Previous Article |
Table of Contents
| Next Article 
BIOLOGICAL SCIENCES / BIOPHYSICS
Determination of solvent content in cavities in IL-1
using experimentally phased electron density
*Institute of Molecular Biology, Howard Hughes Medical Institute, and Department of Physics, University of Oregon, Eugene, OR 97403-1229; and Protein Expression Laboratory/National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health, Building 6B, Room 1B 130, Bethesda, MD 20892-2775
Contributed by Brian W. Matthews, October 30, 2006 (received for review October 17, 2006)
 |
Abstract |
|---|
 Top Abstract Results and DiscussionMethodsAcknowledgementsReferences |
|---|
(IL-1), four independent structures solved by x-ray crystallography indicate that water is not present in the central apolar cavity. In contrast, results from NMR spectroscopy suggest that water has high occupancy within the cavity but is positionally disordered, making it undetectable by standard crystallographic methods. A theoretically based crystallographic-phase refinement technique also suggested that there was the equivalent of two fully occupied water molecules within the apolar cavity. To resolve these discrepancies we sought to obtain an experimentally phased electron density map that was free of possible bias caused by mathematical modeling of the protein or the solvent. By combining native diffraction data with multiple wavelength anomalous data from a platinum derivative, accurate phases were obtained. Using these experimental phases, we estimate that occupancy of the apolar cavity in IL-1 by solvent is close or equal to zero. Polar cavities in the protein that contain ordered solvent molecules serve as internal controls.
hydration | protein folding | water
Although the solvent content of apolar cavities has been examined in numerous proteins, both native (7–10) and engineered (12–16), the protein that has been subjected to the greatest scrutiny is IL-1. This protein contains an apolar cavity, 
40 Å3 in volume, that is centrally located within the -trefoil fold. Among the four structures of IL-1 that have been determined independently by x-ray crystallography (17–20), none contains a water molecule bound within this cavity. In contrast, characterization of bound water in IL-1 by NMR (21) provided evidence for at least one water molecule that is highly occupied, but positionally disordered, within the cavity (22). This discrepancy between crystallographic and NMR results was proposed to result from the high mobility of the water molecule(s), which would give a weak, highly diffuse peak in an electron density map while still producing a strong NMR signal. This interpretation of the NMR data was called into question by Matthews et al. (23) who were able to account for the observed NOE cross-peaks by using other water molecules that occur within the IL-1 structure but are not in the central cavity.
To determine whether water located within the central cavity of IL-1 could be visualized with a more detailed crystallographic analysis, Yu et al. (24) recollected diffraction data for IL-1 in an effort to measure low-resolution reflections more accurately, then proceeded to refine the phases for these low-resolution terms through iterative solvent density modification. Using this procedure, they calculated the total number of electrons within the central cavity to be 18, which they ascribed to two high-occupancy water molecules. Because the validity of this solvent density modification method has not been established, however, it is unclear whether their result is an accurate representation of the electron density within the apolar cavity in IL-1 or an artifact of the density modification procedure.
Recent developments in crystallographic-phase determination have improved the quality of experimental electron density maps dramatically (25). In the present study, we have obtained accurate experimental phases for IL-1 without relying on density modification or refinement techniques and have used the resulting electron density to quantify the solvent content of both the polar and apolar cavities that occur within the protein.
|
Results and Discussion |
|---|
|
TopAbstractResults and DiscussionMethodsAcknowledgementsReferences |
|---|
crystals diffracted to a resolution limit of 1.5 Å (Table 1), derivative crystals diffracted to somewhat lower resolution because of increased absorption and/or damage from heavy atom soaks and the shorter exposure times necessary to minimize decay during multiple wavelength experiments. These effects also give rise to the reduced I/
(I) ratios observed for derivative data sets. During data collection, care was taken to ensure that all data sets were as complete as possible and that low-resolution data were accurately measured. The redundancy of the data ranged from nine for the native data set to seven for each wavelength in derivative datasets. In all data sets, the 
I2
/I2 ratios were within 1 SD of 2.0, the value expected for untwinned data (see Methods). In addition, the twin fractions (
) calculated from mean and mean-square values for H were only 2% for the native data set and <5% for all derivative data sets (data not shown).
|
, which were based on only one or two mercury sites and two or three platinum sites (17, 20), six mercury and seven platinum sites were found in the present study (Table 2). Whether the reason for this discrepancy lies in improvements in data collection techniques or heavy-atom refinement is unclear. The validity of each of the sites was confirmed by proximity to residues expected to react with the heavy metal compounds with the exception of the two mercury sites (sites 4 and 5) that were found within the apolar cavity (cavity 5). Because the presence of these atoms could give erroneous estimates of electron density and inclusion of mercury data did not appear to improve phases significantly, the mercury data were not used in the final phase determination. Despite this omission, the phases derived from the combination of native with platinum multiwavelength anomalous dispersion (MAD) data (Table 3) were of high quality. The mean figure of merit was 0.84 to 2.1-Å resolution without relying on solvent flattening or other forms of density modification.
|
|
ranging in volume from 16 to 45 Å3 (Fig. 1 and Table 4). In addition to the central apolar cavity, there are four hydrated cavities that are considerably more polar, with three or four polar contacts per water molecule. Cavities 1 and 2 contain two water molecules each and have larger volumes (45 and 40 Å3, respectively) than cavities 3 and 4, both of which contain a single water molecule and have a volume of 16 Å3.
|
|
, has a volume of 39 Å3, nearly as large as the cavities containing two ordered water molecules. This cavity differs from the others in that there are no polar contacts with a putative water molecule placed inside the cavity. Instead, the cavity wall is composed entirely of hydrocarbon side-chain atoms contributed by Leu-10, Leu-18, Leu-26, Phe-42, Leu-69, Ile-122, and Val-132.
Distributions of experimentally phased electron density within each cavity in IL-1 are shown in Fig. 1, with numerical data in Table 4. The minimum electron density values within each cavity range from –0.4 to –0.2 e/Å3 and, as expected, tend to be located near the walls of the cavity or between adjacent water molecules. In the cavities that were previously observed by x-ray crystallography (17–20) and NMR (21, 26) to contain ordered water, maximum values are 1.0–1.4 e/Å3 near the expected sites.
Even though the x-ray data set is essentially complete and we believe the experimentally determined phase angles to be reliable, the resolution limit of 2.1 Å still needs to be considered. At this resolution, diffraction ripples from atoms surrounding the different cavities are expected to contribute false electron density to the cavity regions. To estimate the magnitude of this effect we calculated a model electron density map at 2.1-Å resolution with both amplitudes and phases from the reference IL-1 structure described in Methods. The mean figure of merit for the experimentally phased electron density map was 0.84. Therefore, all density values for the model map were multiplied by 0.84 to put the two maps on equivalent scales. Three different variants of the model map were calculated (Table 4), the first assuming no water molecules in the central cavity, the second assuming a single water molecule, and the third assuming two water molecules 2.8 Å apart in the cavity.
Because the presence of solvent in cavities 1–4 (Table 4) is not in dispute, these cavities can be used as internal controls for the calculations. There are 10 electrons in a single water molecule. In the model-phased electron density calculation cavities 1 and 2 contain two water molecules. Allowing for the 0.84 scaling factor, the corresponding integrated electron density is expected to be 2 x 10 x 0.84 = 16.8 e. In the model-phased map (Table 4) the values obtained are somewhat higher (22.2 and 21.0 e). This discrepancy must be caused by the series termination effects noted above. Because the diffraction data are limited to 2.1 Å, some electron density will "bleed" from the atoms surrounding a cavity into the cavity itself. If the cavity contains solvent, some density will also bleed from the solvent into the cavity wall. Because the number and location of atoms within and surrounding a cavity are different, these effects are not equal. The difference will be greatest when a cavity is empty. In this case, electron density will bleed into the cavity from atoms within the cavity walls, but because no atom is present in the cavity, there is no compensation from density bleeding out. We believe that this is the situation with the model calculation with zero solvent for cavity 5. Here, even though no solvent is present, the model calculation suggests that the integrated density in the cavity is 8.2 e. This density must be caused by series termination effects.
Because of these complications we suggest that the best way to evaluate the results is to compare the experimentally observed electron density with that calculated from the respective models (Table 4). Series termination effects should influence the experimental and model calculations equally.
In the central apolar cavity of IL-1 (cavity 5) the integrated experimental density (7.6 e) corresponds most closely to the value for the model density calculated for an empty cavity (8.2 e). (As noted above the fact that neither of these values is close to zero is presumably caused by series termination effects, which should affect the model and experimental density essentially equally.) The integrated model density assuming either one or two water molecules to be present in the cavity is significantly higher (13.4 and 13.9 e, respectively) and disagrees substantially with the experimental value (the fact that the two-solvent value barely exceeds that for one solvent is presumably also caused by series termination). The validity of this approach is confirmed by the generally good agreement between the experimental integrated density and the model-phased electron density for cavities 1–4 (Table 4). In summary, the experimental electron density map confirms that there is appropriate electron density in the four known solvent-containing cavities (cavities 1–4). The map also corresponds to a 2.1-Å resolution image of IL-1 with zero solvent in cavity 5. We conclude that the occupancy of solvent in cavity is equal or close to zero.
Our results do not support the assertion of Yu et al. (24) based on solvent density modification that the apolar cavity in IL-1 contains two high-occupancy water molecules. One of the most obvious potential reasons for this discrepancy is that Yu et al. (24) defined the cavity region to correspond to a sphere of radius 6 Å. Such a sphere has a volume of 905 Å3, which is 23-fold larger than the cavity volume estimated by standard procedures (Table 4). As stated by Yu et al. (24), "To integrate the total number of solvent electrons inside the protein, the cavity region was defined by that portion of the solvent map within a central sphere of 6-Å radius." If this integration were performed over the actual cavity volume of 39 Å3 rather than the sphere volume of 905 Å3, the estimated number of electrons in the cavity would presumably be much less than the quoted value of 18. Also, if there is any error in the "zero" of the map (i.e., in the F000 term), this error will be 23-fold larger when multiplied by 905 Å3 than when multiplied by 39 Å3. Other possible sources of error in the procedure of Yu et al. (24) include the use of a series of ad hoc assumptions. These include: (i) use of a temperature factor of B = 15 Å2 to smooth out high spatial frequency ripples, (ii) the assumption that the solvent density is completely flat and has a B factor of 100 Å2, and (iii) a requirement that the phase angles beyond 4.5-Å resolution must correspond to those from the starting atomic model and cannot change during subsequent cycles of solvent density modification.
Recent thermodynamic simulations (4) suggest that the smallest number of water molecules that can stably occupy an apolar cavity is three (satisfying their hydrogen-bonding potential by forming a hydrogen-bonded trimer). In agreement with this, Collins et al. (27) recently showed that application of pressure to a cavity-containing mutant of T4 lysozyme caused a cluster of approximately four water molecules to occupy the apolar space. Given that the energetic cost of placing a single water molecule into an apolar cavity was estimated to be 20–30 kJ/mol, a value on the order of the free energy of unfolding of IL-1 (22 kJ/mol; ref. 28), it is highly unlikely that only one water molecule is present within the IL-1 cavity. Furthermore, mutations that introduce hydrogen-bonding amino acids into the apolar cavity do not result in the detection of ordered water molecules (28), as might be expected if solvent were present.
|
Methods |
|---|
|
TopAbstractResults and DiscussionMethodsAcknowledgementsReferences |
|---|
(29) was dissolved in 100 mM Tris (pH 7.8) and purified by gel filtration. Fractions containing IL-1 were pooled and concentrated to 8 mg/ml, then flash-frozen in liquid nitrogen and stored at –80°C. Immediately before crystallization, samples were thawed and centrifuged at 14,000 x g for 15 min. Crystals were grown by hanging-drop vapor diffusion against a well solution of 2.2–2.8 M ammonium sulfate, 100 mM Tris, pH 7.0–8.0, using drops consisting of 5 µl of protein solution and 5 µl of well solution (17–20). Square bipyramidal crystals with dimensions of 0.5–1.0 mm were obtained after several weeks at room temperature. Crystals suitable for data collection were harvested into 3.2 M ammonium sulfate, then transferred incrementally into cryoprotectant solutions with increasingly higher concentrations of glycerol. When a final concentration of 20% glycerol (vol/vol) in 3.2 M ammonium sulfate was attained, the crystals were frozen in a nitrogen cold stream at 100 K and stored under liquid nitrogen until needed for data collection.
Heavy Atom Derivatization.
A platinum derivative of IL-1 was prepared as described with slight modification (17). Briefly, 4 mg of potassium tetrachloroplatinate (II) (K2PtCl4) was dissolved in 1 ml of 3.2 M ammonium sulfate to give a 10 mM stock solution. A total of 0.5 µl of this stock solution was added to hanging drops of successful crystallization experiments and allowed to react in the dark for 5 days at room temperature. Derivatized crystals were then back-soaked for 30 min against 3.2 M ammonium sulfate, then transferred into cryoprotectant solution and frozen as described above. Mercury derivatives were prepared in a similar fashion, except that 3 mg of mercuric acetate [Hg(C2H3O2)2] was dissolved in 1 ml of 3.2 M ammonium sulfate to give a 10 mM stock solution, and the incubation period was reduced to 16 h.
Data Collection and Processing. Diffraction data were collected on frozen crystals at the Stanford Synchrotron Radiation Laboratory (Menlo Park, CA) beamline 9-2 using a Quantum 4 CCD detector (Area Detection Systems Corporation, Poway, CA). To obtain reliable intensities to the diffraction limit of 1.5 Å, native data were measured in two sweeps. For the low-resolution sweep, 60 images were collected at a wavelength of 1.000 Å with an exposure time of 8 s and an oscillation range of 2°. For the high-resolution sweep, the same images were recollected with a longer exposure time of 24 s. Reflections with saturated pixels in the high-resolution sweep were ignored during scaling.
MAD data sets were collected for both platinum and mercury derivatives (Table 1). At each wavelength, 180 images were collected in 30° wedges by using an inverse-beam protocol. Each image had an exposure time of 5 s and an oscillation range of 1°. Diffraction data were integrated and reduced by using the HKL package (30).
Exclusion of Twinned Crystals.
The presence of merohedral twinning in crystals of IL-1 has been reported (17). Although MAD phasing of twinned crystals of IL-1 has been successful (31), to avoid artifacts that may result from the superposition of electron density from different twin domains, crystals that exhibited twinning were not used for phase determination in the present work. Two different methods were used to detect the presence of twinning. In the first method, the ratio of mean squared intensities to squared mean intensities (I2/I2) was calculated in thin shells as a function of resolution. For acentric reflections, this ratio is 2.0 for untwinned data and approaches 1.5 for twinned crystals (32). If the ratio calculated for any thin shell deviated from 2.0 by more than a single standard deviation the data set was not included. (The standard deviation was defined as the variance of the thin-shell ratios. There is no theoretical basis for choosing one standard deviation but it is presumably a conservative criterion.) The second method involves calculating a parameter H for each pair of acentric, twin-related reflections. This parameter is the ratio of the absolute difference in intensities to the sum of the intensities [i.e., H = |I1 – I2|/(I1 + I2)]. The twin fraction may be calculated from the mean value of H or H2 by using the following relationships: H = 1/2 – and H2 = (1 – 2)2/3 (32). Data sets for which either estimate of exceeded 5% were also excluded from phasing.
Experimental Phasing.
Heavy atom refinement and phase determination were performed by using SHARP (33). Initially, two heavy atom sites were found by using anomalous difference Patterson maps derived from mercury MAD data. After preliminary refinement of heavy atom parameters, a third mercury site was added based on peaks in residual maps. Data from the platinum MAD experiment were then incorporated with four heavy atom sites modeled initially. Analysis of residual maps yielded three additional mercury sites and three additional platinum sites, giving a total of six mercury sites and seven platinum sites. At this point, a native data set was included to improve the isomorphous signal present in the MAD data. Superposition of heavy atom positions on the reference structure for IL-1 revealed that two of the mercury sites were located within the central apolar cavity (cavity 5). This realization lead to the exclusion of mercury data from phase determination because it was anticipated that the proximity of the heavy atom sites might give rise to artifacts in the electron density map. To avoid possible sources of systematic error, at no point during phase determination was solvent flattening or any other form of density modification applied.
Electron Density Calculations.
For the purposes of comparison, a reference IL-1 structure was obtained by molecular replacement with EPMR (34) using Protein Data Bank entry 5I1B (19) as a search model, followed by refinement to 1.5-Å resolution with TNT (35). The origin of the refined model was shifted to be consistent with the origin of the experimental phases. Cavity surfaces were calculated by using MSP (36) with a probe radius of 1.40 Å and default atomic radii (37) adjusted for hydrogen bonding distances (38) as follows: hydroxyl and carbonyl oxygen atoms, 1.54 Å; carboxyl oxygen atoms, 1.45 Å; and peptide nitrogen atoms, 1.57 Å.
To place the electron density maps on an absolute scale the observed amplitudes were scaled to the amplitudes calculated from the reference IL-1 structure. To determine the value of the F000 term, the number of electrons in the protein (3.74 x 104) was combined with the contribution from mother liquor (calculated electron density of 0.383 e/Å3 and solvent fraction of 64.7%, determined from the refined coordinates). This gave a total amplitude of 9.28 x 104 electrons for F000. Electron density values were calculated on a 0.1-Å grid by using FFT (39) and analyzed with SLICED (M. L. Quillin, unpublished FORTRAN program).
|
Acknowledgements |
|---|
|
TopAbstractResults and DiscussionMethodsAcknowledgementsReferences |
|---|
|
Footnotes |
|---|
Abbreviations: MAD, multiwavelength anomalous dispersion.
To whom correspondence should be addressed. E-mail: brian{at}uoregon.edu
Author contributions: M.L.Q. and B.W.M. designed research; M.L.Q., P.T.W., and B.W.M. performed research; M.L.Q. and P.T.W. contributed new reagents/analytic tools; M.L.Q. and B.W.M. analyzed data; and M.L.Q. and B.W.M. wrote the paper.
The authors declare no conflict of interest.
Data deposition: The atomic coordinates and structure factors have been deposited in the Protein Data Bank, www.pdb.org (PDB ID code 2NVH).
© 2006 by The National Academy of Sciences of the USA
|
References |
|---|
|
TopAbstractResults and DiscussionMethodsAcknowledgementsReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg What's this?
This article has been cited by other articles in HighWire Press-hosted journals:
|
|
 |
|
  J. Qvist, M. Davidovic, D. Hamelberg, and B. Halle From the Cover: A dry ligand-binding cavity in a solvated protein PNAS, April 29, 2008; 105(17): 6296 - 6301. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
