A Chemical Screening Procedure for Glucocorticoid Signaling with a Zebrafish Larva Luciferase Reporter System

In a previous publication18, we screened a library of 640 FDA approved drugs in a pilot screen to evaluate the performance of the GRIZLY assay. This library contained 12 bona fide GCs, thereby allowing us to determine quality measures for sensitivity and specificity of the assay. Figure 2 shows typical examples of the raw data analysis and normalization taken from this screen. A typical result from a dexamethasone control well is depicted in Figure 2a, illustrating the temporal information given by the data. A peak in bioluminescence that occurs at about 12 hr after the treatment slowly decreases over the following 36 hr of measurement. Figure 2a' highlights the approximation of the AUC value using the trapezoidal method. This calculation is carried out for all wells and each technical repeat. The result of the log transformation is shown in Figures 2b and b'. Here the mean of the positive controls are plotted in a normal Q-Q plot. Log transformation leads to a greater approximation of the data points to the theoretical normally distributed values indicated by the red diagonals. A transformation to achieve normality of the data increases the statistical power of the subsequent analyses. Finally, Figures 2c and 2c' show the effect of robust Z-score normalization on plate-to-plate variability: The different baseline values obtained with different plates (Figure 2c) are normalized in the robust Z-score plot in Figure 2c'. The robust Z-score values can be used to identify systematic errors in the data by visualizing the distribution of the hits across the plates. Figure 3a shows one example of a heatmap of a library plate, in which the robust Z-score values are plotted color-coded into the well positions. One easily identifies column 12 with the in-plate positive controls. Positive scoring library compounds are seen in well F8 and, albeit weaker, E2. Figure 3b shows a plate in which a systematic error is present. One observes a series of compounds with decreasing Z-score values in rows A and E over several columns. None of the compounds in these wells was tested positive in the re-test. Since the wells with this decreasing GRIZLY test activity are placed along the path the pipetting robot takes when aliquoting the library plates, this pattern is indicative of carry-over of a positive compound during the automated pipetting process. The robust Z-score normalization together with the presence of known GCs in the library also enabled us to calculate a receiver operating characteristic (ROC) curve for the screen18. Figure 3c shows a plot of the estimated true positive rate against the estimated false positive rate for different robust Z-score cut-off values. The estimated true positive rate is defined as the percentage of true positive hits (here, the known glucocorticoids present in the library) that are found at a given robust Z-score cut-off value. The corresponding estimated false positive rate indicates the percentage of non-positive compounds hit at this cut-off value. The AUC of this curve is 0.95, indicating a high sensitivity and specificity of the screen and excellent assay performance. The AUC curve was also used to estimate the optimal cut-off value for hit identification by calculating the Youden index for different robust Z-scores. We identified the robust Z-score of 1.49 as having the highest Youden index. This cut-off value is indicated in Figure 2c' as a black line separating the hits from the bulk of the compounds. In our screen18, we were able to identify nearly all known GCs present in the library. Only three of the 12 bona fide GCs were not detected. In the case of melengestrol acetate, this was a false negative finding, since this compound tested positive in the re-test at a higher concentration than the one used in the library. The other two GCs were negative also in the re-test. Corticosterone is the major GC in rodents, but not in fish or humans1, while the prodrug prednisone might not be metabolized well by the larval system. Interestingly, also two drugs not annotated as GCs were identified in the screen, of which one could not be confirmed in the re-test (spironolactone, a mineralocorticoid receptor antagonist). The other compound, pregnenolone, is a precursor in steroid biosynthesis, which is converted to cortisol by the adrenal gland. Indeed, treatment with pregnenolone stimulated cortisol production in the larvae18. All these results confirm the high performance of the assay: only one false negative and one false positive compound were among the primary screen results, and 10 of the 11 confirmed compounds with GC signaling stimulating activity were already identified in the primary screen. Figure 1. General principle of the assay and work flow of the screening procedure. a) Scheme of the GRIZLY assay reporter construct. Luciferase (yellow) expression is controlled by a minimal promoter (Pmin, orange) and four concatemerized GRE elements ((GRE)4, white), which are bound by ligand-activated GR homodimers (GR, blue). Red boxes indicate Tol2 transposase sites that facilitate integration of the construct into the genome. The pink box represents a poly-adenylation site (pA). Transgenic larvae carrying this construct are placed in opaque 96 well microtiter plates for bioluminescence measurements. b) Scheme of the luciferase reaction. Luciferase catalyzes the oxidation of luciferin to oxyluciferin, which leads to light emission (hν). The reaction requires also ATP and Mg2+, which is provided by the larva. PPi, pyrophosphate. c) Workflow of the screen. Working stock dilutions are prepared from an FDA approved drug library in 96 well plates, one column containing a negative within-plate control (DMSO only, green) and one column containing a positive control (Dexamethasone, red) (library design). GRE:Luc larvae are distributed in 96 well plates (5-11 replicates) and treated with the library compounds (drug application). Bioluminescence traces are recorded with a bioluminescence reader for two days (data collection). Bioluminescence traces are integrated (AUC calculations), log transformed and robust Z-score normalized (normalization). Normalized data are used for determining quality metrics and for hit identification (data analysis). Chosen hits are retested for dose-dependent activity in the assay (retest). Click here to view larger figure. Figure 2. Data analysis illustrated with data from a screen of a FDA approved drug library. a) Representative trace from a positive control well. a') The area under the curve (AUC) is approximated with the trapezoid rule. The trapezoids used for the calculation are shown in different shades of red. b) Normal Q-Q plot of raw AUC values of the positive control wells (y-axis) vs. a standard normal population (x-axis) before log transformation. b') Normal Q-Q plot of the AUC values after log transformation. A normal distribution is indicated by the linearity of the log transformed data points. c) Plot of log transformed raw data for all compounds tested in the screen. Blue, bona fide glucocorticoids; grey, all other compounds. c') Plot of screen data after robust Z-score normalization. Systematic errors such as inter-plate variations have been removed from the data. Red lines indicate the mean ± s.e.m. of positive within-plate controls. Black line: hit identification cut-off value as determined by Youden-Index optimization (Figure 3). Reproduced with permission from14, 2012 ACS. Click here to view larger figure. Figure 3. Screen results. a) Heatmap of screen results. Robust Z-score values of library compounds are plotted into the respective well positions. The positive controls in column 12 as well as two positive hit compounds in wells F8 and E2 are visible. b) Heatmap of a plate showing carry-over of a positive compound during library preparation with a pipetting robot. A gradient of activity is visible along the pipetting path taken by the robot. c) Receiver operating characteristic (ROC) curve for the screen. The estimated true positive rates (left-hand y-axis) and false positive rates (x-axis) are plotted for increasing robust Z-score cut-off values (color coded, right-hand y-axis). The AUC value of the resulting curve is close to 1, indicating good assay performance. Reproduced with permission from 14, 2012 ACS. d) Table showing compounds active (yellow) or inactive (blue) in the primary screen (prim.) or in the retest. Bona fide glucocorticoids were not always retested (white). Redrawn with permission from14, 2012 ACS. Click here to view larger figure. General settings: Step 1.1 Procedure type: single liquid mode: blow out dispenses per aspirate: 1 optimize for speed: Yes air gaps system: 10 μl transport: 0 Flush/Wash No Aspirate Type of setting Value remarks Position B7, B10, B13, B16 Aspirate Vol. 10 μl Speed 200 μl/sec Liquid tracking 60% Aspirate Height Labware default 22.28 Dispense Type of setting Value remarks Positions D7, D10, D13, D16 Dispense Vol. 10 μl Speed 200 μl/sec Liquid tracking 100 μl Dispense Height Labware default 17.01 Step 1.2 Procedure type: reagent mode: waste dispenses per aspirate: 3 optimize for speed: Yes air gaps system: 15 μl transport: 0 Flush/Wash No Aspirate Type of setting Value remarks Position A4 Aspirate Vol. 3 x 490 μl Air gaps 15 and 0 Speed 200 μl/sec Liquid tracking 400% Aspirate Height Liquid surface Dispense Type of setting Value remarks Position D7, D10, D13, D16 Dispense Vol. 490 μl Air gaps 15 and 0 Speed 200 μl/sec Liquid tracking 100% Dispense Height Labware default 17.01 Table 1. Settings for Multiprobe II. The protocol in Table 1 describes the preparation of 4 plates (each 96 wells). The plates are located on plate adapters in the working area at positions specified by the robot (e.g. B7, B10, B13, B16). General settings: Number of assay repeats dependent on the length of the run Number of plates unlimited Temperature control 28 °C Protocol Calculations US Lum 96 (cps) read Measurement time 2.5 sec Crosstalk correction Distance between plate and detector 0 mm Glow (CT2) correction factor 0 % Table 2. EnVision settings used for the screen.