Plasticity in leaf and stem nutrient resorption proficiency potentially reinforces plant–soil feedbacks and microscale heterogeneity in a semi-arid grassland


  • Xiao-Tao Lü,

    Corresponding author
    1. State Key Laboratory of Forest and Soil Ecology, Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110164, China
      Correspondence author. E-mail:
    Search for more papers by this author
  • Grégoire T. Freschet,

    1. Department of Systems Ecology, Faculty of Earth and Life Sciences, VU University, De Boelelaan 1085, 1081 HV Amsterdam, The Netherlands
    Search for more papers by this author
  • Dan F. B. Flynn,

    1. Department of Ecology, Evolution, and Environmental Biology, Columbia University, 1200 Amsterdam Avenue, 10F Schermerhorn Extension, New York, NY 10027, USA
    Search for more papers by this author
  • Xing-Guo Han

    1. State Key Laboratory of Forest and Soil Ecology, Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110164, China
    Search for more papers by this author

Correspondence author. E-mail:


1. The potential resorption of substantial amounts of nutrients from all vegetative organs of plants has large implications for the plant nutrient economy and for biogeochemical cycles. So far, most studies have focused on leaf nutrient resorption only. Besides, while evidence is growing that soil fertility changes impact on leaf nutrient resorption at a large spatial scale, hardly anything is known of such coupling at a small spatial scale.

2. Here we show that nitrogen (N) in culms of four dominant grasses of northern Chinese steppes contributed from 17% to 36% to the total pool of N resorbed from above-ground senescing parts and accounted for 25–52% of above-ground litter N. These results demonstrate the tremendous importance of non-leaf organs for plant nutrient economy and ecosystem nutrient cycling.

3. More importantly, we found that even microscale variations in resource availability (soil inorganic N; soil moisture) can strongly impact on both leaf and culm N resorption proficiencies (RP) and absolute leaf N resorption of grasses. Moreover, plasticity was responsible for 86% and 43% of within-site variance in leaf and culm RP, respectively, with the remainder owing to interspecific differences between the four grasses. These results imply a much larger role of plant plasticity in driving ecosystem functioning than previously assumed.

4.Synthesis. Our results suggest that plant litter quality varies even at the microscale with heterogeneity in soil resource availability, thereby potentially feeding back on soil properties and sustaining microscale soil fertility patchiness. In parallel, plants of more fertile patches resorbed a greater absolute amount of N, likely benefiting their competitive and reproductive abilities.


Nutrient resorption from senescing tissues, one of the most important mechanisms of nutrient conservation, enables plants to re-use nutrients and thereby be less dependent on environmental nutrient supply (Killingbeck 1996; Aerts & Chapin 2000). Among seed plants an estimated 50% of total leaf N and phosphorus (P) is resorbed during leaf senescence on average (Aerts 1996; Yuan & Chen 2009), with the remainder routed to the decomposition system via litter fall. These resorption rates are nevertheless highly variable across ecosystems and species (Yuan & Chen 2009). At the ecosystem scale, this process strongly controls carbon and nutrient cycling (Kozovits et al. 2007). At the individual plant level, it also has important implications for plant growth, reproduction and competitive ability (May & Killingbeck 1992; Aerts 1999), most particularly in nutrient-poor ecosystems, where the annual nutrient requirement of perennials is strongly dependent on resorption from senescing tissues (Aerts et al. 2007).

Besides leaves, nutrients can also be resorbed from other plant organs and tissues, such as fine stems (Aerts & de Caluwe 1989; Freschet et al. 2010), fine roots (Kunkle, Walters & Kobe 2009; Freschet et al. 2010) and heartwood (Andrews & Siccama 1995). Among these, photosynthetic stems such as culms of grasses have the highest nutrient resorption capacities, with N and P resorption efficiencies (RE) ranging from 45% to 80% and 55% to 91%, respectively (Aerts & de Caluwe 1989; Freschet et al. 2010). Moreover, culms can represent over 40% of yearly above-ground biomass production of dominant herbaceous species in temperate grasslands (Ren, Zheng & Bai 2009). Their contribution to the whole-plant nutrient economy and to carbon and nutrient cycling in ecosystems is thus potentially substantial (Pahkala & Pihala 2000). Yet, to date, stem nutrient resorption and its implications have been mostly overlooked.

At the individual plant level, leaf resorption is usually higher than fine stem or fine root resorption (Freschet et al. 2010). Consequently, considering the high nutrient concentrations and high turnover rates of leaves, Freschet et al. (2010) suggested that the total amounts of nutrients resorbed from leaves should generally be higher than those from stems and roots. However, no study has so far quantitatively compared the respective impacts of resorption from distinct organs on the whole-plant nutrient budget. Besides, several factors could act against these predictions. In response to competition for light, plants may allocate substantial biomass to stems at the expense of leaves (Tilman 1988). Similarly, in nutrient-poor and water-limited environments, plants may increase root biomass allocation relative to shoot allocation (Tilman 1988). Finally, while the nutrient concentrations in photosynthetic stems of herbaceous plants are always lower than those in leaves owing to their supportive role, the former can still be substantial (Berendse, Oudhof & Bol 1987). Acknowledging the respective importance of leaf and stem resorption in the whole-plant nutrient budget (and thus whole-plant economic strategy) is thus of primary importance, particularly in herb-dominated ecosystems.

As a major mechanism of nutrient conservation, nutrient resorption has been hypothesized to vary across nutrient availability gradients (e.g. Eckstein, Karlsson & Weih 1999). Despite contrasting results for leaf nutrient resorption efficiency (Aerts 1996; Kobe, Lepczyk & Iyer 2005), there is evidence that the nutrient concentration of leaf litter often decreases (i.e. resorption proficiency (RP) increases; Killingbeck 1996) in parallel to soil nutrient availability both between (Wright & Westoby 2003) and within species (Richardson et al. 2005; Norris & Reich 2009). As most studies have compared nutrient resorption across sites with contrasting soil fertility, it is yet unknown whether this pattern occurs at finer scale, that is, between coexisting individuals of a single ecosystem, following fine-scale heterogeneity in soil fertility. Indeed, most ecosystems exhibit some degree of patchiness in their soil fertility conditions (Ettema & Wardle 2002). If such sensitive responses of plants to nutrient availability exist, this would indicate micro-scale changes in the strength of plant–soil feedbacks (e.g. Berendse 1994; Aerts 1999), which could drive increased ecosystem heterogeneity and higher biological diversity (Grime 1998; Bardgett 2002), with potentially large consequences for ecosystem functioning (Wardle et al. 2004).

Here we test the hypotheses that culms can contribute as much as leaves (i) to the total amount of N resorbed from the shoots of senescing grasses and (ii) to the total amount of N lost by grasses via litter and that even (iii) micro-scale variations in soil fertility can impact on both leaf and culm N resorption. To test these hypotheses, we linked within-site variation in soil resource availability (soil inorganic N, soil moisture) to intraspecific variability on N resorption (RE, RP) from leaves and culms of the four grasses that dominate a temperate steppe in northern China.

Materials and methods

Study Site and Plant Sampling

This study was conducted in 2007 at a long-term experimental site (43°33′ N, 116°40′ E, 1250 m a.s.l.) fenced since 1999, managed by the Inner Mongolia Grassland Ecosystem Research Station (IMGERS) of the Chinese Ecological Research Network (CERN). Averaged over the 1982–2005 period, mean annual temperature in this area was 0.3 °C with mean monthly temperatures ranging from −21.6 °C in January to 19.0 °C in July. Annual precipitation was 346 mm with 60–80% falling during the growing season, i.e. from May to August. The soil is Calcic-orthic Aridisol (US Soil Taxonomy) with 48.6% sand, 26.1% silt and 25.3% clay (top 10 cm) with clear N limitation of plant growth (Bai et al. 2010). Mean soil bulk density was 1.3 g cm−3, and soil pH was 7.1. The peak above-ground biomass reached 100–150 g dry matter m−2 in mid-August (Bai et al. 2010). The studied vegetation, defined as Leymus chinensisStipa grandis community, is widely distributed across grassland areas of the Eurasian steppe region. It corresponds to a transition between grazed and long-term (c. 30 year) ungrazed states, which, in term of vegetation biomass and topsoil properties, are characterized by large homogeneous patches and heterogeneous mosaic of small patches, respectively (Steffens et al. 2009). The four most dominant grasses of the community were included in this study: Leymus chinensis (Trin.) Tzvel., S. grandis P. Smirn., Achnatherum sibiricum (Linn.) Keng. and Agropyron cristatum (Linn.) Gaertn. They represent c. 70% of the total plant community biomass (Bai et al. 2010).

Between 10 and 12 August, at full maturity of grasses and before the onset of senescence, 30 quadrats of 1 m2 were randomly laid out within a 10 × 10 m plot, which represented overall plant community structure, biomass and species composition in the studied grassland ecosystem. This setup did not include larger-scale replication as we focused here on capturing heterogeneity in plant and soil parameters at the plant scale rather than providing an integrative picture of ecosystem features and heterogeneity. In each quadrat, eight shoots of L. chinensis with similar stature were randomly selected. Four of them were harvested for the purpose of biomass and [N] measurements of mature shoots (two whole shoots) and leaves and culms (two shoots; two fully expanded leaves per shoot, usually the third and forth visible leaves from the top of each shoot; one culm per shoot). For leaves, only blades were included, while both sheaths and stems were included to represent the culms. The four other shoots were tagged with plastic labels. Among those four, two random shoots were labelled further with small pieces of red plastic foil on two fully expanded leaves per shoot. All tagged organs were harvested after complete senescence on 15 October, in the same way as the green organs. Because similar cohorts of plant material were sampled at full maturity and after complete senescence, we avoided biases resulting from mass loss and area shrinkage occurring during plant senescence (van Heerwaarden, Toet & Aerts 2003a; Aerts et al. 2007). The same procedure was followed for the three other grass species, although on 15 of the 30 quadrats only. To minimize effects of variation in culm size, 10-cm (for L. chinensis, A. cristatum and A. sibiricum) and 5-cm (for S. grandis) long pieces of the basal parts of culms were used.

Organ Biomass and N Resorption Measurements

To measure the biomass of shoot, leaf and culm for both mature and senesced grasses, the two whole shoots and all the leaves (not only the third and forth one) and culms (not only the basal parts) of the two other shoots were oven dried at 65 °C for 48 h and weighed separately. This procedure was repeated on each quadrat separately.

For N resorption calculations, only the cohorts of tagged senesced materials were used (four fully expanded leaves and two culm basal parts), along with their mature counterparts. All plant samples (previously oven dried for biomass quantification) were weighed and the average masses per leaf, culm and whole shoot were calculated. Each sample was then ground with a ball mill (Retsch MM 400; Retsch GmbH & Co KG, Haan, Germany), and a 0.1-g subsample was digested in H2SO4-H2O2 following Bennett, Judd & Adams (2002). Then, [N] of each sample was determined by modified Kjeldahl wet digestion (Gallaher, Weldon & Boswell 1976) using a 2300 Kjektec Analyzer Unit (FOSS, Höganäs, Sweden) and expressed in mg g−1. The total N pool of each individual organ (for both mature and senesced material) was calculated as the product of the individual organ weight (g) and its [N] (mg g−1) and expressed in mg. Following van Heerwaarden, Toet & Aerts (2003b) and Aerts et al. (2007), N RE (%) was calculated as the ratio of the difference in organ N pool between mature and senesced organ to mature organ N pool. Nitrogen RP was defined as the extent to which the nutrient content was reduced in senesced material, that is, the senesced tissue [N] (mg g−1), where low senesced tissue [N] corresponds to high proficiency (Killingbeck 1996).

For each quadrat, the amount of N resorbed from leaves and culm (per individual shoot) was calculated as the product of the total mature organ N pool (total mature organ weight × organ [N]) and the organ N RE (%) and expressed in mg. As leaves and culms were sampled from the same individual, the amount of N resorbed from the whole shoot was obtained by adding up the total amount of N resorbed from both leaves and culms. The amount of N left in senesced organs was calculated as the product of total senesced organ weight and their respective [N]. The total amount of N left in each senesced shoot was obtained by adding up the total amount of N left in leaves and culms.

Soil Resource Availability

We estimated soil resource availability by measuring two variables: total soil inorganic [N] and soil moisture content. In early August, five soil cores (topsoil, 0–10 cm) were sampled in each of the 30 quadrats with a metal tube (5 cm Ø) and pooled to obtain one composite soil sample per quadrat. To measure soil water content, a soil subsample from each quadrat was weighed before and after being oven dried at 105 °C for 48 h. Soil inorganic [N] was determined by extracting 10 g of fresh, root-free soil with 50 mL 2 M KCl. The soil–extractant mixture was shaken for 1 h and then filtered (Whatman No.1 filter paper). [NH4+] was measured with the salicylate method while [NO3] was measured using the cadmium reduction method on a FIAstar 5000 Analyzer (Foss Tecator, Hillerød, Denmark). Total soil inorganic [N] was the sum of [NH4+] and [NO3] and expressed in mg kg−1 soil. While soil inorganic [N] and moisture status were measured only once, before the onset of senescence, and therefore do not capture temporal fluctuations in resource availability, a previous study conducted at the same experimental site showed a positive relationship between soil moisture and soil net N mineralization rates (Wang et al. 2006), indicating a strong potential for those two variables to reflect both water and nutrient availability at this site.

Data Analysis

Data were tested for normality using Levene’s test. All non-normal data were log transformed prior to analysis to comply with the normality assumption of parametric tests. Paired t-tests were used to test the intraspecific differences between leaf and culm green tissue [N], senesced tissue [N], N RE and litter biomass as well as the interspecific differences in green tissue [N], senesced tissue [N], N RE and litter biomass of the four grasses (Table 1). As most of the P-values in the t-tests were <0.001, the risk of Type I error was negligible. Differences between leaf and culm contributions to the total amount of N resorbed from species whole shoots (Fig. 1a) and to the total amount of N left in the whole-shoot litter (Fig. 1b) were assessed using paired t-tests. Single and multiple linear regressions were used to test the predictive power of soil inorganic N and water content (independently and in combination) on green and senesced leaf [N], green and senesced culm [N], leaf N RE and culm N RE, respectively (Fig. 2a,b). All these analyses were carried out with spss (Version 13.0; SPSS Inc., Chicago, IL, USA). Additionally, we used Permutational Multivariate Analyses of Variance (‘Adonis’ function in ‘vegan’ package in r; Anderson 2001) to partition intraspecific vs. interspecific variance in leaf and culm litter [N], respectively (de Bello et al. 2011).

Table 1.   Intra- and interspecific differences in N content, N resorption efficiencies, and biomass of leaf and culm from four dominant grasses of Chinese temperate steppes. P-values are from paired t-tests. n = 30 for Leymus chinensis, n = 15 for Stipa grandis, Agropyron cristatum and Achnatherum sibiricum
SpeciesValues ± SEIntraspecific organ difference (P-value)Interspecific difference (P-value)
Green tissue N content (mg g−1)
 Leymus chinensis24.9 ± 0.510.5 ± 0.3<0.001<0.001<0.001
 Stipa grandis16.9 ± 0.78.9 ± 0.3<0.001
 Agropyron cristatum22.0 ± 0.611.2 ± 0.4<0.001
 Achnatherum sibiricum20.8 ± 0.411.0 ± 0.3<0.001
Senesced tissue N content (mg g−1)
 Leymus chinensis7.2 ± 0.24.1 ± 0.1<0.0010.013<0.001
 Stipa grandis7.6 ± 0.35.5 ± 0.2<0.001
 Agropyron cristatum8.1 ± 0.36.3 ± 0.3<0.001
 Achnatherum sibiricum8.0 ± 0.26.0 ± 0.3<0.001
Resorption efficiency (%)
 Leymus chinensis71.1 ± 0.860.4 ± 1.5<0.001<0.001<0.001
 Stipa grandis54.6 ± 1.037.2 ± 2.6<0.001
 Agropyron cristatum63.3 ± 0.942.1 ± 3.5<0.001
Achnatherum sibiricum61.3 ± 0.845.4 ± 2.2<0.001
Senesced tissue biomass (g)
 Leymus chinensis4.0 ± 0.12.4 ± 0.1<0.001<0.001<0.001
 Stipa grandis5.1 ± 0.43.9 ± 0.40.002
 Agropyron cristatum2.3 ± 0.13.2 ± 0.20.001
 Achnatherum sibiricum3.9 ± 0.24.0 ± 0.20.727
Figure 1.

 Relative contributions of (a) N resorbed from leaves vs. culms to the total pool of N resorbed from grass shoots and (b) N remaining in senesced leaves vs. culms to the total pool of N left in above-ground litter. Black columns represent leaves and grey ones represent culms. Standard errors are shown one sided. Significance of differences in leaf and culm contribution is displayed with P-value.

Figure 2.

 Predictive power of soil inorganic N and soil water contents on (a) leaf and (b) culm green N content, senesced N content and N resorption efficiency of four grasses. Symbols: inline imageLeymus chinensis (L.c.), inline imageStipa grandis (S.g.), inline imageAgropyron cristatum (A.c.) and inline imageAchnatherum sibiricum (A.s.). Only significant regressions are displayed.


Relative Contributions of Leaves and Culms to the Above-Ground Pool of N Recycled and N Returned to the Soil

Across all four grasses, N contents of green leaves were significantly higher than those of green culms (Table 1). Leaf N RE were also significantly higher than culm N RE. In contrast, N RP were stronger in culms than in leaves. While senesced leaf biomass was significantly higher than culm biomass in L. chinensis and S. grandis, culm biomass was larger in A. cristatum and there was no significant difference in A. sibiricum (Table 1). Consequently, leaves contributed significantly more (from 64% to 83%, depending on the species) than culms (from 17% to 36%) to the total amount of N resorbed from the whole shoot, for all four species (Fig. 1a). Leaves of three species of four contributed significantly more to the total shoot litter N than the respective species culms (Fig. 1b), whereas culms of A. cristatum contributed slightly more than leaves to the whole-shoot litter N pool.

Relationships between Soil Resource Availability, Shoot Nutrient Status and Shoot Nutrient Resorption

Soil inorganic N and water contents varied between the 30 plots from 3.7 up to 6.6 mg kg−1, and from 5.8% to 7.1%, respectively. Across the four grasses, intraspecific variation in green and senesced leaf N contents were all positively related to both soil inorganic N and water contents (N RP decreases when soil resource availability increases; Fig. 2a). For culms, only A. cristatum displayed a positive relationship between green culm N content and soil inorganic N and water content, while both L. chinensis and S. grandis showed a positive relationship between senesced culm N content and soil inorganic N and water contents (N RP decreases when soil resource availability increases; Fig. 2b). In contrast, leaf and culm N RE were generally unrelated to soil inorganic N and water contents. Only the N RE of A. cristatum leaves was negatively related with soil water content (Fig. 2a), while the N RE of L. chinensis culms was weakly related with soil inorganic [N] (Fig. 2b). As soil inorganic N and soil water contents were tightly correlated (R2 = 0.82), combining them in multivariate regressions only slightly improved the predictions of green [N], senesced [N] and N RE of leaves and culms (data not shown). Intraspecific variance in leaf litter [N] (86% explained variance) was six times higher than interspecific variance (14%). Intraspecific variance in culm litter [N] (57% explained variance) was only slightly higher than interspecific variance (43%). Species plasticity was thus a generally stronger driver of RP than species identity.


This study is the first to quantify the relative value of stem nutrient resorption for the nutrient budget of grasses as well as its importance for above-ground litter quality in grasslands. We also reveal here a new potential mechanism by which plasticity in resorption can impact on topsoil fertility heterogeneity and feedback on plant nutrient economics.

Relative Contribution of Leaf and Stem N Resorption to Shoot Nutrient Budget and Total Shoot Litter N

Despite having generally lower biomass and N RE than leaves, culms of the four dominant grasses accounted for 20–40% of the total N pool resorbed from the shoot during autumn senescence. Thus, while the contribution of stems to herbaceous plant nutrient recycling can be substantial, it does not reach those of leaves in the studied grassland, contradicting our first hypothesis. Nevertheless, the average stem N RE of 49% measured here represents the lower end of the range of N RE measured by previous studies on graminoid culms such as Molinia caerulea (77%; Aerts & de Caluwe 1989), Carex rostrata, Deschampsia cespitosa and Deschampsia flexuosa (45%, 80% and 79%, respectively; Freschet et al. 2010). The contribution of grass culms to the whole-plant nutrient budget might thus be potentially higher for other grasses of distinct ecosystems. In every instance, stem nutrient resorption is of substantial importance for plant nutrient budget estimation.

In support of our second hypothesis, the culms of A. cristatum contributed to more than half of the above-ground litter N returned to the soil. More generally, across the four dominant grasses of our ecosystem, culms represented an average 45% of the shoot biomass and 36% of the shoot N returned to the soil. These results indicate that culms have a strong impact on both the amount and quality of litter entering the decomposition system of grasslands. As litter quality is the predominant driver of litter decomposition rates (Cornwell et al. 2008), herbaceous stem litter, with lower N content and more structural compounds, should have slower decomposition rates than leaves (G. T. Freschet, R. Aerts & J. H. C. Cornelissen, unpubl. data). The similar amount of culms and leaves being shed as litter, which implies the coexistence of both relatively high- and low-quality litter, should thus have a strong influence on the temporal pattern of nutrient release and thus nutrient availability to plants (e.g. Liu et al. 2010).

Substantial Variation in N Resorption at Micro-Spatial Scale

In support for our third hypothesis, this study revealed large variations in shoot RP at very fine spatial scales (100 m2), largely explained by coordinated variation in both soil inorganic [N] and soil moisture. The strong negative response of leaf RP (all four herbaceous species) and culm RP (two of four species), i.e. positive response of senesced leaf and culm [N], to a relatively small increase in soil fertility parameters suggests a tight control of soil fertility on shoot RP. Several studies have revealed substantial variation in nutrient RP at coarser spatial scales (Kobe, Lepczyk & Iyer 2005; Richardson et al. 2005; Norris & Reich 2009) or following fertilization treatments (van Heerwaarden, Toet & Aerts 2003b; Kozovits et al. 2007; Lü & Han 2010) owing to changes in soil nutrient availability. However, we show here for the first time that strong fertility-driven variations in RP exist between co-occurring individuals of the same species at very fine spatial scales. Moreover, intraspecific variance in senesced leaf [N] was six times greater than the interspecific variance of our four dominant grass species, suggesting that soil fertility may have a much stronger influence than species identity on above-ground RP in grassland ecosystems. Previous studies have demonstrated that an increase in soil fertility leads to the return of higher-quality litter to the soil (via higher N content; Aerts 1997), with potentially positive feedback on soil nutrient status and nutrient availability for the same plants (Aerts 1999; Kozovits et al. 2007). We extend this by demonstrating that such positive feedback may operate at the scale of small soil patches with the same plants producing litter and deriving nutrition from them. Thus, in the absence of disturbances, micro-scale differences in soil fertility are likely to be reinforced through time and promote micro-scale soil fertility patchiness. This is consistent with the observation made in the same grassland, where our plot represents a transition between grazed and long-term ungrazed states, that grazing cessation initiates the heterogenization of topsoil properties (e.g. bulk density, organic matter, total N) and vegetation biomass (Steffens et al. 2009). Besides, along a gradient from highly grazed to ungrazed plots, Steffens et al. (2009) showed that changes in topsoil properties were lagging behind changes in vegetation biomass, suggesting thereby that the plant component was an important factor initiating the recovery of ecosystems after disturbance.

Additionally, our results indicate that, while soil fertility shows little impact on leaf N RE, it drives substantial increase in green leaf N content, which implies larger absolute N resorption at higher fertility. This stronger autumn nutrient retranslocation can be beneficial to plants through competitive and/or reproductive functions (May & Killingbeck 1992). Two parallel feedback mechanisms can thus potentially occur between soil fertility and plant economics at the scale of individual plants, with large consequences for ecosystem functioning. For instance, heterogeneity in soil and plant properties resulting from the coexistence of these positive and negative plant–soil feedbacks is likely to reinforce soil biological diversity (Bardgett 2002) and can potentially create opportunities for settlement or development of otherwise subordinates or transient plant species (Grime 1998; Steffens et al. 2009). However, such progressive shifts in ecosystem properties are likely to be overruled by continuous grazing (Schönbach et al. 2011), a common disturbance in grassland ecosystem, which reduces litter input, creates homogeneous soil compaction and increases wind erosion (Steffens et al. 2009). On the other hand, fire, which also frequently occurs in (semi-arid) grasslands, may reinforce these nutrient resorption–soil fertility feedbacks by stimulating to a greater extent the organic matter mineralization in richer and wetter soils than in poorer and dryer ones and therefore favour soil nutrient availability and grass regrowth in more fertile patches (Xu & Wan 2008).


We thank Wu Wei-Jun, Kong De-Liang, Cui Qiang and Dong Ning for help in field work, Li Qiang and Li Li for assistance with nutrient analysis and Hans Cornelissen and Rien Aerts, as well as two anonymous reviewers, for providing helpful comments on the manuscript. This work was supported by The Knowledge Innovation Project of CAS (No. KZCX2-YW-T06), State Key Basic Research Development Program of China (2007CB106801) and the National Natural Science Foundation of China (30830026 and 30821062).