Heat a residue in a protein, large loops pop out

Woah, I've put up a web-site for the molecular-dynamics simulations that I've been working on for the last 4 years. It's funny, although publishing the method in a journal feels like a substantial accomplishment, it's only now that I feel like I'm finished. When the method has been presented in a journal, it feels somewhat coy, a case of look but don't touch, but when the code is also available for others to touch, use and abuse, then it truly feels like you've set the method free into the wild.

So what does the Rotamerically Induced Perturbation (RIP) give you?

RIP is the only molecular dynamics (MD) protocol that can generate artificially large conformational changes, without the need for any form of constraints. It generates true local perturbations.

Whilst microsecond long MD simulations are common-place now (damn you rich supercomputer centers), you can't actually see much going on in these simulations. You certainly don't see surface loop motions of more than 5 Ångstroms.

Before RIP, to see large motions in a protein, you typically need to use MD protocols such as high-temperature annealing, or replica-exchange, which apply high temperatures to specific regions in the protein. If you don't want the rest of the protein to unravel, you need to put constraints on the rest of the protein. Unfortunately, as these constraints must be manually applied, these methods can't be used automatically on a arbitrary protein.

So four 4 years ago, I started working in the Agard lab, on improving an elegant technique called the Anistropic Thermal Diffusion developed by previous postdoc, Noboyuki Ota. He came up with the idea of cooling down a protein in simulation, and then applying a thermometer on one single residue. By watching the residue knock around neighboring residues, he could follow the energy diffusion through the protein.

In order to get it working, Nobu had to put artificial constraints on all the backbone atoms, and all the surface atoms. My project in the Agard lab was basically working on how to heat a residue without the need for constraints. When I got to the lab, I was told Nobu had implemented ATD as a hack to the source-code to XPLOR and that the code was to be found in an unplugged Silicon Graphics machine that was gathering dust in the corner of the lab. It took 6 months of hell before I could reproduce it using a salad of Python scripts and AMBER.

Eventually I worked out how to heat a sidechain in a local way without the need for backbone constraints. The solution was to only apply heat to the sidechain degrees of freedom. As the heating is applied to the rotameric sidechian chi-angles, which does not affect the backbone at all, I called it the Rotamerically Induced Perturbation. At first, I was mimicking Ota's studies but then one day, I decided to crank up the perturbation, and to my absolute astonishment, I say a coherent 10 Ångstrom motion of an α-helix in a 10 ps simulation.

That's when I realized I could use this method to explore large surface motions of proteins. Anyway, the rest of the project involved testing RIP against a whole bunch of proteins with well-studied dynamics, and figuring out how to interpret the simulations. The studies are done, published, and unlike the ATD, the code is not sitting on an obsolete machine in the back of the room. It has been released into the wild.