Zebrafish behavioural profiling identifies GABA and serotonin receptor ligands related to sedation and paradoxical excitation

Anesthetics are generally associated with sedation, but some anesthetics can also increase brain and motor activity—a phenomenon known as paradoxical excitation. Previous studies have identified GABAA receptors as the primary targets of most anesthetic drugs, but how these compounds produce paradoxical excitation is poorly understood. To identify and understand such compounds, we applied a behavior-based drug profiling approach. Here, we show that a subset of central nervous system depressants cause paradoxical excitation in zebrafish. Using this behavior as a readout, we screened thousands of compounds and identified dozens of hits that caused paradoxical excitation. Many hit compounds modulated human GABAA receptors, while others appeared to modulate different neuronal targets, including the human serotonin-6 receptor. Ligands at these receptors generally decreased neuronal activity, but paradoxically increased activity in the caudal hindbrain. Together, these studies identify ligands, targets, and neurons affecting sedation and paradoxical excitation in vivo in zebrafish.


SUPPLEMENTARY NOTE 1
GABAAR ligands produce paradoxical excitation in zebrafish. Compounds with weak phenoscores (x < 0.51) included one GABAB receptor agonist, one PAM of δ-subunit containing GABAARs, two non-BZ-site ligands, three structurally-related GABAAR orthosteric agonists, and seven BZ-site GABAAR PAMs (Fig. 1g). For these compounds, the average phenoscores were significantly less than the positive controls (P < 0.01, Kolmogorov-Smirnov test, Supplementary  Figure 20a), suggesting that these compounds did not phenocopy etomidate. For example, the highest scoring ocinaplon treatment produced a behavioral profile that resembled the negative controls (Supplementary Figure 2). These data suggest that a variety of GABAergic compounds do not cause sedation and paradoxical excitation.
Compounds with intermediate phenoscores (0.51 < x < 0.71) included several types of GABAAR PAMs including thiopental, carboetomidate, THDOC, alfaxalone, diazepam, and valerenic acid. The highest scoring profiles produced by some of these compounds (including alfaxalone, thiopental, and tracazolate) showed a barely detectable statistically significant difference compared to the positive controls (0.01 < P < 0.05). The highest scoring profiles of animals treated with diazepam, and valerenic acid were significantly lower than the positive controls (P < 0.01, Kolmogorov-Smirnov test, Supplementary Figure 20a), however these treatments produced interesting intermediate effects on sedation and paradoxical excitation. For example, although the highest-scoring diazepam treatment was strongly sedating in most assays, it produced eASRs that were relatively weak and inconsistent (Supplementary Figure 2). These data suggest that a variety of PAMs have intermediate effects on sedation and paradoxical excitation.
Interestingly, although DOC and progesterone are neurosteroid precursors, they were among the most potent compounds tested (Fig. 1g). As expected, progesterone's etomidate-like phenotype was suppressed by dutasteride, a 5-alpha-reductase inhibitor that blocks the metabolic conversion of progesterone to allopregnanolone, suggesting that these compounds were converted to active neurosteroids (Supplementary Figure 5).

SUPPLEMENTARY NOTE 2
Target prediction using SEA. We used the Similarity Ensemble Approach (SEA) to predict targets based on 'guilt-by-association' enrichment factor scores (EFs). These EFs were first developed for predicting adverse drug interactions 1 , and balance the overall strength of a given target-tocompound-set association by correcting for the frequency that specific targets are predicted over random compounds sets in the screen 2 . Here, we used EFs to predict targets for the compounds that caused eASRs in the zebrafish. Table 9, Supplementary Figure 12a). We chose eight of these compounds to reorder and retest and found that four of them reproducibly caused eASRs in vivo (Supplementary Figure 12b, Supplementary  Table 9). Next, we tested five of these compounds as agonists and antagonists for activity at seven human mGluRs (mGluR1-6 and mGluR8). However, none of the compounds showed strong functional effects against mGluRs in vitro (Supplementary Figure 12c), suggesting that the compounds did not act via mGluRs in vivo. To further test the mGluR hypothesis, we tried to phenocopy etomidate in dose-response experiments with a panel of structurally-diverse mGluR ligands and ligand combinations (Supplementary Table 9). Although MPEP, a mGluR5 antagonist, reproducibly caused eASRs, MPEP-induced eASRs were substantially lower in magnitude than etomidate-induced eASRs, and MPEP-induced eASRs only occurred in a narrow concentration range (Supplementary Figure 12b, Supplementary Table 9). Therefore, although MPEP weakly phenocopied etomidate, we found no further evidence that hit compounds targeted mGluRs, as predicted by SEA.

SEA identified 15 compounds with enriched target predictions for mGluRs (Supplementary
SEA predicted that GABAAR was a target of four hit compounds (Supplementary Figure 11). We tested three of them (5658603, 5142031 and 7145248) and found that one (5658603) potentiated GABAAR in vitro (Fig. 2f, red arrow). Curiously, we noted that SEA failed to predict GABAAR as a target for most hit compounds that tested positive in the GABAAR FLIPR assay (Fig. 2f), underscoring the value of behavior-based screens for identifying bioactive compounds with poorly annotated chemical structures.

SUPPLEMENTARY NOTE 3 GABAAR and HTR6 ligands likely converge on a common neural substrate
To determine if HTR6 antagonists activated the same neurons as GABAergic ligands, we took three approaches. First, we looked for overlap between 5HT immunohistochemistry and the eASR substrate neurons (Supplementary Figure 13f). Consistent with previous reports, we observed strong 5-HT staining in the telencephalon, pineal gland, hindbrain, and dorsal raphe nuclei 3 . In addition, we observed bilateral 5-HT staining in tracts that converged on the midline of the caudal hindbrain at the same location of the putative eASR substrate neurons in the caudal hindbrain (Supplementary Figure 13f). These tracts likely originated from the dorsal raphe, but we could not trace their origin definitively. Second, we visualized HTR6 mRNA expression by RNAscope but could not detect reproducible expression patterns (Supplementary Figure 14), suggesting that HTR6 mRNA is not abundantly expressed. Finally, we tested for pharmacological interactions between GABAergic and serotonergic ligands. As expected, pretreatment with the GABAAR antagonist PTX rescued etomidate-treated animals, increasing and decreasing the magnitude of the violet light and eASR phenotypes, respectively (Supplementary Figure 15a). Similarly, PTX rescued the GABAergic compound 5658603, and partially rescued compounds 701338, and 5942595, albeit to a lesser extent than etomidate (Supplementary Figure 15a). PTX also partially rescued the behavioral phenotypes of several HTR6 antagonists including BGC 20-761, 6029941, 6028165, 6030006, and 6013263 (Supplementary Figure 15b). By contrast, EMDT oxalate, a HTR6 agonist, did not suppress eASRs caused by HTR6 antagonists (Supplementary Figure 15c), suggesting that the effects of HTR6 antagonists are not easily reversed. Together, these data suggest that GABAAR agonists and HTR6 antagonists likely cause eASR behaviors via different targets that converge on a common neural substrate in the zebrafish hindbrain. threshold of 0.25 to define clusters and visualized using the scipy hierarchy dendrogram function.
SEA and EF calculations. Here we describe our computational pipeline: 1) Use the reference trace to discover the top 125 hit compounds (most similar phenotypically related to etomidate). 2) Organize hit compounds into hierarchical supersets of increasing numbers of hit compounds. Use SEA analysis to generate target predictions for each of the compounds in the sets. Perform enrichment factor calculations on the sets, which attempt to correct the occurrence of target predictions for a set of compounds by comparing to a background distribution. 2 To do so we generated 10,000 sets of 200 random screening compounds each, and applied the following formula to calculate the enrichment of target y for set x: E_xy = n*N / (A * T), where n is the number of times target y is predicted for set x compounds by SEA, A is the number of times any target shows up for set x, T is the number of times target y shows up for any set, and N is a normalization factor equal to the product of all the targets and all the sets.
Determination of phenotypic thresholds and significance. For each ligand, we selected the dose that gave the highest average phenoscore, and for that dose, we performed a two-sample Kolmogorov-Smirnov (KS) test to calculate the KS statistic against the 12 positive control replicates of etomidate @ 6.25 µM using the scipy function ks_2samp from the scipy.stats package (Supplementary Figure 20a).
To calculate approximate thresholds of phenoscore significance, we performed a statistical simulation. For each score in the space of possible phenoscores (binned in 0.05 increments from 0 to 1), we sampled 12 replicates from a uniform distribution centered around the score ranging from -4σ to +4σ away from the mean, and calculated the KS statistic against the etomidate 6.25 µM replicates. We repeated this simulated procedure 100 times to get robust statistics, and took the average of these P values. However, we realized that the standard deviation of replicates across different GABAAR ligands was not a constant value. It tended to be low for extremely poor phenotypes, peaked for intermediate phenotypes, and decreased again for extremely strong phenotypes. Therefore, we fit the standard deviations for GABAAR ligands as a function of phenoscore with a 10th order polynomial using the Polynomial package in numpy (Supplementary Figure 20b). Using this resulting polynomial, we calculated the KS P values from the simulated uniform distributions as we iteratively stepped along the y-axis ; these P values were smoothly distributed except for a discontinuity around phenoscore 0.5 due to rapidly increasing P values in this range (Supplementary Figure 20c). We derived the threshold phenoscores associated with these P values by fitting another polynomial to the resulting distribution in the smooth region (above phenoscore 0.5) (Supplementary Figure 20d) and calculating the roots of the function at those P values. The resulting phenoscores corresponding to 0.01 and 0.05 P value thresholds were 0.51 and 0.71, respectively. Z' and false positives and negatives calculation. To calculate the Z' (quality of screen coefficient), we use the formula Z' = ! , expressed in terms of the mean and SD of the positive (p) and negative (n) controls. The false positive/negative rates were determined at a threshold of 3 SD. Any positive controls with a phenoscore 3 SD away from the positive control mean (! ) were labeled false positives. Likewise, any negatives controls 3 SD away from the negative control mean (! ) were labeled false negatives.     Animals were treated with hit compounds. For each compound, the cLogP (calculated partition coefficient) (x-axis) and minimum concentration required to cause the eASR phenotype, were plotted (y-axis). Unlike historical Myer-Overton analyses, the minimum effective concentration does not decrease with hydrophobicity. The best-fit line and shading represent the resulting regression line and a 95% confidence interval for that regression.       (a-d) Brain activity maps showing significant ΔpERK signals using the Z-brain online reference tool (n = 5-10 animals/condition). Heatmaps indicate positive (green), negative (purple), and nonsignificant (black) changes in pERK labeling (p < 0.0005, Mann-Whitney U test). All activity maps are comparisons between the indicated treatment conditions. (e) Overlay of average α-pERK signal for BGC 20-761(magenta), and etomidate treated animals (green). (f) Overlay of α-5ΗΤ staining (magenta) and the average α-pERK staining (green) for BGC 20-761 treatment. Abbreviations: tel, telencephalon; mb, midbrain; ot, optic tectum; hb, hindbrain; ha, habenula; ob, olfactory bulb; nm, neuromast; ap, area postrema; pg, pineal gland.       1  carbamazepine  anticonvulsant  19  1  phenytoin  anticonvulsant  20  2  fluoxetine  antidepressant  21  2  trazodone  antidepressant  22  3  diphenhydramine antihistamine  23  3  dimenhydrinate  antihistamine  23  3  promethazine  antihistamine  23  4  buspirone  anxiolytic  24  4  alprazolam  anxiolytic  25  4  diazepam  anxiolytic  26  4  oxazepam  anxiolytic  27  5  quetiapine  atypical antipsychotic  28  5  olanzapine  atypical antipsychotic  29  6  atenolol  beta blocker  30  6  propranolol  beta blocker  31  7  ACPA  cannabinoid  32  7  methanandamide cannabinoid  33  8  zolpidem  hypnotic  34  9  benzocaine  local anesthetic  35  9  lidocaine  local anesthetic  39  9  bupivacaine  local anesthetic  40  9  tricaine  local anesthetic  41  9  procaine  local anesthetic  42  10  ketamine  intravenous anesthetic 38  10  isoflurane  inhalational anesthetic 35  10 propofol intravenous anesthetic 36 10 etomidate intravenous anesthetic 37    OC1=Nc2c ( Cc1ccc(C(C)C)c(OCCNCc2ccccc2)c1 no 5802987