This about a really cool paper from Ian Wilson's group:

Crystal Structure of Human Toll-Like Receptor 3 (TLR3) Ectodomain
Jungwoo Choe, Matthew S. Kelker, Ian A. Wilson
Science, Vol 309, Issue 5734, 581-585, 22 July 2005

Toll-like receptors (TLRs) are a new and explosive field in immunology. We're just starting to understand the signaling that they use, upstream and down, and how they turn on various aspects of the immune response and turn off others. One of the things that we don't understand well about TLRs is a very fundamental question: How do they recognize their ligands?

TLRs recognize and respond to conserved aspects of "danger" - say, particularly bacterial aspects of bacteria, or particularly viral aspects of viruses, that mammals (or flies, if you're a Drosophila TLR, or fish, if you're a zebrafish TLR) don't normally have in their body. That means that there should some lock-and-key type system, something about the shape of a particular TLR that matches the shape of the bacterial or viral target. But that structural match, the way TLRs fit with their targets, really hasn't been understood at all well. That's significant not only for the basic question, but because if you're trying to design some kind of pharmacological inhibitor or trigger, you want to know what you're designing toward, you need to know the shape you want to complement.

Ian Wilson, at Scripps, is one of the crystallographer kings out there; he's determined the structure of a whole bunch of things (151, according to the Protein Data Bank), and TLR3 is his latest triumph.

TLR3 recognizes double-stranded RNA, which normally doesn't occur (much) in mammalian cells, but is common in viral infections. When TLR3 sees dsRNA, it turns on the interferon response, which has a lot of antiviral activities. How does it see dsRNA?

Some of the shape of TLR3 had been predicted from the primary protein sequence -- the fact that it's horseshoe-shaped in general, for example--and that was borne out by Wilson's crystal structure:

Now, that offered an obvious explanation for the RNA binding -- the RNA would slip into the concave side of the horseshoe, like slipping a finger into a ring. The size of the pocket was just about right for a piece of RNA to slip into.

But Wilson finds that that's pretty unlikely, for two reasons. First, TLR3 is heavily glycosylated (sugar groups bound to the protein -- very common in surface molecules like TLR3). Sugar groups take up a fair bit of space ... and there are a bunch of them sticking into that hole in the horseshoe:

Here's the same thing shown in space-filling form, with the sugar groups shown in their full size (this picture is taken from Wilson's paper):

So there's not really room in there for RNA after all.

But it's even worse than that. RNA is acidic (duh, that's what the "A" stands for in RNA). That means it's negatively charged, and so it doesn't like to stick to other negative charges. Well, sugar groups on proteins are negatively charged. But there's even more bad news for the pocket. Negatively-charged amino acids in TLR3 cluster into the pocket of the horseshoe. Here's what it looks like if you paint the negatively charged places red, positives blue (also from Wilson's paper, because the various programs I have for manipulating structure views don't do nearly as nice a job of coloring charges):

RNA really, really isn't going to like that pocket.

So where is RNA going to bind? Well, we'd like to find a place on TLR3 where there aren't sugars getting in the way, and we'd like to find a positive charge to make the RNA happy. Turns out there is such a place. Here's a side view of TLR3; look at the sugars sticking out on the left--but there are none at all on the right side of the molecule:

Now I'll rotate the view to look straight at the "right" side (this is the opposite side from the previous view):

And here's the charge distribution on that side, again with negative in red, positive in blue (Wilson's picture again):

Mmmm. You could just snuggle up in that, couldn't you? Two big positive patches, and almost no negatives at all.

So there's a nice hypothesis for you. Wilson's suggestion is that RNA binds to one of those two positive patches (he leans toward the left-most of the two). He also suggests (with, I think, a little weaker rationale -- but still pretty reasonable) that a region nearby that positive patch could mediate dimerization, but I'll leave that argument for now.

I don't know enough about the field to guess how generalizable this is going to be. I suspect, from my limited understanding, that other TLRs will have their own modes of binding ligands, and that TLR3 may even be an oddball -- I think this horseshoe shape is predicted in some other TLRs, and maybe in other TLRs the pocket will bind their ligand. It'll be interesting to find out, anyway.

If you want to play with the structure yourself, the PDB and the MMDB databases both let you play with the structure online, as well as with free downloads.