Meta-analysis reveals that seed-applied neonicotinoids and pyrethroids have similar negative effects on abundance of arthropod natural enemies

Introduction

Arthropod natural enemies (predators and parasitoids) contribute considerable value to agriculture by suppressing pests that attack crop plants. For example, biological control of the soybean aphid (Aphis glycines) is estimated to be worth at least $84 million per year in just five US states (Zhang & Swinton, 2012). Given their importance to pest management, it is essential to understand how agricultural practices influence natural-enemy communities and their ability to suppress crop pests. Insecticide use is one common agricultural practice that can influence natural-enemy populations and biological control. Insecticides are used to manage pests, however, in some cases they also disrupt biological control, leading to unintended outbreaks of target or non-target pests (Geiger et al., 2010; Settle et al., 1996; Stern et al., 1959). Elucidating how insecticides and natural enemies interact can be useful in designing stable pest management strategies, identifying insecticide products that have greater selectivity for pests versus natural enemies, applying them in ways that conserve natural enemies, and in some cases even using insecticides to enhance the efficacy of biological control (Croft & Brown, 1975; Hull & Beers, 1985).

One use of insecticides that is increasingly widespread is planting seeds coated with neonicotinoid insecticides, especially in large-acreage field crops where these seed coatings are ubiquitous in some crops and regions (Simon-Delso et al., 2015). Seed coatings account for at least 80% of neonicotinoids applied in the US (Douglas & Tooker, 2015) and accounted for roughly 87% in Britain prior to recent restrictions (Simon-Delso et al., 2015). Through their systemic activity, seed-applied neonicotinoids target some early-season soil and foliar pests, and their relatively low cost and ease of handling makes them an attractive option as ‘insurance’ against sporadic pest populations (Douglas & Tooker, 2015; Jeschke et al., 2011). The popularity of neonicotinoids is also related to their toxicological profile; their binding behavior at nicotinic acetylcholine receptors confers potent toxicity against a broad array of insect species, and simultaneously low acute toxicity to mammals (Tomizawa & Casida, 2005).

Despite the broad toxicity of neonicotinoids to insects, some researchers reasoned that by applying them in relatively small doses ‘targeted’ to crop seed, seed-applied neonicotinoids would have high ecological selectivity for crop pests and low potential to reduce populations of beneficial insects (Jeschke et al., 2011). The environmental effects of these products have since been called into question by evidence that non-target species, especially bees, can be exposed to seed-applied neonicotinoids via contaminated soil, planting dust, floral resources, and guttation droplets (Godfray et al., 2014; Godfray et al., 2015). This pollinator controversy has spurred new regulations in several jurisdictions (European Commission, 2013; Government of Ontario, 2015), in turn raising the question of whether other pest management tactics that might replace neonicotinoids, including conventionally applied insecticides, have more or less influence on non-target species. We argue that fully understanding the trade-offs associated with seed-applied neonicotinoids requires attention not only to pollinators but also to natural enemies that suppress crop pests.

Seed-applied neonicotinoids can potentially reduce populations of arthropod natural enemies through at least two mechanisms: toxin exposure and prey scarcity. Neonicotinoids are toxic to many natural-enemy species (Cloyd & Bethke, 2011; Hopwood et al., 2013), but it is unclear whether under field conditions natural enemies encounter meaningful doses of toxins from seed treatments. What constitutes a ‘meaningful dose’ is likely to vary by taxon. In the laboratory, insects are orders of magnitude more susceptible to neonicotinoids than arachnids (Table 1), and at least some evidence suggests that this difference is based on the molecular structure of arachnid acetylcholine receptors (Meng et al., 2015). If natural-enemy populations were reduced by seed-applied neonicotinoids through toxin exposure, we would expect insects to be more strongly affected than arachnids. Alternatively, neonicotinoids could exert indirect negative effects, for instance by reducing the abundance of prey, leading to less aggregation, persistence, or reproduction of natural enemies in crop fields (Croft & Brown, 1975). Under this prey-scarcity scenario, we would expect insects and arachnids to be similarly affected, but functional groups to differ relative to their degree of dependence on pest prey (parasitoid > predator > omnivore). These two types of effects of insecticides on natural enemies are not mutually exclusive, and both can interfere with biological control (Johnson & Tabashnik, 1999). Clarifying the mechanisms at play can nonetheless guide researchers and pest managers toward more successful integration of chemical and biological control.

Field experiments on the influence of seed-applied neonicotinoids on natural enemies have reached mixed conclusions. Some have found no statistically significant effects (Al-Deeb et al., 2003), some a mix of null and negative effects (Albajes, Lopez & Pons, 2003), and others more consistent negative effects (Hallett et al., 2014). Some of this variability may be related to the small sample sizes of most field studies, which limits their statistical power. Conversely, variability in study results may reflect real differences across crop species, natural-enemy taxa, active ingredients, and other factors. One powerful tool to make sense of such apparently mixed results and to untangle the various factors influencing study outcomes is meta-analysis (Cumming, 2012). This approach has been fruitfully applied in similar situations, for instance, to estimate the influence of Bt transgenic crops on non-target organisms (Marvier et al., 2007; Naranjo, 2009; Wolfenbarger et al., 2008). One of the salient benefits of meta-analysis in controversial areas is that it provides a rigorous, quantitative, and replicable method of synthesizing evidence for researchers, policy-makers, and the public (Marvier et al., 2007).

Here, we report results from a meta-analysis of studies investigating under field conditions the influence of seed-applied neonicotinoids on arthropod natural enemies. We used a meta-regression approach to test the hypotheses that seed-applied neonicotinoids: (1) negatively affect natural-enemy abundance relative to untreated controls; (2) reduce natural enemy abundance less than conventional foliar- or soil-applied insecticide treatments; and (3) affect natural enemies through a combination of toxin exposure and prey scarcity. Our results taken together should allow researchers and pest managers to better predict the compatibility of seed-applied neonicotinoids and natural enemies, and to more effectively weigh the use of these insecticides against alternative pest-management options.

Materials & Methods

Meta-analysis is a method for synthesizing observations from independent yet similar studies to characterize the size and variability of an effect—in this case the influence of seed-applied neonicotinoids on natural enemies of crop pests. Our meta-analysis was guided to some extent by previous meta-analyses that characterized the influence of Bt crops on populations of non-target arthropods (Marvier et al., 2007; Naranjo, 2009; Wolfenbarger et al., 2008). We departed from these previous studies in our statistical approach. In particular, as described below, we capitalized on advances in statistical programs over the past several years to better account for the hierarchical nature of the dataset. All of our analyses were conducted within the R statistical program (R Core Team, 2015).

Results

Results of the literature search & characteristics of the meta-analysis dataset

In total we screened 921 titles and abstracts, yielding 62 candidate reports (Fig. 1). After assessing eligibility, filtering for relevant functional groups and geographic regions (North America and Europe), and reducing redundant observations, our final dataset for the no-insecticide controls comprised 607 observations collected over 56 site-years and 20 independent studies (Table S1 and References S1). For the pyrethroid controls, our final dataset comprised 384 observations collected over 15 site-years and 8 independent studies (Table S1 and References S1). Corn (Zea mays) and soybean (Glycine max) were the dominant crops, and insects were better represented in the dataset than non-insect arthropods (arachnids and chilopods; Table 2). Unsurprisingly, given the focus of our study, predators were the dominant functional group (67–68% of observations), although parasitoids, omnivores, and mixed functional groups were also represented. Observations were spread fairly equally among soil and foliar habitats, different active ingredients, and pyrethroid application methods, and most observations were derived from peer-reviewed studies (76–93%; Table 2). Plot size ranged widely from 1 to 110,000 m², and proportion of early season sampling ranged from zero to 100% (Table S1). Most studies had a sample size of three to six replicate plots per treatment (Table S1).

Seed-applied neonicotinoids negatively affected natural enemies compared to no-insecticide controls

Consistent with our hypothesis, seed-applied neonicotinoids reduced abundance of arthropod natural enemies relative to untreated controls (Fig. 2). The mean effect size (d) was −0.30 or −0.26, for the ‘site-year/study’ and ‘null’ models respectively. For context, an effect size of −0.30 would correspond to an approximate reduction of 16% in natural-enemy abundance (−0.30 ×median relative standard deviation [RSD] of 0.53). The estimates of the variance components from the fitted model suggested that site-year explained most of the shared variation among observations within studies (study σ² = 0.0035, site-year σ² = 0.08). The negative effect of seed-applied neonicotinoids on natural enemies appeared to be homogenous (Q = 622.5, df = 606, P = 0.31), with an I² indicating that all but 2.7% of variation in effect sizes could be explained by random sampling error.

Despite the low heterogeneity identified in the initial analysis, we proceeded with the ‘moderator’ analysis to test the influence of various factors on the effect of seed-applied neonicotinoids on natural enemies. We did this because we had planned the moderator analysis a priori, and because the Q test, despite being the most powerful test available, still has low power to detect heterogeneity in datasets like ours where the within-study sample sizes are small (Viechtbauer, 2007b). Indeed, when we fit the meta-regression model with our eight moderator variables, the omnibus test suggested that the moderators taken together did explain significant variation in effect size (Q_M = 21.5, df = 11, P = 0.029). Broad taxonomic group apparently drove this result, as it was the only moderator that was significant when tested individually (Table 3). As predicted under toxin exposure, the negative effect of seed-applied neonicotinoids on natural enemies was stronger for insects than for non-insect taxa (mostly spiders and mites; Fig. 2). When estimated separately, the negative effect of neonicotinoids on insects remained significant, while the effect on non-insect taxa did not differ significantly from zero (Fig. 2). Although functional group did not significantly moderate the influence of seed-applied neonicotinoids on natural enemies (P = 0.13), the fitted slopes for this moderator were fairly large and followed a trend consistent with indirect effects via prey scarcity (parasitoid > predator > omnivore, Table 3). Surprisingly, effect size was not influenced to a significant extent by crop species, neonicotinoid active ingredient, habitat, or any of the methodological variables (Table 3).

Table 3:

Estimates and tests of significance for moderators in a meta-regression model testing the effect of seed-applied neonicotinoids on natural enemies, compared to controls treated with no insecticides (n = 607 observations from 56 site-years and 20 studies).

Moderator	Level	β	QM	df	P value
Intercept	–	−0.23	–	–	–
Taxonomic group	–		8.70	1	0.003
	Insects	−0.11
	Non-insect arthropods	0.11
Habitat	–		1.42	1	0.23
	Aboveground	0.057
	Soil-associated	−0.057
Functional group	–		5.61	3	0.13
	Omnivore	0.186
	Mixed	0.049
	Predator	−0.071
	Parasitoid	−0.164
Crop species	–		0.79	2	0.67
	Corn (Zea mays)	0.072
	Soybean (Glycine max)	0.002
	Other	−0.074
Active ingredient	–		0.99	1	0.32
	Imidacloprid	0.043
	Clothianidin/Thiamethoxam	−0.043
Publication type	–		0.51	1	0.56
	Peer-review journal	0.062
	Dissertation/Thesis/Other	−0.062
ln(Plot size)	–	−0.016	0.34	1	0.56
ln(Early sampling + 0.1)	–	−0.076	0.61	1	0.44

DOI: 10.7717/peerj.2776/table-3

Seed-applied neonicotinoids reduced natural enemy abundance similarly to pyrethroid insecticides

Contrary to our prediction, the effect size for pyrethroids versus seed-applied neonicotinoids did not differ significantly from zero (Fig. 3), suggesting that these two groups of insecticides reduce natural-enemy abundance to a similar extent. The sensitivity analysis revealed that one study (Ohnesorg, Johnson & O’Neal, 2009) had a large influence on effect sizes and confidence intervals, so we present results both with and without this study (Fig. 3).

With all studies in the dataset, the mean effect size (d) was 0.16 or 0.07, for models that did or did not include random effects of site-year nested in study (Fig. 3). Including random effects for site-year and study made the confidence intervals very wide, because of the influence of Ohnesorg, Johnson & O’Neal (2009). Similarly, the estimates of the variance components from the fitted model suggested that study explained most of the shared variability among observations (study σ² = 0.58, site-year σ² = 0.04). The effects of seed-applied neonicotinoids on natural enemies compared to pyrethroid controls appeared to be homogenous (Q = 369.6, df = 383, P = 0.68), with an I² indicating that 100% of the variation in effect sizes could be explained by random sampling error.

When Ohnesorg, Johnson & O’Neal (2009) was excluded from the dataset, the confidence intervals were smaller but the conclusion remained the same: seed-applied neonicotinoids and pyrethroids had similar influences on natural-enemy populations (Fig. 3). The overall mean effect size was very close to zero, −0.02 or 0.03, for models that did or did not include random effects of site-year nested in study. The variance components for study and site-year were small in this model (study σ² = 0.01, site-year σ² = 0.0008). Consistent with the first analysis, there was no evidence of heterogeneity in these effects (Q = 316.9, df = 375, P = 0.99, I² = 0).

Again, we proceeded with the ‘moderator’ model to test whether various factors influenced the magnitude of effect size. In this case, the omnibus test suggested that the moderators did not explain significant variation in effect size (all studies: Q_M = 4.50, df = 6, P = 0.61; excluding (Ohnesorg, Johnson & O’Neal, 2009): Q_M = 3.40, df = 6, P = 0.76). This result is not surprising given the zero estimate of heterogeneity in this dataset, and suggests that the effect size of seed-applied neonicotinoids compared to pyrethroids is fairly consistent across the dataset, except for the observations associated with one outlying study.

Discussion

We performed a meta-analysis of field studies to determine the influence of seed-applied neonicotinoids on arthropod natural enemies of crop pests in North America and Europe. After gathering and synthesizing results from almost 1,000 observations gleaned from 20 studies, we found that seed-applied neonicotinoids: (1) reduced natural-enemy abundance and (2) reduced natural enemies similarly to foliar or soil-applied pyrethroids. Furthermore, the influence of seed-applied neonicotinoids on natural enemies differed by broad taxonomic group: insects were more strongly affected than non-insect arthropods such as spiders and mites. This last result suggests that reductions in natural-enemy populations associated with seed-applied neonicotinoids are at least partly a result of toxin exposure, rather than prey scarcity alone.

Seed-applied neonicotinoids reduced the abundance of natural enemies relative to no-insecticide controls, with an effect size (d =  − 0.30) corresponding to roughly 16% reduced abundance. This result was robust to different modeling choices and unexpectedly consistent across crop species and neonicotinoid active ingredients. For comparison, the mean effect of organic farming (versus conventional farming) on predatory insect abundance was estimated to be d = 0.49 (Bengtsson, Ahnström & Weibull, 2005). Both effect sizes suggest that synthetic insecticides can undermine natural-enemy populations, but the consequences of these reductions for ecosystem services are hard to predict given a lack of research relating predator abundance to biological control function and its economic value (Naranjo, Ellsworth & Frisvold, 2015). The one study in our dataset that explicitly related predator abundance to crop yield was our previous study in a no-till soybean system (Douglas, Rohr & Tooker, 2015). In that study, a 31% reduction in early season abundance of slug predators in neonicotinoid-treated plots corresponded to a 67% increase in slug abundance and an eventual 5% reduction in soybean yield. Incidentally, the season-long reduction in slug-predator abundance was 16%, very similar to the mean effect identified in this meta-analysis, suggesting that a reduction of this magnitude can have economic consequences. Future efforts to relate natural-enemy abundance to crop yield could also make use of the concept of ‘natural-enemy units,’ which help to consolidate diverse natural enemies into a single measure of pest-suppression potential (Bahlai et al., 2010; Hallett et al., 2014). Ultimately, it would be valuable to build the knowledge base necessary to fit models relating natural-enemy abundance to pest abundance and ultimately crop productivity, analogous to those recently developed for pollination services (Koh et al., 2016).

Our finding that seed-applied neonicotinoids can in some cases increase predator–prey ratios further highlights that natural-enemy abundance is not equivalent to biological control function. We stress that a formal analysis of predator–prey ratios was not possible, but generally seed-applied neonicotinoids tended to have a smaller effect on natural enemies than on pest aphids, and a relatively larger effect on natural enemies than on other pest taxa (slugs and thrips). This pattern is further supported by case studies in the literature. We are not aware of any systems in which seed-applied neonicotinoids have been associated with resurgence of target pests such as aphids; however, there are several examples where these seed treatments have been associated with increased abundance and sometimes economic outbreaks of non-target pests, including spider mites (Smith et al., 2013), slugs (Douglas, Rohr & Tooker, 2015), and late-season stem-boring caterpillars (Pons & Albajes, 2002).

Our finding that insects were more strongly affected by seed-applied neonicotinoids than were non-insect groups (mainly spiders and mites) suggests that toxin exposure is at least partly responsible for the overall negative effect we observed, and raises the question of how insect natural enemies are being exposed to these seed-applied toxins. Neonicotinoids can poison natural enemies through ingestion as well as contact with sprays or residues (Lucas et al., 2004; Torres & Ruberson, 2004; Wang et al., 2008). Possible exposure pathways include contact with soil or planting dust (Goulson, 2013), ingestion of contaminated prey (Douglas, Rohr & Tooker, 2015; Szczepaniec et al., 2011), and for some natural enemies, ingestion of pollen, nectar, or other plant products (Lundgren, 2009; Moser & Obrycki, 2009). The relative importance of the various exposure pathways in the field is unclear, but we did see a non-significant trend for soil-dwelling taxa to be more strongly affected than foliar-dwelling taxa. Typically ∼90% of seed-applied neonicotinoids remain in soil, rather than entering the growing crop plant (Goulson, 2013), and recent findings reveal a layer of elevated residues on the soil surface where many species are active (Limay-Rios et al., 2016). Soil exposures, therefore, appear to be an important area for future research, particularly because previous research has leaned toward foliar-dwelling taxa. Finally, although the pattern we observed is consistent with the toxin exposure hypothesis, we cannot rule out that different responses to seed-applied neonicotinoids by insects and non-insect taxa reflect differences in ecology rather than (or in addition to) toxin susceptibility. For instance, many spider species can endure very long periods of starvation (e.g., Anderson, 1974), and there are also important differences in mobility between insects and arachnids. The toxin exposure hypothesis could be tested more directly using semi-field enclosure studies that hold immigration and prey availability constant while changing the exposure of insects and other taxa to neonicotinoid residues.

In contrast to toxin exposure, there was insufficient evidence to conclude that prey scarcity contributed to reductions in natural-enemy abundance by seed-applied neonicotinoids, although this result may change with additional research. While functional group was not a significant moderator of natural-enemy response to seed-applied neonicotinoids, there was a trend in the direction we would expect if prey scarcity were involved (parasitoid >  predator > omnivore). The prey scarcity hypothesis is also supported by a case study on the multicolored Asian lady beetle, Harmonia axyridis. This species is an important predator of the soybean aphid, and its population dynamics in the American Midwest over the past two decades correlated with changes in abundance of its soybean aphid prey, which in turn correlated with use of seed-applied neonicotinoids (Bahlai et al., 2015). More generally, although seed-applied neonicotinoids do not always provide economic control of aphids, they do sometimes reduce their seasonal populations (Hallett et al., 2014; Heidel-Baker, 2012; Johnson et al., 2009; Ohnesorg, Johnson & O’Neal, 2009; Tinsley et al., 2012). In turn, aphids are key prey for many generalist predators in agricultural systems (Donaldson, Myers & Gratton, 2007; Symondson, Sunderland & Greenstone, 2002). Future research could test the relative importance of prey scarcity versus toxin exposure through field studies that manipulate prey density independently of neonicotinoid treatment.

We expected seed-applied neonicotinoids to reduce populations of natural enemies less than foliar or soil-applied pyrethroids, but aside from one outlying study (Ohnesorg, Johnson & O’Neal, 2009), this was not the case. The limited number of independent studies (eight) comparing neonicotinoids to pyrethroids may have affected this result, and we encourage more research in this area. That said, our finding is consistent with previous meta-analyses (Naranjo, 2009; Wolfenbarger et al., 2008) that found a negative effect of pyrethroids on predatory arthropods (versus transgenic Bt varieties) of similar magnitude to the negative effect we found for seed-applied neonicotinoids (versus untreated controls). Pyrethroids are the second most widely used class of insecticides in the world after neonicotinoids (Sparks, 2013), and are important alternatives to seed-applied neonicotinoids in North American and European field crops (Budge et al., 2015; Douglas & Tooker, 2015; Furlan & Kreutzweiser, 2015). Their use is therefore likely to increase if, when, and where neonicotinoid use is restricted. Foliar and some soil-applied pyrethroids have the advantage that they can be applied in response to economic pest populations, and can therefore be more compatible with integrated pest management than seed treatments, which are typically applied to the seed months before planting (Furlan & Kreutzweiser, 2015; Johnson et al., 2009). It is also worth noting that pyrethroids and neonicotinoids overlap in their acute toxicity to mammals (Tomizawa & Casida, 2005), although foliar or soil applications would likely also entail different exposures than seed applications. Additionally, in some cases seed-applied neonicotinoids may be replaced by cultural management tactics or nothing at all. While the full economic, human health, and environmental trade-offs of neonicotinoids versus pyrethroids and other pest management strategies are beyond the scope of this study, our results do suggest that seed-applied neonicotinoids are neither uniquely risky nor benign to an important group of non-target invertebrates.

Prior to our meta-analysis, the statistical results within and across studies in our dataset appeared highly variable, and a narrative review of these findings could characterize them as mixed. In fact, their measured effects were largely consistent with one another, as reflected in the low heterogeneity of effect sizes across our datasets. This apparent contradiction results from the modest size of the effect combined with the high variability of measurements in field studies, and emphasizes the importance of considering statistical power during ecological frisk assessment. Detecting a 20% reduction in natural-enemy abundance with 80% probability requires at least 15 plots per treatment for many predatory arthropod taxa (Prasifka et al., 2008), far more than most studies contributing to our dataset (typically three to six replicates per treatment). While increasing sample size is an obvious solution, logistical and funding constraints make this a challenge. We suggest that researchers interpret null results conservatively in light of statistical power. Periodic meta-analyses may be useful for drawing broader conclusions, as has been the case for transgenic Bt crops (Marvier et al., 2007; Naranjo, 2009). The datasets we compiled for this study and a dataset of neonicotinoid effects on bees (Lundin et al., 2015) together provide a foundation for ongoing meta-analyses on the influence of neonicotinoids on non-target species.

There are several important limitations of our meta-analysis that stem from constraints of the dataset we compiled. Because most studies measured natural-enemy abundance within a single field season, our results do not address the influence of seed-applied neonicotinoids on other important metrics like species diversity, sublethal effects on behavior, reproduction of long-lived species, or long-term effects on natural-enemy populations associated with chronic exposure. Furthermore, our dataset comprises manipulative plot studies that by their nature do not account for movement of natural enemies across landscapes. By focusing on seasonal mean abundance, we may have underestimated important but transient effects that occur only during the period soon after planting. Finally, our study was not able to address the influence of seed-applied neonicotinoids in cropping systems outside of North American and Europe. It is our hope that future meta-analyses will benefit from increased research in these areas.

Conclusion

Using meta-analysis to synthesize the results from field studies in North American and Europe, we found that seed-applied neonicotinoids reduced natural-enemy populations similarly to foliar- or soil-applied pyrethroids. The negative effect of neonicotinoids on natural enemies was d =  − 0.30 ± 0.10 [95% CI], corresponding to a reduction of ∼16%. The patterns we observed suggest that seed-applied neonicotinoids exert their effects mainly on insect (versus arachnid) natural enemies, at least partly through toxin exposure. If restrictions on neonicotinoid use encourage substitution with pyrethroids, our results suggest that there will be little net effect on natural-enemy populations. In fact, the results of neonicotinoid restriction for natural enemies are likely to be complex, particularly because some pyrethroids can more easily be saved for those situations in which economically damaging pest populations occur. Finally, translating natural-enemy abundance into biological control function is not possible given current knowledge, and is an important area for future study.

Supplemental Information