We reviewed literature published in peer-reviewed journals up to January 2016 that focused on nematode communities in soil contaminated by metals. Google Scholar and the ISI Web of Knowledge were used to collect relevant publications. We selected only studies that presented, for a given soil, both nematode community indices and the soil’s total metal contents of arsenic (As), cadmium (Cd), chromium (Cr), cobalt (Co), copper (Cu), mercury (hg), molybdenum (Mo), nickel (Ni), lead (Pb), thallium (Tl) or zinc (Zn). Eco-toxicological studies on water nematode communities or at nematode species level (e.g. Caenorhabditis elegans) were excluded. Thirty-seven studies, published between 1991 and 2015, were selected (Table 1). Of these, 28 were published in international journals and 9 as institutional reports. The 37 studies were conducted in 13 countries: France (10 sites), China (9 sites), the Netherlands (5 sites), Spain (1 site), Germany (1 site), the United States (2 sites), Hungary (1 site), Scotland (1 site), Uzbekistan (1 site), Britain (1 site), New Zealand (1 site), Slovakia (1 site) and South Korea (1 site). Most investigated sites were contaminated with heavy metals and/or metalloids following mining or industrial activities or accidents. In 11 studies, the heavy metal contamination was performed experimentally in pots or in field experiments.

We selected 15 nematode parameters for this meta-analysis:

• The abundance of six nematode trophic groups: opportunistic bacterial feeders (cp1, i.e. value 1 on the coloniser-persister scale [21] ), other bacterial feeders (cp2, cp3 and cp4), fungal feeders (cp2, cp3 and cp4), omnivores plus carnivores (cp2, cp3, cp4 and cp5), plant feeders (cp2, cp3, cp4 and cp5) and total soil nematodes [20].

• Four nematode indices: Maturity Index (MI) [21], Nematode Channel Ratio (NCR) [58], Structure Index (SI) and Enrichment Index (EI) [59],

• Two diversity indices calculated on genus level: Number of genera (S) and the Shannon diversity index (H')

• Three metabolic footprints: enrichment footprint (EFOOT), structure footprint (SFOOT) and community footprint (COMFOOT) [22].

Table 2 presents the nematode indices used in this study and their corresponding calculations.

The soil and nematode parameters were extracted directly from the tables, text or figures of each study using PlotDigitizer 2.6.4 software. The software ELIPTO© developed by the French company ELISOL environnement was used to calculate the nematode parameters that were not included in the studies (e.g. abundance of trophic groups, MI, EI, SI, NCR, Shannon diversity index and metabolic footprints) using data on nematode taxa abundance included in the studies. Depending on the available data, it was not always possible to calculate all 15 selected nematode parameters.

a) Ke, Ks, Kb: coefficients relative respectively to the enrichment, struture and basal fauna; n: abundance of the considered familly; b) Ba: number of bacterial feeders; Fu: number of fungal-feeders; c) Vi: cp value; Pi: proportion of the nematode of cp-class considered; d) N: total nematodes abundance; Ni: abundance of nematodes within genera; e) Ni: abundance of nematodes within family i; Wi: average weight of a nematode from the family i; Vi: cp value of family i.

For each study, mean values were extracted for the different parameters; the number of replicates per treatment in each study was also recorded and used for the statistical analyses. If several nematode analyses were conducted on different dates, results from each date were included in our analysis. In most cases, nematofauna analyses were made in the top layer of soil; however, if several soil layers were investigated in the studies, results from each soil layer were included in our analysis.

The source and type of soil contamination in each study included in the meta-analysis are presented in Table 1. In most cases, the soil was contaminated by several heavy metals or metalloids. In some studies, only certain concentrations of heavy metals were measured.

We used generalized linear mixed models (GLMMs) to estimate the nematode community responses to soil contaminated with metals; more specifically, we selected a Poisson distribution with a logarithmic link. These models were chosen because this model is appropriate for analyzing count data and for handling the variability of the observations both within and between studies [63]. The fixed effect measured the average effect (across the 37 studies) of metal contamination; the random effect measured the variation in responses across the studies. Model validity was verified by visually examining the residuals and by testing the significance of the estimated parameters. The GLMMs were fitted by restricted maximum likelihood assuming a Laplace approximation to the restricted maximum likelihood function. The appropriate linear contrasts between contamination classes were estimated and their significance statistically tested, checking multiple comparisons over metal contamination classes using the glht() function in R language. To visually illustrate the fitted results, we created box plots of nematode community parameters; outliers were not depicted. Finally, we assessed the size of the effect of high contamination (c2 vs c0) on all the nematode indices. This was calculated as the median (+/− 95% confidence interval) of (c2 − c0)/c0 values measured on each site. We used lme4, multcomp and ggplot2 libraries for the statistical analyses and creation of the graphics.

From the 37 studies used for this meta-analysis, 355 situations (defined as a nematofaunal community in a specific place on a specific date, with a certain level of contamination) were extracted. These 355 situations corresponded to 1475 individual nematological analyses, with an average of four replicates per situation. The dataset available per nematode parameter (averaged over replicates) was variable since the different studies did not allow the calculation of all 15 selected nematological variables in every case. As a result, the number of values in the dataset per nematode parameter ranged from 37 to 150 depending on the parameter considered. On average, the datasets for c0, c1, c2 and c3 respectively contained 54, 53, 107 and 41 units of data, meaning that c0, c1 and c3 had a similar quantity of data, whereas about twice as much data was available for c2.

Overall, the GLMMs showed significantly negative effects of metal contamination of soil on nematode parameters. Fitted total nematode abundance decreased in line with increasing soil contamination (Figure 1). Of the nematode trophic groups, bacterial feeders cp1, other bacterial feeders (cp2 to 4), omnivores/carnivores and plant feeders also showed a significant decrease in abundance in line with an increase in the level of contamination from c0 to c3.

In contrast, the abundance of fungal feeders showed more erratic variation. In this trophic group, compared to the c0 level of contamination, abundance was significantly lower at c1 and c2, but significantly higher at c3. Moreover, the abundance of fungal feeders was lower in c1 (low contamination) than in c2 (high contamination). However, differences in the median abundance of fungal feeders between contamination classes was weak (<40 nematodes 100 g⁻¹ dry soil) and, overall, abundance was low (100 - 200 nematodes/100g dry soil) compared to that of bacterial feeders.

The effects of metal contamination on selected nematode indices are presented in Figure 2. The Enrichment Index (EI) did not decrease in line with an increasing level of contamination: c1, c2 and c3 showed similar EI values; however, these values were lower than EI values for c0. The Structure Index (SI) decreased significantly with an increase in the level of soil contamination; differences were significant between c0, c1 and c2, but not between c2 and c3. The Maturity Index (MI) followed the same pattern as the EI: this index did not decrease significantly with an increasing level of contamination—c1, c2 and c3 had equivalent MI values, but these were lower than the value for c0. The Nematode Channel Ratio (NCR) was not affected by metal contamination: whatever the contamination level, the decomposition pathway was dominated by bacterial activity. The Shannon diversity index (H') decreased in line with an increase in soil contamination. Differences in the H' value were significant between c0, c1 and c2, but not between c2 and c3. Taxonomic richness also decreased in line with the level of soil contamination. It reached the lowest value in the c3 class (<15 taxa).

The metabolic footprint of nematode community was strongly impacted by metal contamination (Figure 3). The global nematode community footprint

(COMFOOT) decreased in line with an increase in contamination level; the median COMFOOT of the c3 class was two times lower than the median COMFOOT of the c0 class. The plant feeder footprint (PLTFOOT) and structure footprint (SFOOT) followed the same trend, reflecting a strong negative effect of soil contamination on the metabolic activity of plant-feeding nematodes and nematodes with long lifecycles (cp3, cp4 and cp5). The enrichment footprint (EFOOT) was smaller in contaminated areas, but the effects were dependent on the contamination level. Indeed, the c2 class showed a higher EFOOT than the c1 class; however, the lowest EFOOT was found for the c3 contamination class.

The main goal of this study was to identify the nematode parameters most sensitive to soil contaminated by metals. To this end, we compared the size of the effect between c0 (no contamination) and c2 (high contamination) classes, which represented the majority of the contaminated soil situations.

Of the 15 nematode parameters analysed in this study, we observed three groups of responses to soil contaminated by metals (Figure 4). Group 1 included plant feeder abundance, non-opportunistic bacterial feeder abundance, Maturity Index (MI), diversity (S), Shannon diversity index (H') and community footprint (COMFOOT). These parameters showed a significant decrease (10% to 20%) in c2 compared to c0.

Group 2 included omnivore/carnivore abundance, total nematode abundance, Structure Index (SI) and Nematode Channel Ratio (NCR). These parameters also showed a significant decrease (about 20% on average) in c2 compared to c0. Structure footprint (SFOOT) showed an even stronger decrease (about 30%). For most of these parameters, the confidence intervals were higher, indicating higher variability between studies.

Group 3 included opportunistic bacterial feeder abundance, fungal feeder abundance, Enrichment Index (EI) and enrichment footprint (EFOOT). These parameters showed no significant difference or low increased percentages on average in c2 compared to c0. In particular, opportunistic bacterial feeder abundance and EI showed an increase lower than 5%, while fungal feeder abundance and EFOOT showed no significant variation between c0 and c2.

Nematodes have been used as bioindicators of soil quality for more than 20 years ( [14] [18] [58] [64] ). In the context of soil pollution, the advantage of bioindicators is that they can integrate the effects of all the contamination parameters (type, concentration, pollutant interactions, bioavailability of each pollutant, etc.), which are difficult to characterize comprehensively using physico-chemical analyses.

In this study, we analysed soil contamination based on measurements of the total content of heavy metals. Other analytical methods to quantify contaminants exist, such as extraction with CaCl₂ (which mimics the quantity extractible with rains [65] ) or extraction with the chelating agent EDTA [66]. However, methods based on extractable elements cannot reflect with accuracy the bioavailability of metals for organisms in the soil or their potential impact on the environment. The advantage of bioindicators is that they allow the measurement of the combined effect of metallic contaminants (as well as organic pollutants) on ecosystem functioning and can also measure variation in biological activities by integrating all the accumulated individual effects [67]. In many polluted areas, multiple contaminations with heavy metals occur concurrently and could thus induce varying and complex nematode responses [17] [25] [44]. For example, a study by Georgieva et al. [17] has shown the additive effect of zinc and copper. It has also been demonstrated that acidifying compounds can create optimal conditions for the increased mobilization, bioavailability and thus toxicity of metals stored in soils ( [9] [40] [68] [69] ). The effect of metal pollution on nematodes can result from a number of different parameters, such as the concentration of the metal, the type of metal, its bioavailability, and the duration of contact with nematodes [70].

The contamination classes defined for this study and based on total metal content revealed general trends of nematode community responses to various levels of soil contamination, whatever the metallic nature. Although measurements of total content are not always pertinent, they are the most frequently available and, often, the only measurements reported in publications.

It should be noted that at low concentrations, some metals, such as copper (Cu), zinc (Zn) and molybdenum (Mo), are micronutrients that are essential to plant growth (at respective levels of about 2.5, 2.2 and 0.15 mg∙kg⁻¹ soil or less, depending on soil pH). But at higher concentrations, these elements have toxic effects on living organisms and are considered contaminants, as are other metals such as cadmium (Cd), lead (Pb), chromium (Cr), nickel (Ni), mercury (Hg) and arsenic (As) ( [71] [72] [73] ). If micronutrient concentration exceeds the binding capacity between a metal and the soil matrix, then metals can contaminate soil interstitial water and become bioavailable for the organisms living in the soil. As nematodes have a permeable cuticle, they are in direct contact with xenobiotics in interstitial water and may be adversely affected by these contaminants [21]. Trace metals have the potential to bind to proteins and to alter protein functionality [74]. This can cause metabolic disorders in nematodes, leading to a decrease in fitness, low motility, reduced juvenile body length, or even mortality. The effects depend on the type and concentration of metal [70]. For example, it has been shown that cadmium (Cd) has no acute effect on nematode communities up to 160 mg∙kg⁻¹ soil, while nickel (Ni), copper (Cu) and zinc (Zn) decreased the proportion of omnivores and predators at a level of 100 mg∙kg⁻¹ soil after 1 - 2 weeks exposure in experimental studies [40]. At this concentration, total nematodes abundance was also significantly decreased by nickel (Ni), while other metals did not significantly impact total abundance. Another crucial factor potentially impacting nematodes response to metal contamination in soil is the duration of contact, as some nematodes can develop tolerance and/or adapt to long-term soil pollution [9] [75].

Employing a meta-analysis method allowed us to combine the results of a number of studies that characterize the impact of both short-term and long-term effects of real or experimental metal pollution on soil nematode communities. We were able to integrate results obtained from diverse conditions: different soil types, climates (e.g. temperate or continental) and land uses (e.g. grassland, arable land, industrial, mining or urban areas). As nematode community composition is known to vary in response to such environmental factors [76] [77], aggregating this information permitted the statistical power to reveal robust trends of nematode responses to soil contamination. And this, independently of the duration of the pollution, the nature of the metal contamination (e.g. experimental, single or multiple metals, etc.), the climate and the land use.

As reported in several studies, soil contamination by metals has an overall negative effect on nematodes abundance, diversity and community structure [17] [20] [28]. Our meta-analysis revealed that nematode parameters have different responses to the metal content of soil.

All nematode parameters apart from NCR showed sensitivity to metal pollution from the lowest level of contamination (c1). But certain did not reflect the level of soil contamination. These parameters should not be use as indicators of pollution impact.

In our meta-analysis, the abundance of plant feeders, non-opportunistic bacterial feeders (Ba2-4) and omnivores/carnivores showed an acute decrease in line with an increase in soil contamination (across the four contamination classes). Shao et al. [48] also observed a decrease in abundance of plant-parasitic nematodes in line with increased soil contamination by lead and zinc. These authors highlighted that the abundance and proportion of these nematodes were linked with the recovery of the vegetation. As stated by Bongers and Ferris [14], the abundance of plant-parasitic nematodes is mostly determined by the community structure and biomass of plants. Thus, a decrease in plant-parasitic nematodes in a polluted area is probably related not only to the direct toxic effect of heavy metals, but also to reduced plant production [29] [35].

Some studies have reported no significant effect of Cu, Pb or Zn on cp2 bacterial-feeding nematodes, leading to doubts about using these nematodes to assess the effect of heavy metal contamination [17] [27]. However, our results clearly showed that the abundance of cp2-4 bacterial feeders (usually dominated by cp2) decreased in contaminated soil in the upper contamination classes (c2 and c3). The decrease was less notable between c0 and c1 (the fitted medians were similar). Studies have observed a high proportion of bacterial-feeding nematodes in contaminated areas, and some authors have suggested a higher resistance to metal contamination of some taxa in this functional guild, for example, Acrobeloides ( [43] [57] ). Georgieva et al. [43] also suggested that indirect long-term effects such as less competition and predation could favour bacterial feeders in contaminated areas.

Our study confirmed what many others have described: that omnivore and carnivore nematodes are the most sensitive to contaminated soil, and their populations decrease in proximity to the pollution source ( [21] [24] [25] [42] ). The decrease of the number of genera in line with the increase in soil contamination in our results could reflect the loss of the most sensitive nematodes, mainly cp3, cp4 and cp5 nematodes and the omnivore/carnivore trophic group (Zhang et al., 2007; Nagy et al., 2004). The SI, whose calculation is based on the relative abundance of these nematodes, did not decrease between high (c2) and very high (c3) classes, whereas the community footprint (COMFOOT) and structure footprint (SFOOT) did. This indicates that despite a relatively similar nematode community composition in c2 and c3 contamination classes, the abundance and biomass of nematodes decreased in line with the increase in contamination between these two classes, differentiating them. Nematode biomass, used to calculate the nematode footprint, is considered a good indicator of soil pollution in several studies ( [9] [25] [78] ).

Few relationships were observed between the nematode parameter indicators of enrichment status and soil contamination levels. The EI was weaker in contaminated areas, but was not impacted by increased levels of contamination. Cp1 bacterial feeders were more abundant in the c2 contamination class than in c1 and showed the lowest abundance in c3. Enrichment footprint values showed the same patterns as the EI. The latter is a ratio of nematode abundance and does not take into account absolute abundance of nematodes, whereas EFOOT does. Cp1 bacterial feeders are enrichment indicators and are linked with the nitrogen mineralization of organic matter in soils ( [79] [80] ). They are r-strategists with a tolerance to soil pollution ( [17] [81] ), and their development is strongly linked to the nutrient status and organic content of soil ( [48] [79] ). As a result, the abundance of cp1 bacterial feeders, the EI and the EFOOT cannot be considered good indicators of soil pollution and should not be the focus of investigation in contaminated conditions.

The MI showed lower values in polluted areas (c1, c2 and c3) compared to uncontaminated areas (c0), but did not vary between contamination levels. This index has been successfully used to characterize the effects of different levels of pollution on nematode communities ( [21] [25] [36] [44] ), but some authors have underlined the limitations of this indicator when nematode abundance is low [48]. Our results show that the MI is sensitive to pollution, as it has minimum values at the low contamination level (c1), indicating that the maturity of the nematode community was lower at any level of contamination. One interpretation might be that an increase in contamination level impacts nematode abundance, but not the proportion of cp3-5 nematodes in the community.

We found that fungal-feeding nematodes were the least affected by metal contamination. These nematodes actually showed higher abundance in the very high contamination class (c3), leading to an increase in total nematode abundance. The relative insensitivity of fungal-feeding nematodes to pollution has been reported in several other studies ( [16] [27] [28] [45] ). It seems that high levels of soil pollution promote fungal growth ( [9] [82] ). However, as bacterial feeder abundance is much higher than fungal feeder abundance in most soils, the NCR was not strongly modified with the increase in soil contamination, indicating that this index should not be used as an indicator of metal pollution in soil. This finding does not confirm the results from Zhang et al. [57].