This study was conducted in Wanmulin National Reserve, Jian’ou, Fujian, China (27°02′28″-27°03′32″N, 118°02′22″-118°09′23″E). This area has a subtropical monsoon climate, with an average annual temperature of 18.8°C and a frost-free period spanning over 277 days. The mean annual precipitation amounts to 1673 mm and primarily occurs during the period from April to June. The mean annual relative humidity is 80%. The terrain consists of low mountains and hills having an altitude of 234-556 m. The subtropical evergreen broad-leaved forest studied here covers an area of 1.89 km², but it exhibits remarkable species richness, with a total of 1205 vascular plant species representing 581 genera and 161 families (30 families of ferns, 7 families of gymnosperms, and an impressive array of angiosperm taxa comprising 124 families) (Zhu et al., 1997). The most representative plant families in this forest are Fagaceae, Magnoliaceae, Lauraceae, Symplocaceae, and Theaceae. Dominant species include Cinnamomum camphora, Castanopsis carlesii, Phoebe zhennan, Castanopsis faberi, etc.

During June and July 2018, based on the results of the plant community biodiversity survey in this Nature Reserve, 60 dominant woody tree species ( Supplementary Table S1 ) were selected based on the dominance in tree and shrub layers in a representative 40 m × 80 m plot within the forest. For each species, three individuals with similar size (diameter at breast height or ground diameter) were selected, and then the paired leaf and root samples were collected from each individual. Five analogous leaf and root traits were measured, 10 functional traits in total ( Table 1 ), including 6 core traits (LMA, LN, RD, SRL, RTD, RN) which were commonly used to define the LES and RES axes (Weigelt et al., 2021). Fifty fully expanded (current year) and undamaged leaves were collected from the upper crown of each individual. Leaf thickness (LT) (mm) was measured with a digital display vernier caliper (with an accuracy of 0.02 mm), and leaf area was quantified using a leaf area meter (LI-3000C, USA). Leaves were then dried at 65 °C for at least 72 h to a constant weight and weighed. Leaf area and dry mass were then used to calculate leaf mass per unit area (LMA) (g m⁻²). Leaf tissue density (LTD) (g cm^-3) was calculated as leaf dry mass/(leaf area × leaf thickness). Total leaf C and N concentrations (mg g^-1) were determined using an elemental analyzer (Vario EL III, Elementar, Germany).

Root samples were collected according to the method described by Guo et al. (2008). The main root branch near the tree stem was partially excavated just enough to trace its lateral root clusters. Then root clusters with intact branch orders were cut from the main lateral woody roots by extracting an intact soil block (about 20×20×20 cm) and immediately transporting it to the laboratory for further morphological and chemical analyses. In the laboratory, after careful cleaning of adherent soil particles and organic matter, the root branching clusters were dissected into different branch orders following the description of Pregitzer et al. (2002), being kept moist with deionized water throughout. Here, we focused only on first-order roots. Root segments of each order were scanned by the Epson scanner, and the root diameter, length, and volume were measured using Win RHIZO (Pro 2009b). All scanned roots were then oven-dried at 65°C for 72 h and weighed to determine their root dry matter. SRL (m g^-1) was calculated as root length/root dry weight, and RTD (g cm^-3) was calculated as root dry weight/root volume. Total root C and N concentrations (mg g^-1) were determined using an elemental analyzer (Vario EL III, Elementar, Germany).

Species name and taxonomic nomenclature were standardized and corrected according to the Plant List (http://www.theplantlist.org) using the “plantlist” package (Zhang, 2018). The phylogenetic tree was constructed based on the comprehensive angiosperm species-level phylogeny of Zanne et al. (2014), updated by Qian and Jin (2016), using the “V.PhyloMaker” package (Jin and Qian, 2019). The mega-tree implemented in “V.PhyloMaker” (i.e. GBOTB. extended.tre), which included 74533 species and all families of extant vascular plants, was the largest dated phylogeny for vascular plants to date (Jin and Qian, 2019). All 60 species were included in this phylogenetic tree ( Supplementary Figure S1 ).

All data analyses in this study were conducted at the species level. Thus, species mean trait values are used throughout. All trait data was log10-transformed to meet the assumption of normality.

The mean value (Mean), extreme values (Min and Max), median value (Median), and coefficient of variation (CV) were calculated for each trait using the “stat.desc” function in the “pastecs” package. To assess the phylogenetic conservatism in both above-and belowground traits, we calculated phylogenetic signal in all traits by performing Pagel’s λ testing (Pagel, 1999). Pagel’s λ provides a reliable metric for discriminating between complex models of trait evolution (Münkemüller et al., 2012). Pagel’s λ ranges from 0 to 1. A larger value indicates greater phylogenetic conservatism for the given trait. Significance was tested by comparing standardized contrast variables to random values obtained by shuffling trait data across the tips of the tree 999 times. The significance test was performed using the “phylosig” procedure in the “phytools” R package.

We used the “principal” function in the “psych” package (Venables and Ripley, 2002) to perform principal component analysis (PCA) on leaf, first-order root, and whole plant traits (leaf and root traits combined). Phylogenetic principal component analysis (pPCA) was also performed using the “phyl.pca” function in the “phytools” package to identify the dominant dimensions of trait covariation (Revell, 2012). pPCA has been widely used in the determination of independent axes of functional specialization (e.g., Liu et al., 2015; Wang et al., 2018b; Montesinos-Navarro et al., 2020). Student’s t-tests were used to assess differences between growth forms (shrubs and trees). These tests were performed using the “ggpubr” package.

To evaluate the impacts of phylogenetic autocorrelation on trait relationships between leaves and roots, we compared generalized least squares (GLS) regressions with phylogenetic generalized least squares (PGLS) analyses. The latter method accounted for evolutionary association among species and yielded unbiased regression coefficients and significance levels (Revell, 2010). For phylogenetic analyses, we first assessed different phylogenetic correlation structures (Brownian, Martin’s, and Pagel’s), and then we selected the best method by comparing Akaike’s information criterion (AIC). These preliminary analyses showed that Pagel’s λ was the best phylogenetic correlation structure (lowest AIC, Supplementary Table S6 ). We thus applied this methodology, in which phylogenetic regression was performed with a phylogenetic tree whose internal branches were multiplied by λ, leaving the tip branches at their original length, in all tests of the trait relationships in our study (Revell, 2010). Here, λ was estimated with maximum likelihood using the “gls” function from the R package “nlme”.

All statistical analyses were performed in the R4.1.2 statistical platform (R Core Development Team, http://www.r-project.org/), and all results were visualized using the “ggplot2” package.

Among leaf traits in the subtropical evergreen broad-leaved forest, CV was highest for LT (61.8%) and lowest for LC (7.8%). Among root traits, SRL displayed the highest CV (75.1%) and RC showed the lowest (5.1%) ( Table 2 ). The CVs were generally higher for morphological traits than for chemical traits ( Table 2 ). Among leaf traits, LC (λ = 0.89, p < 0.001) exhibited significant, and LN (λ = 0.766, p = 0.058) displayed marginally significant phylogenetic signal, but no phylogenetic signal was found for any leaf morphological trait. In contrast, all root traits, except for RC, showed significant phylogenetic signals ( Table 2 ).

The first two principal components of both pPCA and PCA explained more than 70% of the total variation in leaf traits ( Figure 2A ; Supplementary Figure S2A ; Supplementary Table S2 ). The PC1 axis was strongly and positively correlated with LTD and negatively correlated with LT, thus reflecting an adaptive trade-off in leaf construction in response to light heterogeneity within the forest community ( Figure 2A ; Supplementary Figure S2A ). The PC2 axis correlated positively with both LN and negatively with LMA in pPCA, and the relationships had the opposite signs in PCA. These reflect the dimension of the leaf economics spectrum ( Figure 2A ; Supplementary Figure S2A ). The results of root functional trait pPCA and PCA both showed two dimensions of RES ( Figure 2B ; Supplementary Figure S2B ; Supplementary Table S3 ). The first axis was positively correlated with RN and negatively correlated with RTD, thus representing the root conservation gradient. The second PCA axis correlated positively with RD and negatively with SRL, thereby representing the root collaboration gradient ( Figure 2B ; Supplementary Figure S2B ). Shrubs were mainly distributed at the “do-it-yourself” end of the root collaboration gradient (mainly located on the high SRL side) ( Figure 2B ; Supplementary Figure S2B ). Trees and shrubs differed significantly in their PC2 axis scores ( Supplementary Figure S3D , p < 0.01), with shrubs having a significantly higher SRL than trees ( Supplementary Figure S4G , p < 0.05).

The pPCA and PCA analyses of leaf and root traits combined (10 traits in total) revealed that the first principal component represented the whole plant conservation gradient, with species exhibiting higher RN and LN at one end of the axis, and species displaying higher LMA, RC, and RTD positioned at the opposite end ( Figure 3 ; Supplementary Figure S5 ; Supplementary Tables S4 and S5 ). The LT-LTD trade-off axis and the root collaboration gradient (SRL-RD) were independent of the whole plant conservation gradient and were located in the second and third principal components, respectively ( Figure 3 ; Supplementary Figure S5 ; Supplementary Tables S4 and S5 ). The PCA and pPCA analyses of the six core leaf and root traits also showed that RN, LN, LMA, and RTD were predominantly aligned with the whole plant conservation gradient on the first axis, while the second axis represented the root collaboration gradient ( Supplementary Figure S6 ).

No correlation was found between any other pair of analogous leaf and root traits, that is, between LMA and SRL, LT and RD, LTD and RTD, or LC and RC ( Figures 4A–D , p > 0.05). On the same whole plant conservation gradient, LMA and RTD had a weakly positive correlation ( Supplementary Figure S7A ). However, there was a positive relationship between leaf and root N concentrations, which persisted even after adjusting for phylogenetic relatedness ( Figure 4E , p < 0.001).

The results of this study provide evidence for the presence of LES and a two-dimensional RES within the community of this subtropical evergreen broad-leaved forest ( Figure 1 ), which is in line with our first hypothesis. Additionally, a leaf construction trade-off axis, independent of both LES and RES, also emerged for LT-LTD ( Figure 5 ).

Our leaf economics axis composed of LMA and LN was consistent with the prevailing LES found in global scale studies (Díaz et al., 2016; Wright et al., 2004) and in many regional scale studies (Xiang et al., 2013; Li et al., 2015; Hu et al., 2021; Li et al., 2022). This trait variation reflects the trade-off between resource acquisition and conservation (Reich, 2014; Díaz et al., 2016). Acquisitive species possess a high LN and low LMA, exhibit a short leaf lifespan, and demonstrate rapid growth rates. However, species with high LMA, such as the Fagaceae and Lauraceae tree species dominant in this study, exhibit slow photosynthetic efficiency and a slow growth rate. Their strategies responding to environmental stress are typically characterized by resistance and tolerance, with defense structure and extended leaf lifespan (Wright et al., 2004; Poorter et al., 2009). The variance of LMA in the present study accounted for 15% of the total global variance ( Supplementary Table S8 ). Furthermore, the variance of LN in this current study was lower than that observed in any larger-scale studies ( Supplementary Table S8 ), which contributed to 24% of the total global variance ( Supplementary Table S8 ). This may be attributed to the effect that external environmental filtering had on species composition under similar climate and soil conditions, which resulted in narrowed variations of LMA and LN within the community.

LT and LTD are the two components of LMA, and both contribute to variation in LMA (Villar et al., 2013; Li et al., 2015). In the present study, the leaf construction axis of LT-LTD contributed more to the total variation in leaf traits than did the leaf economics axes (LMA-LN) ( Figure 2A , Supplementary Figure S2A ). This might be associated with the extremely significant negative correlation between LT and LTD within the community ( Supplementary Figure S7B ). Furthermore, the variances of LT and LTD in this current study was significantly higher than those observed in any larger-scale studies ( Supplementary Table S8 ). Such large variations in LT and LTD within a forest community may be attributed to niche differentiation among species in response to aboveground environmental heterogeneity (especially light availability) (Shen et al., 2019; Jin et al., 2024). Plants primarily adapt to their light environment by adjusting leaf morphology and structure (Coble and Cavaleri, 2014). LT, which is determined by the number and length of palisade cell layers and is predominantly influenced by light conditions, represents a leaf’s investment in capturing light. A thicker LT is correlated with a higher photosynthetic rate per unit area of the leaf (Niinemets, 2001). Conversely, LTD is determined by cell wall thickness, cell size, and intercellular air space (Niinemets, 1999). An increase in LTD is attributed to the thickening of the cell wall and to smaller and more tightly arranged cells, which reduce the conductivity of cell CO₂ and lower the photosynthetic potential of leaves per unit dry mass (Niinemets, 1999). LTD is also associated with leaf robustness and physical defense (Poorter et al., 2009; Méndez et al., 2019). Within a forest community, leaves in the upper part of the canopy generally exhibit thicker LT due to high light availability, which maximizes light utilization. Conversely, leaves in the lower part of the canopy display increased LTD as a result of low light availability, enabling them to acclimate to low-light conditions, to prolong leaf longevity, and to protect against insect herbivory. The study conducted by Jin et al. (2024) also showed a trade-off between LT and LTD in the Parashorea chinensis canopy. Such a trade-off facilitates the coexistence among species within a community as they adapt to light heterogeneity. However, this unique trade-off may not be evident on a larger scale. For instance, studies have shown that LT and LTD can vary independently at the regional scale, without exhibiting a trade-off relationship (Nadal et al., 2023). Consequently, the current study demonstrates that, while the external filtering effect of similar climatic and soil conditions results in narrowed LMA variation within a community, the heterogeneity of light availability leads to significant variation in and trade-offs between the two components of LMA (LT and LTD).

In the current study, the pPCA and PCA analyses of the five functional traits of first-order roots ( Figure 2B ; Supplementary Figure S2B ; Supplementary Table S3 ) showed a two-dimensional RES. The first dimension, which represented the root conservation gradient, consisted of RN, RC, and RTD. Tree species with low RN and high RTD are generally considered to conserve resources more efficiently (Weemstra et al., 2016). Conversely, species with high RN and low RTD tend to prioritize resource acquisition at the expense of longevity (McCormack et al., 2012). The second dimension, which represented the collaboration gradient ranging from “do-it-yourself” (with high SRL and low RD) to “outsourcing” (with low SRL and high RD) strategies (Bergmann et al., 2020). In the present study, the arbuscular mycorrhizal (AM) species represented by Lauraceae and Magnoliaceae were generally distributed on the high RD side. In contrast to previous studies (e.g., Bergmann et al., 2020), the Fagaceae species here were mainly located at the conservative end of the conservation gradient, and shrubs were mainly distributed at the “do-it-yourself” end of the collaboration gradient ( Figure 2B ; Supplementary Figure S2B ). The understory shrubs had lower photosynthetic rates under a lower light environment than the trees, so they could hardly afford the huge carbon cost associated with mycorrhizal symbiosis. It has been reported that mycorrhizal fungi consume up to 10% (Řezáčová et al., 2017; Řezáčová et al., 2018) and 20-50% of photosynthesis fixed C (Hobbie, 2006; Řezáčová et al., 2017; Hawkins et al., 2023) in AM and ectomycorrhizal mycorrhizae (EM) trees, respectively. Therefore, having a higher specific root length can significantly enhance soil exploration efficiency per unit carbon input and reduce reliance on mycorrhizae (Caplan et al., 2017).

Our results showed that most of the total variation (about 70%) in first-order root traits within the studied forest community could be explained by the two-dimensional RES, which was comparable to the findings of previous studies both at regional (de la Riva et al., 2021; Ding et al., 2020) and global scales (Bergmann et al., 2020; Weigelt et al., 2021; Carmona et al., 2021). This may be related to the fact that soil heterogeneity is high both at large and small scales, thus putting roots and their symbionts under similar evolutionary selection pressures at different scales. However, the relative importance of the two axes of RES in explaining total variation of root traits still differed across scales. At both the regional (de la Riva et al., 2021; Ding et al., 2020) and global scales (Bergmann et al., 2020; Weigelt et al., 2021; Carmona et al., 2021), the root collaboration gradient (SRL-RD) contributed more to total variation than did the root conservation gradient (RTD-RN), which contrasted our results within the forest community where the root conservation gradient predominated. This might be attributed to the fact that, at larger scales, the root collaboration gradient is primarily influenced by species sorting caused by environmental filtering, and can be narrowed within a forest community having similar climate and soil conditions; whereas resource heterogeneity or ecological niche differentiation may exert a greater influence on the root conservation gradient at a smaller scale within the community. Whether it was RD or SRL in the current study, the variance was much smaller compared to that of larger scales ( Supplementary Table S8 ). RD and SRL accounted for 15% and 10% of the total global variance, respectively ( Supplementary Table S8 ). Conversely, the variances of RTD and RN in the current study was each comparable to those of the first-order roots at the global scale (Ma et al., 2018), and accounted for 78% and 75% of the total global variance, respectively ( Supplementary Table S8 ).

In our study, although the conservation gradients of roots and leaves were correlated, the CV for each root trait was higher than that for each leaf trait on the conservation gradient. This may be attributed to variation in the different spatial scales at which the driving factors influenced root and leaf traits. The driving factors influencing the leaf conservation gradient (or LES) (e.g., air temperature, precipitation, etc.) usually vary on a large scale (Poorter et al., 2009), whereas the drivers for the root conservation gradient (e.g., soil factors) are usually very heterogeneous at small scales (Defrenne et al., 2019; Weemstra et al., 2021). Hence, within a small-scale forest community, variation may be much greater in root traits than in leaf traits due to minimal variability in climatic factors and substantial variability in the soil micro-environment. This could potentially weaken the root-leaf trait correlation at small scales (e.g., within a community).

This study found a correlation between the axes of the leaf economics spectrum and the root conservation gradient ( Figure 3 ; Supplementary Figure S5 ; Supplementary Tables S4 and S5 ). Besides, in PCA and pPCA analysis of the six core leaf-root trait pairs, the conservation gradient axis correlated highly with LN, RN, LMA, and RTD, and orthogonally to the root collaboration gradient axis represented by SRL and RD ( Supplementary Figure S6 ). These findings suggest that the conservation gradient was coordinated above and below ground and could be integrated into a whole plant conservation gradient. This finding is consistent with that of a previous global-scale study on woody plants (Weigelt et al., 2021). Thus, plant species at the “fast” end of the “whole plant conservation gradient” had both larger LN and RN, thus indicating the consistent nitrogen acquisition strategy employed by the plant’s organs. In contrast, the conservative, “slow” species are characterized by higher construction costs per unit leaf area (high LMA) and longer leaf longevity (Poorter et al., 2009), as well as by higher RTD and longer root longevity (Ryser, 1996) ( Supplementary Figure S7A ).

We found a positive correlation between LN and RN within the community of this subtropical evergreen broad-leaved forest, and it was not influenced by phylogeny ( Figure 4E ). This result is consistent with previous studies on woody plants, both at regional- and global-scales (Freschet et al., 2013; Valverde-Barrantes et al., 2015; Wang et al., 2017; Weigelt et al., 2021) and reflects the strong coordination between leaves and roots in nutrient capture and utilization, as well as in metabolism (Osunkoya et al., 2010). Plant tissue N concentration represents the metabolic activity of an organ to a certain extent (Reich et al., 2006). The high metabolic activity of roots requires high leaf photosynthesis to provide carbohydrates to roots. Conversely, the high photosynthesis of leaves requires high root metabolic activity to acquire necessary water and nutrients (Reich, 2014). This result echoes the notion that tissue nitrogen concentrations can highlight the whole plant trade-offs in intrinsic physiology and life history (Kerkhoff et al., 2006; Valverde-Barrantes et al., 2015).

In contrast to tissue nitrogen concentrations, our results showed zero or very weak correlation between analogous leaf and root morphological traits ( Figure 4 ), thus supporting our second hypothesis. Previous studies have also pointed out that morphological traits exhibit decoupling patterns between aboveground and belowground organs (Fortunel et al., 2012; Yang et al., 2021), or that the correlation between analogous above and belowground morphological traits disappears after phylogenetic correction (Wang et al., 2017). Furthermore, global-scale studies have demonstrated that the correlation between analogous leaf and root traits is much weaker in morphological than in chemical traits (Weigelt et al., 2021). The decoupling between leaf and root morphological traits allows plants to independently adapt their morphology in response to diverse above- and below-ground environmental filters, thereby increasing the multitude of ecological strategies (Freschet et al., 2013; Laughlin, 2014). This may facilitate species adaptations to diverse niche dimensions, promote species coexistence and enhance ecosystem stability (Laughlin, 2014; Weigelt et al., 2021).

The decoupling between aboveground and belowground morphological traits may be attributed to multiple causes. Firstly, there are disparities between above- and below-ground organs in functional diversity, distinct selective pressures and environmental constraints (Mommer and Weemstra, 2012; Laughlin, 2014; Wang et al., 2017; Liese et al., 2017). Leaf traits are primarily driven by the main functions of light and CO₂ capture (Poorter et al., 2009), and leaf growth and development are mainly limited by light availability and water supply. In contrast, roots are required to perform multiple functions such as water and nutrient uptake, which includes the uptake of different nutrient elements in different forms (e.g., inorganic vs. organic, free ion vs. mineral- or organic-bounded). The abiotic (e.g., soil nutrients, water, and soil structure) and biotic (soil microorganisms including mycorrhizal fungi and pathogens) stresses to which they are exposed are even more complex (Craine et al., 2005; Bardgett et al., 2014; Mccormack et al., 2015; Weemstra et al., 2016). In a forest community, leaf morphological traits may be more susceptible to microclimate conditions, such as light availability, while root morphological traits may be more sensitive to heterogeneity in soil nutrients and water (Yang et al., 2021). Therefore, different responses to the same growth environment can result in poor coordination between above- and belowground organs. Secondly, in contrast to leaves, roots can outsource the task of resource acquisition to beneficial soil microorganisms such as mycorrhizal fungi. This delegation can reduce the need to develop an acquisitive root, and the root systems of plants dependent on mycorrhizal fungi thereby undergo morphophysiological alteration (Makita et al., 2012). Thus, the coordination between aboveground and belowground morphologies may become decoupled under the influence of the root collaboration gradient. Additionally, Valverde-Barrantes et al. (2015) proposed that the decoupling between leaf-root morphological traits may be attributed to root traits having a stronger phylogenetic conservatism than leaf traits. Our results further suggested that root traits may be more phylogenetically structured than leaf traits ( Table 2 ).