Loading...
Please wait, while we are loading the content...
Similar Documents
Water structure in cubic insulin crystals ( x-ray diffraction / protein hydration / crystallographic refinement / hydration force )
| Content Provider | Semantic Scholar |
|---|---|
| Author | Badger, John Caspar, Donald L. D. |
| Abstract | The electron density distribution of the solvent in the cubic insulin crystal structure, which occupies 65% of the volume, has been mapped from 1.7-A resolution diffraction data by an iterative difference Fourier method, using the previously determined protein structure as the refinement restraint. Starting with phases from the protein and a flat solvent model, the difference map calculated from the data was added outside the protein envelope, and the modified map was then used to recalculate phases for the iterative refinement. Tests of the method with model data, with the experimental data and a variant protein model, and by carrying out a partial refinement of the solvent map demonstrate that the refinement algorithm produces reliable values for the solvent density within the noise level of the data. Fluctuations in density are observed throughout the solvent space, demonstrating that nonrandom arrangements of the water molecules extend several layers from the well-ordered hydration shell in contact with the protein surface. Such ordering may account for the hydration force opposing close approach of hydrophilic surfaces and other long-range water-dependent interactions in living structures. x-ray diffraction data extending from the 002 reflection to 1.7-A resolution has been measured (10), and atomic coordinates for the insulin monomer and 108 bound water molecules in the asymmetric unit were refined to an R factor of 0.20 (8). The 24 insulin molecules within the unit cell are arranged into rows parallel to the cube edges, leaving continuous interconnected solvent channels of diameter -30 A running through the crystal. The 65% volume fraction of the crystal unit cell occupied by solvent corresponds to =440 water molecules per insulin molecule (Mr 5778). Thus, there is slightly more than one water molecule for each nonhydrogen protein atom. In addition, at the pH and ionic strength of the mother liquor, these crystals should contain about three Na' counterions and about one molecule of salt per insulin monomer. The low salt concentration and large solvent volume within these crystals provide a suitable system for examining the structure of water in the vicinity of a protein surface under near-physiological conditions. Our analysis of the crystal data demonstrates that the average distribution of water molecules beyond those in immediate contact with the protein surface is recognizably nonuniform. Water constitutes at least half of the volume in a wide variety of protein crystals (1), and much of this water is not uniquely ordered. Fourier transformation of the correctly phased Bragg diffraction amplitudes, measured by x-ray or neutron crystallography, maps the average distribution of scattering matter in the crystal asymmetric unit, regardless of the ordering of this matter. Phases for the Bragg reflections from well-ordered protein crystals can be calculated from atomic models (including ordered water) which have been refined (2, 3) to fit observed intensities within a small residual. (R factors for data between 6and 2-A resolution are typically =0.150.2; cf. Fig. 1 and ref. 4.) However, omission of disordered solvent in the models leads to large discrepancies between calculated and measured data at low resolution. Blake et al. (5) improved the agreement with the low-resolution x-ray diffraction data from lysozyme crystals by filling the voids in their atomic model with uniform density solvent, but refinement of the average solvent structure was not attempted. Schoenborn and Cheng (6, 7) modeled the disordered solvent in myoglobin crystals as closely spaced pseudoatoms whose weights and temperature factors were adjusted to fit neutron diffraction data, but the three-dimensional solvent structure was not described. We have applied refinement methods to map the average solvent density distribution in cubic insulin crystals, using the previously determined atomic model (8) as a phasing restraint. Cubic crystals ofpig insulin (space group I213 with a = 78.9 A) grow in zinc-free alkaline solutions (pH 9) at low salt concentrations (0.2 M Na2HPO4 or NaCl) (9). Although the protein occupies only about a third of their volume, these crystals are well ordered. A nearly complete set of Bragg Crystallographic Methods When the distribution of a substantial portion of the scattering matter within a crystal has been determined, approximately correct phase angles can be calculated from such a partial model; the missing structure appears with reduced weight in the electron density map computed from these phases and the measured amplitudes. This is the basis of the heavy atom method for solving crystal structures. Using the protein molecule as the initially known part of the structure (i.e., the "heavy atom"), we have developed an iterative procedure to determine the average density distribution of the solvent. An electron density model of the protein structure was constructed (11) from the refined atomic coordinates (8). Hydrogen atoms were placed according to stereochemical criteria and assigned temperature factors 10 A2 higher than the atoms to which they were connected. Hydrogens in -CH3 groups were assumed to be rotationally mobile and assigned temperature factors of60 A2. The protein region was demarked by the electron density contour around this model which gave a molecular volume equal to that calculated from the sum of the canonical residue volumes (12, 13). Electron density outside this volume was set to the value for liquid water and smoothed at the protein/solvent boundary (5). The R factor for Bragg reflections with d > 7.3 A was reduced from 0.69 to 0.30 by this flat solvent density model and the R factor for the higher-resolution data remained almost unchanged. Density fluctuations needed to improve agreement with the observed data were introduced into this flat solvent model map by an iterative density modification procedure involving (i) calculation of a conventional difference Fourier map by using the experimentally determined structure factor amplitudes and amplitudes and phases calculated from the protein 622 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. §1734 solely to indicate this fact. Proc. Natl. Acad. Sci. USA 88 (1991) 623 and solvent electron density model; (ii) addition of this difference map to the current model density map outside the protein volume; and (iii) calculation of a new set of model amplitudes and phases from this modified map for further refinement cycles. In real space, the refinement of the solvent map was restrained by the requirement that the density distribution within the envelope of the protein model should remain unchanged, and in reciprocal space, convergence was driven by the requirement that the observed and calculated structure factors should agree as closely as possible. Similar methods, restraining well-defined parts of a crystal density map to fit a model, have been used in the refinement of small molecule structures (14, 15). A related procedure has been proposed for improving protein electron density maps which are partially interpretable in terms of an atomic model (16). Refined Electron Density Map During the first four refinement cycles, only the lowresolution phases (d > 4.5 A) were treated as variables. The iterative refinement was then progressively extended to 1.7-A resolution, using all the data. After each refinement cycle, scale factors (a single multiplier and temperature factor) between observed and calculated structure factors were determined from the complete data set. The difference amplitudes for unrecorded data were taken to be zero, but the calculated values from the model maps for these terms naturally could change at each iteration. The refinement was terminated after 12 cycles, when the largest feature in the difference Fourier map appeared within the protein volume and the final R factor (0.06) was close to the expected noise level of the data (Fig. 1). An electron density map was then calculated from the experimentally determined structure factors and the refined phases. Structure factors obtained from the final model electron density map were used to substitute for the small number of unrecorded data (1% of terms to 4.5 A, 7% to 2.0 A, and 13% to 1.7 A). A map containing only the solvent density was obtained by subtracting the model protein density from this refined electron density map. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.pnas.org/content/88/2/622.full.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
