Animal studies generally are based on far fewer individuals than clinical trials. If somehow we justify requiring hundreds of thousands of cases (say, diabetics) and at least as many unaffected controls, from multiple global studies, to claim to identify the genes 'for' a trait (GWAS and other similar mapping efforts), and even then don't find much, how is it that we expect to learn about this by mapping studies in a small study of mice?
There are many examples in which serious and clear high-risk mutational effects found in humans do not arise in mice with similar mutations induced transgenically. Altered BRCA1 and breast cancer is one of legions of examples. Much is learned even in those cases, about aspects of the gene's biology, but not necessarily about why the difference in the trait between mice and humans. Mouse skulls develop differently in many ways from those of humans, in particular involving the sutures (joints) between the bones in the skull vault that protects the brain. Yet mutations causing abnormal suture closure in humans have quite similar effects in mice--that is, the mouse model seems to work, but why that is is somewhat unclear. We study dental or limb development in mice, but their teeth and limbs are quite different from ours--however some of the same genes and gene interactions apply in both cases.
When one gene on its own doesn't usually account for human traits like disease (not even those with strongest effect, such as BRCA1 and breast cancer) why do we think we'll understand diabetes by making single-gene knockouts? The rationale, and it's true to some extent, is that we are trying to work out, with animal models, how the gene works and why a mutant version can lead to disease. But this is so very incomplete that it's curious how much we invest in the approach (which, we must immediately acknowledge, we have been doing in our own lab for many years). The challenge is to know when and why the differences exist--and not to over-extrapolate from mouse to human.
Couzin-Frankel has identified many issues and reports on NIH efforts to look at them. They are serious and perhaps should threaten funding of such studies and diverting funds to better ways--and making the environment suitable for people to find them. There are elements of scandal and misrepresentation in the drug area that she reports. Perhaps at least more transparency can result, but forgetting that, the scientific issues are serious in their own right.
Of mice and not-men in many more ways
This article and those cited in it don't begin to touch many of the serious issues, that go way beyond small sample sizes, non-randomization, and so on. Here are some others, that we have seen in our own personal experience with our own work, or those of colleagues:
1. Doing very unpleasant things to mice and getting them approved by your local research ethical review board ("IRB"), torment if not torture often justified on the grounds of preventing disease, an argument often stretched to the limit to justify things with only the remotest connection to health.
2. Reporting the most extensive transgenic effect because that one (author says either privately or even in the paper) is the 'most representative' of the engineered effect.
3. Using statistical techniques to get more results than there is blood in a turnip from small samples, which have all sorts of unreported but potentially relevant nuances.
4. Assuming that an experimental effect done on one mouse strain represents 'the mouse' and, worse, 'the human'. In fact, many if not most transgenic manipulations yield different results for different test strains, and even worse than that, often the chosen strain is one chosen because it has relevant characteristics--more likely to get cancer, more responsive to genetic manipulation, and so on.
5. Assuming that inbred mice are genetically homogeneous--that is, they have no variation within their own genomes (that is, both copies of their chromosomes, inherited one from each parent, are of identical sequence) and, just as bad, no variation among individuals of the same inbred strain. Those who pay attention know that this is not accurate. Getting the whole genome sequence of one mouse from a strain and using that to represent all mice of the strain is an example.
In our own lab we have used our computer program ForSim to do simulations of the inbreeding and inter-breeding process that is involved in work to identify transgenic effects or to identify ('map') genes whose variation affects some trait of interest. It is easy to show that mouse strains, crosses, and representative sequences are not reliable indicators of all the potentially relevant variation that may exist in a particular study design.
There is no easy answer. First, despite reservations and some public opposition, the rationale that we'll save children from horrible diseases if we give those diseases to mice is compelling. We are, after all, in charge. We eat pigs and cows, make chickens live shoulder to shoulder indoors for their entire lives, and so on, so what we do to mice isn't all that different.
Second, despite knowing that mice are different from humans (we usually don't yell "Eeeeek!!" when we see another human), we have a very understandable tendency to assume they're the same. That leads us to make dramatic discovery announcements to the press (and granting agencies). Sometimes these discoveries do, in fact, pan out as advertised. Again, there is no obvious way to know in advance of, or even after, an experimental result is in how or whether it will work on humans.
But at least we should be vastly more circumspect about this sort of animal research. That's only fair to the voting public that pays for it based on what we promise them....and for the poor mice who have no vote in the matter.