Water, water, everywhere… It's time to stop and think

Highlights

• Six computational methods for water molecule placement and analysis reviewed alongside relevant case studies.

• Field is approaching maturity.

• Adoption into mainstream drug discovery programs will increase as more literature-based case studies are published.

Despite the numerous methods available for predicting the location and affinity of water molecules, there is still a degree of scepticism and reluctance towards using such information within a drug discovery program. Here, I review some of the most common and popular methods to assess whether these apparent concerns are justified. I suggest that the field is approaching maturity and that some methods are capable of giving quantitative predictions, which are confirmed experimentally. This suggests that water-placement methods should be used more widely to help direct chemistry efforts, although more successful examples are required to help validate the techniques.

Introduction

One of the axioms of 21st-century drug design and development is to consider the role of water molecules in the active site of the protein target. Built upon the initial studies by Poornima and Dean 1, 2, 3, together with those of Ladbury [4] and Dunitz [5] during the mid-1990s, this axiom broadly states that weakly bound water molecules in proteins should be targeted for ligand replacement, whereas strongly bound waters are ideal candidates for ligand stabilisation. As a result of this thinking, a plethora of papers has been published in the subsequent years that seek to explain or understand the nature of hydration in proteins. As shown in Fig. 1, a literature search conducted with SciFinder for ‘Water molecules in protein binding sites’ revealed a dramatic increase in the number of annual publications from the mid-90s to 2010.

Crystallography provides direct evidence for the presence of water molecules within proteins, although limiting factors, such as resolution and ambiguous electron density, can affect both the number and temperature factors of water molecules added into the structure 6, 7, 8. Studies have suggested that some solvent-accessible binding sites can either be partially filled or even devoid of water 9, 10, and crystallographic approaches are stretched to reliably predict the position of such water molecules that typically exhibit favourable entropy compared with highly localised water molecules that are most frequently observed in crystal structures [11]. As a result, computational approaches for locating water molecules have been developed to further advance our knowledge of active-site hydration.

Such approaches have helped to mature our understanding of the role of water in binding processes, and have built upon the original postulation that water is the ‘third party’ [4] in protein–ligand interfaces [12]. Computational methods have helped to highlight cases where water molecules can influence protein–ligand binding kinetics 13, 14, 15, affect thermodynamic binding signatures 16, 17, 18, or even govern selectivity within kinases 19, 20. They have also been used in conjunction with experimental procedures, such as isothermal titration calorimetry and X-ray crystallography, to explore the nature of the hydrophobic effect. Notably, the Whitesides group has highlighted that water networks can contribute to enthalpy/entropy compensation upon protein–ligand binding using computational approaches 21, 22, while also using them to understand hydration pattern changes upon the association of anions and proteins [23]. Despite the promising results and examples in the literature, it is still often unclear how these techniques can be used in a practical and prospective sense within a medicinal chemistry program and, crucially, what benefit they can bring over traditional approaches.

It can be argued that there still exists a great deal of scepticism within the medicinal chemistry community concerning the value of analysing water distributions. During a lead optimisation program, the decision to target a particular water molecule can be, and typically is, taken without knowing exactly how strongly bound it is to the complex, while introducing lipophilicity in hydrophobic regions to displace weakly bound or ‘unhappy’ water molecules is intuitive to a trained medicinal chemist. Such scepticism could arise from the literature; most publications simply use water distributions to retrospectively explain and rationalise unexpected trends in structure–activity relations (SAR), while the number of publications in which they are used to actually drive and direct the medicinal chemistry program is significantly fewer.

In this article, I look at some of the most successful and current computational methods for locating water molecules within protein-binding sites, and highlight examples where they have been beneficial to drug discovery projects. From this, I assess whether these methods have now evolved to a state where they can be relied upon to help direct medicinal chemistry efforts, or if further work and validation is still required.