Date Published: January 01, 2019
Publisher: International Union of Crystallography
Author(s): Nigel W. Moriarty, Paul D. Adams.
A set of restraints for an iron–sulfur cluster based on small-molecule structures was generated and tested in structure refinement. Additionally, the small-molecule structures also provided bond and angle restraints for linking the cluster to the coordinating cysteine residues.
Using accurate geometric restraints is essential in macromolecular crystallography in order to arrive at chemically meaningful atomic models. The experimental data, even when available at very high resolution, are typically unable to unambiguously define the exact conformation, and therefore prior chemical knowledge is included in the form of geometric restraints. Relying on quantum calculations to help to define these restraints can be very productive (Moriarty et al., 2009 ▸), but for metal clusters the challenge usually exceeds the available resources because of the high basis-set levels that are required for accurate calculations, not to mention the variability in possible geometries. Therefore, the use of high-quality experimental data, typically from small-molecule crystallography, to generate restraints and subsequent validation using a large number of refinements is a common paradigm. This procedure generally makes uses of the r.m.s. deviation between the target restraints and the refined models as a metric. We have used this approach to define accurate restraints for iron–sulfur clusters.
When developing accurate experimental ligand geometries, there are two main sources of information in the field of macromolecular crystallography. One choice is small-molecule structure databases such as the Cambridge Structural Database (Groom et al., 2016 ▸) or the Crystallography Open Database (Gražulis et al., 2009 ▸). The other choice is the very high-resolution macromolecular structures in the Protein Data Bank (PDB; Berman et al., 2000 ▸). Both have their pros and cons (Long et al., 2017a ▸), but in this study both the CSD and the PDB were used.
The three structure searches of the CSD (Groom et al., 2016 ▸) for bond and angle values for the Fe4S4 cluster SF4 resulted in the values and statistics given in Table 1 ▸. The bond distance for S—Fe for the strictest search is 2.29 ± 0.02 Å, which is essentially the same as the values for the other two searches, X(0.05) and X(0.1), and for the coordinating bonds (Fe—SAA). The bond distance from the PDB search is 2.30 ± 0.03 Å, which is in close agreement with the CSD results. This agrees well with the results reported in Tan et al. (2013 ▸), which list all bond lengths for clusters ligated to sulfur as spanning these values. The value is also in good agreement with that posted by Oliver Smart to the CCP4 bulletin board in 2014: 2.298 Å.
The 239 PDB entries cover the resolution range from 0.5 to 3.4 Å, with the best coverage from 1.3 to 3.0 Å. Most of the SF4 geometries were rhomboid but 23, or nearly 10%, were cubic, with an additional 15 (6.3%) of input models having incorrect atom naming. Most metrics such as R factors, Ramachandran, rotamer and clashscore are similar, with some noise in the limits. However, the bond and angle r.m.s.d. values show significant variations. The bond and angle r.m.s.d. values for the entire models (dashed lines) are shown in Fig. 3 ▸. The r.m.s.d. values for the entire model change very little with respect to resolution owing to the limited impact of a small number of deviations corresponding to the metal clusters, but as expected there is a small increase at high resolution because the data provide more information to define the final geometry.
New restraints using a rhomboid geometry have been added to the GeoStd restraints (N. W. Moriarty & P. D. Adams; https://sourceforge.net/projects/geostd/) for use in all PHENIX programs from v.1.13. The restraints can also be loaded into Coot. Both the CSD values and the PDB values were accurate for macromolecular refinement, with the former being demonstrated to provide improved geometries.