Section 23.4.2. Determination of water molecules

The most prominent method by which the structure of water molecules on the surface of macromolecules can be observed at the atomic level is X-ray crystallography. The information classically available from this methodology is on bound water molecules, characterized by a high probability density and reduced mobility relative to the bulk solvent, which results in clearly observed electron density. Information on water structure at larger distances from the protein is available in the low-resolution reflections, but difficulty in modelling the solvent in these areas has led to the common practice of discarding the very low resolution data. This chapter focuses on the water molecules for which there is information at high resolution (>3 Å), although great progress has been made in recent years in modelling the disordered water structure at the protein–solvent interface, enabling more effective use of the low-resolution data (Badger, 1993; Jiang & Brünger, 1994; Lounnas et al., 1994). Typically, one is interested in studying solvent structure because of the effects that it has on the protein. Lounnas et al. (1994) gave a particularly interesting focus on the effect of the protein on the solvent structure surrounding it. Using a combination of molecular-dynamics simulations of explicitly solvated myoglobin and the low resolution X-ray data from myoglobin crystals, they devised a method to describe the effect of the protein on the solvent structure to a distance of 6 Å from the surface. They found that the mobility and probability density of water molecules perpendicular to the protein surface varied considerably depending on the particular composition and three-dimensional structure of the amino-acid residues at the particular area of interest (Lounnas & Pettitt, 1994; Lounnas et al., 1994).

There are a variety of criteria that have been used in placing crystallographic water molecules in electron-density maps. For tightly bound water molecules, with low B factors, the placement involves little or no subjectivity, but the choice of whether or not to include the more disordered waters (or those with low occupancy) can be rather subjective. It generally involves picking the electron-density contour level and B-factor cutoffs as well as making a choice of whether to use a simple difference electron-density map () or to use a higher-order difference electron-density map ( or ). One criterion, applied consistently in placing water molecules on the surface of elastase structures, is the simultaneous presence of electron density at the 3σ contour level in an electron-density map and at the 1σ contour level in a electron-density map. After refinement, those waters are kept that have a B factor of 50 Ų or less. A few exceptions do occur, where there is clear electron density for a water molecule with a B factor of up to 60 Ų. Virtually all of the water molecules placed by these criteria have at least one hydrogen bond to a protein atom or to another water molecule and are mainly part of the first hydration shell on the protein surface.

A method that has provided information on solvent structure complementary to that obtained by X-ray crystallography is based on D₂O − H₂O neutron difference maps (Shpungin & Kossiakoff, 1986). The main advantage of this methodology is in locating partially ordered water molecules whose electron-density peaks may be at the limit of the signal-to-noise ratio allowed for confidently determining positions of water molecules by X-ray diffraction. Scattering of neutrons by H₂O and D₂O is quite different, while scattering from the protein remains the same. Therefore, difference maps based on the two data sets should average to zero where the protein is present and result in peaks only where water molecules are found. Neutron scattering is particularly suited to this because of the threefold greater scattering power of deuterated water molecules relative to light water, providing a larger signal-to-noise ratio in assigning water positions. This method is particularly useful in detecting the second hydration sphere on protein surfaces (Kossiakoff et al., 1992).

NMR spectroscopy can also serve as a complementary technique, providing dynamic information on the lifetime of interaction of a single water molecule on the protein surface. The fact that, with few exceptions, no cross-relaxation peaks are observed at the protein–water interface is an indication that the motion timescale for water molecules in contact with protein is close to that in bulk water at room temperature. The NMR data suggest that water molecules observed in crystal structures have lifetimes of the order of tens of nanoseconds or less (Bryant, 1996). A small number of relatively long-lived structural waters (with residence times in the range 10⁻² to 10⁻⁸ s) can be detected by modern NMR techniques (Otting et al., 1991). Four water molecules have been detected by NMR in bovine pancreatic trypsin inhibitor (BPTI) (Otting & Wuthrich, 1989) and six have been observed in complexes of human dihydrofolate reductase with methotrexate (Meiering & Wagner, 1995). Observation of these waters in the corresponding crystal structures reveals that they are tightly bound waters, with three or four hydrogen bonds to protein atoms, and many are found to bridge between secondary-structure elements or are found to mediate protein–ligand interaction (Meiering & Wagner, 1995). It is important to understand, then, that water molecules near protein surfaces occupy energy minima favoured by hydrogen bonding and ion–dipole effects, which results in water molecules being present in these positions more often than in others. Although when looking at a crystallographic protein structure it is easy to think of a given site as being occupied by a single water molecule, it is in fact only the site that is single, with an enormous number of different individual water molecules sampling it during the time of data collection. This was qualitatively understood from the beginning, but NMR experiments have played a key role in setting quantitative upper boundaries to the residence times of water molecules on the protein surface.

Finally, mention must be made of the computational efforts invested in representing and understanding solvent structure on macromolecular surfaces. The computational work encompasses a variety of methodologies, including integral equation methods (Beglov & Roux, 1997), molecular dynamics (Brooks & Karplus, 1989; Hayward et al., 1993; van Gunsteren et al., 1994), thermodynamic understanding through free-energy simulations (Roux et al., 1996) and statistical-mechanics calculations (Lazaridis et al., 1995). The results of these studies are often complementary to the experimental information already available and provide an important component to the current insight on solvent structure (McDowell & Kossiakoff, 1995) and function (Pomes & Roux, 1996; Oprea et al., 1997). Furthermore, these techniques often provide the only means of obtaining an energetic understanding of some aspects of protein–water interaction.

The question of how the different techniques used to observe the location and properties of water molecules on the surface of proteins relate to and complement one another has been discussed in two short review articles (Levitt & Park, 1993; Karplus & Faerman, 1994). Karplus & Faerman discuss the reliability of each of the methods, illustrating their strengths and weaknesses, while Levitt & Park present the state of our understanding of protein–water interactions as it was in 1993, based on the synthesis of results obtained from the various methods discussed above.