Why does it matter that low-dose effects can't be predicted from high dose experiments?
It matters because it invalidates the main approach used in toxicology to develop risk standards.
All current health standards are based on the assumption that effects occurring at low doses will only grow stronger as the dose increases. So toxicologists wanting to find out the impacts of a chemical do their experiments at high doses, because it's more efficient. Why? If this assumption is true, a high dose will make it easier to see an effect. A larger percentage will die or have birth defects or develop cancer. You can use smaller sample sizes in your experiments, so testing costs are much lower.
Then once toxicologists find an effect, they carry out experiments at lower and lower levels until experimental animals are no different than controls. That's the safe dose, the no-observed-adverse-effect level (NOAEL). The 'only reason' in traditional toxicology to test at lower doses is to find the level at which exposure is low enough to no longer matter.
This experiment, published by vom Saal et al. in 1997, showed that male mice exposed in the womb to diethylstilbestrol (DES) have the strongest reponse to the exposure at an intermediate dose, 0.2 ng/g (delivered to the mother).
The experiment is highly unusual because of the very wide range of doses used... 5 orders of magnitude, from 0.002 to 200 ng/g, and because of how low they actually went, down to 2 parts per trillion.
Standard protocols would have missed this effect. At the highest dose, prostate weight is significantly lower than control weight. At the next dose down, 20 ng/g, prostate weight doesn't differ from control weight. This experiment under standard procedures would have identified 20 ng/g as the NOAEL. The effect of prostate enlargement at 0.2 ng/g would never have been seen.
But these results are revealing that different types of effects, some not seen at higher levels, can be caused at lower levels. This means that high doses can't predict low dose effects.
Here are more examples (click on the image to read more about the experiment):