Accurate calculation of the density of proteins

On the basis of theoretical calculations, Andersson & Hovmoller have recently suggested that the long-established value of 1.35 g cm(-3) for the mean density of proteins should be revised to 1.22 g cm(-3) [Andersson & Hovmoller (2000), Acta Cryst. D56, 789-790]. To substantiate their assertion, these authors used the Voronoi algorithm to calculate the mean atomic volume for 30 representative protein structures. The Voronoi procedure requires that atoms of interest be bounded on all sides by other atoms. Volume calculations for surface atoms that are not surrounded or are only sparsely surrounded by other atoms either are not possible or may be unreliable. In an attempt to circumvent this problem, Andersson & Hovmoller rejected atoms with calculated volumes that were indeterminate or were greater than 50 Angstrom(3). In the present study, it is shown that this criterion is not sufficiently restrictive to ensure accurate volume determinations. When only strictly buried atoms are included in the volume calculations using the Voronoi algorithm, the mean density is found to be 1.47 +/- 0.05 g cm(-3). In addition, an alternate procedure based on the Connolly algorithm that permits all protein atoms to be included in volume calculations gives 1.43 +/- 0.03 g cm(-3) for the mean density of the same set of proteins. The latter two calculated values are mutually consistent and are in better agreement with the experimental value.