A Meta-Analysis of Effects of Bt Cotton and Maize on Nontarget Invertebrates

See allHide authors and affiliations

Science  08 Jun 2007:
Vol. 316, Issue 5830, pp. 1475-1477
DOI: 10.1126/science.1139208


Although scores of experiments have examined the ecological consequences of transgenic Bacillus thuringiensis (Bt) crops, debates continue regarding the nontarget impacts of this technology. Quantitative reviews of existing studies are crucial for better gauging risks and improving future risk assessments. To encourage evidence-based risk analyses, we constructed a searchable database for nontarget effects of Bt crops. A meta-analysis of 42 field experiments indicates that nontarget invertebrates are generally more abundant in Bt cotton and Bt maize fields than in nontransgenic fields managed with insecticides. However, in comparison with insecticide-free control fields, certain nontarget taxa are less abundant in Bt fields.

Public debate regarding risks and benefits of genetically modified (GM) crops continues unabated (15). One reason for the unrelenting controversy is that disagreements about new technologies often have little to do with scientific uncertainty but instead arise from differing personal values and differing levels of trust in public institutions (6, 7). However, in the case of GM crops, scientific analyses have also been deficient (4). In particular, many experiments used to test the environmental safety of GM crops were poorly replicated, were of short duration, and/or assessed only a few of the possible response variables (8). Much could be learned and perhaps some debates settled if there were credible quantitative analyses of the numerous experiments that have contrasted the ecological impact of GM crops with those of control treatments involving non-GM varieties.

Here, we describe a meta-analysis of field studies involving Bacillus thuringiensis (Bt) crops, which represent the predominant modification entailing the novel production of pesticidal substances (Cry proteins) in crop plants. The incorporation of bacterial-derived cry genes into plants means that a wide variety of species are exposed, on a relatively continuous basis, to pesticidal Cry proteins. We restricted our analyses to lepidopteran-resistant cotton expressing Cry1Ac protein, lepidopteran-resistant maize expressing Cry1Ab protein, and coleopteran-resistant maize expressing Cry3Bb protein, because the aggregate collection of field experiments assessing these Bt crops is large enough to draw some compelling conclusions (911).

The standard approach to assessing nontarget effects entails measurements of abundance, survival, or growth of nontarget species when exposed to a GM variety versus when exposed to the same or similar variety lacking the genetic modification. We focused on field studies, and the response variable we analyzed is the abundance of nontarget invertebrates, sampled in a variety of ways. For each experiment, we recorded many attributes, including locations, durations, plot sizes, and sample sizes (12) (table S2). Experiments relied on two different types of control treatments, each reflecting a different philosophy of risk assessment: (i) controls entailing non-GM varieties grown under identical conditions but treated with insecticides and (ii) controls entailing non-GM varieties grown under identical conditions and with no insecticides applied. A third type of comparison, in which both Bt and control plants were treated with insecticides, was occasionally used.

We report a weighted mean effect size, Hedges' d, calculated as the difference between the means of the Bt and the control treatments divided by the pooled standard deviation and weighted by the reciprocal of sampling variance. Negative values indicate lower abundance (whereas positive values indicate higher abundance) in Bt plots compared with abundance in control plots.

The mean abundance of all nontarget invertebrate groups lumped together is significantly reduced in Cry1Ac cotton fields compared with mean abundance in non-GM, insecticide-free fields [Fig. 1A, white bars; 95% confidence intervals (CI) do not overlap with d = 0]. However, the abundance of nontarget invertebrates is significantly higher in Bt cotton compared with that of control fields sprayed with insecticides (Fig. 1A, hatched bars). There was no significant difference in the abundance of nontarget invertebrates for studies where both the Bt and the control fields were treated with insecticides (Fig. 1A; gray bars). Thus, the different types of experimental comparison revealed significantly different effects of Bt crops [Fig. 1A, left; between-groups heterogeneity (Qb) = 49.96; degrees of freedom (df) = 2; P < 0.001]. Results were qualitatively similar when analyses were restricted to the related transgenic events MON531 and MON757 (Fig. 1A, right).

Fig. 1.

Meta-analysis of field studies assessing abundance of nontarget invertebrate species for (A) lepidopteran-resistant Cry1Ac cotton, (B) lepidopteran-resistant Cry1Ab maize, and (C) coleopteran-resistant Cry3Bb maize. Effect size is Hedges' d, and error bars represent bias-corrected 95% CI. Values below each bar indicate the number of different papers or reports and, in parentheses, the number of lines of data summarized (each line of data represents a comparison of agroup's average abundance in a Bt versus control treatment). White bars compare the abundance of nontarget invertebrates in Bt and non-GM varieties, without insecticide applications. Gray bars compare the abundance of nontarget invertebrates in Bt and non-GM varieties, both treated with insecticides. Hatched bars compare the abundance of nontarget invertebrates in insecticide-free Bt varieties versus non-GM varieties managed with applications of [(A)] any chemical insecticide and [(B) and (C)] pyrethroids.

For all Cry1Ab maize events, the overall mean abundance of nontarget invertebrates was significantly lower in Bt compared with that in control fields that lacked insecticide applications (Fig. 1B; leftmost white bar). However, the mean abundance of nontarget invertebrates was greater in Cry1Ab maize than in non-GM maize sprayed with pyrethroid insecticides (Fig. 1B; leftmost hatched bar). Effects measured by using these two different types of control treatments differed significantly (e.g., Fig. 1B, left pair of bars; Qb = 19.36; df = 2; P < 0.001). Qualitatively different patterns emerged when analyses were restricted to single transgenic events. For MON810, effect sizes measured using controls with versus without insecticides did not significantly differ (Fig. 1B, middle bars; Qb = 0.71; df = 2; P = 0.39). For Bt176, the two control types yielded significantly different effects (Fig. 1B, right bars; Qb = 9.41; df = 2; P = 0.012), but there was no significant reduction in abundance observed with the insecticide-free controls (white bar, 95% CI overlaps with d = 0).

For Cry3Bb maize, the mean abundance of nontarget invertebrates was not significantly different in Bt fields compared to abundance in non-GM maize either with or without pyrethroid applications (Fig. 1C; for the left pair of bars, Qb = 0.37 and P = 0.51). This same pattern held when analyses were restricted to event MON863.

The general indication of our analyses is that if agriculture with insecticide applications is the standard of comparison and if adoption of Bt crops truly reduces insecticide applications, then Bt crops may increase the abundance of nontarget invertebrates overall. Alternatively, if the comparison is made to farming systems without insecticides, some nontarget groups are significantly less abundant in Bt than in control fields (Fig. 2). Not surprisingly, the mean abundance of nontarget lepidopterans is significantly reduced in Cry1Ac cotton (Fig. 2A), which targets related lepidopteran pests. There were insufficient data to test this question in Cry1Ab maize. However, the mean abundance of nontarget coleopterans does not appear to be reduced in coleopteran-resistant Cry3Bb maize (Fig. 2C).

Fig. 2.

Comparison of effect sizes across orders of nontarget arthropods in (A) Cry1Ac cotton (Qb = 12.41; df = 5; P = 0.014), (B) Cry1Ab maize (Qb = 19.58; df = 7; P = 0.012), and (C) Cry3Bb maize (Qb = 5.64; df = 6; P = 0.146). Control treatments are restricted to non-GM varieties without insecticide application. Orders are those for which data were derived from a minimum of three distinct reports or publications. Effect size is Hedges' d, and error bars represent bias-corrected 95% CI. Values above each bar indicate the number of different papers or reports and, in parentheses, the number of lines of data summarized. An asterisk indicates that the 95% CI does not overlap with 0.

Coleopterans and hemipterans appear to be slightly less common in Cry1Ac cotton than non-GM, insecticide-free cotton (Fig. 2A). Although these groups both include a wide variety of functional groups (herbivores, predators, detritivores, etc.), we found no indication that some functional groups exhibit stronger effect sizes than others [see Supporting Online Material (SOM) text for additional details]. Lastly, hymenopterans are less common on average in Cry1Ab and Cry3Bb maize compared with hymenopterans in non-GM, insecticide-free controls (Fig. 2, B and C, respectively). For the Cry1Ab comparison, data on hymenopterans mostly comprised parasitic wasps of the braconidae and ichneumonidae. For Cry3Bb maize, data included parasitic wasps and ants. It is unclear whether the reduced abundance of these groups (coleopterans, hemipterans, and hymenopterans) is due to direct toxicity or is a response to reduced availability of prey in Bt crops. A significant reduction of collembolans in Cry1Ab maize is based on too few observations to be credible at this point (Fig. 2B).

To investigate the sensitivity of our findings to how we designed our comparisons, we performed additional analyses with use of several different subsets of experimental comparisons as the foundation (12). For example, we used only studies in peer-reviewed journals or only studies that identified invertebrate taxa to at least the level of family. With only minor exceptions, results using these alternative queries were qualitatively similar to those reported here (SOM text).

To facilitate additional syntheses, we have created a publicly accessible, searchable database, detailing methods and results of lab and field studies examining nontarget invertebrates and Bt crops ( While assembling this database, we found numerous studies that did not report measures of variance to accompany treatment means (40% of 64 reports of field studies), did not clearly present the sample size (20%), or improperly used subsamples to calculate measures of variance (22%). By corresponding with authors, we were often able to resolve these issues. If regulatory agencies were to require researchers to enter details regarding their study methods and results into a similar database, it would be easy to spot omitted information and postpone approval of pesticidal crops until complete records were submitted.

Our analyses provide some support to the claim that GM plants can reduce environmentally undesirable aspects of agriculture, particularly the nontarget impacts of insecticides. However, we examined only one type of genetic modification, and most of the underlying studies entailed controlled field experiments with small spatial scales as opposed to actual farming systems, where continued insecticide use sometimes occurs with Bt crops. Secondly, the conclusion that adoption of Bt cotton or maize may entail ecological benefits assumes a baseline condition of insecticide applications. In reality, both types of control treatment reflect farming practices: in 2005, insecticides were applied to 23% of maize acreage cultivated in 19 states surveyed by the U.S. Department of Agriculture (USDA) (13). Moreover, the vast majority of Bt maize acreage comprises varieties used for silage or processed foods (e.g., corn syrup) for which insecticide use has typically been limited. Insecticides are more commonly used in cotton production, with 71% of surveyed cotton acreage treated in 2005 (13).

Studies such as those synthesized here investigate whether changes in invertebrate abundance are statistically significant. Whereas the lack of a difference is generally considered a signal of environmental safety, it is harder to interpret whether statistically significant differences in abundance translate into ecologically important changes. Regardless of one's philosophical perspective on risk assessment for GM crops, enough experimental data has accumulated to begin drawing empirically based conclusions, as opposed to arguing on the basis of anecdote or hand-picked examples.

Supporting Online Material

Materials and Methods

SOM Text

Tables S1 and S2



References and Notes

View Abstract

Stay Connected to Science

Navigate This Article