Dealing with Flexible Receptors

In today's world of high-throughput screening, chemists often use computers to audition thousands upon thousands of small molecules for coveted roles as new drugs, based on the molecules' talent for binding with a target protein.

There's a problem, though: Despite advances in computational chemistry, so-called docking programs can still overlook some of the most important potential drug candidates. And that's because those programs still aren't fully capable of modeling the true biochemical picture.

More than 100 years ago, chemist Emil Fischer envisioned ligand-protein binding as the fitting of a rigid key into a rigid lock. Since the 1950s, however, chemists have recognized that neither proteins nor ligands are static. In reality, they're dynamic performers, flexing and twisting in their watery biological baths.

Over the years, as computer power and theoretical sophistication have grown, computational chemists have been able to begin to address the problems posed by molecular flexibility. In fact, most high-throughput docking programs now take into account the flexibility of the small ligands.

But the last hurdle--the flexibility of the protein itself--is a large one. Adding protein, or receptor, flexibility into the binding simulation drastically increases its complexity, because the computer must figure out how the protein will move and then test myriad ligands against a host of protein conformations.

It's not that chemists don't know how to model what's going on. In fact, they can describe most of the flexible interactions of an individual protein and ligand quite well. But to dock a library containing hundreds of thousands of molecules into a wriggling protein is a computational nightmare, beyond the limits of today's computers.

"We say receptor flexibility is the bleeding edge," says Brian K. Shoichet, an associate professor of pharmaceutical chemistry at the University of California, San Francisco, and a docking expert. "It is a frontier for this aspect of molecular recognition."

Recently, chemists have made inroads on the problem, devising strategies for handling receptor flexibility while tailoring the process for high-throughput drug discovery. About 10 years ago, chemists introduced "soft docking," a more computationally feasible strategy, where the receptor was artificially made "spongy" and less discriminating to ligands.

STILL ANOTHER improvement came with the modeling of protein side-chain rotations. Libraries of rotameric states were used to create different receptor conformations into which the ligands could then be docked.

However, these techniques still don't treat the motion of protein backbones, and that omission can have serious consequences. For example, protein kinase enzymes, which are potentially important targets for anticancer drugs, have hinged structures that close up when they bind.

But now, a number of research groups are taking the computational complexity even further, focusing on various ways of sampling the different conformations a protein might assume--including backbone motions.

It's thought that proteins can flip between a number of conformational states and that a chance meeting of the right receptor conformation and the right ligand can lead to unexpectedly tight binding.

One way to explore those conformations is with molecular dynamics simulations, using programs such as AMBER or CHARMM. "It's possible to generate a lot of snapshots of what the protein might look like," says J. Andrew McCammon, chemistry professor at UC San Diego. A technique developed by McCammon, known as "accelerated molecular dynamics," can generate an even greater variety of snapshots. Another way is to use structures, determined from X-ray crystallography or nuclear magnetic resonance (NMR) spectroscopy, of a single protein bound to different ligands. Chemists then rank the binding ability of the different ligands in terms of the energy or geometry of the receptor-ligand complex.

Five years ago, in what's regarded as pioneering work, the lab of Irwin D. Kuntz Jr., emeritus chemistry professor at UCSF, used NMR and crystal structures to generate an ensemble of protein conformations. The group tested the binding of numerous ligands to the ensemble using the program DOCK.

Since then, the strategies have proliferated. Both McCammon and Heather A. Carlson, chemistry professor at the University of Michigan, Ann Arbor, who was also a postdoc of McCammon's, use molecular dynamics simulations to generate suites of protein conformations.

McCammon's "relaxed complex method" docks entire ligands into the suite. He selects a subset of the most favorable conformations and then performs even more computationally intensive calculations of binding free energy to whittle down the candidates further. His group, including graduate student Alexander L. Perryman, found that recent studies of the immunologically important FK506-binding protein FKBP-12 compared well with crystallographic data.

Carlson's "multiple protein structure pharmacophore" method involves overlaying all the snapshots from the simulation and looking for regions where chemical features, such as a hydrogen bond or an aromatic group, remain the same. Her group then screens ligand databases for their ability to bind to small collections of features, known as pharmacophores. Docking to these simple pharmacophores is much faster than docking to complex protein surfaces.

Leslie A. Kuhn, biochemistry professor at Michigan State University, East Lansing, takes a unique approach to docking, representing receptor backbone flexibility by sampling motions consistent with the network of covalent and noncovalent bonds. "The major difference is that no potential is involved, while low-energy diverse conformations are guaranteed [to be found]," Kuhn says. "Calculating the interatomic potential is what takes so much computer time."

THE AMOUNT of cross-linking of hydrogen bonds and hydrophobic interactions in the receptor helps determine which bonds are fixed and which are free to rotate. The program ROCK, developed with graduate student Ming Lei and physics professor Mike Thorpe at Arizona State University, Tempe, calculates the large-scale backbone motions of a protein. When Kuhn's group combines these calculations with their own high-throughput docking program, SLIDE, they can model additional receptor and ligand side-chain flexibility.

Computer science professor Thomas Lengauer at Max Planck Institute for Computer Science in Saarbrücken, Germany, and his colleagues have created FlexE, a modified version of their docking program FlexX, to model receptor flexibility based on crystallographic structures.

UCSF's Shoichet is another major player who uses ensembles of experimentally determined structures as a jumping-off point for exploring receptor conformations. His group recently discovered that flexibility simulations can sometimes create absurd receptor conformations. The only way to spot these "bad eggs" is by also considering the receptor's conformational energy, Shoichet says.

Alfredo Di Nola, chemistry professor at the University of Rome, and colleagues developed a method they call molecular dynamics docking to simulate flexible receptor binding in solution.

The software company Schrödinger recently combined its docking program Glide with its protein structure prediction program Prime to treat flexibility, which has led to accurate prediction of a number of receptor-ligand complexes. "What needs to be modeled is the conformation of the protein when it binds the ligand--not how it got there, and not how fast," says Ramy Farid, Schrödinger's vice president for scientific development and program management.

All these approaches weren't possible even a short time ago. "The problem was that computers weren't powerful enough to do docking on several different protein conformations," Carlson notes. "I think [increased computer power] is part of the reason why these techniques are taking off."