2.1 Study site and species selection
This study was conducted in the Dashanchong National Forest Park (28°23′58″–28°24′58″ N, 113°17′46″–113°19′8″ E), Changsha County, Hunan Province, China. This region is typical of a low hilly landscape 55‒217.40 m above sea level. The climate is humid mid-subtropical monsoon climate with an annual mean temperature of 17.30 °C and mean monthly temperatures of ˗10.30 °C in the coolest month (January) and 39.80 °C in the warmest month (July). The mean annual precipitation is 1416 mm (Ouyang et al., 2016; Wu et al., 2019). The soil type was well-drained clay loam red soil developed on slate and shale rocks. The regional climax vegetation was subtropical evergreen broadleaved forest.L. glaber–C. glauca forest, a well-preserved evergreen broadleaved forest is a multi- dominant, unevenly aged forest with structural stability (Zhao et al., 2015) that represents the succession direction of vegetation in the subtropical hilly region.
One-hectare permanent plot (100 × 100 m horizontal distance) of L. glaber–C. glauca evergreen broadleaved forest was established in 2009. The plot was divided into 100 10 ×10 m subplots to conduct a census. The second survey and leaf sampling were conducted in 2019. In this study, 18 dominant species with importance value ≥1.00% were examined accounting for 89.38% (Table 1).

2.2 Leaf sample collection and functional traits measurement

For each leaf trait, ten replicates were obtained from individuals of three standard trees of each dominant species (Pérez-Harguindeguy et al., 2013) growing on a 10 m × 10 m subplot. Fully expanded sun leaves above the crown were collected and packed in sealed moist plastic bags with damp paper that were transported to the laboratory in a cooler with ice to preserve the water saturation of the leaves until the time of the measurements. We measured the following structural functional traits: leaf thickness (LT, mm), leaf area (LA, mm²), specific leaf area (SLA, cm²·g-1), leaf dry matter content (LDMC, mg·g-1), and chemical functional traits: leaf carbon (LC, mg·g-1), leaf nitrogen (LN, mg·g-1), and leaf phosphorus (LP, mg·g-1) contents; leaf nitrogen:phosphorus ratio LN:LP and leaf carbon:nitrogen ratio (LC:LN) (Fig. 1).
We used different methods to measure the leaf thickness of the broad-leaved and coniferous trees. For broad-leaf species, thickness values were measured at five points per leaf from the front, middle, and end with digital calipe(Dasqua 150 mm Special Glass Grating Big Screen Digital Calipe,Sichuan, China, accurate to 0.02/0.001 mm), avoiding the mid-vein and secondary veins to reduce sampling variation, and then the average value was taken as LT. For conifer species, LT was obtained from the middle of the needles (He et al., 2020). The blades were then laid on A4 papers, flattened with a transparent plastic sheet to ensure that each blade was fully unfolded, and scanned to obtain a plane image. Adobe Photoshop CS 6 software (Xiao et al., 2005) was used to calculate the pixel points occupied by each leaf, and the average value was taken as the LA of broad-leaf species. For conifer species, leaf length (LL) and middle leaf width (LW) were determined to calculate as LA = .
Ten leaves per species were randomly selected and soaked in deionized water for 12 h in the dark. After removing the leaves, the excess water was immediately absorbed, and the saturated fresh weight (g) was weighed using an electronic balance (AR2140, OHAUS, Guangzhou Jingbo Electronics Co., Ltd,accurate to 0.0001). The leaves were then oven-dried to a constant weight at 65 °C for at least 72 h and then weighed using an electronic balance. Leaf area and dry mass (g) were then used to calculate the leaf dry matter content (LDMC = leaf dry weight/saturated fresh weight) and specific leaf area (SLA = leaf area/leaf dry weight) (Yang et al., 2021).
After measuring structural traits of leaf, we oven-dried the leaves at 105 °C for 5 min to eliminate the green and then dried them again at 85 °C to a constant weight. The leaves were milled and sieved through a 0.15 mm mesh. LC, LN, and LP were measured using the potassium dichromate heating, Kjeldahl, and molybdenum- antimony colorimetric methods, respectively (Wu et al., 2020). Two sets of parallel experiments and two blank tests were performed simultaneously for all parameter measurement to reduce the experimental errors. The average value was taken as the value of the parameter of the subplot and species and was considered to calculate the LN:LP and LC:LN.

2.3 Variables used in redundancy analysis (RDA) and variation partitioning

Environmental and spatial variables were generated as descriptors, and functional and phylogenetic compositions were generated as response variables before RDA and variation partitioning. Four topographic variables (mean altitude, slope, convexity, and aspect) and seven soil variables (mean soil moisture; pH; and concentrations of total carbon (TC), total nitrogen (TN), total phosphorus (TP), total potassium (TK), and available phosphorus (AP)) were measured as environmental variables. All topographic variables were calculated based on the elevations at the four corners of each subplot. Edaphic variables were measured using systematic and random sampling approaches to collect soil samples near the central point of each subplot. Detailed descriptions and formulas of the topographic and edaphic sampling methods can be found in the study by Zhao et al. (2015).
Principal coordinates of neighbor matrix (PCNM) eigenvectors were generated to describe the spatial variables and were then used as indirect proxies of dispersal-based processes (Borcard & Legendre, 2002). First, we used the central coordinates (X and Y) of each subplot to derive the Euclidean distance matrix. Second, a truncated matrix was generated by retaining and replacing the values in the distance matrix using a threshold (Jiang et al., 2018). Finally, the PCNM variables were generated by performing principal coordinate analysis on the truncated distance matrix (Dray et al., 2012). A total of 57 PCNM variables with eigenvalues higher than zero were generated as spatial variables at a fine spatial scale (10 m × 10 m) using the PCNM function in the vegan package.
To compare the effects of different response variables on the detection of ecological processes, we used 11 response variables, including multivariable, single functional, and phylogenetic variables. First, we calculated the community-weighted mean (CWM) (Bruelheide et al., 2018) taking the relative importance value of tree species within each subplot as the weight. The formula used is as follows: Traitc = ΣPi × traiti , where Traitc represents the CWM of i functional trait of leaf. Traiti and Pi are the values of functional traits of each species and the important value for species i in a subplot. Second, we used phylogenetic dendrograms and employed the phylogenetic fuzzy weighting (PFW) method to generate phylogenetic compositions (Jiang et al., 2018). Detailed descriptions of the calculations and background of the PFW approach can be found in the study by Duarte et al. (2016). This analysis was conducted using principal coordinates of phylogenetic structure (PCPS) (Debastiani & Duarte, 2014). Before analysis, all environmental variables (except convexity) and CWM of traits were log-transformed, and phylogenetic compositions were Hellinger-transformed.

2.4 Statistical analysis

To test the intraspecific and interspecific variations in leaf functional traits, we performed descriptive statistics on the leaf trait data of each species, calculated the mean, range, standard deviation (SD), standard error (SE), coefficient of variation (CV), and skewness of each trait and then tested the difference of each trait using the multiple comparison method of least significant difference (LSD).
To detect the phylogenetic signals of traits, phylogenetic independence analysis (phylogenetic independent contrasts, PICs) was performed and a phylogenetic dendrogram combined with functional traits was constructed. The species details (family/genus/species) were input into Phylomatic (http://www.phylodiversity.net/ phylomatic) to construct a phylogenetic tree. For this, the Angiosperm Phylogeny Group’s APG III consensus tree was taken as the pedigree skeleton and evolutionary branch length using Figtree was obtained (Webb et al., 2008). Next, we used Blomberg’s K test, which compares the variance of PICs with that expected in a Brownian motion (random) model (Blomberg et al., 2003). This analysis implemented the “phylosignal” function in the “picante” package (Kembel et al., 2010) and the phylogenetic tree of the species with 999 randomizations. A significant p value (< 0.05) in this analysis indicates that a phylogenetic signal exists and that phylogenetically close species are more similar than random pairs of species.
To explore the relationships among the nine traits, we determined correlations existing between traits and trait CWMs. The Pearson’s correlation coefficients among leaf traits at the species and community levels were determined. Then we examined the species-level trait relationships using PICs and community-level trait relationships using partial regression analysis by controlling the phylogenetic variables.
Further, RDA analysis was used to explore the influence of topography and soil on leaf functional trait and phylogenetic compositions at community level using the vegan package (Oksanen et al., 2013). To distinguish the effect of stochastic and niche processes on community assembly, variation partitioning was used to assess the relative importance of environmental and spatial variables as explanatory variables for the variations of functional traits (CWMs) and phylogenetic compositions using the rdacca.hp function in the rdacca.hp package (Lai et al., 2022).
All analyses were performed using R version 3.6.0 (R Development Core Team, 2016).