The two data sets collected using the X11 beam line were chosen so as to obtain contrast in the between the two X-ray energies. Due to the large spectral bandpass of this X-ray line it was expected that the maximum obtainable contrast would be significantly less than that obtainable using the X31 beam line. The difficulties involved in accurately calibrating the X-ray energy on this line introduce an additional uncertainty into the magnitude of the contrast attainable. As in the previous section a number of procedures were used in the phasing. In two cases data taken on X31 were included into the phasing so as to assess the effect of the additional optimised anomalous information. The following phase sets were calculated:
The results produced by phasing methods 7 to 11 are summarized in Table
Table: Results of phasing if X11 data in used alone (7) and along with X31 data (8-10). This table complements Table for X31.
and the resulting electron density maps for phase sets 7, 8, 9 and 11 are shown in Figs. , and .
Figure: Alpha helical region extending from residue 87 to 101. Electron density maps for MAD phasing solutions 7, 8, 9 and 11 are shown.
Figure: Beta strand region extending from residue 40 to 53. Electron density maps for MAD phasing solutions 7, 8, 9 and 11 are shown.
Figure: Trp 28 residue. Electron density maps for MAD phasing solutions 7, 8, 9 and 11 are shown.
The solution produced by the X11 data only (method 7) has the lowest correlation coefficient of all data set combinations explored. Solution 11 produces the lowest mean phase deviation and the highest correlation coefficient out of all phase sets calculated. After the application of SQUASH the correlation coefficient improves significantly for this solution. The mean phase deviations as a function of for solutions 7 to 11 are shown in Fig. .
Figure: Breakdown as a function of of the average phase deviation between different MLPHARE phasing solutions utilising data collected on the X11 beam line and the equivalent refined structure solution.
The calculation of phases using all seven data sets (solution 9) allowed comparisons to be made concerning the relative contributions of each of the data sets to the phasing. The Cullis R-factors defined in Sec. were calculated for each data and plotted as a function of for acentric and centric reflections in Fig. .
Figure: Cullis R-factor plotted against resolution () for dispersive differences in acentric reflections (top-left) and centric reflections (top-right) and for Bijvoet differences in acentric reflections (bottom). The reference data set (4) is excluded from the dispersive Cullis R-factor plots.
The usefulness of the Cullis R-factor is that it gives a direct indication of the relative sizes of the observed anomalous differences (dispersive and Bijvoet) and their respective lack of closure errors. A striking result is the similarity, to within , between the values of the anomalous Cullis R factor (bottom) for data sets 3 and 6 which have values of and respectively. The lack of closure error for data set 6 is evidently times lower than that for data set 3. The absolute scale factors of data sets 3 and 6 are 2.37 and 0.64 respectively, i.e. data set 6 has on average a factor of 3.7 more counts contributing to each reflection. This suggest that the statistical error in set 6 is a factor of times lower.
Data set 5 has consistently low Cullis R factors compared with the other X31 data sets (1 to 4) as does data set 6 measured on X11. Data set 3 () displays a poor isomorphous Cullis R factor for acentric reflections when compared with data set 2 (). The experimental errors in data set 2 are however approximately of those for data set 3 as demonstrated by Fig . This would suggest a larger average lack of closure error for data set 3 which is what we observe.
There is a significant difference observed between the quality of solution 3 (data sets 3, 4 and 5) and solution 11 where data set 3 is replaced by set 6. The values of for data sets 3 and 6 are comparable thus the explanation for the difference in the quality of solutions must be the large differences in and . This is due partly to the lower statistical errors in data set 6 and partly due to the large difference in for the two data sets.