Hydrophobic patches on protein surfaces

Abstract

Hydrophobicity is a prime determinant of the structure and function of proteins. It is the driving force behind the folding of soluble proteins, and when exposed on the surface, it is frequently involved in recognition and binding of ligands and other proteins. The energetic cost of exposing hydrophobic surface is proportional to its area, and the question arises to what extent proteins can tolerate large hydrophobic patches on their surfaces. The current thesis is a study into such patches. Chapter 1 provides a general introduction into protein surface hydrophobicity. Chapter 2 describes a numerical algorithm for calculating the solvent accessible surface area. It samples the protein surface in Shrake & Rupley fashion: representing atoms as spherical distributions of points and summing the points that are not buried by any atoms. A number of optimization strategies is applied, yielding an exceptionally fast method. The quality of spherical point distributions is assessed, and a novel, optimal tessellation of the unit sphere is found. The accessible surface calculation method developed in Chapter 2 forms the basis of the hydrophobic patch detection algorithm called QUILT, presented in Chapter 3. The assumption is that hydrophobic surface area is synonymous with solvent accessible carbon and sulfur atoms. Connecting contiguous apolar atoms is not enough to delineate hydrophobic patches, because the relatively strong hydrophobicity of the protein surface (around 60%) results in one large hydrophobic surface. This surface spans the entire protein, and is dotted with polar 'islands' formed by the hydrophilic atoms, with hydrophobic connections through variously sized 'channels' between these islands. To delineate the hydrophobic patches, the channels are closed off by temporarily expanding the solvent-accessible polar atoms. This way, the hydrophobic surface neatly divides into proper patches which are subsequently identified, and adjacent surface area lost due to the polar expansion is added back to the patches thus obtained. Only the largest patches, having sizes exceeding expectation (based on randomizing the protein's surface), are deemed meaningful. The method is applied to a small number of structures to demonstrate the validity and utility of the method. In Chapter 4, the QUILT method is applied to a large sample of monomeric proteins, in order to survey general trends in the distribution of patch sizes on proteins. The largest patch on each individual protein averages around 400 Å2, but can range from 200 to 1200 Å2. Interestingly, these areas do not correlate with the sizes of the proteins, and only weakly with their apolar surface fraction. Trends regarding patch size distribution, amino acid composition and preference, sequential vicinity, secondary structure and mobility are discussed as well. Chapter 5 is devoted to a survey similar to that described in Chapter 4, but here, the interfaces of obligate oligomeric proteins are studied. As before, trends regarding amino acid composition and preference and patch size distribution are described. The largest or second largest patch on the accessible surface of the entire subunit was involved in multimeric interfaces in 90% of the cases, in agreement with interfaces being generally more hydrophobic than the rest of the protein surface. However, hydrophobic patches are not complementary: they are not preferentially in contact across associating subunits. This is perhaps surprising, but is to be expected, because the free energy of subunit association, as far as the hydrophobic patches are concerned, is largely due to the shielding of apolar area from the solvent, rather than from gaining hydrophobic contacts. To gain insight into the dynamic behaviour of hydrophobic patches, QUILT is applied to molecular dynamics simulations of three different protein structures. This is the subject of Chapter 6. The analysis requires an additional method to relate QUILT-patches across time frames of the trajectory, which is described as well. The resulting patch runs show that the area fluctuations are considerable, at around 25% of their size. The most frequently occurring mean patch size is approximately 50 Å2, but can reach around 400 Å2. An uninterrupted patch run can last up to 150 picoseconds, but, owing to protein mobility, is generally much shorter at around 4 ps. There is no clear relation between patch run durations and their average size, but long-lasting patch runs have smaller fluctuations. Although the formalism would allow this, the patches do not 'wander' over the protein surface, indicating that they are genuine surface features. When the patch runs are clustered, the truly persistent patches called recurrent patches are obtained. Only about 25% of them have a strong 'liveness', that is, are represented by an actual patch run most of the time. In amicyanin, the method detects the hydrophobic patch known to be involved in the binding of methylamine dehydrogenase. In phospholipase A2, a large persistent patch consisting of Leu58 and Phe94 is found, the likely functional relevance of which appears to be novel.