Meta-Analysis of Herbicide Non-Target Effects on Pest Natural Enemies

Abstract

A critical component of integrated pest management is minimizing disruption of biological control by reducing the use of pesticides with significant non-target effects on natural enemies. Insecticide non-target effects testing for natural enemies has become increasingly common, but research examining the non-target effects of herbicides on natural enemies is scarce, and recommendations regarding herbicide selectivity are non-existent. We used meta-analysis to summarize laboratory bioassays testing non-target effects of herbicides on arthropod natural enemies and identify patterns in taxon susceptibility and active ingredient toxicity. Data were extracted from 78 papers representing 801 total observations. Herbicides increased natural enemy mortality and decreased longevity, reproduction, and predation. Mesostigmatan mites and hemipterans were the most sensitive to herbicides, and spiders, neuropterans, and hymenopterans were the least sensitive. Mortality was higher in juvenile predators versus parasitoids but did not differ between adults; parasitoid juveniles are likely better protected within the host. In terms of acute mortality, metribuzin, glufosinate, and oxyfluorfen were the most harmful herbicides. Only nicosulfuron, rimsulfuron, pendimethalin, phenmedipham, atrazine, and urea did not increase natural enemy mortality. The large effect size of glufosinate is particularly concerning, as it is the most likely replacement herbicide for glyphosate in many crops. Many active ingredients remain under-studied. Our analysis indicates that herbicides have a strong potential to disrupt biological control in cropping systems.

Keywords: beneficial insects; biological control; glyphosate; herbicide; meta-analysis; natural enemy; parasitoid; pesticide non-target effect; predator.

Figure 1

Effect sizes (d) for variables tested in non-target effects assays. Error bars indicate 95% bootstrapped confidence intervals. Where these intervals do not overlap with d = 0 (dashed line), the effect is significant and indicated with an asterisk. Sample size (n in blue, number of records/observations) is indicated underneath each confidence interval in blue. Positive effect sizes are indicated by black triangles pointing upwards, and negative effect sizes are indicated by red triangles pointing downwards. A negative effect for sex ratio indicates a decrease in females/males compared to the control.

Figure 2

Effect sizes (d) of herbicide-caused mortality on adult and juvenile stages of predators (black circles) versus parasitoids (open squares). Error bars indicate 95% bootstrapped confidence intervals. Where these intervals do not overlap with d = 0, the effect is significant. Where these intervals do not overlap with each other, they are considered significantly different and marked with an asterisk. Sample size (n, number of records/observations) is indicated underneath each confidence interval in blue. “n.s.” indicates that none of the groups compared statistically differed from each other.

Figure 3

Effect sizes (d) of herbicide-caused mortality on both juvenile and adult stages compared between taxonomic groups at the level of (A) order (shapes), (B) families within orders, and (C) genera within families. Error bars indicate 95% bootstrapped confidence intervals. Where these intervals do not overlap with d = 0 (dashed line), the effect is significant. Where these intervals overlap with each other, they are not considered significantly different and were assigned the same letter. Sample size (n, number of records/observations) is indicated underneath each confidence interval in blue. “n.s.” indicates that none of the groups compared statistically differed from each other. Symbol shade and shape for a taxonomic order are consistent between panels.

Figure 4

Effect sizes (d) of herbicide-caused mortality on both juvenile and adult stages compared between herbicide (A) mode of action (MOA) group, (B) chemical class and (C) active ingredient. Error bars indicate 95% bootstrapped confidence intervals. Where these intervals do not overlap with d = 0 (dashed line), the effect is significant. Where these intervals overlap with each other, they are not considered significantly different and were assigned the same letter. Sample size (n, number of records/observations) is indicated underneath each confidence interval in blue. Symbols shade and shape for a MOA group are consistent between panels.