Agricultural intensification and urbanization negatively impact soil nematode richness and abundance: a meta-analysis

The Web of Science core database was systematically searched for relevant publications on October 7, 2016, with the following combination of search terms: (‘nematode communities’ or ‘soil nematodes’ or ‘nematode diversity’ or ‘nematode abundance’ or ‘nematode biodiversity’) and (‘grassland’ or ‘forest’ or ‘agriculture’ or ‘prairie’ or ‘urban’), which resulted in 1,613 articles. Criteria for including an article in the analysis were: studies were conducted in forest, grassland, urban, or agriculture ecosystems; studies identified nematodes to family or genus level; studies reported mean abundance or richness expressed per grams or cm³ of soil; soil samples were collected from natural conditions; and studies reported sample size. Criteria for excluding an article were: studies conducted in controlled conditions like microcosms, mesocosms, pots, or greenhouses; studies expressing abundance of nematodes as relative abundance instead of absolute abundance; and studies reporting data for total free-living nematodes instead of each trophic group. Among the 1,613 articles, 598 relevant articles that contained data on richness and abundance of nematodes in different ecosystems were selected by examining titles and abstracts. Among the 598 articles, 111 articles (Supplementary Material) met the inclusion criteria and were selected for data extraction. Among the 111 articles, 91 expressed data in grams and 20 expressed data in cm³. The first 200 articles from a Google Scholar search were examined using the above search terms, which did not produce additional articles. A spreadsheet was constructed by extracting data from each article on authors, title, year of publication, unit of soil, richness and abundance of nematodes of each trophic group and each c-p guild, overall richness and overall abundance of nematodes, treatment, sample size, and type of ecosystem. Overall richness and overall abundance of nematodes were calculated by adding the number of genera/families and abundance of nematodes of either all trophic groups or all c-p guilds, respectively. Richness and abundance of nematodes under each trophic group and each c-p guild were calculated by adding the number of genera/families and abundance of nematodes corresponding to each guild and each trophic group, respectively. If there was more than one treatment in an article, they were considered as distinct studies in the meta-analysis. For example, there were two treatments, conventional-conservation tillage and organic-conservation tillage in Sánchez-Moreno et al. (2009), these two treatments were considered as two distinct studies. Based on these criteria, a total of 667 studies were subjected for meta-analysis of which 449 studies conducted in agriculture, 28 conducted in DGL, 74 conducted in forest, 36 conducted in NGL, and 80 conducted in urban ecosystems. Soil units in nematode studies are typically expressed as grams (Briar et al., 2007) or in cm³ (Wang et al., 2006). Therefore, the richness and abundance of nematodes expressed per 100 g of soil and 100 cm³ of soil were analyzed separately. Richness and abundance of nematodes per 100 g of soil were compared across all five ecosystems. However, the data expressed per 100 cm³ of soil were compared across only four ecosystems as no urban ecosystem studies using 100 cm³ were available. Abundance of nematodes that was not expressed per 100 g or cm³ of soil was converted to 100 g or cm³ of soil. However, richness of nematodes was not converted because increase in richness cannot be assessed with increase in the quantity of soil.

Effect size typically represents the strength of the relationship between two variables or two groups (treatment and control) but can also refer to the estimate of a single group or value such as richness or abundance of each study (Borenstein et al., 2009). Summary effect size is defined as weighted mean of richness or abundance of all studies in each ecosystem. Meta-analyses were conducted to compare the summary effect sizes of overall richness and overall abundance of nematodes and nematodes of each trophic group and each c-p guild per soil weight and volume basis among different ecosystems such as forest, NGL, DGL, agriculture, and urban ecosystems. Overall richness and overall abundance of nematodes per grams and per cm³ of soil were considered as four main effect sizes; richness and abundance of nematodes per grams and per cm³ of soil in each trophic group and each c-p guild were considered as subgroup effect sizes.

The types of ecosystems, forest, NGL, DGL, agriculture, and urban, were considered as moderator levels. These five ecosystems were assumed to have different regimes of disturbance where forest and NGL are considered less disturbed, whereas agriculture and urban ecosystems are considered highly disturbed from continuous human intervention. The moderator was chosen to determine the influence of disturbance on soil health.

The procedures and terminology of Borenstein et al. (2009) were followed in this analysis. Comprehensive meta-analysis (CMA) software was used to estimate effects of different levels of moderator on nematodes based on their confidence intervals, P _hetero values, Q statistics, and I ² values where Q is heterogeneity, and I ² is a measure of inconsistency across the studies (Version 3, Biostat, Englewood, NJ, USA; 2014). Random effects model was used rather than fixed effects model for meta-analyses as it considers within-study variance along with between-studies variance. Each study was weighted by the inverse of non-parametric variance. Non-parametric variance was calculated using the formula 1/n, where n is the sample size adjusted by using the following formula: V = ( 1 / n × ( 1 + ( t − 1 ) × 0.5 ) ) × ( m / t ) 0.5 , $$V = (1/n \times (1 + (t - 1) \times 0.5)) \times {(m/t)^{0.5}},$$ where m is the number of studies in a paper; and t the number of time-points within a year (Borenstein et al., 2009, equation 24.6). Studies within a paper are generally considered as not independent (Mengersen et al., 2013), therefore, studies were down-weighted by a factor of m ^0.5, (assuming 0.1 correlation among studies). After estimating different summary effects using CMA, the results were plotted in forest plots using SigmaPlot version 13.0 (Systat Software, San Jose, California). The summary effects along with their confidence intervals (CIs) from the meta-analyses were graphically depicted in forest plots.

Q is a weighted squared deviation used to evaluate heterogeneity, defined here as real differences among summary effect sizes. It separates observed variation from true variation. Total variation (Qt) consists of Qw (expected variation, within-study variation, or sampling error) and Qm (excess variation, between-study variation) (Borenstein et al., 2009). I ² is an estimate of the ratio of heterogeneity to total variation across the observed effect sizes (Higgins and Thompson, 2002; Huedo-Medina et al., 2006). It is the proportion of total variation due to heterogeneity in true effect size. I ² is computed as 100 × (Qt–;df)/Qt %, where degrees of freedom (df) measures within-study variation and Qt–df is true heterogeneity or between-study variation. I ² reflects the percentage of variation due to real differences in outcomes among studies (Borenstein et al., 2009). I ² values of 25, 50, and 75% may be considered as low, moderate, and high, respectively (Higgins et al., 2003). In meta-analysis, a significant heterogeneity P value (P _hetero value<0.05) or positive I ² indicates that there were real differences among studies; however, the converse is not true. A non-significant P-value (P _hetero value> 0.05) does not indicate that there were no real differences among studies because the non-significance could be due to low statistical power and/or large real dispersion of effect sizes and/or large within-study variance (Borenstein et al., 2009).

A sensitivity analysis was conducted to assess the stability and consistency of the summary effects. The summary effect was recalculated by removing one study at a time. This measures how sensitive the results are to any one study. The potential presence of publication bias was tested using the Begg and Mazumdar rank (Kendall) correlation test and graphically by examining summary effect sizes vs their standard errors in funnel plots (Begg and Mazumdar, 1994; Borenstein et al., 2009).

A total of 44 summary effect sizes were tested in the meta-analysis performed, of which 40 summary effect sizes were significantly heterogeneous (P _hetero <0.05) and all summary effects had positive I ² values (Table 1). The five summary effect sizes that were not significantly heterogenous included overall richness, c-p 4 richness and abundance, predator richness and omnivore richness from 100 cm³ soil samples (P _hetero > 0.05) (Table 1).

Sensitivity analysis indicates the contribution of each study to the summary effect, which is measured by the change in the summary effect in its absence. The summary effect size of overall abundance per 100 g of soil was most affected by the removal of treatment B4 at Bohemia in the study conducted by Čermák et al. (2011). This study reduced the summary effect size from 1,208.00 to 1,186.23 (Supplementary Material, Table 1). Similarly, the summary effect size of overall richness per 100 g of soil was most influenced by the removal of Renčo and Baležentiené (2015), grassland (control) treatment, reducing the summary effect size from 27.35 to 27.21 (Supplementary Material, Table 2). The summary effect size of overall abundance per 100 cm³ was most affected by the removal of the Bulluck et al. (2002), cotton-gin trash (harvest) treatment. This study reduced the summary effect size from 649.22 to 634.56 (Supplementary Material, Table 3). The summary effect size of overall richness per 100 cm³ soil was most influenced by the removal of the control treatment from Kapagianni et al. (2010) from 28.97 to 28.70 (Supplementary Material, Table 4). These results indicated that no single study changed any of the summary effect sizes to any important degree. Funnel plots did not show any observable patterns between standard errors and point estimate values, indicating no publication bias in this meta-analysis. In addition, the Begg and Mazumdar rank correlation test gave absolute Kendall tau values for all four summary effect sizes of less than 0.22, suggesting no publication bias.

Overall nematode richness expressed per 100 g soil was highest in forest compared to NGL, DGL, urban, and agriculture (P _hetero <0.05) (Fig. 1). However, the overall richness expressed per 100 cm³ of soil was not significantly heterogenous among ecosystems (P _hetero > 0.05) (Fig. 2). The nematode richness of c-p 2, c-p 3, and c-p 4 guilds per 100 g of soil was higher in forest ecosystems than in other ecosystems but richness of c-p 1 nematodes per 100 g of soil was highest in agricultural ecosystems along with forest and NGL ecosystems (P _hetero <0.05). The richness of c-p 5 nematodes per 100 g of soil was higher in forest ecosystems than in agriculture and DGL ecosystems (P _hetero <0.05) (Fig. 3). On the other hand, the richness of c-p 1 (P _hetero <0.05) nematodes per 100 cm³ of soil was higher in DGL ecosystems than in other ecosystems, whereas the richness of c-p 2 (P _hetero <0.05) nematodes per 100 cm³ of soil was higher in NGL, DGL and agricultural ecosystems than in forest ecosystems. The richness of c-p 3 (P _hetero <0.05) nematodes per 100 cm³ of soil was highest in DGL and forest ecosystems. However, richness of c-p 4 (P _hetero > 0.05) nematodes per 100 cm³ of soil was not significantly heterogenous among ecosystems. The richness of c-p 5 (P _hetero <0.05) guild nematodes per 100 cm³ of soil was higher in agricultural ecosystems than in forest ecosystems (Fig. 4).

Figure 1

Figure 2

Figure 3

Figure 4

The richness of bacterivores, fungivores, and predators per 100 g of soil was higher in forest ecosystems than in the other ecosystems and the richness of omnivores per 100 g of soil was higher in forest ecosystems than in disturbed ecosystems. The richness of herbivores per 100 g of soil was higher in forest ecosystems than in agricultural ecosystems (P _hetero < 0.05) (Fig. 5). The richness of bacterivores (P _hetero < 0.05) and fungivores (P _hetero < 0.05) per 100 cm³ soil was higher in DGL ecosystems, whereas richness of herbivores per 100 cm³ soil was lower in agriculture than the other ecosystems (P< 0.05). Richness of predators and omnivores per 100 cm³ soil was not significantly heterogenous among ecosystems (P _hetero > 0.05) (Fig. 6).

Figure 5

Figure 6

The overall abundance of nematodes per 100 g of soil was similar in forest, NGL, DGL, and agricultural ecosystems. However, overall abundance of nematodes per 100 g of soil was higher in agricultural ecosystems than in urban ecosystems (P _hetero < 0.05) (Fig. 7). The overall abundance of nematodes per 100 cm³ soil was highest in DGL ecosystems compared to other ecosystems, NGL and forest (P _hetero < 0.05) (Fig. 8).

Figure 7

Figure 8

The abundance of c-p 1 guild per 100 g of soil was higher in agriculture ecosystems than in NGL ecosystems; abundance of c-p 2 guild per 100 g of soil was higher in DGL and agriculture ecosystems than in urban ecosystems and abundance of c-p 3 guild per 100 g of soil was higher in agriculture ecosystems than in NGL and urban ecosystems In contrast, the abundance of c-p 4 and c-p 5 guilds per 100 g of soil was higher in undisturbed ecosystems than disturbed ecosystems (P _hetero < 0.05) (Fig. 9). Likewise, the abundance of c-p 1 per 100 cm³ soil was higher in agricultural ecosystems than in forest ecosystems. The abundance of c-p 2 and c-p 3 guilds per 100 cm³ soil was higher in DGL ecosystems than the other ecosystems, while the abundance of c-p 5 guild per 100 cm³ soil was higher in forest ecosystems, which are relatively undisturbed (P _hetero < 0.05). Abundance of c-p 4 nematodes per 100 cm³ soil was not significantly different among ecosystems (P _hetero > 0.05) (Fig. 10).

Figure 9

Figure 10

The abundance of bacterivores per 100 g of soil was higher in agriculture than in NGL and urban ecosystems; abundance of fungivores per 100 g of soil was higher in agriculture than in urban ecosystems and abundance of herbivores per 100 g of soil was higher in DGL ecosystems than in urban ecosystems. On the other hand, abundance of predators and omnivores per 100 g of soil was higher in undisturbed ecosystems than in disturbed ecosystems (P _hetero < 0.05) (Fig. 11). The abundance of bacterivores per 100 cm³ of soil was higher in agriculture and DGL than in NGL and forest ecosystems; abundance of fungivores per 100 cm³ of soil was higher in DGL than in forest, NGL and agriculture ecosystems and abundance of herbivores per 100 cm³ of soil was higher in DGL and agriculture than in forest ecosystems. Conversely the abundance of predators and omnivores was higher in forest than other ecosystems (P _hetero < 0.05) (Fig. 12).

Figure 11

Figure 12