations play a significant role in the niche segregation of co‐occurring woody plants in the Euphorbiaceae. Secondary metabolites, however, may enhance diversity at a finer spatial scale by allowing closely related species with similar functional traits to partition biotic niche space within shared habitats in tropical rainforest.


| INTRODUC TI ON
Co-occurrence of numerous closely related species at a local scale is a hallmark of diverse tropical forests (Gentry, 1989). Understanding the mechanisms that maintain such diversity in the face of intense competition for resources remains a long-standing challenge in ecology. Closely related species are often phenotypically and ecologically similar due to phylogenetic conservatism and are likely to occupy similar niches (Harvey & Pagel, 1991;Wiens et al., 2010).
Classical niche theory predicts that ecologically similar species should not stably coexist due to habitat overlap, resource competition, and shared natural enemies (Chesson, 2000;Gause, 1934;Holt, 1977). The idea that natural enemies with highly specialized host ranges may maintain plant diversity through conspecific negative density dependence (Connell, 1971;Janzen, 1970) has been proposed to an attractive potential solution to the paradox. Yet many insect herbivores and microbial pathogens are not single-host specialists (Gilbert & Webb, 2007;Novotny et al., 2002;Ødegaard et al., 2005) and hence are likely to mediate competitive exclusion among plants within their host ranges (Chesson & Kuang, 2008;Sedio & Ostling, 2013). Fortunately, while plant lineages with high local species richness challenge our understanding of coexistence, the very tendency toward phylogenetic niche conservatism that makes the high local diversity of these lineages such an apparent paradox also makes them excellent study systems in which to tease apart the niche axes that underpin their diversity.
Identifying the niche differences that distinguish co-occurring, closely related plants requires the measurement of traits as well as species interactions with the abiotic and biotic environment. Recent research on 'functional' traits of plants suggests that variation in life-history strategy and environmental distribution may be highly multivariate in the space of measurable morphological and physiological traits (Condit et al., 2013;Laughlin & Messier, 2015;Rüger et al., 2018;Trisos et al., 2014), even among closely related species (Sedio et al., 2012). Focusing on a single dimension may overlook niche differentiation in other dimensions of trait or niche space and limit explanatory power, yet the integrated study of multivariate trait space and the interaction of multivariate dimensions with multiple axes of variation in the abiotic and biotic environment has the potential to reveal niche segregation that would not be reflected in a single dimension (Burton et al., 2020;Futuyma & Moreno, 1988;Yang et al., 2018). Furthermore, it is increasingly recognized that physiological tradeoffs that characterize interspecific variation in stress tolerance, resource use, and life-history strategy are best described by multivariate trait dimensions rather than individual trait (Díaz et al., 2016;Messier et al., 2018;Mouillot et al., 2021). However, many previous studies have been limited to consider trait relationships within a leading dimension, but have not considered whether the trait dimensions represent independent axes of variation in plant physiology and resource use (Chave et al., 2009;Wright et al., 2004).
To overcome this problem, correlations among multiple traits or trait dimensions can be expressed as a trait network, in which the network nodes represent traits and edges reflect highly correlated traits in the network (Kleyer et al., 2019;Messier et al., 2017). Such trait networking approaches can enhance our understanding of trait correlations and interactions between trait dimensions and further revealing a small number of high-centrality traits that represent interspecific variation in physiology and life-history strategy.
Understanding the coexistence of closely related species within ecological communities may require the consideration of variation with respect to multiple fundamental eco-physiological strategies (Silvertown, 2004), such as photosynthetic traits, hydraulic traits, resource acquisition traits, and physical defensive traits. In addition to morphological variation, much of the functional trait variation of plants is a result of small organic molecules that comprise the metabolome. The plant metabolome includes primary metabolites involved in core metabolic pathways and the molecular building blocks of large organic polymers, such as nucleotides, amino acids and mono-and disaccharides. However, much of the interspecific variation in plants is a result of the astonishing diversity of secondary metabolites (SMs) with specialized functions (Sedio et al., 2021;Walker et al., 2022). SMs can mediate plant responses to abiotic stressors, such as ultraviolet radiation and freezing temperatures (Rasmann et al., 2014), and can serve as antinutritive agents or acute toxins against herbivores and pathogens (Coley, 1983) and play an important role in shaping natural enemy host ranges (Pagare et al., 2015;Salazar et al., 2018). Much like the shared resources, shared natural enemies such as insect herbivores and pathogens can mediate competitive exclusion of host plants (Chesson & Kuang, 2008;Sedio & Ostling, 2013). But unlike abiotic stressors, 4. Synthesis. Our results suggest that resource-utilization traits and the habitat associations play a significant role in the niche segregation of co-occurring woody plants in the Euphorbiaceae. Secondary metabolites, however, may enhance diversity at a finer spatial scale by allowing closely related species with similar functional traits to partition biotic niche space within shared habitats in tropical rainforest.

K E Y W O R D S
closely related species, convergent evolution, divergent evolution, Euphorbiaceae, functional trait dimensions, herbivores, plant secondary metabolites natural enemies are capable of reciprocal coevolution in response to the evolution of chemical defenses on the part of their plant hosts, which may make them strong agents of selection for divergence in chemical composition and the evolution of novel chemical defenses (Ehrlich & Raven, 1964;Schemske et al., 2009;Volf et al., 2020). The vast diversity of plant SMs has long precluded the study of metabolomics at the community scale. However, the recent rapid rise of ecological metabolomics (Sedio et al., 2021;Sedio, Boya, & Rojas Echeverri, 2018;Walker et al., 2022) promises to illuminate the role of plant SMs even in species-rich and understudied communities such as tropical forests.
Closely related species are derived from a recent common ancestor and hence expected to exploit a limited range of trait space.
Furthermore, closely related species are expected to be more ecologically similar than distantly related species (Ackerly, 2004;Burns & Strauss, 2011). For these reasons, divergence along trait or niche axes among closely related species should be observable against a phylogenetically conserved background and help to reveal the niche dimensions along which interspecific differentiation has contributed to the diversification of, and maintenance of diversity within, the lineage (Ackerly, 2004;Ackerly et al., 2006;McKown et al., 2016;Sedio et al., 2012;Swenson, 2011). Traits are likely to evolve as correlated suites or syndromes that reflect ecological tradeoffs in function (Arnold, 1983). Hence, consideration of phylogenetic patterns with respect to multivariate trait space may reveal conservation or divergence in multivariate ecological strategies (Rüger et al., 2020). phylogenetically conserved resource-use strategies, but phylogenetic divergence in defenses. There is growing evidence of widespread phylogenetic divergence in SMs within tropical tree lineages (Becerra, 1997;Kursar et al., 2009;Sedio, 2017). However, few studies have integrated the comprehensive study of morphological and physiological traits related to resource-utilization strategy with a metabolomics-based study of variation in SMs in a community and phylogenetic context.
Here, we assessed interspecific, ecological, spatial and phylogenetic variation in morphological, physiological and chemical traits to identify the key axes of variation that contribute to the high local diversity of trees in a single plant family in a local community. We measured 40 functional traits related to resource acquisition, photosynthetic capacity, hydraulic conductivity and efficiency and secondary-metabolite profiles for all 26 free-standing woody species of Euphorbiaceae in tropical seasonal rain forest in Xishuangbanna, southwestern China. We examined the correlation structure of interspecific variation among these 40 traits using a trait networking approach (Messier et al., 2017) and interspecific variation in leaf SMs through the use of untargeted metabolomics (Sedio et al., 2021;Sedio, Boya, & Rojas Echeverri, 2018). We coupled these traits with detailed measurements of variation in soil nutrients, light environment, soil water content and herbivore pressure to identify the axes of trait variation that may define niche differences among cooccurring woody Euphorbiaceae with the potential to facilitate ecological coexistence through segregation along key abiotic and biotic gradients. Specifically, we asked the following questions: (i) Could a few traits with central correlational relationships reflect interspecific variation in ecological strategy among co-occurring Euphorbiaceae?
(ii) Do interspecific variation of multivariate trait dimensions contribute to niche partitioning by segregating Euphorbiaceae in time and space? and (iii) Whether major axes of trait variation differ in phylogenetic signal, and hence the phylogenetic scale at which they contribute to niche differentiation among the Euphorbiaceae in a tropical tree community? We expected a few traits with central correlational relationships to represent interspecific variation in ecological strategy among co-occurring Euphorbiaceae species. In addition, we expected the local Euphorbiaceae to exhibit evidence of niche partitioning along multiple trait and environmental axes. We also expected that morphological and physiological trait axes and the abiotic niches to show greater phylogenetic signal than SMs, showing evidence of trait divergence among closely related species.

| Study site
The study was conducted in a seasonal tropical rainforest dynamics plot (FDP) in Xishuangbanna, southwestern China (101°34′E, 21°36′N; Figure 1a). The most dominant family in the plot is Icacinaceae, followed by Lauraceae and Euphorbiaceae, based on the importance index (Lan et al., 2008). The mean annual temperature is 21.8°C and the mean annual precipitation is 1492.9 mm in the plot. The forest is influenced by a tropical monsoonal climate, with 84% of mean annual precipitation (1246 mm) occurring during the rainy season from May to October, and a long dry season that lasts from November to next April. Soil type in the plot is mainly laterite with deep soil layers and thin humus (Cao et al., 2006). Habitat heterogeneity was caused by the three perennial streams which traverse the plot and merge together at the southeastern corner. The 2012 census recorded a total of 392 tree species belonging to 196 genera and 69 families represented by individuals with ≥1 cm diameter at breast height.

| Focal species
All 26 species of Euphorbiaceae (Appendix S2: Table S1) in the plot were selected as focal species for the following reasons: (1) Euphorbiaceae is one of the largest families of all flowering plants (Ernst et al., 2015) and nearly global in distribution with the exception of boreal areas, although it is more abundant in tropical regions (Rahman & Akter, 2013). (2) Euphorbiaceae ranked third with respect to the importance value among families in our plot, including 9827 individuals with DBH > 1 cm and 25.51% of the total basal area (Lan et al., 2008, Figure 1b). (3) Euphorbiaceae includes both pioneer and late successional species, components of both the canopy layer and understory, and is distributed from valley to ridge, reflecting variations in resource acquisition ability, photosynthesis capacity, shade tolerance and water requirements (Davies et al., 1998). (4) Most species of the Euphorbiaceae have extraordinary chemical diversity and these chemical compounds are thought to play an important ecological role through herbivore feeding deterrence and antimicrobial activity (Vasas & Hohmann, 2014). Thus, Euphorbiaceae provides an excellent system to examine niche segregation and assembly among closely related species.

| Functional traits measurements
To explore the species coexistence with respect to multiple niche axes, we classified functional traits into five multivariate dimensions based on specific ecological functions (Appendix S2: Table S2).
We collected 13 traits for resource acquisition, 8 traits for photosynthetic ability, 11 hydraulic traits, 8 physical defensive traits and secondary-metabolite profiles that likely include interspecific variation in chemical defenses against herbivores and pathogens ( Figure 1c; Appendix S2: Table S3). See Appendix S1 for detailed functional trait measurements. Descriptive statistics of 40 functional traits are reported in Appendix S2: Table S4.

| Environmental variables
Numerous environmental parameters have been proposed to be major drivers of species distributions. We measured soil nutrient properties, soil water content, light environment and insect herbivory to provide environmental context for species variation in traits that mediate resource acquisition, hydraulic ability, photosynthetic capacity and defensive ability (Both et al., 2019;Rosas et al., 2019;Sun et al., 2016). Soil available nitrogen (N), extractable phosphorus (P), extractable potassium (K), total carbon (C) and soil water content were published in Yang et al., 2014. Light environment was measured using a digital camera (Nikon Coolpix 4500; Nikon Corporation) with a fisheye lens (Nikon FC-E8 Fisheye Converter; Nikon Corporation, Japan) to take hemispherical photographs at low light conditions, with no strong direct light in the sky before sunrise or after sunset (Hale & Edwards, 2002) in each quadrat. We used the software Gap Light Analyzer Version 2.0 to analyze all images and calculated the canopy gap fraction for each photograph, in which light environment was quantified as the fraction of the image not occupied by vegetation cover (Frazer et al., 2000).
To quantify herbivore pressure, we measured herbivory on all species of Euphorbiaceae encountered (Halpern et al., 2010). We randomly selected five mature individuals with height ranging from 5 to 6 m for each species (Caldwell et al., 2016), and for each individual, three branches were taken from each direction and 10 leaves per branch were selected beginning from the tip. All collected leaves were scanned (Epson Co.), and leaf area was calculated using ImageJ (Abramoff et al., 2004). We measured the percent loss in area for each leaf by comparing the damaged leaf area to the area of the inferred intact leaf shape using the scanned images (Kurokawa & Nakashizuka, 2008). For each leaf, we calculated the herbivory ratio as the ratio of the damaged area to the estimated undamaged area of the leaf (i.e. leaves that suffered greater herbivore damage have a higher herbivory ratio). We classified herbivore damage as hole feeding and marginal feeding based on a guide of >140 distinctive patterns of damage caused by chewing and mining herbivores, excluding fungal and mechanical leaf damage (Labandeira et al., 2007).
Principal component analysis (PCA) was conducted on each type of environmental factor to reduce the trait data into major orthogonal axes. We used the first three principal components for further analysis of environment distance (Appendix S2: Tables S5 and S6).

| Network analysis: Exploring trait and traitdimension correlations
We evaluated relationships between measured traits and broad, multivariate trait dimensions representing resource acquisition, photosynthetic, hydraulic and physical defensive traits using network analysis in which we calculated the connectivity and distance properties of interconnected traits. We used the igraph package to construct the functional-trait network (Csardi & Nepusz, 2006).
In this form of network analysis, nodes represent distinct traits, which are linked by edges that represent correlations among traits (Messier et al., 2017). We used Pearson's correlation to calculate the observed trait correlations. Pairwise trait correlations with r > 0.2 were significant at p < 0.05 and were shown in the network.
All correlations below this threshold were set to zero. Because some trait correlations were negative, network connection strength between pairs of traits were weighted by the absolute correlation strength (Kleyer et al., 2019). Moreover, based on network analysis, we calculated network centrality and then found traits representing strong effects on the plant phenotype in each trait dimension.
We calculated the indicator of 'degree' to denote the centrality of each trait following Messier et al. (2017), which is the number of connections leading to a trait. Through this network analysis, we screened traits with large degree values one by one and selected the top four traits with large degree values in each trait dimension.
In subsequent analyses, we explored the environmental and spatial turnover with respect to these top four traits with the greatest 'centrality' in each dimension (centrality) and with respect to all traits in each trait dimension (all).
To explore the degree of trait integration that characterizes interspecific variation with respect to major dimensions representing resource-acquisition, photosynthetic, hydraulic, and physical-defense traits among the Euphorbiaceae, we used a network-based expectation-testing framework following Messier et al. (2017). For each of the four major trait dimensions, we compared two alternative expectations: the expectation that four traits with high network centrality describe interspecific variation (E CENTRALITY ) and the expectation that interspecific variation is better described using all measured traits (E ALL ). For each expectation test, we used standardized Mantel's tests, which calculate the Pearson's correlation coefficient for two correlation matrices (Zuur et al., 2007): the expected correlation matrix (E CENTRALITY or E ALL ) and the observed trait correlation matrix (D) calculated from our empirical data (Cheverud et al., 1989). We repeated these expectation tests for E CENTRALITY and E ALL for each of four major trait dimensions (Appendix S2: Tables S8-S11). These comparisons thus allowed us to test the expectation that each trait dimension is defined by its most central traits as opposed to a wider range of measurable traits. We did not have a specific expectation regarding the relative strengths of trait correlations, so we only included values of −1, 0 and 1 in the expectation matrices. For all traits, if the relationship between traits was expected to be positive, it was set to 1, if it was expected to be negatively correlated, it was set to −1, and if no relationship between traits was expected, it was set to 0, with expectations based on Messier et al. (2017). For E CENTRALITY , the correlation value was set to 0 if one of the traits was not a highcentrality trait and −1 or 1 for pairs of high-centrality traits (Yang et al., 2019;Yao et al., 2021). Note that these values do not test whether the correlations are perfect, but simply specify the signs of the correlation.

| Functional-trait turnover along environmental gradients
To test functional-trait turnover along environmental dimensions, we first calculated functional beta diversity between subplots at local scale on 20 m × 20 m. For each of the five trait dimensions, we calculated the functional dissimilarity between each pair of subplots using the trait distance (D pw ) of the four traits with the greatest centrality values and all functional traits (Ricotta & Burrascano, 2008). D pw was calculated as follows: where n k1 represents the number of species in community k1; n k2 represents the number of species in community k2; ik2 is the mean pairwise trait distance between species i in community k1 to all species in n k1 + n k2 , community k2 and jk1 is the mean pairwise trait distance between j species in community k2 to all species in community k1.
We used generalized additive models (GAMs; Wood, 2006) to test for significant nonlinear pattern of our hypothesized relationships between environmental drivers and turnover in functional traits in each dimension. We treated soil nutrients, light environment, soil water content, and leaf herbivory ratio as environmental factors. For the GAMs, we restricted the number of knots to three to avoid locally overfitting the data but still allowing unimodal or slightly more complex model fits. To identify the most relevant and best fitting environmental driver for each functional trait dimensions, we calculated GAMs for the turnover of each functional trait with environmental distance.
We also estimated functional-trait turnover with respect to spatial distance. Thus, we considered distance with respect to environmental factors and space as the independent variable and similarity with respect to trait dimensions as the dependent variables. Cubic splines were used to fit non-linear relationships in regression models.

| Phylogenetic signal of traits and trait dimensions
We to test for phylogenetic signal of functional traits and trait dimensions. We calculated the K mult from Adams (2014), which provides a useful means of evaluating phylogenetic signal in high-dimensional multivariate traits. The K mult is found from the equivalency between statistical methods based on covariance matrices and those based on distance matrices that is not possible with alternative procedures based on likelihood (Freckleton et al., 2002). We used this approach because it does not rely on the inversion of covariance matrices, thus it is not restricted to cases where the number of trait dimensions is less than the number of species in the phylogeny (Adams, 2014). This is particularly important for our traits, where the number of trait dimensions frequently exceeds the number of species in a phylogeny. Using computer simulations based on Brownian motion, values of K mult < 1 imply that taxa resemble each other phenotypically less than expected under Brownian motion whereas values of K mult > 1 imply that close relatives are more similar to one another phenotypically than expected under Brownian motion.

| Turnover in functional-trait dimensions along environmental gradients
Similarity with respect to trait dimensions that reflect photosynthetic traits, hydraulic traits, resource-acquisition traits and physical defenses significantly decreased with increasing environmental and spatial distance (Figure 3a-d, p < 0.001), indicating that these functional traits turned over along environmental gradients and spatial distance to a greater extent than expected by chance. In contrast, similarity of SMs increased with average herbivory distance in the local scale and then stabilized (Figure 3e, p < 0.001). In addition, the turnover with respect to axes representing functional traits was significantly greater with environmental distance than spatial distance, especially for physical defensive traits and SMs at large spatial scales.

| Phylogenetic signal in functional trait dimensions
Most resource-acquisition, photosynthetic and hydraulic traits and trait dimensions showed significant phylogenetic signal (

| DISCUSS ION
The co-occurrence of numerous closely related species challenges expectations of species coexistence, but also presents an opportunity to better understand the mechanisms that generate and maintain diversity in tropical forests. Here, we have explored multiple dimensions of variation in morphological, physiological, and chemical functional traits to identify key axes that may contribute to niche segregation among co-occurring confamilial species. Our results revealed substantial differentiation in trait dimensions related to photosynthetic, hydraulic, resource-acquisition and defensive strategies with the potential to contribute to species coexistence by allowing species to segregate with respect to variation in resource availability and herbivore pressure over time and space. Whereas, resourceacquisition traits exhibited phylogenetic signal, the diversity of closely-related Euphorbiaceae within the Xishuangbanna forest is likely further enhanced by phylogenetic divergence among the closest relatives with respect to SMs. We conclude that differentiation in chemical anti-herbivore defenses among closely related species may define another key trait axis that elevates species community richness beyond what would be supported by resource and habitatdefined niche partitioning alone.

| Resource-utilization and defense traits exhibit contrasting community patterns
Interspecific competition for resources is expected to result in the competitive exclusion of inferior competitors from a community (Palmer, 1994 Notes: E CENTRALITY -Centrality defined trait dimensions: only four traits with large degree value. E ALL -All defined trait dimension: all traits in each dimension. For each trait dimension, the best supported expectation is highlighted in grey.
competitors (Chesson, 2000). Species differ in intrinsic fitness; the greater the differences in fitness, the greater niche differences must be to stabilize coexistence among competitors (Adler et al., 2007).
In a forest, the most obvious opportunities for niche segregation among tree species are habitats defined by spatial heterogeneity in edaphic resources, soil moisture, and light. Variation in species abundances over environmental gradients are mediated by morphological and physiological traits. Turnover with respect to traits can reflect filtering that habitat variation exerts on local assemblages, which provides a window on niche segregation among co-occurring species (Condit et al., 2000;Le Bagousse-Pinguet et al., 2017;Ravenscroft et al., 2014). A previous study demonstrated trait turnover with respect to both geographic and environmental distance within the Xishuangbanna plot (Yang et al., 2015). Here, we observed greater turnover in mean trait values of subplots with respect to environmental distance than simple geographic distance among subplots

F I G U R E 3
Differentiation of traits with large centrality values and secondary metabolites with respect to environmental and spatial distance. Panels (a)-(f) illustrate distance decay of (a) photosynthetic traits with light environment and spatial distance, (b) hydraulic traits with soil water content and spatial distance, (c) resource acquisition traits with soil resource content and spatial distance, (d) physical defensive traits with average herbivory ratio and spatial distance, (e) secondary metabolites with average herbivory and spatial distance, (f) all traits with all environmental and spatial distance. To calculate environmental distance with respect to soil variables and all variables, we used the first PC of variation in soil variables and in all traits, respectively. The contour lines represent the density of the traits and distance values.
In addition to abiotic resource requirements, trees exhibit a fundamental trade-off between growth rate and survival (Rüger et al., 2018(Rüger et al., , 2020, which manifests at the extremes as species that grow fast in high-resource (especially high-light) environments but invest comparatively little in defense and those that grow slowly in resource-poor environments but invest heavily in defense (Coley, 1983). We observed significant heterogeneity in herbivore pressure within the forest plot ( Figure S1), likely driven by variation in light availability and hence productivity. Likewise, the turnover we observed in physical defenses was greater with respect to variation in herbivory than with spatial distance (Figure 3d), most likely because physical defenses were low and herbivory was high in high-resource environments, particularly canopy gaps (Coley, 1983).
Competition mediated by shared natural enemies is equivalent to resource competition in its capacity to mediate competitive exclusion (Chesson & Kuang, 2008), hence species differences with respect to SMs that shape insect and pathogen host ranges can define niche differences that stabilize coexistence (Connell, 1971;Janzen, 1970;Sedio & Ostling, 2013). The turnover we observed in SMs with either spatial distance or distance defined in terms of similarity of herbivory rates was much less than that observed for other traits (Figure 3). This is likely because natural enemies that respond to the density of host plants promote local neighborhoods of chemically dissimilar individuals, which tends to reduce turnover in chemistry at larger spatial scales (Sedio & Ostling, 2013).

| Resource-utilization and defense traits exhibit contrasting phylogenetic signal
Evaluating the phylogenetic signal in functional traits can provide an opportunity to assess the interaction between trait evolution and community assembly (Kembel & Hubbell, 2006 In contrast to physiological traits related to resource acquisition, species similarity with respect to SMs did not exhibit phylogenetic signal (Table 2; Figure 4). This result is consistent with a growing number of studies of SMs in tropical tree lineages, including Bursera (Burseraceae) in Mexico (Becerra, 2007), Eugenia (Myrtaceae), and Protium (Burseraceae) in Panama (Kursar et al., 2009;Sedio, Parker, et al., 2018), Ficus (Moraceae) in Papua New Guinea (Volf et al., 2018), Piper in Costa Rica (Salazar et al., 2016) and Inga and Protium in Peru (Endara et al., 2017;Salazar et al., 2018;Vleminckx et al., 2018 Overall, the study found substantial differentiation in trait dimensions related to photosynthetic, hydraulic, resource-acquisition and defensive strategies with the potential to contribute to species coexistence by allowing species to segregate with respect to variation in resource availability and herbivore pressure over time and space. However, it remains unclear how these traits and SMs impact the demographic processes underlying the mechanisms discussed.
Therefore, future studies should focus on how these traits, including SMs that shape the host use patterns of herbivores and pathogens, directly impact species coexistence through their effects on growth, survival and reproduction.

| CON CLUS IONS
Identifying the factors that permit closely related species to co-exist in species-rich tropical forests continues to be a major challenge in ecology. Traditional functional trait-based approaches have explored species differences with respect to a small number of traits that may not reflect the range of abiotic and biotic niche dimensions along which co-occurring species may segregate. Our results indicate that a small number of traits with high network centrality within the Euphorbiaceae reflect interspecific variation in ecological strategy with respect to global trait dimensions, representing resource acquisition, photosynthetic capacity, drought resistance and hydraulic efficiency and physical defense. Furthermore, analyses of turnover with respect to environmental gradients suggest that species differences among multiple axes of trait variation may contribute to species coexistence in the Xishuangbanna forest by allowing Euphorbiaceae to exploit distinct microhabitats defined in terms of light, moisture and soil nutrients. These physiological niche differences exhibit phylogenetic signal, whereas SMs do not, possibly as a result of diversifying selection by insect herbivores and pathogens.
Our results suggest that phylogenetic divergence among closely related species with respect to SMs may enhance the local diversity of Euphorbiaceae beyond that supported by resource-based niche segregation by promoting coexistence among close relatives with similar habitat preferences in tropical rainforest. Academy of Science. We thank Yazhou Zhang and Lu Sun for assistance with data analysis.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/1365-2745.13984.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available from the Dryad Digital Repository https://doi. org/10.5061/dryad.jwstq jqcw (Wang et al., 2022).