Introduction

Nitrogen (N) and phosphorus (P) are important nutrients for plant growth and ecosystem productivity in terrestrial ecosystems (Vitousek and Howarth 1991; Elser et al. 2007; Vitousek et al. 2010). Although there are multiple ways that reactive N enters ecosystems such as atmospheric N deposition, agricultural fertilization and N fixation by legumes, the available P for plants is mainly derived from mechanical rock weathering (Vitousek et al. 2010). The biogeochemical cycles of carbon (C), N and P elements are considered biologically coupled (Finzi et al. 2011; Sistla and Schimel 2012), and the unbalanced N and P inputs into the biosphere are expected to decouple nutrient cycling and shift terrestrial ecosystems from N limitation toward P limitation (Vitousek et al. 2010; Peñuelas et al. 2013). A better understanding of the influences of N-P imbalance on biomass production is important for predicting C dynamics as well as ecosystem production and functioning, especially in the context of anthropogenic N inputs.

The theory of ecological stoichiometry predicts that plant growth rate is characterized by a specific ratio of RNA to protein which has been associated with the organisms’ N:P ratio (Elser et al. 2010). Recently, abundant studies focused on the global patterns of N inputs on plant N-P stoichiometry, and they revealed that N addition generally increased plant N:P ratio (Peñuelas et al. 2013; Yuan and Chen 2015; Li et al. 2016; Sardans et al. 2017; Soons et al. 2017; Yue et al. 2017). The increase in tissue N:P ratio is usually recognized as an indicator of P limitation for plant growth. This is because plants would allocate more resources to ribosomal RNA than in proteins under rapid growth according to the growth rate hypothesis (Elser et al. 2010), and the increased N:P ratio illustrates an enhanced production of N-rich proteins, but not that of P-rich ribosomal RNA (Vitousek et al. 2010; Sistla and Schimel 2012; Peñuelas et al. 2013), thus leading to P limitation in terms of growth. Nevertheless, linkage between the responses of N:P ratio and plant biomass to N enrichment has not been established at broad scales. The N-induced P limitation may be a progressive process. At low levels of N addition, P may not be a limiting factor because phosphatase activity is highly responsive to N changes (Marklein and Houlton 2012), and N enrichment increases soil P availability and thus plant C assimilation. However, as N continues to increase, the N-induced increase of P availability is insufficient to balance the increasing P demand (Li et al. 2016), and P limitation may gradually become predominant (Peñuelas et al. 2013; Peng et al. 2017b), and thus weaken the positive response of plant growth. Based on above deductions, we predict that the N-induced changes in plant biomass exhibit an asymptotic relationship with the changes in N:P ratio across the globe (Fig. 1).

Fig. 1
figure 1

A framework showing that ecosystems shift from N limitation toward P limitation as N availability increases

Full size image

To test the above hypothesis, we performed a meta-analysis to assess the N effects on plant aboveground biomass (AGB), tissue N and P concentrations and N:P ratio at the global level. The response ratio (RR) was adopted to indicate the magnitude of variable responses to N addition. Then, we explored the relationship of the RR of AGB (AGB-RR) and RR of plant N:P ratio (N:P-RRs) across all studies. Further, we evaluated responses in soil inorganic N and P concentrations with increasing N rate to test whether the P limitation resulted from imbalanced N and P availabilities.

Materials and methods

Data collection

We searched published papers reporting the influences of N addition on AGB, plant N and P concentrations and N:P ratio, and soil inorganic N and P concentrations using ISI-Web of Science (Thomson Reuters, New York, NY, USA) and Google Scholar (Google Inc., Mountain View, CA, USA). Keywords and combinations that used for the searching were as follows: “nitrogen / phosphorus / nutrient / fertilization” and “grassland / meadow / steppe / forest / tree / wetland / bog / fen / tundra / soil” in the title, keyword or abstract. In addition, we screened datasets in previous meta-analyses that examined plant biomass or nutrients in response to N enrichment (Li et al. 2016; Peng and Yang 2016; Yue et al. 2017). In total, our database was established with 760 entries including the above-mentioned parameters from 51 publications (see Table S1). Studies were identified based on the following criteria: 1) all experiments should report both plant N:P ratio and AGB data in control and N addition treatments simultaneously. For AGB, various methods of biomass estimation (direct harvest for most experiments or calculated from nutrient pools divided by their concentrations, or estimated from other proxy variables, e.g., tree size and herbaceous cover) were accepted; 2) N was directly added in the experiments. Several studies examined the interactions of N addition with other manipulations (e.g., changing precipitation regimes, P addition and elevated ozone, etc.). In such cases, only observations from the single-factor treatments (i.e., N addition) were collected, excluding values from the combined treatment; 3) Studies that used different plant species in the same experiment or performed at distinct sites or with different N rates were treated as independent (Janssens et al. 2010; Peng and Yang 2016); and 4) N:P ratios were either directly extracted from the selected papers or calculated using N and P concentrations (Deng et al. 2015). Raw data were either obtained from tables or extracted from figures using Origin Pro 8.0 (OriginLab, Northampton, MA, USA). Ancillary site information, such as site coordinates, mean annual temperature (MAT), mean annual precipitation (MAP), ecosystem type, N fertilizer form, N addition rate, and year of literature publication were included.

Meta-analysis

The effects of N addition on various parameters were quantified by calculating the natural logarithm of the ratio between treatment (Xt) and control groups (Xc): RR = ln (Xt / Xc) (Hedges et al. 1999). Effect size was usually weighted by the inverse of the pooled variance (LeBauer and Treseder 2008; Xia and Wan 2008; Janssens et al. 2010) or replicate (Lam et al. 2012; Peng et al. 2017a) in previous meta-analyses. The studies collected in our database did not always include published variance. In addition, some N:P ratios were calculated from N and P concentrations when both parameters were recorded in the same papers. In such a case, the variance cannot be readily estimated (c.f. Wang and Taub 2010). Therefore, the effect size was weighted by sample size using the formula: weightn = ncnt / (nc + nt), where nc is the number of control groups and nt is the number of treatment groups (Adams et al. 1997). This made better replicated studies contribute more to the overall mean effect size (Adams et al. 1997). A random-effect model with resampling techniques was used to calculate the mean effect size and 95% confidence interval (CI). The mean RRs and 95% CIs were generated by bootstrapping using MetaWin 2.1 (Sinauer Associates Inc., Sunderland, MA, USA). The N effect on a response variable was considered significant if the 95% bootstrap CI did not overlap zero. Otherwise, N was considered to have no significant effect on those variables.

Since the aim of the current analysis was to examine how the N-P imbalance influences plant biomass production, we explored the relationship between the AGB-RR and N:P-RR. The Akaike information criteria (AICc) score was used to assess the goodness-of-fit for various models (i.e., linear function, quadratic function and Michaelis-Menten function), and the lowest AICc value indicates the best model (Burnham and Anderson 2002). We also quantified the relative contribution of various factors (i.e., MAT, MAP, N rate and N:P-RR) to AGB-RR variation using the following model:

$$ \mathrm{AGB}-\mathrm{RR}={\beta}_0+{\beta}_1\mathrm{MAT}+{\beta}_2\mathrm{MAP}+{\beta}_3N\ \mathrm{rate}+{\beta}_4N:P-\mathrm{RR}+\varepsilon $$

where β0-β4 and ɛ are the coefficients and the residual, respectively. The standardized regression coefficients were calculated using General Linear Model. Because of a nonlinear relationship between AGB-RR and N:P-RR (i.e., Michaelis-Menten function, see Results), the N:P-RR was natural logarithm transformed to improve the linearity in General Linear Model. Although using data transformation may lead to some biases, this approach has been widely adopted to address the nonlinear relations in multiple regression analysis (Burnham and Anderson 2002). All statistical analyses were conducted in software package R 3.2.1 (R Development Core Team 2015).

Quality criteria for meta-analysis

To ensure robust results in our meta-analysis, we have made a checklist to address the methodological criteria developed by Koricheva and Gurevitch (2014) (see Table S2). In particular, we first tested the between-group heterogeneity (Qb tests) of the effect size among ecosystem types and N fertilizer forms. In our analysis, ecosystem types included forests, grasslands, wetlands and tundra (Fig. S1; Table S1). The N fertilizer forms included NH4NO3, urea, NH4NO3 + urea and NO3―-N (Table S1). A significant Qb indicates that the effect sizes showed different responses among these conditions (Hedges et al. 1999). In this case, we further compared the data of AGB, plant N, P and N:P ratio. A subgroup with fewer than two observations was excluded due to the lack of statistical power.

Second, we checked publication bias. Meta-analysis may suffer from publication bias, because statistically significant results tend to be published more likely than insignificant results (Koricheva and Gurevitch 2014). Publication bias was checked using funnel plots (Egger et al. 1997). A funnel plot can offer a visual assessment of the effect size against sample size or precision (the reverse of standard error of the RR). In absence of publication bias, the observations should be symmetrically distributed in a ‘funnel’ shape around the mean effect size. Visual assessment is subjective, but the commonly used methods for statistically testing publication bias (e.g., Begg’s test, Egger’s test, “trim and fill” method, or fail-safe number; Koricheva and Gurevitch 2014) required the effect size, its associated variance (i.e., standard deviation) and sample size. As mentioned above, variables in some studies, especially the N:P ratio data did not include variance values; hence, it is not possible to perform a statistical test of publication bias in our analysis.

Third, we evaluated the temporal changes in effect size. It has been suggested that the magnitude of effect sizes may change over time due to the gradual accumulation of insignificant results caused by longer publication process, methodological improvement or real biological changes in the magnitude of effect size (Koricheva and Gurevitch 2014). Here we tested the temporal changes in AGB, plant N, P and N:P ratio using publication year as a moderator. Our results indicated no significant temporal changes in the effect sizes of the above variables (Fig. S2).

Results

External N input significantly increased AGB across all studies. On average, elevated N enhanced plant growth by 32.4%. Meanwhile, N addition, in aggregate, increased N concentration and N:P ratio by 20.5% and 28.2%, respectively. In contrast, the P concentration was markedly reduced by 4.4% under N enrichment (Fig. 2 insert).

Fig. 2
figure 2

Relationship between the response ratio (RR) of aboveground biomass (AGB) and RR of plant N:P ratio across all studies. Insert figure shows the overall effects of N addition on AGB, plant N and P concentrations and N:P ratio. Error bars indicate 95% confidence intervals. Sample size for each variable is shown above the bars. To make the graph more intuitive, the labels in x and y axes were presented in percentage change. However, the RR was used in regression analysis due to its conformance to normal distribution

Full size image

Among the explanatory variables, only N:P-RR significantly explained the variation in AGB-RR as determined by General Linear Model (Table S3), indicating that N:P ratio was the best predictor for plant biomass production in response to N addition. Further, we used three models to explore the relationship between AGB-RR and N:P-RR (i.e., linear, quadratic and Michaelis-Menten model), and results showed that the Michaelis-Menten model was the best one to fit the relationship (Fig. 2; AICc = 133.5 for linear model, AICc = 123.2 for quadratic model and AICc = 118.3 for Michaelis-Menten model, Table S4).

The changes in plant N, P and N:P ratio were positively correlated with changes in soil inorganic N, inorganic P and N:P ratio (Fig. 3). Changes in soil inorganic N and N:P ratio were proportional to N addition rate (Fig. 4a, c), while the soil inorganic P remained unchanged with increasing N rate (Fig. 4b).

Fig. 3
figure 3

Relationships between the response ratio (RR) of pant N and RR of soil N (a), RR of plant P and RR of soil P (b), and RR of plant N:P ratio and RR of soil N:P ratio (c) across all studies. To make the graph more intuitive, the labels in x and y axes were presented in percentage change. However, the RR was used in regression analysis due to its conformance to normal distribution

Full size image
Fig. 4
figure 4

Changes in the response ratios (RRs) of soil N (a), soil P (b) and soil N:P ratio (c) as a function of N addition rate across all studies. To make the graph more intuitive, the labels in y axis were presented in percentage change. However, the RR was used in regression analysis due to its conformance to normal distribution

Full size image

Discussion

Our meta-analysis showed that additional N inputs significantly increased AGB and N:P ratio in plants, and the magnitude of the responses were within the ranges of previous meta-analyses (AGB: 32.4% vs 18–54%, Elser et al. 2007; LeBauer and Treseder 2008; Xia and Wan 2008; Lee et al. 2010; Fay et al. 2015; plant N:P ratio: 28.2% vs 25–32%, Yuan and Chen 2015; Li et al. 2016; Sardans et al. 2017; Yue et al. 2017). The discrepancy between our study and others may partly result from the differences in sample size and weighting method of the RR. Although both AGB and plant N:P ratio increased by N addition, our results revealed a saturating response of AGB to increasing plant N:P ratio (Fig. 2). This finding is in line with our primary hypothesis. It is well recognized that N and P are two essential nutrients for plant growth (Marschner 2011). A majority of leaf N is associated with photosynthetic processes, including light harvesting and carbon reaction enzymes, and thus plays a fundamental role in leaf photosynthesis (Taiz and Zeiger 2006). A positive photosynthetic rate response to leaf N concentration in different species is well established (Reich et al. 1997). P also plays a pivotal role in various plant functions, such as the formation of phospholipids which are used in energy metabolism (e.g., NADPH and ATP) in light and dark reactions (Taiz and Zeiger 2006). Leaf P concentration also positively influences light-saturated photosynthesis across various vegetation types (Crous et al. 2017). Biological stoichiometry theory considers the balance of multiple chemical elements in living systems (Elser et al. 2010; Sistla and Schimel 2012). Low levels of N input stimulate biomass accumulation. As N continues to increase, the increased tissue N:P ratio indicates that ecosystems gradually shift from N to P limitations (Vitousek et al. 2010; Peñuelas et al. 2013). Excess N addition at this stage may not further increase the rate of leaf photosynthesis and C assimilation because P limitation may constrain the ribulose-1,5-bisphosphate (RuBP) regeneration and N partitioning to Rubisco for Amax, causing a declining trend (Reich et al. 2009; Peng et al. 2017b). This is supported by a previous study, which showed that the slope of the relationship between photosynthetic capacity (Amax) and leaf N is sensitive to the foliar P concentration across 341 plant species, with low P values associated with a lower slope of Amax vs N concentration, indicating that the Amax – N relationship may be constrained by relatively low P in leaves (Reich et al. 2009).

A question arises as to how the N-induced plant P limitations occur? Our analysis revealed positive relationships of N, P and N:P ratio RRs between plants and soils (Fig. 3), indicating that plant N and P status was influenced by the changes in soil N and P availabilities. In general, soil N availability increases with increasing N rate (Högberg et al. 2006; Peng et al. 2017b; Fig. 4). Although plants also have developed several strategies to acquire additional soil P under N enrichment, such as changing above- and belowground C allocation, root architecture, and increasing root exudates and phosphatase activity (Hinsinger 2001; Marklein and Houlton 2012), inputs to P pools from mineral weathering or from slowly recycling biotic pools may be insufficient to maintain balance as N input continues to increase (Vitousek et al. 2010). We observed no significant correlations of soil inorganic P with N rate, indicating unparalleled changes in soil N and P availabilities with increasing N additions. In addition, the proportional increase in the RR of soil N:P ratio with N rate (Fig. 4c) further demonstrated the N-P imbalance in the soils with elevating N availability. Overall, the N-induced un-equivalent increases in soil available N and P concentrations may contribute to a gradual P limitation for plant growth.

Our meta-analysis highlights the importance of changes in plant N:P ratio in predicting AGB response to N additions. Nevertheless, previous studies showed that other factors, such as MAT, MAP or N rate may also affect the AGB responses (Lu et al. 2011; Soons et al. 2017). Although no significant changes occurred in AGB-RR with MAT in binary plot (Fig. S3a), there were weak but significant positive relationships of AGB-RR with MAP and N rate (Fig. S3b, c). The insignificant explanation of MAP and N rate in multiple regression analysis may result from the multi-collinearity among explanatory variables, which would result in excluding the MAP and N rate from multiple regression models. This is because the N:P-RR explained the variation in AGB-RR better in statistical terms (MacNally 2000). It is worth mentioning that the positive AGB-RR response leveled off or even declined when N rate was higher than 30 g m−2 year−1 (Fig. S3c). This result, to some extent, supported the increased P limitation with increasing N additions.

The current meta-analysis provided a statistical evaluation of the central tendency of plant biomass production and N:P status to experimental N addition at the global scale, and evaluated how the changes in N:P ratio affected plant biomass production response to N enrichment. However, our study may have several methodological and statistical limitations. For instance, we found between-group heterogeneity of the data, i.e., distinct variable response among different ecosystem types and N fertilizer forms (Table S5). The magnitude of increase in plant N in grasslands was higher than that in forests, while forests had a greater plant P reduction than grasslands (Fig. S4a). This is likely because forests often have a larger biomass and the N-induced increase in plant C accumulation would result in a greater ‘dilution effect’ than grasslands. Thus, the P concentration in woody plants may be lower than that in herbaceous plants under N enrichment. We also observed diverse responses of plant N:P ratio among N fertilizer forms (Table S5; Fig. S4b). This is possibly caused by the small sample size of NH4NO3 + urea and NO3―-N treatments (Fig. S4b), which would limit the statistical power. In addition, we only used funnel plots to assess publication bias. Although funnel plots for AGB, plant N, P and N:P ratio were roughly symmetrical (Fig. S5), the publication bias cannot be statistically tested by the commonly adopted methods, due to the lack of sampling variance. Nevertheless, since the overall effect sizes of AGB and plant N:P ratio were within the ranges in previous meta-analyses (AGB: 32.4% vs 18–54%; plant N:P ratio: 28.2 vs 25–32%), we expected that publication bias had a low influence in the current study.

To our knowledge, this study for the first time, links the responses of plant N:P ratio and biomass production to N addition at the global level. Our analysis revealed three knowledge gaps that deserve further investigation. First, growing evidence points to global changes, particularly increasing N inputs, may significantly alter nutrient stoichiometry in plants, microorganisms and soils (Peñuelas et al. 2013; Yuan and Chen 2015; Li et al. 2016; Sardans et al. 2017; Yue et al. 2017). The shifts in N:P ratio are considered to have significant impacts on the ecosystem structural and functional traits as well as C sequestration potential (Peñuelas et al. 2013). Therefore, apart from plant biomass, it is imperative to know how the N-P imbalance affects other important ecological processes (e.g., soil respiration, litter decomposition, root dynamics, etc.). Second, meta-analysis is the principle method in regional- and global-scale data synthesis. However, considering its methodological and statistical limitations, coordinated distributed experiments, which involve standardized protocols (Fraser et al. 2013), is thus called for to advance our understanding of the ecological processes and nutrient stoichiometry relationships. Third, great uncertainty in predicting future global terrestrial C sequestration may partly result from the poorly understanding of the C-N-P interactions under global change scenarios. Although terrestrial C cycle models have been developed to include N limitation in C sequestration, few of them consider the influences of P limitation and the N-P balance (Peñuelas et al. 2013; de Vries 2014). Modeling studies need to consider the biomass-N:P ratio relationship and the underlining mechanisms to allow more reliable predictions of the future global C dynamics considering anthropogenic N inputs.