Toward ab initio refinement of protein X-ray crystal structures: interpreting and correlating structural fluctuations

The refinement of protein crystal structures currently involves the use of empirical restraints and force fields that are known to work well in many situations but nevertheless yield structural models with some features that are inconsistent with detailed chemical analysis and therefore warrant further improvement. Ab initio electronic structure computational methods have now advanced to the point at which they can deliver reliable results for macromolecules in realistic times using linear-scaling algorithms. The replacement of empirical force fields with ab initio methods in a final refinement stage could allow new structural features to be identified in complex structures, reduce errors and remove computational bias from structural models. In contrast to empirical approaches, ab initio refinements can only be performed on models that obey basic qualitative chemical rules, imposing constraints on the parameter space of existing refinements, and this in turn inhibits the inclusion of unlikely structural features. Here, we focus on methods for determining an appropriate ensemble of initial structural models for an ab initio X-ray refinement, modeling as an example the high-resolution single-crystal X-ray diffraction data reported for the structure of lysozyme (PDB entry “2VB1”). The AMBER force field is used in a Monte Carlo calculation to determine an ensemble of 8 structures that together embody all of the partial atomic occupancies noted in the original refinement, correlating these variations into a set of feasible chemical structures while simultaneously retaining consistency with the X-ray diffraction data. Subsequent analysis of these results strongly suggests that the occupancies in the empirically refined model are inconsistent with protein energetic considerations, thus depicting the 2VB1 structure as a deep-lying minimum in its optimized parameter space that actually embodies chemically unreasonable features. Indeed, density-functional theory calculations for one specific nitrate ion with an occupancy of 62% indicate that water replaces this ion 38% of the time, a result confirmed by subsequent crystallographic analysis. It is foreseeable that any subsequent ab initio refinement of the whole structure would need to locate a globally improved structure involving significant changes to 2VB1 which correct these identified local structural inconsistencies.