Table 1.
Gedankenexperiment cocktail screens created to test the distance metric across individual cocktail components.
Table 2.
Summary of crystal parameters, data collection and refinement.
Figure 1.
Heat map representing distance metric data generated from varying sodium chloride concentration from 0.01 to 4.41(A) the C6 metric and (B) the CDcoeff.
In (C) the metric is weighted to the changing variable, concentration. The darker blue colors at the extremes represent the greater metric distances produced when cocktails are compared with those further away in the series; in this case representing the difference between 4.41 M sodium chloride and a solution that contains 0.01 M sodium chloride. The diagonal compares identical cocktails and each heat map is symmetric with the top left corner comparing cocktail 1 to 12 (in this case) and the bottom right comparing 12 to 1.
Figure 2.
Heat map representing distance metric data generated from the pH-variant screen.
The C6 metric is shown in (A) with the CDcoeff in (B). Finally, (C) shows the CDcoeff weighted to only explore the changing term, pH. The screen contains fourteen identical cocktails, with the pH being increased incrementally by 0.5 units as the cocktail identification numbers increase. The white line diagonally bisecting the figure represents the region where each cocktail is being compared to itself. The darker blue colors at the extremes represent the greater metric distances produced when cocktails are compared with those farther away in the series; the darker sections in the corners correspond to the comparison between cocktails 1 and 14, which have pH values of 3.4 and 9.9, respectively.
Figure 3.
Heat map for (A) the C6 metric (B) the CDcoeff and (C) the CDcoeff focusing only on the cocktail fingerprint term for the PEG molecular weight screen.
All cocktails in this screen are identical, except for the PEG component, which samples ten of the molecular weight PEGs used in the standard 1,536 crystallization screen in our laboratory. The cocktails are ordered by increasing PEG molecular weight which makes the trend clear.
Figure 4.
All cocktails in the screens are identical, except the identities of their cations and anions are changed for each successive cocktail, according to the Hofmeister series [33].
(A) Results of the C6 metric on the cation screen (B) the CDcoeff on the cation screen (C) and the CDcoeff weighted to the cation screen (C). Similarly, D, E and F, show the same results with the anion screen.
Figure 5.
Each of these heatmaps represents a metric comparison between two consecutive generations of a screen from the Hauptman-Woodward Medical Research Institute [8].
The screens have 1,536 cocktails, and the heatmaps can be viewed as overlays of the 1,536-well plate in which these screens reside, with the colors of each block representing the metric difference between the successive cocktails in that particular location on the plate. Each square unit of color corresponds to the comparison between the cocktails in the successive generations in that location on the plate. In the top, the C6 metric is used while the CDcoeff is shown below. Both metrics were able to highlight two rows of cocktails that were altered considerably between generations 8 and 8A, in the form of a line of darker wells in the lower third of figures. The C6 metric, however, identified that cocktails outside of these two rows were slightly different, when they were actually identical. This discrepancy most likely arises from the C6 metric's use of penalties in its PEG and salt terms.
Figure 6.
Pairwise distance matrix for the 1,536 cocktails in the generation 8 crystallization screen.
Maximum similarity is denoted as blue with minimum as red. The cocktail identification numbers are given in the axis with the information mirrored across the diagonal. The light blue areas represent salt based conditions with the checkerboard red incorporating PEG as the precipitation agent.
Figure 7.
Heatmaps illustrating the results of the Hierarchical Clustering Algorithm.
Several clusters are labeled and identified through different colors. The large cluster, C20, represents conditions that contained PEG. The other clusters are those that did not but where the majority of crystals described later formed. The dashed line represents the default max cophenetic distance cutoff of one standard deviation.
Figure 8.
Regions of crystallization space where hits for BfR192 were found.
Out of the 28 clusters, 11 were identified containing at least 1 crystal hit. The full list is given in Table 4.
Table 3.
Cocktails that produced visually recognizable crystals in the clusters identified in Figure 7.
Table 4.
Clusters analyzed as a function of hits and percentage of sodium, potassium or phosphate present in the chemical cocktails.
Figure 9.
Cluster 13 isolated from Figure 7.
Cocktail numbers with an asterisk (*) are from those cocktails where human classification indicated a crystal hit. Within each grouping the cocktails are arranged from high to low ID number, a default within the software. While the overall ordering of images could be interpreted in terms of a crystallization phase diagram human intervention is required in the final analysis.
Table 5.
Chemical cocktails in the selected crystallization regions (Cluster 13) of the cluster diagram.
Figure 10.
Structure of the BfR192 exopolyphosphatase-related protein showing the two domains and highlighting the cleft containing the sodium, potassium and four phosphate ions.
Figure 11.
Stereo picture showing detail of the active site of the BfR192 exopolyphosphatase-related protein and identifying residues with which the phosphate ions interact.