Fig 1.
Experimental timeline for chemical exposures at concentrations {0uM, 0.0064uM, 0.064uM, 0.64uM, 6.4uM, 64uM} added at 6hpf, with n = 32 embryos per concentration. At 24hpf, a nondestructive assay including two one-second light perturbations at 30s and 40s was performed (Details about this 24hpf behavioral assessment can be found at Reif et al. 2015). At 120hpf, behavior was measured under environmental conditions of 7 minutes continuous light exposure followed by an 8 minutes of dark. Behavioral and developmental (morphological) assessments were then recorded.
Fig 2.
Step 1: Visualize movement index; Step 2: Remove annotated dead fish for every concentration of a chemical; Step 3: Propose a statistical modeling method; Step 4: Check for artifacts, such as technical issues, global plate and position effect, and remove any bad plates; P1: Plate 1; P2: Plate 2; C1: Column 1; C12: Column 12; Step 5: Apply statistical modeling method to provide dose-response patterns for analysis; Step 6: Statistical analysis pipeline; Step 7: Assess reproducibility of our computational framework.
Fig 3.
A: Morphological overview of controls. Incidence rate assessment (Y axis) for all chemicals by endpoint (X axis). For mortality, rate was calculated with a sample size of 32. For other endpoints, rate was calculated conditionally on alive zebrafish larvae. The red line was drawn at 10% for visualization. B: Movement index overview: Plot of healthy (i.e. no annotated morphological endpoints) zebrafish larvae in control wells for all plates. Red line was drawn by connecting the mean of each experimental time point. Y axis: Movement Index; X axis: Time (minutes).
Fig 4.
Performance of Our Novel Statistical Modeling Method.
A: Example controls (having similar survival rates) illustrate the transformation. These two separate controls were plotted by different colors. Blue: TX000769 (Propoxycarbazone-sodium); Black: TX000900 (Methamidophos). Top: Movement index of each time point, and the line was drawn by connecting the mean movement indexes. Y axis: Movement index; X axis: Time. Bottom: Lines were drawn using our method, which connects the differential entropy of each time point. Y axis: Differential entropy (Nats); X axis: Time. B: All 1,060 control groups were plotted. Top: Each line represents a chemical. Line was drawn by connecting mean movement indexes. Y axis: Movement index; X axis: Time. Bottom: Each line represents a chemical. Line was drawn by connecting differential entropy across time. Y axis: Differential entropy (Nats); X axis: Time. C: Coefficient variation of each time point using all control groups and various statistical modeling methods. Y axis: Coefficient of Variation; X axis: Time.
Fig 5.
Statistics for checking artifacts.
P value and Cohen’s d from each permutation was plotted (Black represents plate; Blue represents position). Y axis: Student’s t test p value; X axis: Cohen’s d. Horizontal red line was drawn at a significance level of 0.05. Vertical red line was drawn at 0.2 to represent the general rule of thumb of effect size.
Fig 6.
Differential entropy of each concentration was plotted across experimental time. Color key is shown in Fig 1. Y axis: Differential entropy (Nats); X axis: Time (m). Red segments represent light condition from 3m to 9m. A: Inactive: TX000888 (Terbacil) was inactive at all concentrations. B: Hypoactivity (L) and Hypoactivity (D): TX001406 (Cyclanilide) shows significant hypoactivities at 64uM for both light and dark intervals. C: Inactive (L) and Hypoactivity (D): TX001412 (Fipronil) is inactive at light interval and shows significant hypoactivity at dark interval at 0.064uM, 0.64uM, 6.4uM, and 64uM. D: Hyperactivity (L) and Inactive (D): TX007214 (Dieldrin) shows significant hyperactivity at light interval but it is inactive at dark at 64uM. E: Hypoactivity (L) and Inactive (D): TX003357 (44’-Oxydianiline) shows significant hypoactivity at light interval and inactive pattern at dark interval at 0.064uM, 6.4uM, and 64uM. F: Hyperactivity (L) and Hypoactivity (D): TX006644 (Haloperidol) shows significant hyperactivity at light interval and significant hypoactivity at dark interval at both 6.4uM and 64uM. In addition, at 0.64uM, it shows significant hyperactivity at light interval. G: Hyperactivity (L) and Hyperactivity (D): TX005098 (4-Pentylaniline) shows significant hyperactivity at 64uM for both light and dark conditions. H: Inactive (L) and Hyperactivity (D): TX005080 (44’4”-Ethane-111-triyltriphenol) is inactive at light and shows significant hyperactivity at dark at 6.4uM and 64uM.
Table 1.
Summary of statistically significant behavioral profiles.
Fig 7.
Relationship between morphological profiles and behavioral profiles.
Disulfiram significantly affected 13 endpoints starting at 0.64 uM. Disulfiram also caused significant hypoactivity in both intervals with a lowest effect level of 0.64 uM. For morphological profiles, the panels represent (from top left) Aggregate Entropy, mortality, summation of any endpoint, then each of the specific endpoints (see ‘Methods‘). The X axes show concentration (0uM, 0.0064uM, 0.064uM, 0.64uM, 6.4uM, 64uM from left to right). The Y axes show Aggregate Entropy for the first panel, then incidence counts for all other panels. Red indicates statistical significance for each measure (p < 0.05).
Fig 8.
Summary of results by the whole experimental system.
This Venn diagram provides the summary of the total number of significant chemicals detected by each assay. It also provides statistics regarding the benefits of including all assays. Missing rate: the number of chemicals that would have been missed using a subset of these assays.