So what are these cancer animal models that I was speaking of so poorly? On the face of it, they actually seem like they'd be pretty good, other than being rather disgusting. (That said, it's important to keep in mind that they're not as disgusting as watching people die from cancer when you could be doing something about it.)
What you do is take human cancer cell lines and implant them in a mouse, a process called xenografting. When they form a tumor, then you treat the mouse with your drug candidate and see if the growth rate slow, stalls, or reverses compared to untreated controls. Sounds pretty simple.
But the complications show up very quickly as you look closer. For one thing, these human cancer cells are often cell lines that have been propagated for a while in vitro, and there's room to wonder about how much they've changed since their days as primary tissue. There's also the issue of the number of different cancer cell types you could use - hundreds, thousands, more? We know what tissue they came from, and we know some of the biochemical differences between them, but nowhere near all. Not even most of the important differences, if you ask me, since we don't even know what some of those important differences are yet.
What we have are characterizations like "Cell line such-and-such, non-small cell lung cancer, resistant," or "colon, slow-growing, responds to everything." Each cell line has its own reputation. At least the fact that these reputations are pretty constant gives you some confidence that we're all talking about roughly the same cells, which is no small thing. (More than once in the history of cellular research, people have realized that cell lines which were all supposed to be the same thing had drifted apart.)
Another level of difficulty is that these things are implanted, rather than growing in situ in the tissue of interest. Any cell biologist will tell you that the matrix a cell grows in is one of the fundamental variables of cell culture. Now, once the tumor has formed, the cells are surrounded by other cancer cells, which is closer to the real situation. But they're still being vascularized by mouse blood vessels, which obviously respond to mouse signals and carry mouse blood. That's the fundamental animal model problem, and it's a tough one.
Finally, these aren't any old mice. In order to get the cells to "take" when they're injected, these mice have a severely compromised immune system. They mostly have no thymus, for starters (and no hair, either, as a side issue.) Here's one - if you find hairless dog and cat breeds cute, you probably won't mind these guys, either. They don't make very good pets, though, because (as you'd imagine), they will catch every disease available, and likely as not die from it.
At bottom, these models are probably too permissive. As I mentioned the other day, they can make compounds like Iressa look just fine, when we now know that they confer no real benefit in humans. (If our market were nude mice with good health insurance, we'd be set, though, as would the mice.)
So what good are they, and are we really doing a good thing by running them? Well, it's hard to imagine that your compound is going to do any good in humans if it doesn't at least work in the nude mice, so they serve a screening function. It's true, though, that for some years now, if the compound hasn't worked in the mice it's never gotten to humans, so we don't have as many checks on that idea as we'd need to be sure of that assumption. But we see a lot of disconnects like Iressa, which argues for false positives being more of a problem than false negatives.
And I'm not sure how good the models are at rank-ordering compounds, either. I can justify their use as a pass/fail, but that's about it. We should be doing better, and people are trying to. And a lot more are trying behind closed doors - better animal models would simultaneously help large numbers of desperate patients and save the drug industry about a billion dollars. More on all this in another installment.