Nitrogen inputs accelerate phosphorus cycling rates across a wide variety of terrestrial ecosystems

Introduction

Chemical elements – most commonly nitrogen (N) and phosphorus (P) – constrain plant productivity and microbial functioning in the vast majority of terrestrial ecosystems (Vitousek & Howarth, 1991; Elser et al., 2007). Human activities have substantially modified N and P cycles, however, thereby altering the pattern, magnitude, and extent of nutrient limitation on land (Falkowski et al., 2000; Schlesinger, 2009). Of particular importance has been the massive rise in anthropogenic N fixation, roughly doubling natural levels of N inputs to the terrestrial biosphere, resulting in widespread eutrophication of terrestrial and aquatic ecosystems (Vitousek et al., 1997). Further, this perturbation can result in divergent outcomes for the terrestrial biosphere: whereas excess N appears to enhance carbon dioxide (CO₂) uptake in some cases, it decreases plant biomass and diversity in others (Aber et al., 1989; Clark & Tilman, 2008; Thomas et al., 2010). While mechanisms underlying these differences are complex, they hinge broadly on organisms’ responses to the excess N – particularly whether N leads to enhanced P conservation across terrestrial ecosystems (Vitousek et al., 2010). Here we argue for widespread couplings between N and P cycles, leading to accelerated rates of P cycling in responses to elevated N across many areas of the terrestrial biosphere.

Liebig’s law of the minimum posits that the single scarcest resource in relation to plant demands is most limiting; hence, according to this law, only one resource can limit plant productivity at a time (Liebig, 1842). From an ecophysiological perspective, however, plants (and microbes) are expected to allocate their resource reserves toward strategies that increase the acquisition of the most limiting ones – thus moving toward a point in which all resources simultaneously limit productivity and growth (Bloom et al., 1985; Chapin et al., 2002). This economic analogy would seem to contrast with the idea of single nutrient limitation, as evidenced by N and P co-limitation in terrestrial and aquatic ecosystems (Elser et al., 2007; Harpole et al., 2011). Mechanistically, the preponderance of N and P co-limitation suggests strong couplings between these two nutrients in cells, arising via interactions that link N and P in biomolecules. For instance, mRNA, a relatively P-rich compound, transcribes proteins, which are rich in N and serve important functions such as CO₂ uptake (i.e. the enzyme RuBisCO). Consequently, P fuels investment of N at local scales, which then allows for uptake of more carbon (C) at larger ones. However, while these cellular level mechanisms are important, additional large-scale constraints that extend beyond those of individual cells are necessary for understanding long-term controls on N and P cycles, nutrient limitation, and organism-element cycle feedbacks in ecosystems.

An example of larger-scale feedback between N and P involves plant and microbial investments in phosphatase enzymes (McGill & Cole, 1981; Duff et al., 1994). These enzymes are rich in N and appear to be necessary for all forms of life (Duff et al., 1994), allowing recycling and acquisition of P by organisms (Colvan et al., 2001). Extracellular varieties of phosphatase are particularly relevant to nutrient cycling; once exuded into the soil by organisms, these enzymes act to mobilize P from ester-bonded forms, generating phosphate ions that are available for uptake and assimilation. In this way, phosphatases can reduce organisms’ short-term P deficits, and can be used as a proxy for decomposition and nutrient demand (Sinsabaugh et al., 2008).

However, phosphatase enzymes are rich in N, meaning that organisms must sacrifice N to get P via this path (Olander & Vitousek, 2000; Treseder & Vitousek, 2001; Allison, 2005; Houlton et al., 2008). As a result of this coupling between N and P cycles, phosphatase activity is probably highly responsive to changes in both N and P, representing an important strategy by which organisms might be able to adjust to human alternations of global biogeochemical cycles (Gusewell & Freeman, 2005; Wang et al., 2007; Vitousek et al., 2010).

Nevertheless, previous work has not sufficiently demonstrated relationships among N, P and phosphatase across biomes, despite strong theoretical predictions suggesting their existence (Spiers & McGill, 1979; McGill & Cole, 1981; Chen et al., 2002). This is partly attributable to the diversity of phosphatase enzymes (Richardson et al., 2005; Turner, 2008), as measurements of phosphatase activity may only measure a certain type of phosphatase or substrate (Richardson et al., 2005). Additionally, some ecosystem models are beginning to incorporate nutrient dynamics and even phosphatase activity, which has been shown to be important for understanding and predicting climate change at ecosystem to global scales (Wang et al., 2007; Houlton et al., 2008; Wang & Houlton, 2009; Wang et al., 2010b). Thus, understanding and testing hypotheses related to C, N, P and phosphatase are not simply an academic exercise; they have implications that extend well beyond the interactions themselves.

Here we examine the generality of phosphatase-mediated interactions between N and P cycles, using meta-analysis to explore responses of phosphatase to fertilizations with N, P and N×P across different terrestrial ecosystems (Fig. 1). In particular, we tested the following hypotheses.

Materials and Methods

To select appropriate studies for our meta-analysis, we searched Web of Science, entering combinations of the keywords nitrogen, phosphorus, fertilization, deposition, phosphatase, extracellular enzyme*, and enzyme-activity. We restricted our results to studies performed in natural terrestrial ecosystems. All studies measured phosphatase activity in experimental treatments with either N or P fertilization and in control treatments without fertilization; a few studies were also factorial in N and P. Finally, we selected only those studies that included the quantity of N and P fertilizer used for each treatment, as well as the mean, standard error of the mean, and number of replicates for each treatment. For studies that included phosphatase activity for multiple sites or fertilization levels, multiple data points were collected. For studies where phosphatase activity was measured at multiple time-points, only the final value was used to maintain consistency between studies. Where data were not presented in tables, DataThief (Tummers, 2006) was used to acquire numbers from figures.

We used meta-analysis to examine the core hypotheses (stated in the Introduction) (Hedges & Olkin, 1985; Hedges et al., 1999). Each data point in the meta-analysis includes two measures of phosphatase activity: the experimental treatment with fertilization vs the control treatment without fertilization. Each data point is summarized by a response ratio (RR), the log of the response ratio (L_RR), and the sampling variance (V). RR was calculated as the experimental mean divided by the control mean, representing an index of response magnitudes; L_RR was calculated as the log₁₀ of RR. Positive values of L_RR represent an increase in fertilized relative to control conditions, whereas negative values indicate suppressed phosphatase activity. The use of L_RR is favorable because it equally weighs the negative and positive responses and facilitates statistical analysis. V was calculated using the experimental mean, the standard error of the experimental mean, the number of experimental replicates, the control mean, the standard error of the control mean, and the number of control replicates (Hedges et al., 1999).

We divided the database into three fertilization classes: N, P, and N×P. Each class was summarized by the weighted mean of L_RR (L_RR*), the standard error of L_RR* (SE(L_RR*)), and the Q statistic (Q) (Hedges et al., 1999). L_RR* is calculated by giving greater weight to data points with a lower standard deviation and higher precision. The use of L_RR* increases the precision of the central tendency statistic, and L_RR* was used in place of the unweighted mean, because V differed between individual data points (Hedges et al., 1999). We calculated SE(L_RR*) to determine the standard error of the weighted mean (Hedges et al., 1999). The Q statistic was calculated and tested at the 95% confidence interval to account for differences in standard error of the individual studies (Hedges et al., 1999). Between-experiment variation is statistically significant when the Q statistic exceeds the critical value of the χ² distribution. These summary statistics were calculated for each type of fertilization: N, P, and N×P (Table 1). In addition, we performed a single sample t-test to determine if the mean of L_RR for each fertilization class is significantly different from zero (Table 2).

The data set was unbalanced; and so we used a linear mixed effects model. The fixed effects were the fertilization treatments and the random effects were the different studies. We performed a Tukey test in the statistical package R using ‘multcomp’ to factorially compare the different fertilization treatments (R Development Core Team, 2007). The Tukey test determined the estimates, standard error, z-values, and P-values for each comparison (Table 3).

We also subdivided each group into extracellular phosphatases measured on the root and in the bulk soil. Phosphatase activity measured on roots is assumed to reflect the plant and its symbionts, while phosphatase activity measured in the soil is assumed to account for both plant and microbe sources. We performed our analysis on roots, soils, and roots + soils across treatments (Table 1), and used the Tukey test to quantify differences between root and soil responses (Table 3).

Results

The data set spans eight major biomes, including grassland, shrubland, temperate coniferous forest, temperate deciduous forest, tropical dry forest, tropical rainforest (including subtropical and lower montane), tundra, and wetland (Supporting Information Table S1). Observations with N or P fertilization span all such biomes, while observations with N and P added together covered grassland, tropical rainforest, and wetland biomes. In all, we collated 195 observations from 34 separate studies. Of these observations, 83 measured phosphatase activity on plant roots; the remaining 112 measured phosphatase in the bulk soil. Twenty-two studies added N alone, four studies added P alone, four studies added N and P separately, and four studies added N and P in factorial combinations (Fig. 1).

Root and soil phosphatase activity increased substantially with added N and decreased with added P and N×P in combination (Fig. 2, Table 2). Under N fertilization, added N stimulated phosphatase activity by 46% relative to controls. However, Q was 771.08, indicating substantial experimental variation. Under P fertilization, added P depressed phosphatase activity by 29% relative to controls. Q was 425.29, showing that between-experiment variation was significant, although less than for added N. When N and P were added together, phosphatase was depressed by 39% compared with controls, similar to the case of P alone. Q was 89.25, showing less between-experiment variation than for added N or P. These trends were apparent in both plant and soil phosphatase measures (Fig. 3, Table 1), and across biomes (Fig. S2).

Positive responses to soil and root phosphatase were observed at relatively low N fertilization (Fig. 4a). N additions ranged between 1.525 and 20 g N m⁻² yr⁻¹, with positive responses apparent throughout the entire range beyond the lowest N input site (i.e. 1.525 g N m⁻² yr⁻¹). Negative responses of phosphatase were also observed, although in many fewer cases than for positive responses (Fig. 4a). Plant phosphatase responded significantly more positively to N fertilization than did soil phosphatase (Fig. 3, Table 3). Overall, phosphatase activity in response to N fertilization was significantly different when compared with P and N×P fertilization (P-value < 0.0001).

Rates of P fertilization ranged between 2.7 and 40 g P m⁻² yr⁻¹. With the exception of the highest level of fertilization, depression of phosphatase activity in response to P inputs was apparent over the entire range (Fig. 4). Plant and soil phosphatase both responded negatively to P fertilization (Fig. 3). Similarly, negative responses of phosphatase activity to N×P fertilization were observed for all data points (Fig. 4).

We plotted L_RR of phosphatase activity over the ratio of N added to P added in each study to visualize combined effects; the N : P ratio was 1 : 1, 2 : 1, or 4 : 1 in our synthesis (Fig. 4c). Plant and soil phosphatase both responded negatively to N×P fertilization (Fig. 4). Phosphatase activity under added P alone was not significantly different from that of N×P combined (P-value = 0.681).

Discussion

This meta-analysis demonstrates that extracellular phosphatase enzyme activities are highly responsive to changes in N and P supplies, among plant roots and bulk soils, and across a wide array of terrestrial ecosystems. Our findings support the first two hypotheses: added N stimulated while added P decreased phosphatase substantially. Such widespread regulation of phosphatase activity by N and P is consistent with conceptual (McGill & Cole, 1981; Olander & Vitousek, 2000) and numerical models (Wang et al., 2007; Houlton et al., 2008). Our results also support our fourth hypothesis: N×P led to decreased phosphatase activities in plants and soils, similar to effects of added P alone. We therefore conclude that increased N stimulates the activity of P mineralizing enzymes, that added P suppresses this activity, and that the effect of P appears to be stronger than the N effect, on average – with qualitatively consistent patterns for plants and soils, tropical to polar latitudes, and across many biomes (Fig. S1). The significant response ratio and large between-sample variance (Q > 0) show that phosphatase response to fertilization is consistent despite species, environmental, and methodological variation. Moreover, these couplings between nutrients and phosphatase are consistent with systemic regulation of the stoichiometry of N and P cycles, offering a mechanism to help explain N and P co-limitations on land (Elser et al., 2007; Harpole et al., 2011).

Although enzymes are thought to be affected by factors other than nutrients, our meta-analysis did not reveal strong effects of soil pH, mean annual temperature, or mean annual precipitation on the response of root and soil phosphatase to N and P fertilization (Fig. S2). Nor did our results depend on measurement technique, or type of phosphatase (i.e. acid or alkali; mono- or di-ester phosphates). For example, most studies only measure mono-phosphates, whereas both mono- and di-ester phosphates are cleaved by phosphatase. Additionally, most studies only measured acid or alkaline phosphatases, but in soils with heterogeneous pH conditions, both may exist. One concern is that phosphatase is generally measured in the laboratory at standardized pH; however, optimal pH varies between enzymes and soils (Turner, 2010; German et al., 2011). Nevertheless, despite a variety of environmental and experimental conditions, our results are qualitatively consistent with a model of enzyme production that assumes a set C and N costs of enzyme production under varying N and P levels (Allison & Vitousek, 2005).

Our results also point to some fundamental differences between plant and microbial responses to changes in N and P. Phosphatase activity measured on plant roots and in the soil showed similar trends, but root effects were stronger than bulk soil effects for single N or P addition (Table 3). These results suggest that plant phosphatases can respond more strongly to changes in resource availability than microbial phosphatases, and are consistent with those of previous studies of phosphatase activity (Colvan et al., 2001). This notion fits with our present understanding of stoichiometry: whereas terrestrial plants show relatively diverse C, N and P ratios and modes of interaction (Vitousek, 1984; McGroddy et al., 2004; Townsend et al., 2007), the stoichiometries of microbial biomass (Cleveland & Liptzin, 2007) and microbial extracellular enzyme activities (Sinsabaugh et al., 2008, 2009) are relatively conserved. Interpretation of soil phosphatase activity is complicated by the presence of inactive extracellular enzymes, and may be inhibited by inorganic phosphate (Nannipieri et al., 2011); enzyme turnover or enzyme affinity at low concentrations could help in quantitatively understanding measures of phosphatase activity.

Although this analysis revealed strong trends, we also noted deviations in phosphatase responses to N and P additions. In the case of N fertilization, only two of 47 data points indicate a suppression of plant phosphatase, or L_RR < 0. These two data points were for laboratory bioassay seedlings of Agrostis capillaries that were transplanted into experimental grassland sites (Johnson et al., 1999, 2005). For soil phosphatase with added N, 26 of 85 data points have L_RR < 0. These exceptions span temperate deciduous forest, grassland, tropical rainforest, tundra, and wetland biomes. Soil phosphatase activity showed a stronger negative response to added N than did plant phosphatase, pointing to either steeper sensitivities of the rhizosphere to nutrient fertilization or fundamental differences among free-living microbial and root/rhizosphere community responses, or both.

It is not clear why some observations deviate from the average responses of phosphatase to N and P enrichment, although neither soil pH nor climate appears responsible (Fig. S2). Environmental factors that affect nutrient availability, including duration of fertilization, soil depth, soil organic matter content, C : N : P ratio, microbial community composition, distance to rhizosphere, and the form of N added, are likely to affect phosphatase activity and may explain negative responses to N fertilization (Tarafdar & Jungk, 1987; Johnson et al., 1998; Olander & Vitousek, 2000; Sinsabaugh et al., 2005; Keeler et al., 2009; Naples & Fisk, 2010; Weand et al., 2010). Particularly important may be soil acidification resulting from N addition, which could change nutrient availability, microbial communities, and optimal phosphatase activity. Deviations in phosphatase response to added P were less than for added N, although soil phosphatase levels in soils beneath stands of N₂-fixing red alder (Alnus rubra) did not decrease with added P (Compton & Cole, 2001). Here, phosphatase activity was positively correlated with labile and organic bicarbonate P, rather than P fertilization, and the constancy of high levels of phosphatase supports the idea that N fixers invest substantially in phosphatase (Giardina et al., 1995; Zou et al., 1995; Houlton et al., 2008; Venterink, 2011).

Allison & Vitousek (2005) suggested that phosphatase responds more strongly to complex forms of P than phosphate additions – despite potential phosphatase inhibition by inorganic phosphate (Nannipieri et al., 2011). Additionally, phosphatase is active even in P-rich environments, and may be important for other processes, such as C acquisition (Heath, 2005). This might explain why other studies that measured phosphatase activity in soils without P fertilization did not find a negative relationship between P availability and phosphatase activity (Speir & Ross, 1978; Spiers & McGill, 1979; Speir & Cowling, 1991; Adams, 1992; Allison & Vitousek, 2005; Naples & Fisk, 2010). The contradictory results have been attributed to differences in study sites that affect background N and P availability (Spiers & McGill, 1979; Olander & Vitousek, 2000; Allison et al., 2007a; Sinsabaugh et al., 2008) or differences in species that affect nutrient assimilation and allocation (Gusewell & Freeman, 2005; Naples & Fisk, 2010). Allison & Vitousek (2005) fertilized soil in the lab and did not see a significant decrease in microbial phosphatase activity, possibly because microbes continually produce at least low levels of phosphatase.

The smaller sample size of 21 data points from four studies and the range of N : P ratios of fertilization somewhat limit our analysis of N×P interactions. The fertilization N : P ratios were 1 : 1, 2 : 1, or 4 : 1, much lower than that of plants (28 : 1; McGroddy et al., 2004), litter (45 : 1; McGroddy et al., 2004), soil (13 : 1; Cleveland & Liptzin, 2007), and soil microbes (7 : 1; Cleveland & Liptzin, 2007). A portion of the added P becomes unavailable through soil complexation, so the actual ratio of bioavailable added N : P is unknown and warrants further research (Havlin et al., 2005). Fertilizer with a high proportion of available P may conceal the threshold at which N and P co-limitation ceases.

The results of this study are consistent with those of correlative studies of phosphatase activity in the absence of nutrient fertilization. Our results show that as N increases or P decreases, phosphatase activity increases. Therefore, we predict that ecosystems subject to atmospheric N deposition will increase phosphatase activity, all else remaining equal. Several studies have shown increasing phosphatase activity across sites with increasing N availability or increasing N deposition (Turner et al., 2002; Bragazza et al., 2006; Gress et al., 2007; Chung et al., 2007), even in sites that are limited by P, or N and P together (Olander & Vitousek, 2000; Treseder & Vitousek, 2001; Hedin et al., 2003). Other studies have looked at phosphatase activity across gradients of P availability, and generally show a trend of increasing phosphatase activity with decreasing P availability (Wright & Reddy, 2001; Allison et al., 2007b; Wang et al., 2010a; Turner, 2010).

While these results strongly support theoretical understanding and show the importance of phosphatases, our analysis is bounded by two key issues. First, P uptake by organisms and phosphatase production cannot be directly inferred by measuring maximum potential phosphatase activity. Especially important is the fact that phosphatase is measured in plants and soils in the laboratory. Although phosphatases are clearly advantageous, the actual return per unit of phosphatase investment is largely unknown (but see Treseder & Vitousek, 2001; Wang et al., 2007). Future work with isotope labeling could help to address this question. Second, data are scarce in the tropics, despite evidence showing that this ecosystem has broad significance for global change (Townsend et al., 2011), with outcomes affected by the kinds of N×P×phosphatase interactions examined here (Houlton et al., 2008). Experiments in tropical sites, and across a range of soils and plant communities, are particularly warranted.

Nevertheless, our meta-analysis unequivocally shows that increasing N availability tends to increase P cycling rates, offering a path by which plants and ecosystems can adjust to changes in N and P supplies. These results have several implications for ecosystem responses to global environmental change, especially N deposition and elevated atmospheric CO₂. Increased atmospheric CO₂ may have a fertilization effect on net primary productivity, stimulating growth of plants (Norby et al., 2005). The extent to which plant growth increases in response to CO₂ concentrations depends largely on available nutrients, especially N and P (Shaw et al., 2002; Reich et al., 2006). N deposition is increasing globally, which may induce P limitation (Gress et al., 2007; Vitousek et al., 2010); however, the up-regulation of phosphatase activity suggested by this meta-analysis may delay the onset of P limitation (by years to decades), and could help to explain why P limitation is not immediately observed in response to N fertilization (Finzi, 2009). Finally, these interactions between N and P could help to explain the existence of nutrient co-limitations (see Elser et al., 2007); our results support the notion that added N enhances P conservation, thereby promoting nutrient equilibrations in a way that complements those between P and N fixation (e.g. Vitousek et al. 2010). These nutrient interactions suggest the complicated nature of coupled resource limitation: changes in the cycling of one nutrient can alter the availability of another.