A co-worker put me on to an interesting paper earlier this year by Harvard's George Whitesides (with a co-author credit going to a well-known chem-blogger). Whitesides, a perennial favorite in Nobel betting, does a lot of absolutely first-tier physical organic chemistry, an area that I love to read about (and one that I'd probably be an awful practitioner of).
Almost all drugs bind to sites on proteins. Some proteins have only one site (that we know of) that a small molecule will fit into, while others have several. There have been a lot of attempts over the years to go after the latter group by hitting more than one site at the same time - but with only one drug. Imagine two different drug molecules, each fitting into a different site on a single (multi-sited) protein. Now imagine combining them into one compound, by attaching some sort of linking chain between them, and you've got one (larger) molecule that can reach around and fill two binding sites.
This has worked in some cases, at least on a research level (I'm not aware of any drugs that have yet made it to market by taking advantage of this effect, though). (Update: there is a marketed protein, bivalirudin, that binds to two sites on thrombin, but I'm still not aware of any small molecule drugs in this category). You can pick up huge amounts of affinity by this trick, though, to the point that neither of the original "business ends" of the molecule need to be particularly good binders on their own. And since we in the industry are distressingly good at producing molecules that don't bind to things very well, the idea of combining some of these into multivalent wonders is appealing.
But there are a lot of unknowns. Figuring out how to modify the original structures in order to tie them together is, as they say, non-trivial. (If you hang around scientists and engineers much, you know to head for cover when you hear that expression). And what kind of chain should you use, anyway? How long does it have to be, and what happens if it's too long or too short? And what's the linking chain doing, anyway - sticking to the surface of the protein, waving around by itself, or what?
Whitesides and his people have used carbonic anhydrase as a model system, which is an enzyme whose structure and behavior is as well known as these things get. They find, not unreasonably, that when the linking chain is too short the activity of your wonder molecule just gets killed: you're stuck with one end bound to the protein, and a big tail flopping around uselessly, unable to reach the next binding site. The "just-right" chain length is the best, naturally. But (interestingly) you don't pay much of a penalty for being longer than necessary, even several times longer. Apparently the chain will coil around and find something to do with itself as long as the two ends are bound.
And while it's doing this, it doesn't appear to be contacting the protein in any meaningful way. This took a lot of careful experimental thermodynamics to check, but there's no extra binding energy involved with any of the common chains. So if you're going to try this trick, Whitesides's advice is not to worry about what chain to use. Stick with a plain-vanilla linker, as flexible as possible, make it a bit longer (at least at first) than you think you'll need, and you've improved your chances right there. And he has the numbers to back this up, which is what physical organic chemistry is all about: opinions made solid by data. It's good stuff.