One of the most promising recent developments in computational biochemistry is the development of potential capability to computationally design entirely new enzymes from scratch that can perform reactions inaccessible to naturally occurring proteins. Such enzymes can be of great utility as novel biofuels, synthetic reagents and new drugs. A particularly noteworthy set of publications in this regard was from David Baker’s and Ken Houk’s groups in Seattle and Los Angeles. In 2008, the groups designed an enzyme for performing a base-catalyzed Kemp elimination, a reaction which converts a benzisoxazole into a cyanophenoxide by proton abstraction.
A couple of months back, they again made news by designing a Diels-Alderase from scratch, an enzyme catalyzing the DA reaction whose natural counterpart does not exist. Although the catalytic rates they obtained were relatively modest compared to the best rates seen in nature, this is still an important and remarkable step forward.
The studies hinged on two tools- quantum mechanical design of a transition state for the enzymatic reaction, and buildup of the protein architecture around this ideal transition state using the program Rosetta. In the first step, a transition state for the reaction was surrounded by specific amino acid residues and optimized in an ideal geometry. This arrangement is called a “theozyme”, an ideal, minimal theoretical construct. In the second step, Rosetta was used to ‘embed’ this theozyme in a protein framework borrowed from existing protein structures in the PDB. Iterative cycles of optimization of the amino acids around the reactants led to several designs. Some of these designs turned out to be active, and crystal structures revealed the remarkable similarity between the computer and real-life counterparts. However, there was no easy way except actual testing to distinguish active and inactive designs beforehand (the inactive designs could not be crystallized since by definition they were probably too unstable to form crystals).
Now a new analysis nicely looks at the difference between the active and inactive designs obtained for the Kemp elimination. The authors first try to use static quantum chemistry calculations to resolve the difference between the two sets. Unfortunately this does not work very well since the energy difference between the sets are quite small, about 2 kcal/mol, and QM methods for such complex systems can often produce comparable errors.
However, enzymes are dynamic creatures, and it’s probably not too surprising that one has to resort to dynamical studies to discern the differences between active and inactive structures. To this end, the authors used 20 ns MD simulations. They compared the results with two designed Kemp eliminators (including one antibody) whose crystal structures are available. Firstly, they simply observed the mobility of the residues in the active sites and found out that in general, residues in the active designs don’t move around as much as those in the inactive ones, indicating stable packing. Then they looked at the hydrogen bonds holding the reactants together. In general, they found a tighter distribution of hydrogen bond angles in the active and crystallized structures compared to the inactive ones. This observation would be in keeping in line with the optimized hydrogen bonding networks in active sites. Lastly, they looked at accessibility of water in the active site. The base involved in the Kemp elimination is a carboxylate. Ideally, this carboxylate would not be solvated so that it is free to serve as a base. Indeed, analysis of water molecules surrounding this carboxylate indicated that while the carboxylates in the active, crystallized proteins are almost completely free of water molecules, the carboxylates in the inactive designs are typically surrounded by a couple of water molecules. This also again confirms the ‘looser’ packing in the inactive sites.
This is a nice study because it not only validates MD as a possible tool to distinguish and rank active and inactive designed enzymes, but it also provides insights into the basic physical features of optimized enzyme active sites. Compact packed side-chains, optimal hydrogen bonding geometries and relative inaccessibility of key residues to water is about what you would expect evolution to do when asked to come up with good enzyme designs.
Kiss, G., Röthlisberger, D., Baker, D., & Houk, K. (2010). Evaluation and ranking of enzyme designs Protein Science DOI: 10.1002/pro.462