Legume Shrubs Are More Nitrogen-Homeostatic than Non-legume Shrubs

Legumes are characterized as keeping stable nutrient supply under nutrient-limited conditions. However, few studies examined the legumes' stoichiometric advantages over other plants across various taxa in natural ecosystems. We explored differences in nitrogen (N) and phosphorus (P) stoichiometry of different tissue types (leaf, stem, and root) between N2-fixing legume shrubs and non-N2-fixing shrubs from 299 broadleaved deciduous shrubland sites in northern China. After excluding effects of taxonomy and environmental variables, these two functional groups differed considerably in nutrient regulation. N concentrations and N:P ratios were higher in legume shrubs than in non-N2-fixing shrubs. N concentrations were positively correlated between the plants and soil for non-N2-fixing shrubs, but not for legume shrubs, indicating a stronger stoichiometric homeostasis in legume shrubs than in non-N2-fixing shrubs. N concentrations were positively correlated among three tissue types for non-N2-fixing shrubs, but not between leaves and non-leaf tissues for legume shrubs, demonstrating that N concentrations were more dependent among tissues for non-N2-fixing shrubs than for legume shrubs. N and P concentrations were correlated within all tissues for both functional groups, but the regression slopes were flatter for legume shrubs than non-N2-fixing shrubs, implying that legume shrubs were more P limited than non-N2-fixing shrubs. These results address significant differences in stoichiometry between legume shrubs and non-N2-fixing shrubs, and indicate the influence of symbiotic nitrogen fixation (SNF) on plant stoichiometry. Overall, N2-fixing legume shrubs are higher and more stoichiometrically homeostatic in N concentrations. However, due to excess uptake of N, legumes may suffer from potential P limitation. With their N advantage, legume shrubs could be good nurse plants in restoration sites with degraded soil, but their P supply should be taken care of during management according to our results.


INTRODUCTION
Symbiotic nitrogen fixation (SNF) is a strategy of plants to acquire nitrogen (N) from the atmosphere, i.e., a trait shared by a large number of species in the Fabaceae family (legumes). SNF broadens potential N resources and generally increases N absorption, resulting in higher N concentration ([N]) and nitrogen to phosphorus (P) ratio (N:P) in N 2 -fixing legumes than non-N 2 -fixing plants (Güsewell et al., 2003). For example, previous studies in the tropics have reported higher leaf [N] in legumes than in other plants (Townsend et al., 2007;Nasto et al., 2014;Bhaskar et al., 2016), which could be related to higher photosynthetic capacity and water use efficiency (McKey, 1994;Adams et al., 2016). SNF also helps legumes to keep stable [N] in N-limited soils (Lambers et al., 2008;Hobbie, 2015). Therefore, N 2 -fixing legumes can be more abundant in N-limited habitats with less competitive exclusion from non-N 2 -fixing plants (Rastetter et al., 2001;Menge et al., 2008). Such an ability of plants to maintain their nutrient composition despite nutrient variation in their resource supplies was regarded as stoichiometric homeostasis (Sterner and Elser, 2002;Elser et al., 2010). In this sense, N 2 -fixing legumes can be more N-homeostatic than non-N 2 -fixing plants.
Because of sufficient N and stronger homeostasis, legumes are also likely to differ from non-N 2 -fixing plants in the nutrient scaling relationship among tissues.
[N] and [P] in different tissues are usually dependent and show correlated responses to ecological and evolutionary factors Yang et al., 2014). Soil N availability is the direct cause of plant [N] variation . However, since soil N availability may not be the driving factor for legumes' [N], there could be less variation and co-variation in [N] among different tissues. For example, Liu et al. (2010) found that root and leaf [N] of a legume Caragana microphylla did not co-vary. However, another study found little difference in nutrient scaling relationships between legumes and non-legume plants (Yang et al., 2014). Thus, it is still an open question how SNF affects the nutrient co-variation among tissue types.
Previous studies also found coordinated variation of N and P within tissues (Broadley et al., 2004;Townsend et al., 2007;He et al., 2008;Sardans et al., 2016). Taking leaves for example, this correlated change can be described as: This equation can be ln-transformed (logarithm transformation to the base of the mathematical constant e) to: The slope b is normally positive, showing the positive relationship between ln-transformed leaf [N] and [P]. Even though there is disagreement over the slope b being 3/4 Niklas, 2006) or 2/3 (Reich et al., 2009), the consensus is that [N] increases relatively more slowly than [P] (Wright et al., 2004;Elser et al., 2010).  demonstrated that N-P relationships can differ between life forms, and Ågren (2008) said that the excess uptake of N or P could shift the slopes of N-P relationships. As a result, we may infer that legumes with sufficient N supply have different N-P relationships from non-N 2 -fixing plants often faced with N limitation. Legumes with very high N:P might suffer from P limitation (Güsewell, 2004), so the slopes of N-P relationships can be lower due to the excess N absorption (Ågren, 2008), i.e., the increasing rate of [P] can be even faster.
In summary, we expected stoichiometry of N 2 -fixing legumes and non-N 2 -fixing plants to differ considerably, and that this would influence plant nutrition responses to the variation in soil and change plant stoichiometric regulation among tissues.
In this study, we compared the stoichiometric patterns of N 2 -fixing legume shrubs and non-N 2 -fixing shrubs in northern China. Compared to trees, shrubs are generally more uniform in size, so the plant size effect and the "dilution" of nutrients in structural tissues of large trees is weaker Elser et al., 2010;Yang et al., 2014). In temperate China, species in the genera Lespedeza, Caragana, and Sophora are among the most widespread legume shrubs, while species in the genera Vitex, Corylus, and Spiraea are typical non-N 2 -fixing shrubs. Previous studies have reported the use of legume shrubs in ecological restoration where they act as nurse plants to facilitate the growth of tree seedlings and herbs by increasing soil nitrogen levels (Gómez-Aparicio et al., 2004;Zhao et al., 2007). Our research could provide useful suggestions in legumes' nutritional requirement for this application.
Specifically, we proposed the following hypotheses. . 4. In order to maintain optimal N:P, the increasing rate of [P] relative to [N] in legume shrubs is higher than that in non-N 2 -fixing shrubs, which means the legume shrubs have gentler N-P slopes within tissues than non-N 2 -fixing shrubs.

Study Sites and Sampling
Leaf, stem, and root samples of legume shrubs and non-N 2fixing shrubs were collected in 299 natural temperate broadleaved deciduous shrubland sites between July and September (mostly July and August) from 2011 to 2013. Among these sites, legume shrubs were sampled in 96 sites each with three 25 m 2 plots (Figure 1). Totally, we sampled 19 species of six genera in the Fabaceae family with 105 replicates and 113 species of 64 genera in 32 non-N 2 -fixing families with 534 replicates (see Supplementary Data Sheet 2 for details). Species were identified according to the Flora of China (Wu et al., 2006). Non-legume N 2 -fixing families such as, Elaegnaceae and Coriariaceae were excluded. We classified plant families by the APG III system (APG III, 2009). At each site, fully expanded sun leaves, stems and roots (mainly coarse roots in the top 30 cm of soil) of at least five individuals of each species were collected and assembled in fabric bags then dried in the sun, transported to the laboratory and oven-dried at 65 • C for 72 h. We also investigated soil nutrient concentrations of the sites. To do this, three one-meter-deep (or deep to the base rock) pits along the diagonal of each plot were excavated to collect soil samples. For each profile, soil at the depth of 0-10, 10-20, 20-30, 30-50, 50-70, 70-100 cm was sampled and soil samples from the same depth in one plot were mixed. We also excavated one profile of 100 cm outside each site and collected soil cores in left, right, and front surfaces at the depth of 0-10, 10-20, 20-30, 30-50, 50-70, 70-100 cm to measure the soil bulk density (BD). Soil samples used to test element concentrations were air-dried, roots removed, and then ground to pass through a 100 µm mesh sieve (see Yang et al., 2014 for more details on the sampling protocol).
Climate variables we used in this analysis include mean growing season temperature (MGT) and growing season precipitation (GP). Growing season was defined from May to October (Yang et al., 2014), and mean monthly temperature and precipitation in growing season were extracted from the means of 1970 to 2000 of the WorldClim spatial climate data (Hijmans et al., 2005; available at http://www.worldclim.org).

Measurement of Nutrient Concentrations
[N] and [P] of plant and soil samples were measured at the Measurement Center of the Institute of Botany, Chinese Academy of Sciences. Soil total [N] (STN) and plant tissue (leaf, stem, and root) [N] were determined using an elemental analyzer (2400 II CHNS; Perkin-Elmer, Boston, MA, USA) under 950 • C for combustion then reduced to 640 • C. Soil total [P] (STP) and plant tissue (leaf, stem and root) [P] were measured using the molydate/ascorbic acid method after H 2 SO 4 -H 2 O 2 digestion (Jones, 2001). STN and STP from 0 to 30 cm were used as surrogates of the soil nutrient conditions, because STN and STP were highly correlated at all depths, and soil depths in many sites did not reach 100 cm.

Data Analysis
Student's t-test was applied to compare the ln-transformed [N], [P], and N:P of each tissue between legume shrubs and non-N 2fixing shrubs after comparing their variance by an F-test.
In order to remove the influences of species identity, climate (MGT and GP) and other soil factors (BD and the other nutrient concentration), we first calculated the residuals of tissue nutrient compositions ([N] res , [P] res , and N:P res ) using general linear models. Taking the leaf as an example, [N] res , [P] res , and N:P res can be presented as error terms (ε 1 , ε 2 , and ε 3 , respectively) of regression equations.
The ε 1 , ε 2 , and ε 3 stand for the part of nutrient variance that cannot be explained by other effects but corresponding soil nutrient compositions only. Hence, leaf [N] res , [P] res , and N:P res can be calculated as follows.
Leaf N : P res = ln(Leaf N : P) − (γ 0 + γ 1 Family + γ 2 Genus Reduced major axis (RMA) estimation was applied to explore the relationship of nutrient concentrations residuals between tissues and soil, among different tissues, and within each tissue. An allometric equation was employed to fit the relationship using the ln-transformed stoichiometric traits.
To explore the homeostasis of plant nutrients, we built an equation between the nutrients in plants and those in soils, where X stands for nutrient compositions in soil and ln(Y) stands for [N] res , [P] res , and N:P res of plant tissues, because the calculation of nutrient residuals has included the lntransformation of nutrients. The homeostasis coefficient H, which equals the reciprocal of the slope, was used to quantify the homeostasis. H equivalent to 1 (slope equaling 1) indicates a lack of homeostasis, H between 1 and infinity (slope between 0 and 1) indicates incomplete homeostasis, and an infinite H (slope equaling 0 or insignificant) means strict homeostasis (Sterner and Elser, 2002;Elser et al., 2010). We also explored the relationship for nutrients across different tissue types using an exponential scaling approach, where nutrient residuals of the tissue close to the ground were set as ln(X), whereas those of the tissue close to the top were set as ln(Y). The equation of nutrient scaling relationships within tissues is still the same, while ln(X) and ln(Y) represented tissue [P] res and [N] res , respectively. The exponential slope (b) was used to represent the relative rate of nutrient accumulation. A b equivalent to 1 indicates an isometric allometry, whereas b different from 1 indicates asymmetric allometry, with b > 1 suggesting that Y increases more quickly than X and b < 1 suggesting the opposite Yang et al., 2014). We used likelihood ratio tests when comparing the regression slopes between functional groups. To test if the patterns are consistent among different taxonomic groups, we refined the non-N 2 -fixing shrubs to a certain taxonomic group by selecting the family (Rosaceae with 150 replicates, to confine all non-N 2 -fixing shrubs to a family) and the genus (Vitex with 94 replicates, to confine all non-N 2 -fixing shrubs to a genus) with most replicates. We then repeated the same procedure to examine patterns for species from the family Rosaceae, the genus Vitex and the rest (the remaining 290 non-N 2 -fixing shrubs samples from neither Rosaceae nor Vitex), respectively, but omitted the "Family" and both "Family" and "Genus" terms when calculating residuals for the Rosaceae and the Vitex, respectively.
All analyses were conducted using the basic and smatr (RMA regression and likelihood ratio test) packages of R version 3.2.2 (Warton et al., 2012;R Core Team, 2015).

Nutrient Scaling among Tissue Types of Legume Shrubs and Non-N 2 -fixing Shrubs
[N] res in legume shrubs were not correlated between leaves and non-leaf tissues (stems or roots), but were correlated between stems or roots (slope = 1.02, 95% CI: 0.82-1.27) (Figures 4A-C).

DISCUSSION
In this study, we compared N stoichiometry of legume shrubs and non-N 2 -fixing shrubs to explore the potential influence of SNF on plant stoichiometry. The results showed that N 2 -fixing legume shrubs were richer in N, more homeostatic and more independent in N correlations among tissues, but had larger demand for P, than non-N 2 -fixing shrubs. The patterns also hold when comparing legume shrubs to a single family (Rosaceae), or a single genus (Vitex) of non-N 2 -fixing shrubs or to the remaining non-N 2 -fixing shrubs, strongly supporting the hypothesis that SNF provided higher [N] and stronger homeostasis for N in legume shrubs compared to non-N 2 -fixing shrubs in general. Consequently, higher [N] and stronger homeostasis resulted in weaker N co-variation among tissues of legume shrubs. Nevertheless, owing to the extraordinarily high N:P and flat N-P slopes, legume shrubs need more P to balance their N income. Detailed comparisons between these two functional groups and influences of SNF on stoichiometry of legume shrubs are discussed below.

Nutrient Concentrations and Ratios in Legume Shrubs and Non-N 2 -fixing Shrubs
The comparison results support our first hypothesis about the nutrient concentrations and ratios. [N] and N:P in three tissues were higher in legume shrubs than in non-N 2 -fixing shrubs, but STN and SNP were even lower in microsites with legume shrubs than in microsites with non-N 2 -fixing shrubs ( Table 1), suggesting that SNF provides legume shrubs with a clear N advantage compared to non-N 2 -fixing shrubs. Studies in tropical forests suggested that legume plants also have P advantage in addition to N, owing to their production of N-rich phosphatase to break down organic P (Houlton et al., 2008;Nasto et al., 2014). However, we did not find higher [P] in legume shrubs probably because the strategy of temperate shrubs is different. Another possible reason is that the relatively more P-limited soil in China (Han et al., 2005) suppressed the P advantage of legume shrubs, and the high leaf N:P ( Table 1) also suggests that legume shrubs could face strong P limitation (Güsewell, 2004).

Stoichiometric Homeostasis of Legume Shrubs and Non-N 2 -fixing Shrubs
Strict stoichiometric homeostasis in N was evident for legume shrubs, but not for non-N 2 -fixing shrubs. In agreement with our second hypothesis, STN does not affect [N], indicating strict homeostasis in N, in different tissue types of legume shrubs. SNF provides an additional N supply, which offsets N limitation and allows legume shrubs to keep stable [N] (Lambers et al., 2008;Hobbie, 2015). Therefore, legume shrubs are able to maintain physiological function and gain higher fitness than other plants under N-limited conditions, so that they can have higher dominance on barren soil (Rastetter et al., 2001;Menge et al., 2008). Due to the influence of N homeostasis, N:P res were also more homeostatic in legume shrubs than in non-N 2 -fixing shrubs (Figures 3G-I). In addition, N:P res were only correlated with [P] res but not [N] res in legume shrubs, whereas N:P res were both correlated with [N] res and [P] res in non-N 2 -fixing shrubs (Supplementary Figure 4), suggesting the contribution of N homeostasis to the stability of N in legume shrubs. Consistent with previous studies (Yu et al., 2010;Sardans et al., 2016), soil N availability improved [N] of non-N 2fixing shrubs. However, such effects might be masked by the effects of taxonomy and other environmental factors (Yang et al., 2016). The stronger stoichiometric homeostasis in leaves corresponds to previous studies with other plants (Garrish et al., 2010;Schreeg et al., 2014;Yan et al., 2016). Leaf [N] is closely related to many important physiological activities, such as photosynthesis and respiration (Reich et al., 2003), so plants maintain high and stable [N] in leaves to keep the efficiency of necessary metabolic processes (Güsewell, 2004). Similarly, the more constrained N:P in leaves also reflects that leaves are less responsive to the variation of soil nutrient composition because of the demand to keep efficient carbon fixation (Sterner and Elser, 2002;Schreeg et al., 2014).
In a previous study, Brouwer (1983) proposed that nutrients were transported to distant tissues only after the tissues closer to the nutrient source had met their needs. Based on this hypothesis, the closer the tissue is to the nutrient source, the higher its priority to nutrients will be. In other words, we would expect that the variation of soil nutrients would affect roots most and leaves least. Indeed, we found a steeper slope between STN and root [N] res than between STN and other tissue types' [N] res , and the slope of the root N:P res -SNP relationship was also steeper than between SNP and other tissue types' N:P res in  non-N 2 -fixing shrubs. Our results suggest that changes in root nutrient composition can mirror soil's nutrient variation better than other tissues.
[P] res did not show clear patterns with STP for either legume shrubs or non-N 2 -fixing shrubs, probably because soil available P is highly sensitive to the environment (He and Dijkstra, 2014) and STP may not be a good surrogate for soil P availability.

Nutrient Scaling among Tissue Types in Legume Shrubs and Non-N 2 -fixing Shrubs
The correlations of nutrients among tissue types were stronger in non-N 2 -fixing shrubs than in legume shrubs (Figure 4), after controlling for other factors. Again, the difference may be masked by the effects of taxonomy and environments (Yang et al., 2014). STN was the only driving force of N variation among tissues in our case.
[N] res in each tissue in non-N 2 -fixing shrubs responded to STN in a similar fashion (Figures 3A-C), so we could observe positive [N] res relationships. In contrast, [N] res in legume shrubs were homeostatic to STN (Figures 3A-C), owing to sufficient N supply independent of the soil N pool, and thus they showed weaker [N] res co-variation among tissue types. However, the positive [N] res relationship still existed between structural tissue types (stem and root), probably due to their physiological similarity . Unlike [N] res , we found the coordinated [P] res and N:P res among tissues in legume shrubs (Figures 4D-I), still demonstrating physiological and ecological connections among tissues  for legume shrubs and non-N 2fixing shrubs. The synchronized variation of nutrients among leaves, stems and roots is consistent with previous studies across various plant species Yang et al., 2014). The scaling slopes of [N] res (non-N 2 -fixing shrubs) and [P] res (both legume shrubs and non-N 2 -fixing shrubs) showed that tissues further away from the soil accumulated nutrients more slowly than those closer to the soil, supporting Brouwer's hypothesis that tissue types nearest to the nutritional source are preferential in nutrient allocation (Brouwer, 1983;Yang et al., 2014). Furthermore, tissues further away from the soil had a reduced nutrient variation rate, implying stronger nutrient homeostasis, than those closer to the soil (Yan et al., 2016), in agreement with our results for homeostasis. The scaling slopes of N:P res among tissue types in legume shrubs and non-N 2 -fixing shrubs also followed the same pattern mentioned above, suggesting that leaves are more stable in nutrient ratio than stems and roots (Schreeg et al., 2014;Yan et al., 2016). N vs. P Scaling within Tissue Types of Legume Shrubs and Non-N 2 -fixing Shrubs Consistent with our fourth hypothesis, we found lower N-P slopes in each tissue type in legume shrubs. Correlated N and P scaling relationships of non-N 2 -fixing shrubs can be partly explained by a strong correlation between ln-transformed STN and STP (R 2 = 0.20) (Güsewell and Koerselman, 2002;He et al., 2008). Our results concur with prior studies that P accumulates faster than N in leaves (Wright et al., 2004;Elser et al., 2010), as well as in stems and roots . However, the slopes in this study are not comparable with other studies because residuals (rather than real concentration data) were used in this study. P limitation can reduce the slope, but N limitation will do the opposite (Ågren, 2008). The slopes were flatter in legume shrubs than in non-N 2 -fixing shrubs (Figure 5), suggesting that P increases faster in legume shrubs than non-N 2 -fixing shrubs and that legume shrubs are more P-limited due to excess N absorption of SNF (Ågren, 2008). Meanwhile, higher plant N:P also demonstrates that P limitation is stronger in legume shrubs than non-N 2 -fixing shrubs (Güsewell, 2004). Besides requiring more P to maintain optimal N:P, legume shrubs also need enough P to maintain the functioning of the SNF (Vitousek and Field, 1999;Benner and Vitousek, 2007;Sulieman and Tran, 2015). Therefore, N 2fixing legume shrubs demand more P and may be consequently more P-limited than non-N 2 -fixing shrubs (Vitousek et al., 2010). The differences between N-P slopes ( Figure 5) and between N:P ratios ( Table 1) were smaller in leaves than in other tissue types, suggesting the stronger stoichiometric stability in leaves to sustain normal physiology (Schreeg et al., 2014;Yan et al., 2016).
In conclusion, there are obvious stoichiometric differences between legume shrubs and non-N 2 -fixing shrubs. Extra N supply in legume shrubs enhances their homeostasis to soil nutrient deficiency and provides stronger N stability. However, due to the surplus uptake of N, legume shrubs may suffer from potential P limitation. Legume shrubs are good nurse plants at the early succession stage of restoration habitats, which can provide not only canopy shade but also more N (Gómez-Aparicio et al., 2004;Zhao et al., 2007). Nevertheless, special attention should be paid to the nutrient conditions of nurse legumes, due to their tendency of being P-limited. We were not able to measure the actual SNF, so further studies on how these differences are connected with SNF and plant fitness are now necessary to unravel the underlying evolutionary and physiological mechanisms.