Computational protocol: Applying an Empirical Hydropathic Forcefield in Refinement May Improve Low-Resolution Protein X-Ray Crystal Structures

Similar protocols

Protocol publication

[…] It is obvious and unassailable that current protocols for model-building and refinement based on low-resolution X-ray reflection data produce structural models of poorer quality than those based on high-resolution data. We are testing the hypothesis that these deficiencies can, at least in part, be related to the lack of well-developed hydropathic interaction networks in these models. We have sought to illustrate this point with available crystallographic data, but there is a paucity of directly comparable and unbiased structural data for proteins solved at varying resolutions. Another approach, used in this work, is to synthesize low-resolution data by truncating high-resolution data (vide infra) and evaluate structures refined against these data . In (red circles) we present normalized (relative to the crystallographic structure model) intramolecular HINT scores, calculated for 309 structural models for 25 proteins refined against data truncated at resolutions between 1.48 and 4.88 Å. This score is calculated as the sum of all non-covalent intramolecular atom-atom interactions using the paradigm described above, i.e., higher scores represent in toto more favorable high-quality interactions within the structure. Clearly, there is a trend of an accelerating decrease in HINT score, especially for resolutions worse than 3.0 – 3.5 Å, indicating that, just as we hypothesized, these models indeed have poorer quality hydropathic interaction networks. Another evaluation of structure as a function of resolution can be obtained by calculating non-covalent energies of structure models with a molecular mechanics forcefield. The CHARMM electrostatic term (, red circles) shows a similar trend: between 3.0 and 4.8 Å there is a more than 30% decrease in favorable electrostatic energies, relative to those in the crystallographic models, again in accord with our hypothesis. This theme is repeated with other knowledge-based structural metrics including Ramachandran scores (percentage of residues in the favored regions), as illustrated (red circles) in . All of these data confirm that there is a clear tendency towards decreasing structural quality as the experimental resolution of the data is decreased.One approach to probe, and perhaps ameliorate, the disparity between structural models refined with high- and low-resolution data, is to include electrostatic interactions in X-ray refinement protocols. If electrostatics substantively improves structural quality, we can assert that compromises to polar interaction networks, e.g., hydrogen bonds or weaker, longer-range acid-base interactions, are the dominant source of structural errors in low-resolution structural models. On the other hand, partial or negligible changes in structure quality would strongly suggest that other factors are at play. In (green circles), we present normalized intramolecular HINT scores, normalized electrostatic energies from CHARMM and Ramachandran scores, respectively, for structures refined with the optional electrostatics protocol in CNS, which we are terming “CNS+electrostatics”. While the HINT scores () are higher overall by about 25% after refinement with this protocol, the trend of decreasing HINT score with resolution is essentially unchanged. Electrostatic energy () is likewise stabilized by about 15%, but even this, which essentially reports the same property used in its optimization, trends to lower values (higher energies) with lower resolution. Finally, Ramachandran scores () suggest that refinement with electrostatics only modestly improves structural quality (4% improvement at 3.5–4.0 Å and 2% at ≥4.0 Å) for models from low-resolution data. The lack of significant improvement of the latter is especially notable as it is an independent and universally accepted structural metric. Furthermore, the higher overall HINT scores and lower electrostatic energies, which were both referenced to their deposited high-resolution structural models, suggests that the inclusion of electrostatics in refinement may result in models with non-native (and potentially overweighted) polar interaction networks.In the remainder of this paper we describe the implementation and testing of a structure refinement protocol enhanced with the HINT hydropathic forcefield. It is our view that, because Coulombic electrostatic terms focus exclusively on polar components of interaction networks, refinement with electrostatics is, at best, inadequate for improving the quality of low-resolution structure models. It is important to also include terms that improve the independent and complementary hydrophobic component of the networks. [...] Structural quality can also be assessed by knowledge-based metrics that “rank” a structure with respect to others. Model quality, as reported by indices like the Ramachandran score or MolProbity clashscore, has been shown to worsen with decreasing resolution. Histograms for Ramachandran scores () and clashscores (a measure of the number of unusually short interatomic distances in a structure, ) report the same trend: while inclusion of electrostatics alone has only a modest impact, the inclusion of the HINT representation of non-covalent interactions results in much more significant improvements in structure quality. The HINT potential, which is based on pairwise non-covalent interactions, has no “intrinsic knowledge” of preferred peptide backbone angles, yet the CNS+HINT models have its inclusion a significantly higher fraction (13% larger for resolution ≥4.0 Å) of residues in favored regions of the Ramachandran plot. Clashscores () show an even more dramatic (51% at ≥4.0 Å) improvement for the CNS+HINT structures. In addition, since the clashscores for the native and CNS+electrostatics refined structures are virtually identical, the anomalously low electrostatic energies and increased HINT scores (relative to reference) for CNS+electrostatics models () are, in part, an artifact of abnormally short interatomic distances between polar atoms. In contrast, the better clashscores from CNS+HINT refinement strongly suggests that this protocol results in better-defined interaction networks. […]

Pipeline specifications

Software tools CHARMM, CNS, MolProbity
Application Protein structure analysis
Organisms Dipturus trachyderma