Text

theses

ETD doctoral

As RNA structures continue to be solved at a rapid pace, and as RNA has become a target for therapeutics and has been found to have many different functions other than basic nucleic acid functions, the importance of properly modelled structures continues to grow. With the conventional restraints used in crystallographic refinement, persistent outliers and errors crop up in the structures published. The use of AMBER-derived restraints in the PHENIX crystallographic refinement process has been proven to improve the structures of proteins modelled based on experimental data, and is implemented in RNA structures in this thesis. Further improvement of structural modelling can be made in solvent description. While the most accurate way to model solvent is through explicit solvent molecules in crystal MD simulations, it is also the most computationally expensive. Meanwhile, the faster implicit models, such as the Generalized-Born model, are approximate and can sometimes lead to improper secondary structure in macromolecules. The periodic 3D-RISM method, presented in Chapter 3 and an upcoming paper, calculates densities for each solvent entity. It is thought to be more accurate than general implicit methods, but is faster than explicit methods. In this thesis, crystal MD simulations and periodic 3D-RISM calculations are employed to study the solvation of RNA structures.

In Chapter 2, parallel refinements with conventional and AMBER-derived restraints in PHENIX on RNA molecules are presented. The resultant structures are analyzed via energy calculations and MolProbity analysis. The results show that in a data set of 21 structures, the AMBER restraints lead to improved electrostatic and non-bonded interactions over conventional restraints, which was expected, as this is the main improvement of AMBER restraints over conventional restraints. This leads to overall energetic improvement over the course of the data set, except for the structure at highest resolution. Also, this occurs with little concession to structure factors, as the r-free factors are very similar. There are increases in the r-work, but the r-gap, or the gap between the r-work and r-free factors (an indication of over-fitting when high), is generally the same if not decreased as compared to the conventionally restrained refinements. The geometric outliers are more numerous for AMBER-restrained structures, but analysis and testing of repetitive bond and angle outliers finds that this appears to be due to both a larger distribution of angles and bond lengths due to the interconnectedness of all the energy terms in AMBER, as well as a difference in the ideal values for these terms between AMBER and MolProbity. At low resolution, where the experimental data is poor and the need for external restraints is greatest, there is even greater improvement. This implies greater physical accuracy of the structures produced, and could lead to improved structural understanding.

In Chapter 3, the periodic 3D-RISM theory is presented. The existing 3D-RISM code for non-periodic systems has been expanded to periodic systems, which allows for the possibility of use in refinement description of solvent. Results are presented for experiments in proteins and RNA comparing refinement with the standard flat density, 3D-RISM results, and explicit MD solvent densities, which show that 3D-RISM improves r-factors over the standard density, while being improved upon by MD, which is more time-consuming. Results for different proteins are also presented showing that the number of water molecules produced via 3D-RISM calculations are all very similar to the numbers derived from crystal MD. Further work including calculation of 3D-RISM solvent throughout refinements may be the next step.

In Chapter 4, crystal MD simulations of three of the structures from the PHENIX data set are presented as an opportunity to look at the dynamics of the structures as well as a baseline for testing the accuracy of 3D-RISM code presented in Chapter 3. The 3D-RISM calculations match fairly closely the number of water molecules found by MD. Through comparative 3D-RISM calculations, it is found that solvent composition has an effect on the number of ions produced to neutralize the solute. More sodium ions are used than potassium ions when used at the same concentration in conjunction with magnesium. As sodium’s ionic radius is smaller than potassium’s, it appears that this is due to size differences. The 3D-RISM code provides an appropriate, and less time-expensive, approximation of solvent description and interactions than the standard crystal MD simulations.

In Chapter 5, the sarcin/ricin domain of the ribosomal RNA of E. coli, a well-conserved domain across species with many structures in the PDB, is used as a test molecule for PHENIX refinements with AMBER restraints, periodic 3D-RISM singlepoint calculations, and minimizations with periodic 3D-RISM. These small structures all contain the same solute, so the differences in results should be resolution- or solvent-dependent. When analyzing the PHENIX results, there is a trend in energetics, specifically non-bonded interactions, toward greater improvement over conventional restraints with AMBER restraints as the resolution worsens. Different parameters were tested to determine what sets resulted in the fastest runs.

ETD_11055

Ph.D.

Includes bibliographical references

doi:10.7282/t3-mk47-d911

The author owns the copyright to this work.