FEATURE ARTICLE
Ecology of Transgenic Crops
Genetically engineered plants might generate weed problems and affect nontarget organisms, but measuring the risk is difficult
Michelle Marvier
What Is Significantly Safe?
Transgenic crops that produce insect toxins must undergo two separate reviews of environmental safety before they can be sold commercially in the U.S. (Approval for human consumption is a separate process.) First, the U.S. Environmental Protection Agency (EPA) reviews laboratory studies assessing a crop's effects on particular nontarget organisms, including pollinators, predatory insects and, often, soil invertebrates. Second, after collecting sufficient field data regarding a crop's performance and safety, the crop's developer may petition the U.S. Department of Agriculture (USDA) to allow commercial-scale cultivation. As of July 2000, the USDA had approved 50 petitions. Of those, 14 were for approval of crops with insect-resistance traits, all via Bt toxin. So far, the approved Bt-crop species are corn, cotton, potato and tomato.
An accurate assessment of a transgenic crop's environmental risks requires effective experimental and statistical protocols. Government organizations should label a transgenic crop as safe only if rigorous testing fails to detect any problems. Statistically speaking, a test's power is the probability that it can detect a significant difference between treatments when one exists. In practice, the statistical power of a particular test depends on the magnitude of the measured difference between the experimental groups, the amount of variability among replicates within groups and the sample size, or number of replicates per group. Of those, an investigator can most directly control only the sample size. Consequently, a rigorous test of safety must include a substantial number of replicates. Surprisingly, the vast majority of the examined toxicity studies reported in USDA petitions for deregulation relied on appallingly few replicates, usually just three or four per treatment group.
For a better understanding of assessing results, let's consider an actual example. Calgene, Inc., of Davis, California, submitted a petition (#97-012-01p) to the USDA for a variety of Bt cotton. In experiments designed to test this transgenic crop's impact on soil invertebrates, investigators placed four replicate batches of earthworms—with 10 worms per batch—in soil that included ground leaves from either transgenic or nontransgenic cotton. After 14 days, the investigators noted the weight and survival of the worms. The transgenic cotton leaves did not affect the survival of the earthworms. In fact, only one worm died out of the 80 total worms, but the 14 days represented a rather short portion of an earthworm's potential life span, which often exceeds several years. Accordingly, weight might provide a more sensitive measurement of the crop's impact. In these experiments, investigators weighed the worms in batches of 10 at the beginning and end of the 14 days. In the end, batches of earthworms that lived in the soil exposed to Bt cotton gained 29.5 percent less weight, on average, than the other earthworms. That difference, though, is not statistically significant. So, this study concluded that this particular Bt toxin did not impair weight gain in earthworms.
How believable is that conclusion? As you shall see, this study offered a very low probability of detecting a true difference between the treatments, because it relied on a small sample size (n = 4) and the data included considerable variation among replicates (pooled variance, sp2 = 273.5). In fact, if we set the probability of wrongly rejecting the null hypothesis, or a, to the standard 0.05, and we desire a 90-percent chance of detecting a true difference, or b = 0.10, we can solve for the minimum detectable difference (d) of this study:
d is > or = (2sp2/n)0.5(ta(2),v + tb(1),v)
where the t values are critical statistical values of the t distribution and v = 2 x (n - 1). Applying this equation to Calgene's experiment reveals that it could only detect a difference between treatments that exceeded 56.37 percent, which is almost twice the magnitude of the observed difference.
Moreover, the above equation can be rearranged to solve for the minimum sample size that would be needed in order to attain a particular probability of detecting a specified difference between the two groups. Suppose, again, that we desire a 90-percent chance of detecting a true effect, but now we wish to be able to detect an effect of the observed magnitude, which was 29.5 percent in Calgene's data. To do so requires only eight replicates per treatment. In other words, if the experiment were repeated with eight replicates instead of four and the variance stayed the same, a difference of 29.5 percent would be statistically significant. So, with a few additional replicates, one might have concluded that this Bt cotton did harm the tested species, and in just 14 days.
» Post Comment