Pesticide effects on soil fauna communities—A meta‐analysis

The wide use of pesticides across the globe to improve crop yield raises concerns regarding their impacts on biodiversity (Bernhardt et al., 2017; Pelosi 2021; Wang 2020). Despite recent initiatives reduce in many countries (e.g. European Commission, 2022), it continues increase globally (Sharma 2019), with a broad range substances still being applied, and new continuously put market (Wang It is, thus, crucial synthesize effects identify most detrimental scenarios gain general understanding consequences for biodiversity. So far, syntheses ecological pesticide have focused individual population levels, specific taxonomic groups such as pollinators natural enemies (Desneux 2007; Douglas & Tooker, 2016), or types neonicotinoids (Pisa 2015). To date, no quantitative synthesis has evaluated soil communities. Soil organisms represent about quarter global (Bardgett van der Putten, 2014) play important roles multiple ecosystem services (Schuldt 2018; Soliveres 2016; Wagg 2014). biodiversity, quality fertility are, humanity (Wall However, response communities environmental change remains understudied compared above-ground (FAO, ITPS, GSBI, SCBD, EC, 2020; Phillips 2017). This is particularly case invertebrates (such earthworms springtails) that not only constitute significant fraction (Eisenhauer 2019; FAO, 2020) but are also an essential part terrestrial food webs, animals, including vertebrates, depending this resource (Barnes 2023; Scherber 2010; Wardle, 2002). exposed often reach through spraying, seed coating, fumigation, leaching tillage practices 2018). Several persistent soils (Hladik Navarro 2007), sometimes over course decades (chlordecone: Cabidoche 2009). For instance, can be found up several years after application 2015; Wood Goulson, Pesticides induce direct toxic sublethal invertebrates, even at concentrations commonly field (Pelosi 2014; Pisa 2015), accumulate chemicals tissues (Beaumelle Römbke Direct include responses from level. example, cause DNA damage, alter enzyme activities, growth rates, foraging activity reproduction (Gunstone On top these effects, further affect different indirect mediated by shifts resources biotic interactions microbes, plants arthropods (Chen 2013; Clements Rohr, 2009; Desneux Morgado Zaller A review summarized results 394 studies large proportion reports negative (70.5% endpoints reported, Gunstone 2021). Negative were far more frequent scale individuals populations, while invertebrate community variable. qualitative, vote-counting approaches, cannot quantify magnitude effects. Quantitative next step generalize contexts (soil types, crops, geographic regions, etc.). Although generally expected, factors related type exposure involved As result, unclear whether decrease if certain than others. Indeed, respond differently pesticides, according mode action chemical structure (Biondi 2012; Frampton 2006; Jänsch target (insecticides) other (fungicides herbicides), broad-spectrum simultaneously pests diseases, could, elicit contrasted Additionally, real-world involve combinations targets throughout season 2021), differ they applied alone conjunction (Wood Furthermore, conditions terms dose temporal extent (short vs. long-term exposures) greatly influence fauna may rates time (Amossé Atwood Another moderator relates functional traits. animals contrasting due sensitivities (De Silva expected stronger larger (Liess von Ohe, 2005; Yvon-Durocher 2011). In addition, live close contact surface been sensitive litter-dwelling 2004). Such group-specific could lead diversity abundance, neutral composition Only addressing potential context-dependencies, we generate comprehensive able inform land managers policymakers. Overarching hypothesis (H1): abundance studies, with: (H1.2) depends Hypotheses (H2): applications determine fauna, sensitivity (H3): Functional traits We conducted systematic literature search collect used previously compiled database (Beaumelle, Thouvenot, 2019). fully reported (Phillips Specifically, included keywords retrieve ‘pesticide’, ‘fungicide’, ‘insecticide’ ‘agrochemical’) (combination form ‘soil’ AND (‘fauna’ OR ‘arthropod’ ‘invertebrate’) (‘community’ ‘biodiversity’)). From full search, screening abstracts texts, identified 104 focusing screened texts those against our inclusion criteria final set 54 (PRISMA plot: Figure S2). verify efficacy checked initial retrieved preidentified papers relevant current meta-analysis scope (Burrows Edwards, 2002; Fountain Knacker 2004; Scholz-Starke 2011, Velcheva 2012). each study, collected biomass, richness, indices along moderators. First, defined reference treatment levels study based intensities use. investigating sites located gradient intensities, level was condition lowest intensity concentration, highest concentration. data resolution higher family (except Enchytraeidae). When presented taxa treatments (different products, mixtures pesticides), them all created separate observations taxon treatment. If measured same plot points, extracted control latest point reflect pseudoreplication (Ferlian Hedges 1999; Yue means variances tables, text figures (Rohatgi, test (H1.2), recorded identity (active ingredients) assigned (Table S1). Pesticide herbicides, fungicides, insecticides, (i.e. single substance targeting groups, e.g. fumigants). addressed effect nematicide, insecticides category. products combined into single, applications, categorized ‘multiple substances’. acknowledge categorization does allow us address its strict sense. perspective, realistic tank together, sprayer), well combination season. Both approach enabled together present analysis. hypotheses (H2.1) (H2.2). recommended information provided primary studies. missing, ANSES E-Phy (ANSES, 2023) similar crops pests. application, repeatedly season). reflected last sampling event start experiment (that first application), three categories: (1) ‘short-term’ 3 months application; (2) ‘intermediate-term’ spanned 4 year; (3) ‘long-term’ year years. note testing dynamics communities, outside study. Quantifying would required modelling framework autocorrelation within amount available data, fewer measure changes time. Instead, approach, meta-analyses 2018), enables account any variability mean sizes driven extents moderated (H3), Global Biodiversity Atlas (Orgiazzi 2016) before assigning two Taxonomic presence (arthropods) absence (enchytraeids, nematodes earthworms) exoskeleton body size categories (microfauna, mesofauna macrofauna, Table operate continuum size, ecology, web compartments (Decaëns, Potapov Thakur useful comparing drivers meta-analytical approaches chose Hedge's g metric because interested difference between (Koricheva 2013). represents standardized raw divided pooled standard deviation groups) corrected positive bias (Viechtbauer, 2010). widely making comparable suitable log-response ratio incorporate zero values dataset very low richness treatments; Lajeunesse, Shannon 11 six lacked deviations (SDs) did report them. Based 50 SDs index (from nine studies), regressed SD slope impute missing (Lajeunesse, limitation lack prevented possibility moderators, metric, relationship SDs. metrics, evenness, number insufficient considered. fitted models hypotheses. All random identity, independent. observation-level following Viechtbauer (2022), ensure assume identical Effect weighted inverse variance models, giving weight well-replicated Q-Q plots deleted residuals verified strongly depart normal error distribution Cook's distance influential excluding yielded results. tested interactive moderators using likelihood tests (LRTs). significant, refitted without interaction main Wald-type Chi-square (QM). Main considered when p < 0.05. predicted confidence intervals. shown interval overlap zero. R software package metafor (R Core Team, Viechtbauer, model form: Y ~ PesticideType x CommunityMetric, PesticideType, categorical five (herbicides, substances), levels: (densities biomass data) (richness indices) there too few separately richness. allowed overarching (H1) quantifying marginal grand metrics (H1.1). According substances, spectrum organisms, invertebrates. formally post-hoc analysis model, four pesticides: (4) fungicides herbicides group do 1). Publication assessed funnel indicated clear sign publication (Figure S3). Egger's regression deviates precision errors) covariates interacting indicate bias, none intercepts differed significantly showing asymmetry conditions: ‘recommended rates’ ‘temporal extent’ (H2.2), traits: ‘body size’ (H3.1), ‘exoskeleton’ (H3.2). Those subsequent depend (H1.2). Due (few observations). second-level therefore, abundance. highlighted moderator, differences addition identity. These Moderator, structure: (1|CommunityMetric) + (1|StudyID/ObservationID), Moderator either extent, exoskeleton. structure. Each subsets subset (n = 31 studies; 158 observations), represented (although had excluded 1), which removed 37 12 removed. Primarily, inherently measuring invertebrates). result subsetting, one remained (addressing macrofauna). Similarly, 14 removed, (broad spectrum, substances: Finally, performed supplementary analyses impact pesticides. combining versus belonging type. quantified well-represented (glyphosate neonicotinoids) substances. Mean estimated described above (random structure, glyphosate neonicotinoid studies). Our encompasses publications, total 294 observations. Most experiments 51 277 covering contexts, although mostly Northern Hemisphere Studies 86 active S1) 1): 19; 116), 10; 37), 8; 42) fumigants, n 7; 37). Twenty Multiple 14), 14; 48). 31, 158), varied greatly, sampled 7 days 22 application. 47; 190), 16; 41), 12; 28), 9; 27) evenness 3; 8). They covered 15 1; Nematodes 21 80 followed Acari 13; 50), Collembola 55). (macro-arthropods, micro-arthropods fauna). balance among investigated insecticide much wider 1, 16 (9 6 herbicides). revealed (Hedge's g) −0.30 (CI: −0.47; −0.14; 2a). Richness negatively affected (significant metric: QM(df 1) 3.94, 0.0471), consistent 2a: non-significant interaction, LRT(df 4) 3.43; 0.4886). decreased diversity, (non-significant 8.35, 0.0797), intervals overlapped indicating (or neutral) averaging similarities herbicides. Grouping (QM(df 3) 8.20, 0.0420), broad-spectrum, elicited Supplementary suggested (herbicides fungicides) (broad-spectrum insecticides; S4a). trend, (only avoid confounding substances) statistically 1.04, 0.3086, 15; 30), 0.43 −1.25; 0.39). neonicotinoids, S4). S4b). strong S4c). analysing 2b, 31; (−0.30) (−0.66; 0.06). Responses 5) 9.65, 0.0858), caused declines contrast 2a), 2b). Focusing enough (multiple both (the sampling) 3a; 2) 1.26, 0.5320, 50; 257). Across short- sizes, 3a). Long-term dominated 5; 23) 18), 2; observations, respectively. repeated out 17), intermediate-term 22, 19 intermediate term, respectively). Testing needed data. observed short-, intermediate- 3.01, 0.2219). Single short term (−0.22; CI: −0.44; −0.01; 63) (−0.45; −0.85; −0.04; 11). category, ingredient (diflubenzuron) (0.03; −0.33; 0.39; 15). There evidence presence/absence modulated 3). general, 2.41, 0.2994, 247). associated (effect non-overlapping zero, 3), soft broadly overlapping 1.11, 0.2917, 49; 280). None interacted (Body size: 6) 5.45, 0.4877; Exoskeleton: 3.29, 0.5102). raising widespread intensive agriculture, investigate confirm t