We extracted results for plant biomass and tissue N concentration from N addition studies after building a database by searching the Science Citation Index (SCI) of the Institute of Scientific Information. References cited in a large number of N review articles and books were cross-checked to ensure inclusion of pre-1987 articles, which are not listed in the SCI. Any article published before June 2007 that met the following criteria was included in our analysis: reported responses of biomass and/or N concentration at the species level; species reported in the studies were those occurring in natural terrestrial ecosystems; means, standard deviations or standard errors with sampling sizes for both control and N addition treatments were provided. Articles that reported results on whole plants and parts of plants (e.g. leaf, shoot, root) were also included in our database. For the biomass analyses, different methods of biomass estimation (e.g. direct harvest for most herbaceous species and use of allometric relationships for some tree species) were accepted because we did not consider this to be a significant source of error in this analysis. Tree species at different ages were accepted. On the basis of these criteria, articles that reported responses to N addition at community level (Elser et al., 2007; LeBauer & Treseder, 2008) and agricultural and horticultural species were excluded from our analysis. In addition, results from other proxy variables were not included in our analysis. For example, plant height or size was not included in biomass analysis, and tissue N content (e.g. g per plant) was not used in the analysis of N concentration. For biomass, the preferred metric was biomass per unit area (g m−2), and other biomass data (e.g. g per plant) were transformed if information on plot area was provided in the paper. Otherwise, these data (e.g. g per plant) were also included in our analysis. Nitrogen concentration was transformed as N percentage (%) and used for analysis. Data with means and standard deviations for both control and N addition groups provided in the original articles were used directly. Data (means and some measures of variance) presented in graphs were extracted by digitizing the figures using SigmaScan (Systat Software Inc., San Jose, CA, USA).
In meta-analyses, independence of the data being synthesized is assumed, and including multiple results from a single study violates this assumption, leading to alterations in the structure of the data, inflating samples and significance levels for statistical tests (Wolf, 1986; Vander Werf, 1992). Therefore some researchers have advocated the inclusion of only one result from each study (Vander Werf, 1992; Koricheva et al., 1998; Liao et al., 2008) when considering the lack of independence to be a serious problem for meta-analysis. However, the loss of information caused by the omission of multiple results in each study may become a more serious problem than that caused by violating the assumption of independence (Hedges & Olkin, 1985; Gurevitch et al., 1992). Thus, many researchers have included more than one result from a single study in their meta-analyses (Gurevitch et al., 1992; Wooster, 1994; Curtis, 1996; Curtis & Wang, 1998; Maestre et al., 2005; Wang, 2007). Even though we made efforts to exclude duplicate results in different publications (e.g. some results published as ures in one paper and as tables in another), a large number of comparisons were used in our analysis because individual papers usually provided data from more than one treatment (e.g. varying N concentrations), different plant tissues (e.g. leaf, root), and/or different growing stages. Therefore, our estimates are not wholly independent. However, in order to minimize the degree of nonindependence in our study, we first averaged those data obtained in the same year for the same species under identical N treatment. Then, we conducted the analyses between biological realms and functional types again and found those patterns were unchanged (data not shown) when compared with the results using all data. On the other hand, we compared responses and sample sizes of different tissue types (leaf, root, branch/twigs, and wood) across woody species, in which distribution of tissue types could be more complex than herbaceous species. Although the uneven distributions of tissue types within categories could result in statistical biases, the proportions of each tissue type did not change substantially between categories (Table S4). Thus, all the results reported in our study were obtained by analyzing the data of all growing stages and tissue types.
We divided these data into two datasets: dataset 1 included only N addition without other resources (Text S1), and dataset 2 contained studies on N addition together with other resources (Text S2). In studies in dataset 2, if an effect of additional resources (e.g. CO2 enrichment) on plants was given, we took this effect as the control, and the combined treatment (e.g. CO2 enrichment plus N addition) as the N addition treatment. For dataset 1, because various parts of plants were included, we first roughly divided all data into above-ground tissue, below-ground tissue, and whole plants. We then compared responses between above-ground and below-ground groups. We found that the responses among functional types were not changed except for N concentration among growth forms (Fig. S1). We also detected that responses between above- and below-ground tissues were similar within a plant functional type. For example, biomass responses were greater for above- than for below-ground tissues in most plant functional types, whereas N concentration responses showed no differences between above- and below-ground groups in most functional types (Fig. S1). On the other hand, sample sizes of below-ground tissues were much smaller than above-ground tissues in all functional types. Thus, all the three categories (above-ground tissues, below-ground tissues, and whole plants) of plant biomass and N concentration were included in our database to increase the number of plant categories.
There is substantial variability (from 1 g m−2 in Lowe et al. (2003) to 100 g m−2 in Loveland & Ungar (1983)) in the amount of N added among different studies. In order to test whether the amount of N added has an impact on plant responses, we first roughly partitioned added N into low (< 10 g N m−2) and high amounts (> 10 g N m−2) with similar sample size (Table S5). In addition, the large sample size of some plant functional types (herbaceous species, woody species, grass, forb, tree, and shrub) allowed us to further divide N amount into the following classes: < 5, 5–10, 10–15, 15–20, 20–25, and > 25 g N m−2. For dataset 2, we categorized the N addition treatments into control (added N alone), facilitated (added N together with other resources), and limited (added N but limited by other abiotic factors). Because of the large sample size, we further tested the responses of added N together with phosphorus (P) addition and CO2 enrichment.
We also collected other background information relevant to the data from the papers, including latitude, temperature and precipitation. For those studies with no geographical and climate information, we used data from the study sites where the experiments were conducted. Treatments that contained adding water or changing temperature were not included in the analysis of the impact of climate on plant N responses. For this type of analysis, we divided this background information into the following classes – latitude: 0–10, 10–20, 20–30, 30–40, 40–50, 50–60, and 60–70°; mean annual temperature (MAT): < 0, 0–5, 5–10, 10–15, 15–20, 20–25, and > 25°C; and mean annual precipitation (MAP): < 300, 300–600, 600–900, 900–1200, 1200–1500, and > 1500 mm.