Quantitative High-Throughput Profiling of Environmental Chemicals and Drugs that Modulate Farnesoid X Receptor

The farnesoid X receptor (FXR) regulates the homeostasis of bile acids, lipids, and glucose. Because endogenous chemicals bind and activate FXR, it is important to examine which xenobiotic compounds would disrupt normal receptor function. We used a cell-based human FXR β-lactamase (Bla) reporter gene assay to profile the Tox21 10K compound collection of environmental chemicals and drugs. Structure-activity relationships of FXR-active compounds revealed by this screening were then compared against the androgen receptor, estrogen receptor α, peroxisome proliferator-activated receptors δ and γ, and the vitamin D receptor. We identified several FXR-active structural classes including anthracyclines, benzimidazoles, dihydropyridines, pyrethroids, retinoic acids, and vinca alkaloids. Microtubule inhibitors potently decreased FXR reporter gene activity. Pyrethroids specifically antagonized FXR transactivation. Anthracyclines affected reporter activity in all tested assays, suggesting non-specific activity. These results provide important information to prioritize chemicals for further investigation, and suggest possible modes of action of compounds in FXR signaling.

Despite the growing interest in FXR ligands in drug discovery, little is known with regard to the roles of FXR in mediating or protecting xenobiotic-induced toxicity. Abnormal FXR function leads to numerous disorders such as cholestasis, diabetes, and cancer, and plays a role in liver regeneration 18 . Depending on the dose of FXR ligands, distinct outcomes have been observed across different species. Exposure of GW4064 has been reported to cause hepatobiliary injury to medaka eleutheroembryos 19 . CA and GW4064 have been found to protect mice from acetaminophen-induced hepatotoxicity 20,21 yet a CDCA-rich diet has been reported to induce liver hypertrophy in mice 22 . Treatment with the FXR antagonist tempol or intestine-specific deletion of FXR led to similar anti-obesity effects in mice 23 . Theonellasterol, a recently discovered FXR antagonist from marine sponge, protected mice susceptible to cholestasis from bile acid-induced liver damage 24 . In addition, gastric bypass surgery has emerged as a potential therapy for diabetes mellitus type 2 25 where FXR-dependent increase of circulating total bile acids was observed in mice treated with vertical sleeve gastrectomy 26 . These studies suggest that agonists, antagonists, and modulators of FXR could exert protective or adverse effects depending on health states and exposure doses.
One challenge in defining the role of FXR in mediating xenobioticinduced toxicity is the depth of data on the structural classes of chemicals that act on FXR. Here, we report the profiling of 10,766 substances (8599 unique compounds) in modifying FXR signaling and associated cytotoxicity as part of the Tox21 Phase II program 27 . We utilized the quantitative high-throughput screening (qHTS) data combined with computational methods to identify biological activity patterns of the Tox21 10K compound collection in order to prioritize chemicals for more extensive follow-up studies. Compounds identified as FXR-actives were grouped into several clusters based on similarities in chemical structure, drug class, or known biological target. The representative FXR-active clusters were further compared for their selectivity against other tested human nuclear receptors including the androgen receptor (AR), estrogen receptor alpha (ERa), peroxisome proliferator-activated receptor delta (PPARd), peroxisome proliferator-activated receptor gamma (PPARc), and the vitamin D receptor (VDR) to identify FXR-specific chemical scaffolds.
Results qHTS performance of FXR-bla and viability assays. To identify environmental chemicals and drugs that modulate FXR signaling, we screened the Tox21 10K compound library against the FXR-bla assay in both agonist and antagonist modes. To rule out FXR antagonist response caused by compound cytotoxicity, a cell viability assay was conducted in the same well as the FXR-bla assay. The ratiometric readouts of FXR-bla assay for measuring FXR activity are based on the b-lactamase-coupled fluorescence resonance energy transfer (FRET) technology 28 . CDCA and (Z)guggulsterone, positive controls for agonist and antagonist screening, respectively, yielded an EC 50 value [i.e. concentration calculated to induce a half maximal response with standard deviation (SD)] of 29 6 6 mM and an IC 50 value (i.e. concentration calculated to inhibit a half maximal response with SD) of 50 6 12 mM. The agonist and antagonist screening worked well as evaluated by average signal-to-background (S/B) ratios of 4.4 for both assays, and average coefficients of variation (CV) of 7.0% and 3.5%, respectively. The average Z9 factors of the agonist and antagonist screening were 0.35 and 0.75, respectively. Cytotoxicity screening in the FXR agonist and antagonist screening also showed consistent responses with average S/B ratios of 67.1 and 67.7, average CV values of 13.0% and 12.0%, and average Z9 factors of 0.60 and 0.69, respectively. Data reproducibility of a given compound was assigned as active agonist/antagonist match, inactive match, inconclusive, or mismatch based on average curve rank and percentage of inactive outcomes of the three independent measurements 28 . The triplicate runs of the Tox21 10K compound collection as well as the 88 compounds duplicated in each plate showed low mismatch rates of ,1% in the FXR-bla screening ( Figure 1). The antagonist and agonist screening identified 8% (861 substances) and 2% (215 substances) active matches, respectively, containing FXR-active compounds and positives resulting from assay artifacts.
Identification of FXR agonists and antagonists. After the primary screening, the test compounds were categorized as active agonists/ antagonists, inconclusive, or inactive compounds based on the activities observed in both ratiometric and 460 nm readings 28 . There were 1141 and 2172 compounds that showed activities in the FXR-bla agonist and antagonist mode assays, respectively. Four known FXR agonists, CDCA (EC 50 (Table S1, Figure 2a, Figure 2b). Some positive compounds were further verified and re-assigned as inconclusive compounds by additional criteria to exclude potential false positives as a result of compound autofluorescence (efficacy FXR-bla 460 nm /efficacy FXR-bla ratio or efficacy FXR-bla, 535 nm /efficacy FXR-bla, ratio . 2, PubChem assay identifier: 720681 and 720682) or cytotoxicity (IC 50, viability /IC 50, FXR-bla ratio , 3, p , 0.05). For example, benzo(k)fluoranthene and triamterene are highly fluorescent at 460 nm in the assay medium, producing a concentration-dependent increase in all b-lactamasebased assays (data not shown). Digoxin and bortezomib were inconclusive antagonists of FXR because the two compounds showed FXR antagonist activity at or near cytotoxic concentrations (Table S1). Two hundreds and sixty-six unique compounds including potent FXR-active compounds (EC 50 or IC 50 values , 10 mM) identified from the primary screening and selected structural analogs were re-tested in the same FXR-bla and viability assays, yielding confirmation rates of 67% (73 of 109) and 90% (144 of 160) in the agonist and antagonist screening, respectively. Twentyfive novel and representative compounds with agonist or antagonist activities confirmed in the FXR-bla assay were shown in (Table 1) detailing compound efficacy, potency, curve class, and data reproducibility in the primary and confirmatory screening. The 25 compounds were further tested in a FXR coactivator recruitment  assay to determine whether a given FXR-active compounds is an FXR ligand or a potential FXR signaling modulators ( Table 1). The agonist control CDCA showed an EC 50 value of 29.90 mM in binding of FXR-LBD and inducing coactivator recruitment, and the known FXR ligand ivermectin fully inhibited CDCA-induced coactivator recruitment with an IC 50 value of 0.91 mM. Cyclopamine (EC 50 5 10.57 mM, efficacy 5 94%) and 9-aminoacridine (EC 50 5 11.17 mM, efficacy 5 152%) showed full agonist activity, and both compounds were unable to induce coactivator recruitment to FXR-LBD (Table 1). Several partial FXR agonists including daunorubicin (EC 50 5 1.02 mM, efficacy 5 48%), doxorubicin (EC 50 5 1.35 mM, efficacy 5 68%) and epirubicin (EC 50 5 5.78 mM, efficacy 5 44%) also showed antagonist effects in the FXR-bla assay with IC 50 values of 5.53 mM, 2.80 mM and 17.80 mM, respectively (Table 1). These FXR-active anthracyclines were able to inhibit CDCA-induced coactivator recruitment at potencies similar to their antagonist activity in the FXR-bla assay ( Table 1). Among the confirmed compounds that completely inhibited CDCA-induced FXR-bla activity, actinomycin D (IC 50 5 0.02 mM) was the most potent, followed by flavopiridol (IC 50 (Table 1). Colchicine was also identified as an FXR antagonist in the coactivator recruitment assay with an IC 50 value of 0.03 mM and 52% efficacy (Table 1).
Structural clusters of FXR agonists and antagonists. The Tox21 10K compound collection was first grouped into 1,014 clusters of structural classes using the self-organizing map (SOM) algorithm 29 . As shown in Figure 3, each cluster containing unique and replicated compounds was colored according to the significance of enrichment (negative logarithmic scale of a p-value) in FXR-active agonists or antagonists to assess the probability of a given chemical scaffold to activate FXR-bla or inhibit CDCA-induced FXR-bla transactivation. The FXR-active compounds yielded 189 structural classes in which 56 and 152 clusters were significantly enriched with compounds that activate or inhibit FXR, respectively. The representative FXR-active clusters are described in Table S2.
Mechanism of action and targets in FXR signaling. Compounds which share similar biological functions despite distinct chemical scaffolds may be used to identify potential interactions between their biological target and FXR. Using the known primary targets and mechanism of action of approved and investigational drugs, 35 FXR-active drugs more potent than the natural FXR ligands [e.g., CDCA for agonists; (Z)-guggulsterone for antagonists] were classified into four drug classes and six target classes depending on their primary biological function (Table S3).
Selectivity of FXR-active compounds against a family of nuclear receptors. To detect potential assay artifacts and study compound selectivity, activity patterns of the top twenty-seven FXR-active clusters against AR, ERa, PPARd, PPARc, and VDR, which used the same b-lactamase reporter gene technology (Table S4 and  Table S5), were evaluated. Retinoic acids (k20.7) and alpha-cyano (type II) pyrethroids (k28.24) were found to be the top clusters enriched in FXR selective agonists and FXR selective antagonists, respectively ( Figure S1 and Figure 5). Unlike k28.24, the noncyano (type I) pyrethroid (k28.23) were two-fold more responsive to PPARc than FXR (Table S5). The albendazole-like benzimidazoles (k7.9) and the nocodazole-like benzimidazoles (k7.10) shared similar selectivity patterns in the AR-bla antagonist and the FXR-bla antagonist screening ( Figure 5). The nicardipine-like dihydropyridine cluster (k10.12) showed antagonistic or cytotoxic response in all tested b-lactamase assays for nuclear receptors ( Figure S2). Benidipine, manidipine, and nicardipine exhibited weak FXR agonist activity. The nifedipine-like dihydropyridine cluster (k11.12) yielded several distinct selectivity profiles. Felodipine, lacidipine, and lemildipine activated AR, ER, FXR, and PPARd. Cilnidipine and nitrendipine were found to be FXR agonists, and nitrendipine was also an AR antagonist. Nimodipine, nifedipine, and nisoldipine were identified as antagonists of FXR and AR. The   cholic acids (CDCA, DCA, and LCA; Figure 5) and the FXR-active benzoic retinoic acids (AM80 and AM580; Figure S1) selectively induced transactivation of FXR. Guggulsterones exhibited antagonist activity to AR and FXR ( Figure 5). Some FXR-active pyrethroid insecticides such as cyfluthrin and bifenthrin were highly selective FXR antagonists ( Figure 5). The anthracycline chemotherapeutics, including doxorubicin, daunorubicin, idarubicin, and pirarubicin, displayed mixed agonist/antagonist response against AR, ERa, and FXR, and antagonist response against PPARc, PPARd, and VDR ( Figure S3). The six structural classes of tubulin binders showed distinct selectivity patterns ( Figure 5). Griseofulvin, colchicine, podofilox, and the majority of FXR-active benzimidazoles (i.e., nocodazole, carbendazole, and mebendazole) shared similar activity patterns in inhibiting b-lactamase expression driven by FXR, AR, PPARc, and VDR. Griseofulvin and colchicine also antagonized estradiol-induced ER transactivation. Three FXR-active vinca alkaloid-based tubulin binders acted as mixed agonists/antagonists of ER and antagonists of AR and FXR ( Figure 5).

Discussion
In the present study, we used a qHTS platform in combination with chemoinformatics to profile 8,599 unique environmental chemicals and drugs for their potential to modulate FXR signaling. False positive and false negative rates were minimized by using a format consisting of three replicate concentration-response curves with 15 concentrations for each test compound in primary screening. Additionally, the FXR results were compared with Tox21 qHTS screening results against several additional targets to identify nonspecific effects such as compound autofluorescence, cytotoxicity, reporter gene-dependent response, or other artifacts. Overall, the high data reproducibility of replicate compounds indicated that the screening was robust. The mismatch rates of the 10K triplicate runs and the 88 replicates were both less than 1%, indicating good assay reproducibility. From the primary screening, more FXR antagonists than FXR agonists were identified. Interestingly, many of the FXRactive drugs identified (e.g., k1.10 for anthracycline chemotherapeutics and k7.9/k7.10 for dihydropyridine anti-hypertensive drugs) showed a range of agonist activity between 13% and 70% of CDCA activity as well as antagonist activity to CDCA-induced FXR activation, a pharmacological behavior consistent with classification as a partial agonist 51 . Only a few FXR agonists showed efficacies similar to CDCA, including the topical antiseptic 9-aminoacridine 36 and the teratogenic plant-derived steroid cyclopamine 52 . About 1270 unique compounds from the Tox21 10K compound collection had inhibitory effects on CDCA-induced FXR activity. More than 542 unique of these active compounds appeared to be cytotoxic as detected by viability assays in the same well, thus requiring cluster analysis for chemical prioritization and orthogonal assays for further confirmation.
The FXR-bla assay is able to detect both weak and potent FXR agonists. UDCA, a 7-beta isomer of CDCA previously reported as a weak FXR agonist 5 or a non-FXR ligand 53 , exhibited partial agonist activity at high micromolar concentrations in the FXR-bla assay (Figure 2a and Table S1). The potency of the other known FXR agonist GW4064 in the FXR-bla assay (Figure 2b and Table S1) is comparable (,5-fold difference in EC 50 values) to the literature value measured in a coactivator recruitment assay 13 . The positive control CDCA was the most efficacious FXR agonist compared to the other cholic acids, DCA and LCA, which showed weak agonistic  (Table S1). The induction of FXR activity by these cholic acids is highly specific. It is likely due to the fact that cholic acids have a shape distinct from other endogenous steroids and the ligand binding domain of FXR adopts a unique orientation for binding of cholic acids 54 . Both (E) and (Z) isoforms of guggulsterone were confirmed as FXR antagonists in both the primary screening and confirmatory assays. (Z)guggulsterone was used as the antagonist control in the FXR screening and its (E) isoform also showed antagonist response in the AR-bla assay. Although guggulsterones were first discovered as naturally occurring FXR antagonists, they also bind other steroid receptors including the glucocorticoid receptor, the mineralocorticoid receptor, and the progesterone receptor 55 , making it complicated to interpret the effects of guggulsterones in FXR signaling and lipid metabolism 8 .
The recently discovered FXR agonist ivermectin 10 and the other five avermectin macrolide antiparasitic agents (k3.3) identified in this study potently induced FXR-bla transactivation at submicromolar concentrations (Table S1). Ivermectin not only showed partial agonist activity but also exhibited antagonistic activity in the FXR-bla assay. This observation is consistent with an independent study in which ivermectin was identified in both biochemical and cell-based assays as a potent FXR antagonist 56 . Ivermectin, an avermectin antiparasitic drug used in humans, reduces hyperglycemia and hyperlipidemia symptoms in the diabetic mice model at submicromolar concentrations via FXR-mediated signaling 10 . Human patients with overdoses of ivermectin, abamectin, and emamectin have shown acute cardiotoxicity, acute neurotoxicity, or adverse effects on the gastrointestinal tract 57 although linkage to FXR-mediated effect is not known. Doramectin, an ivermectin analog for anti-parasitic treatment in dogs, was two-fold more potent and more efficacious than ivermectin in activating FXR in our screening. In addition, the two milbemycin macrolides-milbemectin and moxidectin which lack the disaccharide moiety present in avermectins exhibited weak and no activity in the FXR-bla assay (Table S1). These results indicate the importance of the disaccharide in forming hydrogen bonds with FXR as observed in the co-crystal structure of ivermectin-bound FXR-LBD 10 .
We identified 63% (12 of 19) of the tested dihydropyridine-class calcium channel blockers as FXR agonists or FXR antagonists with low to high micromolar potencies (Table S1 and Figure S2). Two recently reported FXR antagonists, nimodipine and felodipine 56 , were identified as an antagonist and an agonist in the FXR-bla assay, respectively. The dihydropyridine drugs in the Tox21 10K compound collection were grouped into two clusters represented by nicardipine (k10.12) and nifedipine (k11.12). The two clusters exhibited distinct FXR activity and selectivity towards the other five nuclear receptors ( Figure S2). The major structural difference of the two clusters is the extra phenyl group in the nicardipine-like compounds which may introduce additional steric hindrance and hydrophobic interactions in binding of FXR and other nuclear receptors. When compared with other functionally related receptors such as the pregnane X receptor (PXR), the majority of the FXR-active dihydropyridines are more potent in activating PXR 58 . Accidents of toddlers over-dosed with nifedipine, a dihydropyridine drug used to treat hypertension and to control angina, resulted in hypotension, lethargy, and/or vomiting 59 . Adults overdosed with nifedipine showed symptoms such as hypotension, sinus tachycardia, and/or sinoatrial and atrioventricular nodal depression 60 . Although therapeutic doses of dihydropyridine drugs normally yield submicromolar serum concentrations 61 and no direct link between FXR and overdose of dihydropyridines has been reported, potential impacts of dihydropyridines on FXR-related physiological function need further evaluation as cases of overdose of these drugs increase.
Sixty-seven percent (12 of 17) of the tested pyrethroid insecticides were identified as potential FXR-specific antagonists (Table S1 and Figure 5). These pyrethroids inhibited CDCA-induced FXR transactivation at high micromolar concentrations. Pyrethroid insecticides kill insects by over-stimulating the voltage-gated sodium channels 62 . Humans are exposed by pyrethroids through ingestion or occupational exposure, and can result in paresthesia 63 , gastrointestinal irritation 63 , or abnormal glucose regulation 64 . However, human exposure of pyrethroids would likely have to reach very high levels to trigger FXR effects, doses at which other modes of activity/toxicity would likely already have been manifested. One interesting finding for pyrethroids is a relationship between structure and FXR activity. The alphacyano pyrethroids (k28.24) were more enriched in FXR antagonists and more FXR-specific than the noncyano pyrethroids (k28.23) ( Table S5). The main structural difference between the two clusters is the substitution of vinylic hydrogen to a cyano group in the k28.24 pyrethroids. While the majority of k28.24 pyrethroids exhibited FXR-specific antagonistic effect, cyhalothrin, fenvalerate and flucythrinate, also showed antagonist activity in other nuclear receptor assays including the AR-bla and the VDR-bla assays ( Figure 5). These three compounds have an aryl group instead of the acid moiety present in most pyrethroids. These data suggest that FXR could be a novel mammalian target of pyrethroids, and that the structural differences of the FXR-active pyrethroids play a crucial role in compound activity and selectivity.
Our data provides interesting evidence that chemicals known to inhibit tubulin are active as FXR antagonists. All tested tubulin inhibitors including benzimidazoles (e.g., nocodazole), colchicine, docetaxel, griseofulvin, podofilox, and vinca alkaloids (vinblastine, vincristine, and vinorelbine) exhibited FXR antagonist activity at submicromolar to low micromolar concentrations in the FXR-bla assay screening ( Figure 5). Despite the fact that most of these FXRactive tubulin inhibitors are mitotic inhibitors, concentrationdependent inhibitory effects of these compounds in the FXR-bla assay were only observed in the antagonist screening. The FXR-active benzimidazoles, vinca alkaloids, colchicine, and podofilox are 5 to 100-fold more potent in inhibiting CDCA-induced FXR activity than activating PXR 58 . Benzimidazoles have been identified as FXR agonists with lipid lowering effects in vivo 65 . However, the FXR-active benzimidazoles identified from the screen were unable to induce FXR-bla transactivation and to alter FXR coactivator recruitment (Table 1). Compared to the reported benzimidazole-based FXR agonists, the nocodazole-like benzimidazoles lack an alkylated N3 functionality important for FXR binding. Therefore these benzimidazoles are likely to modulate FXR signaling via alternative mechanisms. Tubulin inhibitors and microtubules have been implicated in regulating a number of nuclear receptors including AR 66 , the glucocorticoid receptor (GR) 67 , PXR 68 , and the retinoic acid receptor (RAR) 69 as well as the aryl hydrocarbon receptor (AhR) 70 . Those studies suggest that microtubule inhibitors affect nuclear receptor trafficking by not allowing coregulators to move from the cytoplasm to the nucleus. Additionally, ERa was reported to activate histone deacetylase 6 (HDAC6) to deacetylate tubulins in human breast cancer MCF-7 cells 71 . Future work should interrogate the role of microtubules in FXR signaling and evaluate effects of tubulin inhibitors on FXR target genes to determine if these compounds are acting directly or indirectly on FXR activity.
In summary, the present study identified several key scaffolds of FXR-active drugs and environmental chemicals. Because FXR regulates diverse metabolic pathways, it has emerged as an important drug target and a potential toxicity mediator. Complete structureactivity relationships of anthracyclines, avermectins, dihydropyridines, and pyrethroids require further testing of a larger number of tailor-designed analogs to identify the pharmacophore important for FXR binding. Systematic evaluation of compound selectivity of FXRactive environmental chemicals against other functionally related receptors like PXR, liver X receptor (LXR), and constitutive androstane receptor (CAR) 72 , and G protein-coupled bile acid receptor www.nature.com/scientificreports SCIENTIFIC REPORTS | 4 : 6437 | DOI: 10.1038/srep06437 (GPBAR1) 73 is of great importance to understand how these compounds affect lipid and xenobiotic metabolism.
qHTS of beta-lactamase reporter gene and cell viability assays. The online screening procedures to profile the Tox21 compound collection against androgen receptors (AR), estrogen receptor alpha (ERa), farnesoid X receptor (FXR), peroxisome proliferator-activated receptor delta and gamma (PPARd/c), and vitamin D receptor (VDR) were adapted from the previously reported procedures 28 and the results were deposited to the PubChem BioAssay database (Table S4). The detailed protocols of the FXR-bla and viability screening on the Tox21 compound collection are described as follows (Table S6 and Table S7). Five mL of suspended FXR-bla cells at a cell density of 1000 cells/mL was plated in 1536-well tissue culture-treated black clear bottom plates (Greiner Bio One North America, Monroe, NC, USA) using a 8tip multidrop reagent dispenser (Thermo Fisher Scientific, Waltham, MA, USA). The standard procedure to screen FXR agonists starts with cell incubation at 37uC/5% CO 2 for 5 hours, followed by addition of 23 nL of compound solution on a compound transfer workstation (Kalypsys, San Diego, CA, USA). Chenodeoxycholic acid (CDCA) and DMSO (Sigma-Aldrich Corp., St. Louis, MO, USA) were used as positive and negative agonist-mode controls, respectively. To screen compounds that antagonize CDCA-induced transactivation of FXR, an extra 1 mL of CDCA was added on the top of the cell/compound mixtures to achieve a final agonist concentration of 50 mM. After 16 hours of incubation at 37uC/5% CO 2 , 1 mL of CCF 4 -AM substrate reagents was added to each well using a Bioraptr Flying Reagent Dispenser (FRD) workstation (Beckman Coulter, Indianapolis, IN, USA), followed by a 2 hour incubation at room temperature in the dark. Samples were excited at 405 nm and the resulting fluorescence emission intensity values at 460 nm and 530 nm were recorded on an EnVision plate reader (Perkin Elmer, Shelton, CT, USA). The cytotoxicity effects were measured in the same plates by adding 3 mL of CellTiter Glo reagent (Promega, Madison, WI, USA) to each well and incubating the plates at room temperature in the dark for 30 min. The luminescence values were acquired on a ViewLux plate reader (Perkin Elmer, Shelton, CT, USA). The confirmatory screening used two 1536-well compound plates for 266 selected compounds, where the original stock solutions (10 mM or 20 mM) used for primary screening were serially diluted to yield eight concentrations at 154 dilution ratios and each dilution series was plated to the same compound plate. The same assay protocols of FXR-bla and viability assays in the primary screening were used to test the 266 compounds in three independent experiments, in both agonist and antagonist modes, and on an off-line screening platform.
qHTS data analysis. The qHTS data was analyzed according to the previous protocol 28 . Briefly, raw plate reads for each titration point were first normalized relative to the positive control compound (agonist mode: CDCA, 100%; antagonist mode: guggulsterone, 2100%; cell viability: tetra-n-octylammonium bromide (TOAB), 2100%) and DMSO-only wells (0%) as follows: % Activity 5 [(V compound 2 V DMSO )/(V pos 2 V DMSO )] 3 100, where V compound denotes the compound well values, V pos denotes the median value of the positive control wells, and V DMSO denotes the median values of the DMSO-only wells, The data set was then corrected using the DMSO-only compound plates at the beginning and end of the compound plate stack by applying an in-house pattern correction algorithm. The half maximum activity values (AC 50 ) and maximum response (efficacy) values were obtained by fitting the concentration-response curves of each compound to a four-parameter Hill equation 76  for inactive) were converted to curve ranks (1 to 9 integers for increasing activating abilities; 29 to 21 integers for decreasing inhibitory abilities; 0 for inactive) according to the criteria previously described 28 . The activity outcome of a test compound in each readout was first categorized based on the average curve rank from the triplicate runs and the reproducibility calls. The final activity outcome of each compound was determined based on its multi-channel readout activity, for example, compounds with inactive curve ranks in both the 460 nm channel and/or the FRET ratios were concluded to be inactive. A compound was assigned as autofluorescent and inconclusive when the efficacy at 460 nm is two-fold greater than the efficacy of the ratiometric readout. Compounds that antagonize FXR and kill cells at similar potencies (i.e., when AC 50, viability /AC 50, ratio , 3, p , 0.05) were considered as cytotoxic and inconclusive. Data reproducibility was categorized as active match, inactive match, inconclusive, and mismatch according to the previously described criterion based on activity outcome differences between replicates and percentage of inactive outcomes 28,77 . The 10K library was clustered based on structural similarity (LeadscopeH fingerprints; Leadscope, Inc., Columbus, OH, USA) using the selforganizing map (SOM) algorithm 29 . Each cluster was evaluated for its enrichment of active agonists/antagonists and significance of enrichment was determined by pvalues from the Fisher's exact test.
FXR coactivator recruitment assay. The glutathione S-transferase (GST)-tagged FXR-LBD protein, terbium (Tb)-labeled goat anti-GST antibody, and fluoresceinlabeled SRC2-2 coactivator peptide were purchased from the Life Technologies (Carlsbad, CA, USA). The ability of an FXR agonist to recruit a coactivator is reported by FRET between the donor Tb-labeled antibody (l ex 5 340 nm and l em 5 495 nm) and the acceptor fluorescein-labeled peptide (l em 5 520 nm), where the agonist induces conformation changes of FXR-LBD and leads to coactivator recruitment. Six mL of 5 nM GST-tagged FXR-LBD, 0.1% bovine serum albumin (Sigma-Aldrich, St. Louis, MO, USA), 5 nM Tb-labeled anti-GST antibody, and 500 nM fluoresceinlabeled SRC2-2 were plated in a 1536-well black solid bottom plate (Greiner Bio One North America, Monroe, NC, USA) using a Mantis single-channel liquid dispenser (Formulatrix, Inc., Waltham, MA, USA). Twenty-three nL of compound solution in eight concentrations in DMSO was added to the corresponding well using a compound transfer workstation (Wako Automation, San Diego, CA, USA). The reaction was incubated in the dark at room temperature for 30 minutes. Samples were excited at 340 nm, and the resulting emission intensity values at 495 nm and 520 nm with 90 ms delay time and 300 ms integration time were acquired on the EnVision plate reader (Perkin Elmer, Shelton, CT, USA). The antagonist assay was conducted in the presence of 50 mM CDCA to induce coactivator recruitment. The TR-FRET ratios were normalized using DMSO as 0% FXR activity and 50 mM CDCA as 100% FXR activity for both agonist and antagonist mode screening.