The study site was located in the Lingfeng National Nature Reserve (52°15′–52°31′N, 122°41′–123°26′E) in the Greater Hinggan Mountains, NE China (Supplementary Figure 1). Part of the cold temperate zones, the region has a mean annual temperature of about −5°C and a mean annual precipitation of ca. 500 mm. The main soil types in this area include gelic cambosols, permagelic gleyosols, and orthic spodosols (Editorial Committee of Soil Geography in China, 2014). Within Lingfeng National Nature Reserve, as across the cold-temperate region, coniferous forests constitute the dominant vegetation in the terrestrial ecosystems. Forest swamps comprise the second largest vegetation type in the Greater Hinggan Mountains (Editorial Committee of Wetland Vegetation in China, 1999). The Emuer River flows across the Lingfeng National Nature Reserve, forming large areas of forest swamps in the riparian zone (Editorial Committee of Wetland Vegetation in China, 1999). We applied a space-for-time substitution approach and selected three forest swamps from across the riparian zone to represent a successional gradient (early, middle, and late stages) (Supplementary Figure 1). Although the accurate ages for the successional stages of forest swamps were not available, we used a rough estimation based on previous studies of spore and pollen records which was performed not far from our study region (<50 km) (Xia, 1996). The ages of chosen forest swamps ranged from the present day (early stage) to 1300 years old (late stage). The main drivers of forest swamp succession are the expansion of Sphagnum, decreases in soil pH and nutrient availability and changes in water supply (Rydin and Jeglum, 2013). The early successional stage (present day) was characterized by minerotrophic soils with a high soil pH and water-table level and dominant species such as Larix gmelini, Betula platyphylla, and Carex schmidtii (Hu et al., 2019a). The middle successional stage (hundreds of years) had intermediate soil nutrient availability and pH with a moderate water table level and an abundance of L. gmelini, Vaccinium spp., and mosses. The forest swamps representing the late successional stage (ca. 1300 years) had the lowest nutrient availability, soil pH and water-table level, and common species including L. gmelini, Ledum palustre, Vaccinium vitis-idaea, and Sphagnum (Editorial Committee of Wetland Vegetation in China, 1999; Hu et al., 2019a).

We performed all the field investigation and sampling during the peak growing season, between July 1 and 15, 2017. The three forest swamps we chose to represent succession across the riparian zone were located close to one another (<5 km apart). As described in section “Study Region,” there was a significant decrease in soil pH and nutrient availability along the successional gradient (Supplementary Figure 2; Rydin and Jeglum, 2013; Hu et al., 2019a). To determine the species composition, we randomly located five replicated plots (10 × 10 m) within each type of forest swamp (Supplementary Figure 1). Within each plot, we then recorded the percentage cover of the understory shrubs and herbs using two 2 × 2 m quadrats (corners) and five 1 × 1 m quadrats (center and corners), respectively. Based on percentage cover, we collected samples of the dominant and subordinate understory plants at each successional stage. Notably, species composition was determined at quadrat level and averaged into plot-level data, and the sampling and trait measurement was performed for species within each successional stage. In total, we sampled 19 vascular plant species (11, early; 5, middle; and 8, late), which together constituted more than 95% of the total cover per successional stage (Hu et al., 2019a). Both species identity and abundance of understory plants changed during forest swamp succession (see Supplementary Table 1 in Hu et al., 2019b for a species list).

For each species within each successional stage, three batches (replicates) of leaves were sampled; each batch was randomly collected from 5 to 10 adult individuals. Only leaves that were green, mature and without evidence of animal herbivory were collected. In addition, three batches of intact roots of each species were carefully collected at 0–20 cm soil depth. We randomly sampled roots from 3 to 15 adult individuals of each species at each successional stage. The roots of each species were excavated and identified based on the aboveground parts of plants. All sampled plant material was stored in sealed plastic bags, kept cool and measured for traits within 8 h of collection.

Five soil samples (0–20 cm) were randomly collected from each plot after clearing the litter layer on the soil surface. Approximately 10 g of each fresh soil sample was oven dried at 105°C for 6 h to determine soil water content (%). The other soil samples were air dried at room temperature (ca. 20°C) to constant weight and then analyzed for soil pH, soil total C, N, and P contents. Once dry, the soil samples were passed through a 2 mm sieve. To determine soil pH, 5 g of soil from each sample was shaken with 12.5 mL demineralized water in a glass beaker and left to stand for 30 min, after which pH was measured using a pH Meter (PB-10; Sartorius, Germany). Before measurement of soil C, N, and P contents, the soil samples were ground and passed through a 0.15 mm sieve. Soil total C and total N contents (mg g^–1) were measured using an elemental analyzer (vario MICRO cube; Elementar, Germany), whereas soil total P content (mg g^–1) was determined using inductively coupled plasma-optical emission spectrometry (ICP-OES Prodigy 7; Teledyne Leeman Labs, United States). Soil C:N ratio was calculated as the ratio of soil total C content to soil total N content.

For each species at each of the three successional stages, we measured six pairs of analogous leaf and fine root traits related to acquisition and conservation of resources (i.e., carbon, nutrient, and water). A subsample of each batch of leaves and fine roots were used to measure the traits. The fresh weight of each subsample varied due to the size of leaves and fine roots of different species, i.e., 0.2–3 g for leaves and 0.5–2 g for roots. First, we carefully cut the fine roots (diameter <2 mm) off each fresh root subsample and weighed them. We then scanned the fresh leaf and fine root samples using a Cannon scanner (resolution: 400 dpi) and measured total leaf area and fine root length from the scanned images using ImageJ¹ and WinRHIZO (PRO V.2004a; Regent Instruments, Canada), respectively. We also calculated root diameter (mm) using WinRHIZO. The fresh weights of the leaf and fine root samples were recorded, following which they were oven dried at 65°C for 72 h to determined their respective dry weights. Specific leaf area (SLA, mm² mg^–1) was calculated as the ratio of total leaf area to oven-dry weight, while specific root length (SRL, m g^–1) was calculated as root length per unit of oven-dry weight. Leaf dry matter content (LDMC, mg g^–1) and fine root dry matter content (RDMC, mg g^–1) were estimated using the ratio of leaf and root oven-dry weight to fresh weight, respectively. Then, leaf (LTD, g cm^–3) and fine root tissue density (RTD, g cm^–3) were respectively calculated as leaf and root dry mass divided by the volume of the tissues. Leaf volume was calculated as the product of leaf area and thickness, and root volume was estimated as the product of root diameter and length. Lastly, the concentrations of C, N, and P (mg g^–1) in the leaves and fine roots were determined using the same methods as for the measurement of soil C, N, and P contents.

To quantify community-level traits, we used the CWMs of trait values. For each of the five plots at each successional stage, the CWM for each trait was calculated as ∑i=1n(pi×traiti), where p_i is the relative abundance of species i and trait_i, the trait value of species i (Garnier et al., 2004). To quantify the functional diversity of understory plant communities, we first calculated functional dispersion (FDis) for each plot. We chose this index not only because it is a functional diversity index which can integrate both single and multiple traits, but also because it can include the relative abundance of species within a community (Villéger et al., 2008; Mouchet et al., 2010). The FDis index describes the average distance in multidimensional trait space of individual species to the centroid of all species (Laliberté and Legendre, 2010). Higher FDis values show higher variability in plant functional traits across species within communities after taking into account species abundance. We calculated abundance-weighted FDis in each plot for all leaf or fine root traits together, and separately for each individual trait. Because different plant traits have different units, trait values were standardized to zero mean and unit variance before calculating functional diversity. To better explore the successional changes in functional composition, we also calculated another two components of functional diversity, i.e., functional richness (FRic, the volume of trait space occupied by species within the community) and evenness (FEve, the regularity of the distribution of species abundance in the trait space) (Villéger et al., 2008; Mouchet et al., 2010). These two indices provided complementary information for functional diversity. Both CWMs and functional diversity were calculated using the dbFD function of the “FD” R package (Laliberté et al., 2014).

Overall, successional stage had significant effects on CWMs of all leaf and fine root traits (p < 0.05; Table 1). However, we observed contrasting changes in traits related to different resource strategies along the successional gradient (Figure 1). Traits related to resource acquisition, including SLA, leaf N, leaf P, SRL, root N, and root P, tended to decrease as succession progressed, although the CWMs of the middle and late successional stages were not significantly different (p > 0.05; Figures 1, 2). Being related to resource conservation, LDMC, leaf tissue density, leaf C, RDMC, and root C tended to increase along the successional gradient, although LDMC was not significantly different between the middle and late stages (p > 0.05; Figures 1, 2). In contrast, root tissue density increased significantly from early to middle stages but decreased from middle to late stages (Figure 1). These leaf and root traits concentrated mainly on axis 1 of the PCA, which accounted for 86.8% of the total variation (Figure 2). The PC1 of traits decreased significantly with increasing soil pH (R² = 0.84, p < 0.001; Figure 2). In addition, there were similar changes in leaf and fine root traits related to the same resource strategies along the successional gradient (Figures 1, 2).

According to the results of ANOVA, successional stage had significant influences on FDis of both leaf and fine root traits (p < 0.05; Table 1 and Figure 3). There was no significant interaction between successional stage and plant organ on FDis of multiple traits as well as individual traits except for C and P (p > 0.05; Table 1). For multiple traits, both leaf and root FDis tended to decrease along the forest swamp successional gradient, although leaf FDis of the early and middle stages were not significantly different from each other (p > 0.05; Figure 3). As for individual traits, FDis of all leaf and root traits, except for N, decreased similarly along the successional gradient (Figure 3). For both leaf and root N, plant communities at the middle successional stage had the highest FDis (Figure 3).

In contrast to FDis, both FRic and FEve of plant traits were highest for plant communities at the middle successional stage (Supplementary Figure 3). However, both FRic and FEve of leaf and root traits showed similar changes during forest swamp succession, which are consistent with the result of FDis (Supplementary Figure 3).

Soil properties changed during forest swamp succession (Supplementary Figure 2). Soil pH and soil N content showed significant positive relationships with FDis of both leaf and fine root traits (p < 0.05; Supplementary Table 1 and Figure 4). Even after controlling for other variables, soil pH was the best predictor of leaf and fine root FDis (Figure 5). The CV of soil factors varied along the successional gradient (Supplementary Figure 4), but the CV of soil N was not significantly correlated with FDis of leaf and root traits (p > 0.05; Supplementary Table 1 and Figure 4). On the other hand, plant species richness of forest swamps was significantly influenced by successional stage (F_2,12 = 15.61, p < 0.05). Species richness of the early successional stage was significantly higher than that of the other two stages (p < 0.05; Supplementary Figure 2). The contribution of species richness to leaf and root FDis in forest swamps was weak both before and after controlling for the effects of other variables (Supplementary Table 1 and Figure 5).

We found that CWM dry matter content, tissue density and C increased, and SLA, SRL, leaf, and root N and P decreased as the succession progressed in forest swamps. It has been largely known that dry matter content, tissue density and C involve resource conservation and physical defense and resistance to disturbances (Elger and Willby, 2003; Pérez-Harguindeguy et al., 2013), and that SLA, SRL, leaf, and root N and P are closely related to resource acquisition (Wright et al., 2004; Reich, 2014). Thus, those results support our first hypothesis, i.e., community-level resource-conservative traits tend to increase, and resource-acquisitive traits decrease with succession. These findings are consistent with the results of previous studies on primary succession in temperate rain forests (Holdaway et al., 2011), old-field succession (Shipley et al., 2006), and secondary succession in deciduous forests (Wilfahrt et al., 2014) and subtropical wet forests (Muscarella et al., 2016). All these studies concluded that plant communities at later successional stages tended to be more resource-conservative as light or nutrient availability decreased. In this study, the directional shifts in CWM plant traits with succession reflect the adaptation of plants to the decrease in soil pH and nutrient availability along the successional gradient.

Moreover, CWMs of analogous leaf and fine root traits showed similar trends along the successional gradient. This result is consistent with our expectation since leaf traits have been reported to be strongly coordinated with fine root traits at both community (Figure 2) and species level (Valverde-Barrantes et al., 2017; Hu et al., 2019a). It showed that there were similar plant strategy responses reflected in both the leaf and fine root traits. Notably, compared with species-level traits (Hu et al., 2019a), community-level plant traits showed more clear and predictable changes as succession progressed. This indicates the importance of integrating species relative abundance into studying and predicting plant community functional structure changes.

Small changes in community-level functional traits occurred between the middle and late successional stages, although these traits tended to show predictable shifts across all stages. There are two possible reasons for this result. One may be that these two later stages had a more similar species composition and denser shrub layer than the early stage (Hu et al., 2019a). Thus, forest swamps from the two later successional stages had more similar CWM traits. The other one may be related to the context or ecosystem differences. Different ecosystems could have different modes of succession. For example, Lohbeck et al. (2013) found contrasting successional changes in functional composition of dry and wet tropical forest, and Aiba et al. (2016) observed context-dependent changes in the functional composition of tree communities after land-use change along successional gradients. The factors including ecosystem type may increase the difficulty of predicting changes in plant functional composition during succession (Chang and Turner, 2019). Thus, future studies should focus more on examining plant functional traits across different ecosystems, in different contexts, and along environmental gradients, to improve the predictability of plant trait shifts in response to environmental changes (Bruelheide et al., 2018) and the understanding of plant-soil interactions (Holdaway et al., 2011).

Partially consistent with our expectation, multi-trait FDis of forest swamps decreased with succession, and this pattern coincided with the patterns of functional diversity of most individual traits, except for leaf and root N. These results were in agreement with the findings of Mason et al. (2012), but they are inconsistent with the findings of the majority of previous studies which reported that plant functional diversity tended to increase with succession (Lohbeck et al., 2012; Purschke et al., 2013; Zemunik et al., 2015; Craven et al., 2018), or showed no clear patterns (Böhnke et al., 2014; Laine et al., 2018). In contrast to FDis, FRic, and FEve based on leaf and root traits increased from the early to middle successional stage, then decreased from middle to late stage, indicating the dependence of functional diversity patterns on metrics. In our study, during forest swamp succession, some fast-growing and nutrient-acquisitive plant species (e.g., C. schmidtii, Ribes nigrum, and Rubus arcticus), characteristic of the early stage, were replaced by slow-growing and nutrient-conservative species (e.g., L. palustre and V. vitis-idaea) by the late successional stage (Hu et al., 2019a). The apparent decline in functional diversity during succession may reflect a convergence of plant community traits on resource-use strategies by the late stage which had lower soil pH, water and nutrient availability (Supplementary Figure 2). This finding indicates the important role of environmental filtering in limiting functional diversity during forest swamp succession, which is also supported by the strong explanation of soil pH and nutrient availability for functional diversity (Figures 4, 5). Inconsistent with our second expectation, FDis of leaf or root N was highest at the middle successional stage. That is because the most dominant species (i.e. Vaccinium uliginosum and V. vitis-idaea) at the middle stage had more contrasting leaf N (21.41 vs 10.99 mg g^–1) and root N (4.68 vs 6.21 mg g^–1) than dominant species (L. palustre and V. vitis-idaea) at the late stage (leaf N, 14.68 vs 9.10 mg g^–1; root N, 3.95 vs 5.07 mg g^–1). Overall, plant functional diversity showed directional changes during forest swamp succession. This study, focusing on a less studied ecosystem (forest swamp) and a different driver of ecological succession (soil pH), provides new evidence for the constraint of environmental filtering on plant functional diversity.

Similar successional patterns in functional diversity of leaf and fine root traits reflect the coupling of above- and belowground plant functional diversity. This is consistent with the coordination between community-level leaf and fine root traits found in this study. Moreover, functional diversity of leaves and fine roots also showed similar responses to soil properties (Figures 4, 5). As soil pH and nutrient availability decreased during succession, leaf and fine root functional diversity tended to be similarly affected by these soil properties. Our results support the idea that functional diversity of root traits can be predicted by that of leaf traits just as in many case species root traits could be inferred from leaf traits based on the strong correlations between them (Freschet et al., 2010; Liu et al., 2010; Valverde-Barrantes et al., 2017). The findings of congruent patterns for leaf and root functional diversity are inconsistent with the few studies which investigated both above- and belowground plant traits (Mason et al., 2012; Zemunik et al., 2015; de la Riva et al., 2018). Mason et al. (2012) found that functional diversity of aboveground plant traits decreased along a long-term soil chronosequence, and Zemunik et al. (2015) observed functional diversity of belowground traits related to nutrient acquisition increase along the same chronosequence. These different successional patterns for leaf and root traits may be related to the different calculation of functional diversity in these two studies. Meanwhile, de la Riva et al. (2018) found incongruent patterns of leaf and root morphological trait diversity along an arid gradient in Mediterranean woody communities. The different results for economic traits from this study and morphological traits from de la Riva et al. (2018) clearly showed that the patterns of leaf and root functional diversity could also depend on the trait dimension. As root traits have been found to be multidimensional (Marañón et al., 2020), it would be important to examine the coordination of leaf and root functional diversity based on other suites of traits, e.g., regenerative traits.

While very few of previous studies have investigated the successional shifts in functional diversity of belowground traits, our results found that leaf and root functional diversity decreased similarly during forest swamp succession. This finding has important implications for plant community assembly and ecosystem functions. First, similar trends in leaf and fine root functional diversity across succession suggest that assembly processes may operate similarly for above- and belowground compartments, at least for the resource economic dimension. Second, our finding would also aid the understanding of successional changes in ecosystem functions, because functional diversity of economic traits has been shown to determine many important ecosystem functions, e.g., productivity, carbon fluxes, and storage (Zuo et al., 2016; De Long et al., 2019).