Molecular replacement: finding the right search model

Acta Cryst. (2001). D57

The October 2001 issue of Acta Cryst. Section D (volume D57, part 10) is the collected proceedings of the annual CCP4 Study Weekend held this January at the U. of York, UK, and entitled 'Molecular replacement and its relatives'. The most frequent stumbling block to successful ab initio solution of a macromolecular crystal structure by molecular replacement is low structural similarity between the search model and the target structure, and many of the articles addressed this important problem of finding the optimal search model. Owing to space constraints, this is a personal selection of articles that highlight this theme.

Any improvement in the success rate of molecular replacement methods will clearly have an impact on the outcome of the structural genomics projects now underway or planned, given that a significant proportion of the proteins coded by a genome are likely to contain previously unknown folds, or folds having very low structural similarity with known structures.

In 'Pushing the boundaries of molecular replacement with maximum likelihood' Randy Read demonstrates how his new 'BEAST' program utilizes a likelihood-based scoring function that can compensate for a poor model. Likelihood-based scoring functions, which measure the probability that the values of the experimental observations (in this case the X-ray diffraction intensities) would have been obtained if the current model were correct, are now used routinely in model refinement, and their use in molecular replacement is a logical extension of this. The author shows that combining multiple models, weighing them according to their likelihood of similarity with the target structure provides improved results.

On the subject of model quality, in 'Molecular replacement by evolutionary search' Charles Kissinger reports that polyalanine models are frequently better as search models for protein structures than ones that include side-chain atoms, even when the co-ordinates of those atoms are known quite accurately (though normally less accurately than those of the main-chain atoms). This may seem somewhat counter-intuitive, but is in accord with a common observation that less complete but more accurate models are often better than more complete but less accurate ones.

Other articles focus on less traditional types of molecular replacement model, both structural and non-structural (such as an electron density map or a molecular envelope), including models obtained by NMR, EM or solution scattering techniques.

In 'Molecular placement of experimental electron density: a case study on UDP-galactopyranose mutase' Jim Naismith describes use of the subunit-averaged electron density from a P2₁ crystal form of the enzyme as a molecular replacement model for a P2₁2₁2₁ crystal form. The relationship between the crystal forms so obtained allowed use of multiple crystal averaging to solve the structure.

In 'Phasing from an envelope' Quan Hao describes the use of molecular envelopes determined from low-resolution X-ray scattering data from a protein in solution to orient and position the molecule in the crystal by a real-space search technique.

In 'Using electron-microscopy images as a model for molecular replacement' Eleanor Dodson reviews the technical problems associated with using EM models for ab initio solution of crystal structures by molecular replacement. Not the least of these problems are the widely varying conventions for describing data structures (in particular those for electron density maps) used by electron microscopists and X-ray crystallographers!

In 'NMR trial models: experiences with the colicin immunity protein Im7 and the p85α C-terminal SH2-peptide complex' Richard Pauptit describes the non-trivial solutions of two protein structures using ensemble NMR models from which the more flexible portions have been excised. In a similar vein, in 'Solution solution: using NMR models for molecular replacement' Yu Wai Chen describes a protocol for preparation and use of NMR-derived models and results of its application.

Finally, two articles assess the potential of using all available deposited structural information, directly or indirectly to develope better search models. In 'Creating structure features by data mining the PDB to use as molecular-replacement models' Tom Oldfield concludes that more advanced molecular replacement techniques than those currently available will be required to allow use of protein fragments derived by data-mining techniques as search models.

In 'Evaluating the potential of using fold-recognition models for molecular replacement' David Jones analyses all the structures in the PDB solved by molecular replacement in order to assess the minimal degree of structural similarity required for successful solution, and then goes on to use this as a yardstick to judge the current potential of tertiary structure prediction and threading methods to generate useful molecular replacement models. He attempts to estimate the potential impact of these methods on the success rate of a typical structural genomics project, arriving at an estimate that about 50% of the protein structures from a ‘typical’ genome will be soluble by a combination of structure prediction, modelling and molecular replacement.

This issue of Acta Cryst. D provides a comprehensive update of current work in this field, whose overall strategic value will be immense when the goal of ‘complete coverage of all the protein folds’ is realised. This goal is a major funded objective of the NIH structural genomics initiatives in the USA.

Ian Tickle, Astex Technology Ltd, Cambridge, UK