Learning about Biomolecular Solvation from Water in Protein Crystals

Abstract

Water occupies typically 50% of a protein crystal and thus significantly contributes to the diffraction signal in crystallography experiments. Separating its contribution from that of the protein is, however, challenging because most water molecules are not localized and are thus difficult to assign to specific density peaks. The intricateness of the protein-water interface compounds this difficulty. This information has, therefore, not often been used to study biomolecular solvation. Here, we develop a methodology to surmount in part this difficulty. More specifically, we compare the solvent structure obtained from diffraction data for which experimental phasing is available to that obtained from constrained molecular dynamics (MD) simulations. The resulting spatial density maps show that commonly used MD water models are only partially successful at reproducing the structural features of biomolecular solvation. The radial distribution of water is captured with only slightly higher accuracy than its angular distribution, and only a fraction of the water molecules assigned with high reliability to the crystal structure is recovered. These differences are likely due to shortcomings of both the water models and the protein force fields. Despite these limitations, we manage to infer protonation states of some of the side chains utilizing MD-derived densities.

Water structure in solution and crystal molecular dynamics simulations compared to protein crystal structures

“…These results agree with most previous investigations. [8][9][10][11] The recall of 43-70% at 1.4Å obtained here is comparable to the recall scores obtained by Higo and Nakasako, 9 60%, or Altan et al, 10 70% at this distance. These values were obtained at a density threshold of $0.6 e ÀÅÀ3 , a lower threshold than used here, which explains why our lower-bound results are worse than in the previous studies.…”

“…S3 †), as was also observed in previous studies. 10,12 These results show that the MD simulations reproduce the positions of the crystal water molecules rather poorly. These results agree with most previous investigations.…”

“…Following previous studies, this will be termed the water recall statistics. [8][9][10][11][12] We also studied the fraction of MD water peaks that have a corresponding crystal water at the same distance thresholds. As in previous studies, this is dened as the water prediction statistic.…”

“…9 Another MD study in crystal was carried out by Altan et al for a mannose-binding protein and 70% of the crystallographic waters were found to be within 1.4Å of an MD water. 10 Rudling et al managed to reproduce >70% of water molecules in the crystal structures of 12 different proteins within 1Å. 11 However, only water molecules in the binding site of the proteins were studied, which are usually well-ordered.…”

Using Schematic Models to Understand the Microscopic Basis for Inverted Solubility in γD-Crystallin

“…where A 4 is the solvent accessible surface area of Patch 4, ρ w = 3.3 × 10 −2Å −3 is the number density of water in the bulk at room temperature, and g C (r) is the radial distribution function of water around carbon atoms determined in Ref. 48. This estimate thus assumes that (i) the solvent has a radius of 1.4Å (the SASA definition), (ii) the average van der Waals radii of protein heavy atoms is ∼ 1.6Å, and (iii) the first solvation shell ends with the first peak of g(r) at 4.5Å.…”