What explains variation in the impacts of exotic plant invasions on the nitrogen cycle? A meta-analysis


  • P. Castro-Díez,

    Corresponding author
    1. Departamento de Ciencias de la Vida (Unidad Docente de Ecología), Facultad de Biología, Ciencias Ambientales y Química, Universidad de Alcalá. Ctra. Madrid-Barcelona km 33, Alcalá de Henares, Madrid, Spain
    Search for more papers by this author
  • O. Godoy,

    1. Department of Ecology, Evolution & Marine Biology, University of California, Santa Barbara, CA, USA
    Search for more papers by this author
  • A. Alonso,

    1. Departamento de Ciencias de la Vida (Unidad Docente de Ecología), Facultad de Biología, Ciencias Ambientales y Química, Universidad de Alcalá. Ctra. Madrid-Barcelona km 33, Alcalá de Henares, Madrid, Spain
    Search for more papers by this author
  • A. Gallardo,

    1. Departamento de Sistemas Físicos, Químicos y Naturales. Universidad Pablo de Olavide. Ctra. Utrera km 1, Sevilla, Spain
    Search for more papers by this author
  • A. Saldaña

    1. Departamento de Ciencias de la Vida (Unidad Docente de Ecología), Facultad de Biología, Ciencias Ambientales y Química, Universidad de Alcalá. Ctra. Madrid-Barcelona km 33, Alcalá de Henares, Madrid, Spain
    Search for more papers by this author


Exotic plant invasions can notably alter the nitrogen (N) cycle of ecosystems. However, there is large variation in the magnitude and direction of their impact that remains unexplained. We present a structured meta-analysis of 100 papers, covering 113 invasive plant species with 345 cases of invasion across the globe and reporting impacts on N cycle-related metrics. We aim to explain heterogeneity of impacts by considering methodological aspects, properties of the invaded site and phylogenetic and functional characteristics of the invaders and the natives. Overall, plant invasions increased N pools and accelerated fluxes, even when excluding N-fixing invaders. The impact on N pools depended mainly on functional differences and was greater when the invasive plants and the natives differed in N-fixation ability, plant height and plant/leaf habit. Furthermore, the impact on N fluxes was related mainly to climate, being greater under warm and moist conditions. Our findings show that more functionally distant invaders occurring in mild climates are causing the strongest alterations to the N cycle.


Nutrient availability controls primary production, carbon sequestration, water eutrophication and soil fertility, among other ecosystem services. Different global change drivers have caused strong alterations to these cycles. For instance, direct human actions from industry, agriculture and farming have increased nitrogen (N) pools at global scales (Vitousek et al. 1997). Invasions of exotic plants represent another important disruption of nutrient cycles. Invasive plants, sensu Richardson et al. (2000), are exotic, human transported species which attain self-sustaining populations over considerable areas. The pathways by which plant invasions alter nutrient cycles include, among others, changes in the quality and quantity of litter entering the soil, changes in rates of atmospheric N-fixation and plant N uptake, changes in soil macro- and microbial communities, and alteration of soil properties and/or microclimatic conditions for decomposer activity (Vitousek & Walker 1989; Mack et al. 2001; D'Antonio & Corbin 2003; Yelenik et al. 2004; D'Antonio & Hobbie 2005; Mayer et al. 2005; Follstad Shah et al. 2010). The impact of invasive plants on N budgets at global scale is complex to assess because (1) most studies are limited to local scales (Vitousek & Walker 1989; Mack et al. 2001; Ehrenfeld 2010) and (2) the direction and magnitude of the impact may depend on particular species traits, on properties of the recipient site and/or on the interaction of both (Ehrenfeld 2010; Pyšek et al. 2012; Hulme et al. 2013). Accordingly, case studies with different histories of introductions report contradictory effects. For instance, exotic invasive plants have been observed to increase (Vitousek & Walker 1989; Haubensak et al. 2004; Jiang et al. 2009; Follstad Shah et al. 2010), decrease (Christian & Wilson 1999; Scott et al. 2001; González-Muñoz et al. 2013), or have no effect (Castro-Díez et al. 2009) on N pools in vegetation, litter and/or soils. The rates of N transfer between different ecosystem compartments, such as soil, litter or plant tissues, have also been found to increase (e.g. Ehrenfeld et al. 2001; Ashton et al. 2005; Hawkes et al. 2005; Castro-Díez et al. 2009; Leicht-Young et al. 2009), decrease (Johnson & Wedin 1997; Evans et al. 2001; Scott et al. 2001; Godoy et al. 2010) or remain unchanged (Stock et al. 1995; Mahaney et al. 2006; Castro-Díez et al. 2009) after invasion.

Motivated by the huge divergence of results, ecologists have been searching for generalizations. The first major review by Ehrenfeld (2003), following a vote-counting approach, reported that invasive plants tended to increase N pools in ecosystems and/or to speed up N transfer rates. Subsequent reviews by Liao et al. (2008) and Vilà et al. (2011), following a meta-analytical approach, and by Pyšek et al. (2012), using data mining tools, largely agreed with Ehrenfeld's results. Nevertheless, they also revealed large heterogeneity of impacts across studies, part of which was explained by the N-fixing capacity and woodiness of the invasive plant (Liao et al. 2008; Vilà et al. 2011). A recent critique of quantitative field studies on the impacts of plant invasions highlighted this heterogeneity and argued that identifying the sources of variation is of fundamental importance (Hulme et al. 2013). The literature suggests an array of factors that may mediate the impact of exotic invasive plants on the N cycle, including the properties of the invaded site, such as climate, insularity, aspect, age of invasion, soil type or resource availability (Stock et al. 1995; Scott et al. 2001; D'Antonio & Corbin 2003; Daehler 2003; Levine et al. 2003; Jiang et al. 2009; Ehrenfeld 2010; Pyšek et al. 2012; Hulme et al. 2013). An important source of variation can also come from whether particular key traits of the invading species are present in the recipient community (Chapin et al. 1996; D'Antonio & Corbin 2003). For instance, the impact of a N-fixing invader is expected to be greater if the invaded community lacks N-fixing species (Levine et al. 2003), or the impact of an invasive tree will be greater in a treeless community (Mack 2003). Finally, effects can be influenced by phylogenetic relationships among species, an evolutionary factor that can also bias meta-analyses of impacts (Chamberlain et al. 2012). Unfortunately, how these factors modulate the impact of invasive plants on the N cycle has not been tested at a global scale.

The major aim of this study was to identify factors that explain variation in impacts of exotic plant invasions on the N cycle by conducting a structured meta-analysis of published information. Our meta-analytical approach focuses on metrics that cover most of the N cycle in terrestrial ecosystems and can be related to N pools within ecosystem compartments (soil, litter, plants) and with fluxes across them (Fig. 1). Specifically, we tested whether mean effect sizes depend on methodological aspects (i.e. the study approach–experimental or observational–and the criteria to select the native control), on some properties of the invaded site (whether the impact was mediated by disturbance, residence time of the invader, climate, insularity or biome), and/or on biotic properties of both the invader and the recipient community (i.e. the functional and phylogenetic distance between invasive and native species, and/or the position of the invader in the phylogenetic tree). We also explored whether the species pool covered by quantitative field studies is representative of invasive plant species worldwide. Because N-fixation is the most obvious pathway of impact (D'Antonio & Corbin 2003), we specifically explored whether an over-representation of N-fixing invasive plants might influence the results.

Figure 1.

Schematic representation of the nitrogen (N) cycle. Grey diamonds indicate N pools in different compartments. Black arrows indicate rates of N transfer between different compartments. Numbers are explained in Table 1.

On the basis of the above information, we developed several predictions. Given that the physiological processes involved in the N cycle are limited by drought and low temperatures (Schindlbacher et al. 2010; Szukics et al. 2010; Guntinas et al. 2012), we expected stronger impacts when the recipient ecosystems occur in warm and wet climates (hypothesis 1). We also expected that plant invaders that are functionally more distant from the natives would have larger impacts (hypothesis 2). Given that island communities are typically poor in species and usually differ in taxa frequency from those of continents (Whittaker 1998), there were more chances for an invader to be functionally distant from natives, and therefore to cause stronger impacts (hypothesis 3). Given that functional differences between species may strongly depend on their evolutionary relatedness (Daehler 1998; Pyšek 1998; van Kleunen et al. 2007, 2010), we expected phylogenetic distance between the invasive and native species to be proportional to the impact on the N cycle (hypothesis 4). Finally, invaders belonging to particular clades in the phylogenetic tree, such as the N-fixing families Fabaceae, Elaeagnaceae, Betulaceae, Myricaceae, were expected to cause larger impacts than others (hypothesis 5).


Data collection

We searched the literature for metrics related to the N pools and fluxes of the N cycle (Fig. 1). In some cases, we found direct measures of the target pool or flux (i.e. ‘target variables’, e.g. N mineralization rate), but in others we found one or more metrics standing for the target pool or flux (i.e. ‘other accepted metrics’; for instance, litter lignin and litter C/N are proxies for litter decomposition rate). We performed a meta-analysis selecting those variables with enough case studies (≥ 10). Selected variables and the rationale for their inclusion in the analysis are summarised in Table 1.

Table 1. Variables selected to evaluate the impacts of exotic plant invasion on the N cycle. In some cases, other metrics were accepted when the target variable was not available or when both have been found to correlate in the literature (a negative sign means that they correlate negatively with the target variable). The part of the N cycle that each metric is related to is shown in the fourth column. The first column classifies target variable between pools or fluxes
CategoryTarget variableOther accepted metricsRelated witha
  1. a

    The number in parentheses refers to boxes and arrows in Fig. 1.

  2. b

    RGR-plant relative growth rate.

  3. c

    SLA-specific leaf mass.

N poolsAbove-ground plant mass N pools in above-ground vegetation (1)
 Leaf or shoot N concentration
 Litter mass N pools in plant litter mass (2)
 Litter N concentration
 Soil organic matter Soil N pools in organic matter (3)
 Soil mineral N Soil N pools in mineral form (4)
 Soil total N (3) and (4)
N fluxesLitter decomposition rate

Litter lignin (-)

Litter C/N (-)

Litter decomposition rate (6 and to a minor extent 7)
 N Mineralization/nitrification/ammonification rate N transfer rate from soil organic matter to mineral soil (7)
 Plant RGRbSLAcVegetation N uptake rate from soil (8), litter production rate (5)

On 12 March 2010 we searched for articles reporting impacts of invasive terrestrial plant species invasions on any component of the N cycle, using ISI Web of Knowledge, with no restriction on publication year. Three search formulae were used: (1) (Exotic* OR alien* OR invasive*) AND (native*) AND (plant*) AND (leaf OR leaves OR litter) AND [LMA OR SLA OR SLM OR nitrogen or phosphorous OR lignin OR decomposition or C:N or C/N or (life SAME span) or turnover]. (2) (Exotic* OR alien* OR invasive*) AND (native*) AND (plant*) AND (nitrogen) AND (soil*) AND [mineralization OR nitrification OR ammonification OR pH OR respiration OR (microb* SAME biomass*) OR fixation OR (organic SAME matter)]. 3) (Exotic* OR alien* OR invasive*) AND (native*) AND (plant*) AND [RGR OR NPP OR (growth SAME rate) OR (primary SAME product*)] NOT (estuar* or marsh* or seawee* or macrophyte* or demograph* or mosquito* or earthworm* or amphipod*).

All formulae searched for papers comparing native and exotic invasive species or invaded/non-invaded communities. However, the first formula searched for data on leaf/litter properties or decomposition rates, the second searched for soil properties and rates of N transformations, and the third for plant productivity or relative growth rate. We also surveyed the references in relevant articles we retrieved and added other recently published articles which were considered relevant on the topic. This search resulted in a set of 420 papers which was subsequently filtered out to reject those that did not meet the following conditions:

  1. Studies were carried out in terrestrial ecosystems. We excluded aquatic ecosystems because their N cycle is strongly affected by water flow, which moves N within the ecosystem and creates a set of pools (i.e. water, sediments and water-sediment interface) different from those of terrestrial ecosystems.
  2. Studies either examined natural invasions in the field (observational studies) or performed experiments in common gardens or glasshouses (experimental studies).
  3. Studies explicitly mentioned the identity of the invasive species causing impacts, and compared any of the variables in Table 1 between the invasive species and coexisting natives, or between invaded and non-invaded sites.
  4. It was clear that the term ‘invasive’ refers to exotic species, with self-sustained populations and with the potential to spread far beyond the introduction site.
  5. Variables were measured simultaneously in the invasive species/invaded site and in the native species/non-invaded site in the same (micro)environmental conditions, so that the influence of other factors on the effect size was minimised.
  6. The study reported average values of the variables, number of replicates and reliability (standard deviation or standard error) for all measurements.

These criteria reduced the initial number of papers to 100 (see Appendix S1), 26 and 30 of them were also covered by the meta-analyses of Liao et al. (2008) and Vilà et al. (2011) respectively.

Among the selected papers, we adopted the following criteria to select case studies or units of analysis (i.e. the unit for calculating the effect sizes):

  1. If the same study reported data for more than one independent pair of species/sites, each pair was considered as an independent case study (Borenstein et al. 2009), matching the criteria of other meta-analyses (Liao et al. 2008; Benayas et al. 2009; Bertheau et al. 2010; Vilà et al. 2011).
  2. If the study reported more than one value for the same pair (for different sites, treatments or times), we pooled effect sizes and variances among all values reported for the same pair by doing a separate meta-analysis, following the formula to combine effect sizes across case studies in a grand mean effect size (d+, see below). The estimated pooled mean effect size and the mean variance were used in the final data set (Borenstein et al. 2009).
  3. In multispecies studies which explicitly assigned invasive species to native pairs, we used these pairs as independent observations, and noted which of the following criteria met the native control: (1) high abundance, (2) growth form similar to the invasive, (3) same genus or family as the invasive.
  4. When there was one native control for two or more invasive species, each invasive-control combination was considered as a separated case study.
  5. When several native controls were available for the invasive species, different criteria were used to select the native control: (1) random selection, 2) the most abundant, 3) the most similar to the invasive (according to the functional traits listed below) or 4) pooling the effects of all potential invasive-native pairs, (as in van Kleunen et al. (2010)). In those cases where information on abundance was missing or all the available natives were equally abundant and/or shared functional properties, we applied a random selection for criteria 1, 2 and 3, with the restriction that a different species had to be selected by each. By performing all analyses separately on the data sets resulting from each selection criterion, we tested the influence of the native selection criteria on the outcome of the meta-analysis. In the case of criterion 4), hypotheses 2 and 4 were tested with fewer cases, because species pooling prevented the computation of some functional and phylogenetic distances (see below).

For each case study, we recorded information regarding characteristics of the study, abiotic properties of the invaded site, and biotic properties of the invasive plant and native control (see Table 2). Most of these characteristics were obtained from the papers. Whenever average climatic properties (mean annual temperature and precipitation) were not reported, we obtained them from the WorldClim database (Hijmans et al. 2009), on the basis of the coordinates or site names reported in the article. WorldClim provides average climatic values for the period 1950–2000. Although the use of averages of climatic series did not allow us to test the effects of within-site climatic variability, these effects were assumed to be negligible compared to those driven by the broad climatic gradient covered by our data set. Whenever possible, species (both invasive and native) were coded according to the following traits: woodiness (1: woody, 0: non-woody), self-support (1: self-support, 0: vine or climber), height (0: < 1 m, 0.5: 1–5 m, 1: > 5 m), N-fixation ability (1: fixing, 0: non-fixing) and plant/leaf habit (1: evergreen or perennial plant, 0: deciduous or annual plant). Subsequently, we subtracted the value of the native from the value of the invasive in each trait, which gave us a magnitude and a direction of each trait change upon invasion. To obtain an overall magnitude of the ‘functional distance’ between the invasive and the native species (irrespective of the direction) we summed up the absolute values of all trait differences. When the target variable was measured in the ecosystem rather than in a species (e.g. soil properties), the functional distance was only computed if the authors made it clear that the property could be assigned to a single species (e.g. if the soil was collected in a site with a 100% covered by a single species). The information on species traits was obtained from the papers, Floras and online databases (Appendix S2).

Table 2. Factors used to explain the heterogeneity of plant invasion effect size across studies by means of structured meta-analysis. The third column indicates the data set to which it was applied (full data set with all pools and fluxes; dataset with pools; data set with fluxes). Shading helps to keep track of the same piece of information across columns [Correction added on 2 December 2013, after first online publication: ‘Grey rows indicate N pools in different compartments’ has been changed to ‘Shading helps to keep track of the same piece of information across columns’ in the caption of Table 2.]
Structure factorCategories/Values of the factorData set to which it was applied
  1. a

    SOM- Soil organic matter.

  2. b

    Soil N min- Soil N in mineral form.

  3. c

    Only when the invaded site is an island.

  4. d

    Initially we included ‘Savannah’, ‘Subtropical desert’ and ‘Boreal forest’, but we considered that the number of cases was not large enough to obtain representative patterns (N < 11) so we do not show them in results.




Pools + fluxes
Type of pool

In litter

In aboveground vegetation

In soil in mineral form

In soil organic matter

Type of flux



SOMa→ Soil N minb

Study type




Pools + fluxes
Removal experiment

Yes (the control is a site where the invasive plant was removed)

No (the control is a non-invaded site)

Pools + fluxes
Impact mediated by disturbance



Pools + fluxes
Residence time

1: < 50 years

2: 50–100 years

3: 100–200 years

4: 200–500 years

5: > 500 years

Pools + fluxes

The invaded site is an island

The invaded site is a continent

Distance to the continentc(Continuous)All

Tropical forest


Temperate forest

Temperate grassland

Mean annual precipitation(Continuous)All
Mean annual temperature(Continuous)All
Invasive plant is N-fixing



Functional distance(Continuous)All
Woodiness distance(invasive value–native value)All
Self-support distance(invasive value–native value)All
Plant height distance(invasive value–native value)All
Plant/leaf habit distance(invasive value–native value)All
N-fixation distance(invasive value–native value)All
Phylogenetic distance(Continuous)All

Phylogenetic relatedness was assessed by building a pruned phylogenetic tree with all invasive and native species as terminal tips using the maximally resolved seed plant tree (R20120829) available in Phylomatic (http://phylodiversity.net/phylomatic/). We calibrated the resulting tree by dating the nodes in Phylocom 4.1 with the Branch Length ADJustment function (BLADJ), on the basis of the lognormal clade age estimation of Bell et al. (2010). Ferns were excluded. With this phylogenetic tree, we estimated the phylogenetic distance for each invasive-native pair and we additionally tested for the phylogenetic signal of the impacts of the invasive species across the dataset (i.e. whether differences in the effect size of invaders depends on their position in the phylogenetic tree) using the K-statistic (Blomberg et al. 2003) in the R package Picante (Kembel et al. 2010). For both the phylogenetic distance and the phylogenetic signal, we conducted these analyses with the four pairing criteria described above.

Finally, to assess whether N-fixing invasive species were over-represented, we compared the proportion of N-fixing invasive plants and the proportion of case studies focused on N-fixing plants in our database with the N-fixing invasive species reported in several databases/checklists throughout the world (see Appendix S3).

Statistical analysis

The unit of analysis was a pair of invasive (i) and native species (n) or a pair of invaded (i) and non-invaded sites (n). For each case, Hedges' d was used to estimate the effect size because (1) it weighs cases by their number of replication and the inverse of their variance, and (2) it accounts for the effects of small sample sizes (Rosenberg et al. 2000). Hedges' d for a given case study was calculated as follows:

display math

where S is the pooled standard deviation and J is a weighting factor based on the number of replicates (N) in each case. J was calculated as follows:

display math

The variance of Hedges' d was computed as follows:

display math

Hedges' d ranges from −∞ to +∞. Largest effect sizes come from those cases with a large difference between the invasive and the native species (or invaded and non-invaded site) and low variability. A positive value of d indicates that the target variable in the invasive plant/invaded site has a larger value than the native control.

We performed the analysis in the following way:

  1. Synthetic analysis of all pools and fluxes: We used the target variables and other accepted metrics in categories ‘pools’ and ‘fluxes’ (see Table 1). Given that some studies reported more than one variable related with the same pool or flux, we applied the following rules to avoid pseudo-replication: (1) the ‘target variable’ was preferred over ‘other accepted metrics’; (2) we pooled effect sizes and variances for several target variables or other accepted metrics in the same type of pool or flux and in the same case study, by doing separate meta-analyses (see below computation of d+). The variable ‘Soil total N’ was excluded because it cannot be assigned to a single pool category (see Table 1). The final data set contains 113 exotic invasive species and 345 case studies with only one effect size for each case study in each type of pool or flux.
  2. Separate analysis of pools and fluxes: The data set was split into pools and fluxes, and two separate meta-analyses were performed on each subset.
  3. Separate analysis of each target variable: We performed additional meta-analyses for each raw variable (Table 1), including soil total N, which added two new case studies to the initial data set.

In each analysis, effect sizes across all comparisons were combined using the random effects model to provide the grand mean effect size (d+), where the weight of each case is the reciprocal of its sampling variance. A random effects model was preferred because it accounts for the fact that, in addition to sampling error, there is a random component of variation in effect sizes among studies. The effect was considered significant if the bias-corrected 95% bootstrap-confidence interval (CI) of d+, calculated with 999 iterations, did not bracket zero (Rosenberg et al. 2000).

For each grand mean effect size calculation, we computed the total heterogeneity of effect sizes across studies by means of the QT statistic as follows:

display math

where n is the number of cases, di is the effect size of case i and wi is the reciprocal of the sampling variance. A significant QT, tested against a chi-square distribution with n-1 degrees of freedom, indicates that the variance of effect sizes among studies is greater than expected by sampling error and implies that there may be some underlying structure to the data. In those cases, we performed structured meta-analyses, using all factors listed in Table 2 (characteristics of the study, abiotic and biotic properties), to explain heterogeneity of effect sizes. To test whether mean effect sizes of variables differed between the levels of each categorical factor (e.g. between islands and continents, between N-fixers and non N-fixers), we assessed the significance of the between-group heterogeneity (QB) with a chi-squared test (Rosenberg et al. 2000). Finally, we tested whether variation in effect sizes covaried with the continuous factors (precipitation, mean temperature of the invaded site, functional and phylogenetic distances between invasive and native species, etc.) by using weighted least squares regression. The amount of heterogeneity explained by the regression model (QM) was tested against a chi-square distribution with 1 degree of freedom to assess its significance. If the model was significant, we further assessed the slope of the model and its significance.

Meta-analysis results may be affected by publication bias, i.e. the selective publication of articles finding significant effects over those which found non-significant effects (Begg 1994). We explored this possibility by correlating the standardised effect size with the sample size across studies. We found a negative (Spearman = −0.032 to −0.064, depending on the pairing criteria) but non-significant (> 0.24 in all cases) correlation, which indicates that larger effect sizes in one direction are not more likely to be published than smaller effect sizes (Rosenberg et al. 2000). Besides, a plot of effect sizes vs. sample sizes across studies revealed a funnel-shaped distribution (Appendix S4) which is expected in the absence of publication bias (Palmer 1999). A plot of the standardised effect sizes against normal quantiles followed a straight line, indicating that effect sizes are normally distributed (Rosenberg et al. 2000) (Appendix S4). Finally, the fail-safe number (i.e. the number of null results (either non-significant, unpublished or missing studies) that would have to be added to make the overall test of an effect statistically non-significant) was calculated as a measure of the strength of the result (Rosenberg et al. 2000). We found values over 29925, which were larger than 5N+10 = 1735, where N is the total number of cases in our analysis. This means that the observed result can be considered as a reliable estimate of the true effect (Rosenberg 2005).


Characteristics of the database

Our data set covered 345 case studies, which included 113 exotic invasive species (16 N-fixing and 97 not N-fixing). Among these species, the most represented were Robinia pseudoacacia with 13 cases, Acacia saligna with 11, Microstegium vimineum and Bromus tectorum with 10, and Elaeagnus angustifolia, Ailanthus altissima and Berberis thunbergii with 9 cases each. Among the 46 families of invasive species Poaceae with 75 cases was the most common, followed by Fabaceae (54 cases), Asteraceae (30) and Elaeagnaceae (12). The proportion of N-fixing invasive plants across different databases worldwide varied from 2% in New Zealand to 32% in the United States, being 15% on average (Appendix S3). This was similar to the proportion of N-fixing species covered by our data set (14%). However, N-fixing species accounted for 21% of the case studies, suggesting that they were over-selected in studies testing the effects of invasive plants on the N cycle. The geographical distribution of case studies was uneven, 73% of them occurring in North America and Europe. In contrast, Africa was only represented by 5% (all occurring in South Africa), Australia + New Zealand by 5%, Asia by 3% and South America by 1%. It is noteworthy that the Hawaiian Islands contributed 11% of the cases (Appendix S5). The average number of target variables reported for each case study was 2.4, with a maximum of 9. The most frequently reported variable was specific leaf area (SLA) with 58 case studies, followed by litter decomposition rate, soil total N, litter N and soil mineral N. Less often reported were litter lignin and litter mass, with 10 and 14 cases respectively (Appendix S6). In 41% of the case studies comparing traits between species pairs, the native control was reported to be abundant. In 53%, it belonged to the same growth form as the invader and in 28% it belonged to the same genus or family as the invader. Finally, in 70 case studies (20%), more than one native species was available for each invader, and we selected native controls using the four previously defined criteria. Results were consistent across selection criteria (see Appendices S7–S8), thus we focus on the random selection criterion because it represents a balance between all criteria (see Appendices S7–S8 for the full set of results for the four criteria).

Impacts of plant invasion on the N cycle

We found a positive effect of plant invasions on all the N cycle-related metrics considering the full data set (pools + fluxes) (d= 0.63, bootstrap 95% confidence interval 0.41–0.82), although the effect disappeared when the native control was an invader-removed site (heterogeneity between case studies with each type of control (QB) = 7.30, < 0.05, Appendix S7). We also found that the effect size was similar for pools and fluxes (Appendix S7), but as expected, there was large heterogeneity among case studies (QT = 805, < 0.001). Each N pool increased with a similar magnitude following invasion (QB = 3.54, = 0.49, Fig. 2). However, the effect on N fluxes varied (QB = 10.05, < 0.05), so that the largest acceleration was found on the N transfer from soils to plants, while the N transfer from plants to soils was not affected by invasions (Fig. 2). Correlations of effect sizes among target variables reported within the same studies revealed that several plant invasion impacts covaried (Appendix S9). For instance, increases of litter N upon invasion coincided with increases in litter mass, litter decomposition rate and soil mineral N; increases in soil organic matter also coincided with increases in soil mineral N.

Figure 2.

Mean effect size (Hedges' d) of plant invasions impacts on the overall nitrogen (N) pools (first row) and on different N pools across the ecosystem (a) and on the overall N fluxes (first row) and on different fluxes among compartments (b). Lines indicate 95% bootstrap-confidence intervals. A mean effect size is significantly different from zero when its 95% confidence interval does not bracket the zero line. Positive mean effect size indicates that the invasive species possess larger value for the trait than the native species. Sample sizes for each trait are indicated in parentheses.

Determinants of the impacts on the N cycle

Among all factors considered for explaining variation in effect size (Table 2), climate and invader-native functional distance were the most important, insularity and phylogeny had an intermediate effect, and study type, residence time, or if the impact was mediated by disturbance were poor predictors (Appendices S7–8).

In line with hypothesis 1, N fluxes were more accelerated following invasions in moist and warm locations, and this was robust to the removal of case studies with extreme effect sizes (see Fig. 3). Accordingly, effect sizes on N fluxes were larger in the Tropical Forest biome, as compared with Temperate (grasslands or forests) or Mediterranean biomes (QB = 12.6, < 0.05, Fig. 4a, Appendix S7). Functional distance between the invasive plant and the native control did not explain heterogeneity of impacts, but some of its components did so (partial support to hypothesis 2). Specifically, the impact on N pools was larger when the invader was either N-fixing, tall, annual/deciduous or any combination of the three traits, and the native control was non-N-fixing, short and/or perennial/evergreen (Fig. 5, Appendices S7–S8). Plant/leaf habit distance similarly explained the variation of invasions impact on N fluxes (Appendix S8). Effect sizes on N pools + fluxes were also larger when the invaders were N-fixers (QB = 9.75, < 0.01, Appendix S7). Although removing these species from the analysis reduced the effect size, it remained positive and significant (d+ = 0.50, bootstrap 95% confidence interval 0.30 to 0.72).

Figure 3.

Variation of the plant invasion effect size (Hedges' d ) on ecosystem nitrogen (N) fluxes according to the mean annual precipitation (a) and mean annual temperature (b) of the invaded site. QM is the heterogeneity explained by the model and its significance. Arrows mark cases with extreme effect size, which might have a large influence on the result. Therefore, analyses were repeated without each of these cases and without all of them. Results for the mean annual precipitation model ranged from QM = 26.26 to 35.51, < 0.001 in all cases. Results for the mean annual temperature model ranged from QM = 6.756–10.52, < 0.01 in all cases.

Figure 4.

Variation of the plant invasion effect size (Hedges' d) on ecosystem nitrogen (N) fluxes according to the biome (a) and the insularity (b) of the invaded site. Lines indicate 95% bootstrap-confidence intervals. Statistics for these analyses are shown in Appendix S7.

Figure 5.

Variation of the plant invasion effect size (Hedges' d ) on ecosystem nitrogen (N) pools according to three components of the functional distance between invasive and native plants, namely, plant height, N-fixation and plant/leaf habit. QM is the heterogeneity explained by the model and its significance.

The effect sizes of invaders on N fluxes were larger on islands than in continents (Fig. 4b, Appendix S7), partially supporting hypothesis 3. In addition, when impacts on pools and fluxes were considered together, the effect size in islands increased with the distance to the continent (QM = 6.74, < 0.001, Appendix S8). However, this was driven by the over-representation of cases occurring in the Hawaiian Islands.

In contrast to our expectation (hypothesis 4), phylogenetic distance between invasive and native plants did not influence the impacts (QM = 0.30, = 0.58, Appendix S8). We also found no phylogenetic signal on the impact size of invaders either on the overall dataset, in N pools or in N fluxes (Appendix S10). However, when different N pools were assessed separately, we found a significant effect on the N pool of aboveground biomass; grass invaders (Poaceae) had a moderate negative effect, whereas N-fixing families (Fabaceae and Elaeagnaceae) increased the N pool in standing biomass up to four times more than the rest of the invaders (Appendix S10–11).


Effects of invasions on pools and fluxes

We found that exotic plant invasions accelerated N fluxes and increased N pools, in agreement with previous studies (Ehrenfeld 2003; Liao et al. 2008; Vilà et al. 2011). This result was robust to variation in some methodological aspects (such as criteria to select the native control or whether the approach was experimental or observational) and in properties of the invasion context (i.e. residence time, whether the impact was mediated by disturbances). This trend also was robust to the removal of N-fixing invaders from the analysis.

Our results indicate that plant invasions tend to increase N inputs to the ecosystem. The fact that the effect size on N pools notably declined after the removal of N-fixing invaders (Appendix S7) indicates that the main mechanism explaining this pattern is an increase of N-fixation, as previously suggested (Chapin et al. 1996; D'Antonio & Corbin 2003; D'Antonio & Hobbie 2005). However, non-N fixing invaders still tended to increase pools (Appendix S7), suggesting that additional mechanisms may operate. For instance, non-symbiotic N-fixation by microbes of the rhizosphere has been found to increase in some invaded systems (Williams & Baruch 2000), although the reverse has also been reported (Ley & D'Antonio 1998; Mack et al. 2001). Non-N fixing invasive plants may increase N pools in the ecosystem using it differently from natives (e.g. taking up N at different soil depths, at different times or accessing to different forms of N than native species), which may increase total nutrient use in the ecosystem and reduce N losses of mineral forms by leaching (Fargione et al. 2003; D'Antonio & Hobbie 2005; Ehrenfeld 2010). Alternatively, the high N pools of invaded sites may also be the cause, rather than the consequence, of invasions, because some disturbances that increase N availability (e.g. fertilization or N deposition) may promote plant invasions (Davis et al. 2000; D'Antonio & Corbin 2003; D'Antonio & Hobbie 2005). However, this information was rarely reported.

We did not detect changes in the distribution of N across categories of pools, but did find a wide variation of effect size within each ecosystem compartment. In fact, we might expect very different redistribution patterns depending on the nature of the invasive plant and the invaded community. For instance, if invasive herbaceous plants replace native trees, the main increase in ecosystem N would occur in litter or in soil, but not in above-ground plant biomass, whereas the reverse would be expected if invasive trees invade native grasslands. Unfortunately, the low number of case studies with different combinations of invasive-native growth forms did not allow us to test this hypothesis. Our results are in contrast to those by Liao et al. (2008), who reported a higher N increase in plant roots, followed by plant shoots, microbes and soil. However, in the case of N pool in roots, the seven case studies they reported are not sufficiently representative to merit general conclusions.

The overall trend for faster N fluxes following invasion agrees with previous reviews (Ehrenfeld 2003; Liao et al. 2008; Vilà et al. 2011) and is consistent with the hypothesis that invasive plants usually possess traits associated with faster N turnover, such as low leaf construction costs, high leaf N content and short leaf life span (Pattison et al. 1998; Durand & Goldstein 2001; D'Antonio & Corbin 2003; Pyšek & Richardson 2007). The finding that plant N uptake from soil was the most accelerated flux upon plant invasions (Fig. 2) is consistent with reports showing that invasive plants often grow faster (D'Antonio & Corbin 2003; Grotkopp & Rejmánek 2007; Leishman et al. 2007; Pyšek & Richardson 2007; van Kleunen et al. 2010), and therefore have a higher potential for soil N uptake. The non-significant impact of plant invasions on litter decomposition rate contrasts with the low C/N ratio found in invasive litter (Appendix S6), and may be partly explained by a high content of lignin, as this was found in leaves/shoots of many invasive plants (Ehrenfeld 2003; Knight et al. 2007; Godoy et al. 2010). Lignin bonds with protein N and produces complex molecules difficult to attack by microorganisms (Gallardo & Merino 1992, 1993), leading to an increase of N trapped in the litter and in the soil organic matter, which is not readily available for uptake by most terrestrial plants. Contrastingly, plant invasion enhanced N mineralization (Fig. 2, Appendix S6), maybe due to a larger or more balanced supply of resources to microorganisms by exotic litter (Rothstein et al. 2004; Chapuis-Lardy et al. 2006; Castro-Díez et al. 2009; Strickland et al. 2010), to an improvement of the microenvironmental conditions in the soil for microbial activity (Chapin et al. 1996; Mack & D'Antonio 2003; Norton et al. 2004; Marchante et al. 2008), and/or to a change in the composition of the soil biota (Hawkes et al. 2005; Peltzer et al. 2009; Rodriguez-Echeverria 2010).

What explains differences of impacts across studies?

By considering methodological aspects of the studies, properties of the recipient ecosystem and biotic properties of both invasive and native plants, we were able to identify new sources of cross-studies variation of plant invasion impacts on the N cycle, that were not considered in previous reviews (Liao et al. 2008; Vilà et al. 2011). The larger acceleration of N fluxes occurring under warmer and moister conditions (Fig. 3, hypothesis 1), may be explained by the fact that physiological processes and enzymatic reactions (as those involved in nutrient cycles) are generally faster at higher temperatures (Wallenstein et al. 2011), particularly if there is no other limitation, such as water shortage. This may be useful for predicting impacts of invasive plants across locations. However, predictions over time at local scales cannot be made, mostly because we could not include (i.e. poorly described in the original sources) the effect of other variables likely affecting more strongly the size of the impact at local scales, such as history of introduction, human use and soil properties.

The fact that N-fixing plants caused stronger impacts on N pools when they become invasive, especially in communities lacking N-fixers (Fig. 5), had already been suggested by previous reviews (Chapin et al. 1996; D'Antonio & Corbin 2003; Liao et al. 2008; Vilà et al. 2011). Besides, a large plant size may magnify the N-fixing impact to the ecosystem. Moreover, invaders with short-lived tissues (either entire plants or leaves) invest less in defence but more in production and have shorter turnover rates (Herms & Mattson 1992; Reich 1993; Castro-Díez et al. 2000); therefore they tend to accelerate N fluxes when compared to natives with long-lived tissues. This is why the most dramatic examples of invasive plants impacts on N pools involve situations with a combination of at least two of these features (e.g. Vitousek & Walker 1989; Rice et al. 2004; Yelenik et al. 2004). The failure of our functional distance to explain heterogeneity in the N cycle impacts (hypothesis 2) was due to the fact that this metric is an absolute value (we summed up the absolute values of particular distances to avoid they cancel each other out). Indeed, the three components of the functional distance which explained heterogeneity (N-fixation, plant height and plant/leaf habit distances), showed that distances with the same absolute value but different sign corresponded with different values of effect size (Fig. 5).

Consistent with hypothesis 3, the acceleration of N fluxes caused by invasive plants was larger in islands than in continents (Fig. 4b). Also consistent with the hypothesis, the functional distance between invaders and natives was larger in islands (0.93 ± 0.14 vs. 0.52 ± 0.06, Student's = 2.71, = 0.01). However, this cannot explain the larger impacts of invaders in islands because functional distance did not affect the impact size on N fluxes (Appendix S8). An over-representation of cases of N-fixing invaders in islands did not explain this result either, as this proportion was larger in continents (11 and 23%, respectively, in the full data set, and 6 and 21% in the N fluxes data set). Alternatively, this result may be explained by the milder climatic conditions typically found in islands, as compared with inland sites located at equivalent latitudes, and/or an over-representation of cases in Hawaii (i.e. a moist tropical climate). In fact, island sites in this study were on average wetter and warmer than continental sites (mean precipitation ± SE = 2227 ± 135 mm and 751 ± 24 mm, respectively, Student's t = 10.75 P < 0.001; mean annual temperature = 18.9 ± 0.6 and 12.6 ± 0.3 °C for islands and continents, respectively, Student's t = 9.19 P < 0.001). Therefore, more studies in non-tropical islands are necessary to unravel whether the insularity effect was confounded with the climatic effect.

Phylogenetic distance between invaders and natives was not a good predictor of plant invasions impact either on N fluxes or on N pools, not supporting hypothesis 4. The poor correlation between phylogenetic distance and the distance for the three functional traits more relevant to this pattern -N-fixing ability, plant height and plant/leaf habit- could explain this result. Indeed, in our broad tree including disparate families, N-fixing ability was restricted to Fabales and a single species within Rosales (E. angustifolia). Moreover, plant height and plant/leaf habit are labile traits within families, so that several families have tall and short species (e.g. Rosaceae, Solanaceae) and perennial and annual species (e.g. Poaceae). Similarly, impacts on the N cycle did not depend on the position of invasive species in the phylogenetic tree (little support for hypothesis 5). Only the impact on the N pool in the aboveground vegetation showed a phylogenetic signal, because of the opposite effects that N-fixing clades and the Poaceae had on this pool. This reinforces the idea that, although some families may have certain traits that can potentially increase or decrease N pools, their final impact seems to be related more to the interaction between invaders traits and the biotic and abiotic properties of the invaded site.

Limitations of the dataset inherent to published literature

We present the most extensive review conducted to date on the impact of invasive plants on the N cycle. However, our database was sensitive to the bias that the literature on plant invasion impacts has in the selection criteria of species and target variables (Hulme et al. 2013). For instance, our analysis included a small fraction (N = 113) of invasive species worldwide, which is estimated in the thousands (Hulme et al. 2013). In addition, some taxa were over-represented (the seven most studied taxa accounted for 21% of the case studies). Some N-fixing invaders are over-selected, probably because they represent the most obvious and dramatic examples. Despite these limitations, our work shows that, even not considering the effect of N-fixers, invasive plants still generally increase N pools and accelerate N fluxes. We also found a gap of knowledge for several continents, as most studies were conducted in North America and Europe. Therefore, more research effort is needed in the remaining (sub)continents. Finally, we found that the effect size of invaders for several target variables were correlated, suggesting that impacts on the N cycle can be assessed by exploring a few easy-to-measure key effects of the invaders. For instance, an increase in litter N upon invasion may involve impacts in other several ecosystem N pools (e.g. litter and in soil) and in N fluxes (e.g. litter decomposition rate).

Another limitation of this meta-analysis relies on the fact that the effect size of invaders was often calculated using a single native species as representative of the entire native community. A more realistic approach would be to use the average community trait value, weighed by species abundances, as native control. However, detailed information on community structure is rarely reported. This is an important point to be considered by ecologists in the future. Until then, we faced this shortcoming by exploring different rationales to select the native control. The fact that our findings were similar across criteria gives strength to our approach and suggests that all available native pairs for the same invader did not largely differ in their implications to the N cycle.


Using a meta-analytical approach we have shown that the wide variation of plant invasion impacts on the N cycle can be explained by certain characteristics of the studies, plant functional properties and environmental conditions. A relevant outcome was that moist and warm environments are more vulnerable to alterations to the N cycle following invasions by exotic plants, presumably because such conditions exacerbate the acceleration of N fluxes. In addition, we found that when the invasive plants are more functional distant from the native control, in terms of N-fixing ability, plant height and plant/leaf habit, they may cause larger impacts on the N cycle. This knowledge may help to improve current weed risk assessment approaches, which are rarely based on consistent quantification of impacts (Hulme 2012), by considering that both climatic conditions and functional differences between the invader and the native residents may modulate the magnitude of the impact.


This study was initiated during a stay of PC at the Forest Ecology and Forest Management Group (Wageningen University) funded by the University of Alcalá. We appreciate the facilities provided and the enriching discussions and feedback received by members of FEM group. We are indebt with Dr. Lourens Poorter and Dr. Milena Holmgren for their comments and discussion during the early stage of the study, and with Prof. Petr Pyšek, Prof. David M. Richardson and four anonymous reviewers for their comments to improve the manuscript. We are also indebt with Dr. Brad Hawkins for his language corrections. This study was supported by the grants CGL2007-61873/BOS, CGL2010-16388/BOS of the Spanish Ministry of Science and Innovation, POII10-0179-4700 of Junta de Comunidades de Castilla-La Mancha and the REMEDINAL network S2009/AMB-1783 (Comunidad de Madrid). OG acknowledges the post-doctoral financial support provided by the Spanish Ministry for Education and Science and Fulbright Commission (FU2009-0039).


PC, AA designed the study; PC, OG, AA, AG and AS obtained the data and prepared the database; PC performed the meta-analysis; OG performed the phylogenetic analysis; PC led the writing and all the authors contributed to revisions.