Today I read a paper titled Transient Exposure to Low Levels of Insecticide Affects Metabolic Networks of Honeybee Larvae. While the title is impressive, the experiment and conclusion are much less so.
The study examined the expression for 10,736 genes in bee larva from 6 hives. In addition to 3 control hives, 3 hives were exposed to plausible quantities (2 ppb) of a neonicotinoid insecticide called imidacloprid.
Statistical analysis found 300 genes that were differently expressed between the larva from exposed vs. control hives:
For example, the greatest difference was seen in a gene called GB19113. That sounds pretty significant, right? Let's take a look at the data:
What's up with control hive 3??
It seems to have very different gene expression going on than the rest. Probably, some accident of timing, temperature, pheromones, or wild food supply resulted in some biological process going on in there, but not in the other hives.
This begs the question, what would be the expected variance between any two hives? We get 300 statistically significant genes different expressed when we bin based on exposure vs. control. But what if we bin the larva differently? What if we bin the larva based on other permutations of their hives?
This would seem a necessary control to determine the systemic error caused by any between-hive variance. From what I know about genetics, and about bees, we might expect this variance to be quite large.
Larval development is a time in a bees life where genes involved in growth are turning on and off quite often. One bee hive might have accelerated larval breeding compared with another. This might easily have nothing to do with imidacloprid exposure in the parts per billion.
Since the authors don't supply all the data, but only the data for the 300 genes in their list, I can't calculate the extent of this systemic error.
Without knowing the systemic error, no conclusion can be reached with regard to the safety of imidacloprid use on bee populations.
Another flaw with the study is the lack of testing for dose-response. In studies of this type, it is typical to expose different doses. This allows the support of a causal relationship by correlating increased dose with increased effect.
I trust the reader will not be fooled by any such deception.