We first assessed the ability of a diverse set of compounds to activate human PXR by determining detailed concentration-response curves for activation of human PXR for 25 androstane steroids (Table 1), 11 estrane steroids (Table 1), 29 pregnane steroids (Table 2), 50 bile salts (Additional file 2; some bile salts were previously published 15), and 50 additional diverse compounds that included xenobiotics and vitamins (Table 3) (see Figure 1 for selected chemical structures of an androstane steroid, a pregnane steroid, two bile salts, and two additional compounds). These activation data further confirm the broad ligand specificity of human PXR, with most compounds only activating at micromolar concentrations.

In two prior studies that compared PXRs from different species, human and zebrafish PXRs were found to share some activating ligands, including pregnanes, androstanes, and a few xenobiotics such as nifedipine and phenobarbital 1215. Activation of zebrafish PXR by the much larger set of 165 compounds tested on human PXR was determined in this study, and these two species showed considerable overlap in their ligand specificity (Tables 1, 2, 3, Additional file 2). Human PXR has very broad specificity for steroid hormones and their synthetic intermediates (Figure 2A) albeit mostly at micromolar concentrations likely to exceed typical physiologic concentrations 28.

Zebrafish PXR was activated by far fewer steroid compounds which were essentially a subset of those that activate human PXR (Figure 2B). For both human and zebrafish PXRs, pregnane steroids showed the highest activity (Figure 2, Table 2). Human and zebrafish PXRs showed more differences in regard to bile salt activators with zebrafish PXR being activated by very few of the bile salts tested (Additional file 2). In terms of the evolution of bile salts, human PXR is activated by both evolutionary 'early' bile salts 262728 (e.g., 27-carbon bile alcohol sulfates such as 5α-cholestan-3α,7α,12α,26,27-pentol [cyprinol] 27-sulfate) and 'recent' bile salts (e.g., cholic acid) (Additional files 2 and 3). Zebrafish PXR is activated only by early bile salts, including 5α-cyprinol sulfate and 5β-scymnol (5β-cholestan-3α,7α,12α,24,26,27-hexol) 27-sulfate (Additional files 2 and 3). The results are consistent with crystallographic studies of human PXR that show a large, flexible ligand-binding pocket 293031323334. This pocket can accommodate bile salts of both 5α (A/B trans) and 5β (A/B cis) orientation (Figure 1), as well as those with differing side-chain lengths and conjugation. This is in contrast to studies of farnesoid X receptors (FXRs; NR1H4) and VDRs, two other NRs that are activated by bile acids 35363738. In particular, FXRs are antagonized by 5α-bile alcohol sulfates 39 while VDRs are essentially only activated by the smallest bile salt, lithocholic acid (5β-cholan-3α-ol-24-oic acid), and its metabolites 384041.

In a comparative study, we determined concentration-response curves for a common set of 16 compounds (steroids, bile salts, xenobiotics) in a set of PXRs from six species (human, zebrafish, mouse, rat, rabbit, and chicken; Tables 1, 2, 3, 4, Additional file 2). The pharmacophores generated are shown mapped to two of the generally more active ligands, 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and 5β-pregnane-3,20-dione (Figure 3). Human, rat, and mouse PXRs showed very similar pharmacophores with 4–5 hydrophobic features and multiple excluded volumes (Figure 3A,C,D). The pharmacophores for these three PXRs all suggest generally large ligand-binding pockets with differences only in positions of the features. It is interesting that compared with previous pharmacophores for human PXR 424344 which contained 4–5 hydrophobic features and at least 1–2 hydrogen bonding moieties, there are no hydrogen bonding features in the current human PXR pharmacophore. This could be due to the molecules used in the current training set being mostly bile salts and having active and inactive compounds with similar features. As the Catalyst pharmacophore generation method looks for differences between the extremes of activity to describe the features contributing to the pharmacophore, this may represent a limitation of the method. While a single universal pharmacophore for human PXR (and perhaps PXRs from other species) may be impossible due to the size and flexibility of the binding site, it is likely in the current study that the 16 selected molecules may just be a sub-section of the binding pocket. For example, this may be where steroidal compounds fit 33 as modelled previously with a pharmacophore 45. Therefore, the pharmacophores serve as a novel qualitative method for analysis of PXR ligand specificity across the species.

Zebrafish PXR showed the most constrained pharmacophore based on the 16 ligands, suggesting a small binding pocket compared with the other PXRs, consisting of 3 hydrophobes, 1 hydrogen bond acceptor, and excluded volumes (Figure 3B). Rabbit PXR had a similar pharmacophore model to zebrafish PXR but no excluded volumes as in the former (Figure 3E). Chicken PXR had a pharmacophore qualitatively different from the other PXRs, with the model indicating a symmetrical array of features that contribute to activity (Figure 3F); it is perhaps noteworthy that this PXR has a smaller 'insert' sequence between helices 1 and 3 of the LBD than that of human, mouse, rat, and rabbit PXRs 912. The pharmacophore models for both chicken and zebrafish PXRs also show a hydrogen bond acceptor not found in the models for PXRs from other species (Figure 3); this hydrogen bonding interaction may contribute to the relatively high activity of TCDD in chicken and zebrafish PXRs. Pharmacophore statistical summaries are presented in Additional file 4.

Whereas other vertebrates such as human, mouse, rat, chicken, and zebrafish have a single PXR gene in their respective genomes, two PXRs have been identified in the African clawed frog (Xenopus laevis) 746. This is likely a consequence of the tetraploidy of the X. laevis genome 47. The phylogeny confirms that these two genes are bone fide orthlogs to mammalian PXR; however their pharmacology and tissue expression pattern is markedly different 7464849. Xenopus laevis PXRs are alternatively termed 'benzoate X receptors' (BXRs) due to their activation by endogenous benzoates (such as 3-aminoethylbenzoate; Figure 1) that play a role in frog development 7. Similar benzoates have yet to be characterized in other animals, suggesting that these may be unique to amphibians. In addition to PXRs, other gene families show divergence in Xenopus laevis relative to other vertebrates. Per-ARNT-Sim (PAS) proteins such as the aryl hydrocarbon receptor (AHR) nuclear translocator are an example 50.

Our search of the sequenced genome of the related Western clawed frog (Xenopus tropicalis; an animal with a diploid genome) revealed only a single PXR gene. Cloning of the LBD of this PXR from adult female ovary and expression in a GAL4-LBD chimeric construct allowed for determination of ligand specificity. Similar to studies of the Xenopus laevis PXRs, the Xenopus tropicalis PXR was insensitive to steroids, vitamins, and xenobiotics that activate mammalian or chicken PXRs, but was activated by two benzoates described as activators of the Xenopus laevis PXRα (Additional File 5) 749.

Sequencing of the Ciona intestinalis genome revealed a single gene with similarity to vertebrate NR1I genes VDR, PXR, and CAR 1819. We previously reported a preliminary analysis of the Ciona VDR/PXR 51 and now present more detailed data. While the DBD of the Ciona VDR/PXR can be easily aligned to the corresponding sequence of vertebrate VDRs, PXRs, and CARs, alignment of the LBD is difficult in some regions (Additional file 1). As summarized in Table 5, the LBD of Ciona VDR/PXR has low sequence identity to vertebrate VDRs, PXRs, and CARs (17.1%–26.8%). In the DBD, the Ciona VDR/PXR has the highest sequence identity to sea lamprey and zebrafish VDRs (Table 5). The phylogeny of the Ciona VDR/PXR, as inferred by maximum likelihood analysis, does not clearly group this receptor with either VDRs or PXRs (Figure 4). This likely indicates that more sequences are needed, especially additional NR1I receptors (if present) in basal vertebrates (such as Agnatha) and chordate invertebrates. The low sequence identity between the Ciona VDR/PXR and vertebrate VDRs, PXRs, and CARs may be a result of rapid evolution, which has been detected in some gene families (including developmental regulators) in Ciona intestinalis and other tunicates 18195253.

The LBD of Ciona VDR/PXR was cloned from cDNA fragments generously provided by Professors Yuji Kohara and Norituki Satoh, and then inserted into the PM2-GAL4 plasmid to create an LBD/GAL4 chimeric receptor. Unlike similar constructs derived from the vertebrate VDRs, this Ciona VDR/PXR was not activated by any vitamin D derivatives, vertebrate bile salts, or steroid hormones (Additional file 5). Screening of a 76-compound nuclear hormone receptor ligand library (BIOMOL International, Plymouth Meeting, PA, USA) revealed that 6-formylindolo-[3,2-b]carbazole activated Ciona VDR/PXR in the low micromolar range (Additional file 5). 6-Formylindolo-[3,2-b]carbazole (20 μM) was chosen as the reference maximal activator of Ciona VDR/PXR. Screening of an additional 90 compounds comprising steroid hormones, vitamins (other than vitamin D), benzoates, and xenobiotics revealed that carbamazepine and n-butyl-p-aminobenzoate also activated Ciona VDR/PXR in the micromolar range (Additional file 5). Interestingly, 6-formylindolo-[3,2-b]carbazole, carbamazepine, and n-butyl-p-aminobenzoate are all planar molecules.

The Catalyst pharmacophore approach can also be used to generate common feature (HIPHOP) alignments 54 of the three molecules that active Ciona VDR/PXR. In this case the pharmacophore consisted of 1 hydrogen bond acceptor and 2 hydrophobic areas (Additional file 6). This pharmacophore is generally quite different compared with the models for other PXRs described above and in many ways reflects the very narrow substrate selectivity compared with the other six species.

Compared to other vertebrate NR subfamilies, the evolutionary history of the NR1I subfamily is difficult to reconstruct due to a high degree of functional and sequence divergence 101222. Some studies speculate that an ancestral gene duplicated early in vertebrate evolution (or possibly even prior to evolution of vertebrates), with subsequent divergence to become separate PXR and VDR genes 910121517202251. Later in vertebrate evolution, a single PXR gene duplicated, with subsequent divergence to form separate PXR and CAR genes 10. Throughout this manuscript, we follow the convention of designating the non-mammalian PXR/CAR-like genes as PXRs 12, although it is not certain that the ancestral PXR/CAR-like gene is actually the same gene now called PXR in mammals 91020.

Using 49 amino acid sequences of extant VDRs, PXRs, and CARs, we inferred phylogeny by maximum likelihood (Figure 4). Several clusters are clearly evident and supported by bootstrap analysis in the phylogeny presented in Figure 4: vertebrate VDRs, mammalian CARs, and mammalian PXRs. The major difficulty is the placement of the frog PXRs, which are quite different from other PXRs in function, tissue expression, and sequence 7464849. The chicken PXR clusters with the CARs in Figure 4; however, by many measures, chicken PXR is equally related to mammalian CARs and PXRs 91012.

We also utilized maximum likelihood to infer the amino acid sequence of three 'ancestral' receptors, indicated as nodes in Figure 4 labelled as 'AncR1', 'AncR2', and 'AncR3' (Additional file 7). AncR1 represents the ancestral single receptor gene prior to duplication and subsequent divergence to VDRs and PXRs. AncR2 represents the PXR gene ancestral to the split between fish PXRs and mammalian CARs/PXRs. AncR3 represents the ancestral single receptor gene prior to duplication and subsequent divergence to mammalian PXRs and chicken PXR/mammalian CARs. It should be pointed out that ancestral reconstruction based on receptors that are markedly divergent in sequence, particularly when there are insertions or deletions of receptors relative to one another, is subject to significant uncertainly and should be interpreted cautiously. The inter-helical regions of the LBD are particularly difficult to predict. For the LBD, the percentage of amino acid residues with posterior probability greater than 0.7 is only 56.4%, 80.4%, and 65.0% in AncR1, AncR2, and AncR3, respectively (Additional file 7). These overall posterior probabilities are significantly lower than reconstruction of ancestral sex and mineralocortoid NRs (in the NR3 family) 555657, where the cross-species sequence divergences are much less than for the NR1I subfamily of receptors. These uncertainties make homology modelling (or even functional expression) of the LBDs of reconstructed NR1I ancestral sequences unreliable. Therefore, we focused on cross-sequence comparisons of amino acid residues identified as interacting with ligands in crystal structures of human VDR 585960, rat VDR 6162, zebrafish VDR 63, human PXR 293031323334, human CAR 6465, and mouse CAR 66. In this subset of 'ligand-binding residues', the percentage of amino acid residues with posterior probability greater than 0.7 is 56.8%, 90.2%, and 80.4% in AncR1, AncR2, and AncR3, respectively (Additional file 8); each of these values is higher than for the overall LBD sequence indicated above.

At the amino acid residue positions identified as ligand-binding residues, we compared Ciona VDR/PXR, AncR1, AncR2, and AncR3 to mammalian PXRs (human, mouse, rat, rabbit), chicken PXR, Xenopus laevis PXRα and PXRβ, zebrafish PXR, human CAR, human VDR, and sea lamprey VDR (Figure 5). As with overall sequence comparisons in the LBD (Table 5), sequence identities at ligand-binding residues for Ciona VDR/PXR compared to the other receptors were overall low (< 25%). Interestingly, AncR1 showed the highest sequence identity to human VDR (64.7%); all other sequences were less than 51% identical to AncR1 (Figure 5 and Additional file 8). This would be consistent with VDR being the ancestral NR1I receptor 51. The differences between Ciona VDR/PXR and AncR1 at the ligand-binding residue positions may be explained by rapid evolution of the Ciona gene, as discussed above. AncR2 had the highest sequence identity at ligand-binding residues to zebrafish PXR (56.9%), compared to ~ 45–50% for other PXRs and only 31.4% to human CAR. AncR3 had highest sequence identity at ligand-binding residues to mammalian PXRs (66.7% to 70.6%) compared to only 37.3% to human CAR (Figure 5 and Additional file 8). The results for AncR2 and AncR3 both suggest that CAR has diverged the most from the ancestral sequence at ligand-binding residues and would be consistent with PXR being the ancestral gene.

A key factor in protein interactions with ligands or other proteins is presence of intrinsic structural disorder 6768. To assess whether disorder may account for pharmacological differences between the PXRs from different species, intrinsic disorder of the amino acid residues were predicted using the PONDR VL3H algorithm 68 and summarized by the percentage of residues with probability of disorder greater than 50%. Disorder probabilities were analyzed by domain (DBD or LBD) or total protein sequence (Additional files 9 and 10). Rabbit PXR was shown to possess lower predicted intrinsic disorder in the LBD compared with human, mouse, rat, chicken, and zebrafish PXRs. The African clawed frog PXRα (BXRα) had the lowest predicted intrinsic disorder in the LBD of any PXR (Additional files 9 and 10); as discussed above, this receptor has very restricted ligand specificity, essentially responding only to benzoates (and their close structural analogs) shown to be important in early frog development 7. In terms of intrinsic disorder, Ciona VDR/PXR was closer to PXRs than to VDRs in the LBD (Additional files 9 and 10). The chicken PXR was distinct from the other PXRs in terms of low intrinsic disorder in the DBD; in this regard, chicken PXR is much more similar to CARs (Additional files 9 and 10). This is consistent with the hypothesis that an ancestral gene very similar to chicken PXR duplicated, with the two genes ultimately diverging into separate CAR and PXR genes (chicken PXR has about equal sequence similarity to mammalian CARs and PXRs) 91012. In the DBD, chicken PXR may have structural features more similar to mammalian CARs than PXRs. The results are consistent with differences in structural disorder possibly contributing to differences in pharmacologic specificity.

The sources of the chemicals were as follows: n-butyl-p-aminobenzoate, n-propyl-p-hydroxybenzoate, nifedipine, rifampicin (Sigma-Aldrich, St. Louis, MO, USA); 5α-cholanic acid-3α,7α,12α-triol (allocholic acid; Toronto Research Chemical, Inc., North York, ON, Canada); Nuclear Receptor Ligand Library (76 compounds known as ligands of various nuclear hormone receptors; BIOMOL). 5α-cyprinol sulfate (5α-cholestan-3α,7α,12α,26-tetrol-27-sulfate) was isolated from Asiatic carp (Cyprinus carpio) bile 78, 5β-scymnol sulfate (5β-cholestan-3α,7α,12α,24,26-pentol-27-sulfate) was isolated from the bile of Spotted eagle ray (Aetobatus narinari) bile, and 5β-cholestan-3α,7α,12α-triol-27-oic acid, taurine conjugated was isolated from the bile of the American alligator (Alligator mississippiensis). Bile salts were purified by extraction and Flash column chromatography. Bile alcohol sulfates were chemically deconjugated using a solution of 2,2-dimethoxypropane:1.0 N HCl, 7:1 v/v, and incubating 2 hours at 37°C, followed by the addition of water and extraction into ether. Completeness of deconjugation and assessment of purity was performed by thin-layer chromatography using known standards. Other than those described above, steroids and bile salts were obtained from Steraloids (Newport, RI, USA).

Xenopus tropicalis frogs were obtained from NASCO (Fort Atkinson, WI, USA). All animal studies were performed in conformity with the Public Health Service Policy on Humane Care and Use of Laboratory Animals, incorporated in the Institute for Laboratory Animal Research Guide for Care and use of Laboratory Animals. All vertebrate animal studies were approved by the University of Pittsburgh Institutional Animal Care and Use Committees (approval number 0601348) or Committee on Animal Studies of the University of California, San Diego.

The LBD of Xenopus tropicalis (Western clawed frog) PXR (xtPXR) was cloned by PCR from RNA extracted from ovary of an adult female frog. Ciona VDR/PXR was cloned from cDNA fragments ciem829d05 and cilv048e18 provided by Professor Yuji Kohara (Center for Genetic Resource Information, National Institute of Genetics, Research Organization of Information and Systems, Mishima, Japan) and Professor Noriyuki Satoh (Kyoto University, Kyoto, Japan), with analysis of the cDNA clones supported by Grant-in-aid for Scientific Research on Priority Area "Genome" of Ministry of Education, Culture, Sports, Science and Technology, Japan. The LBD of xtPXR (residues 103–390) and Ciona VDR/PXR (residues 57–391) were inserted into the pM2-GAL4 vector to create a GAL4/LBD chimera suitable for study of ligand activation 17.

The creation of a HepG2 (human liver) cell line stably expressing the human Na⁺-taurocholate cotransporter (NTCP) has been previously reported 17. HepG2-NTCP cells were grown in modified Eagle's medium-α containing 10% fetal bovine serum and 1% penicillin/streptomycin (Invitrogen, Carlsbad, CA, USA). Plasmids containing cDNAs for 8 PXRs from 7 species (human, mouse, rat, rabbit, chicken [also called chicken X receptor, CXR], African clawed frog [also termed Xenopus laevis benzoate X receptors α and β, BXRα and BXRβ ], zebrafish), as well as the reporter constructs tk-UAS-Luc and CYP3A4-PXRE-Luc, and 'empty' vectors pCDNA, PsG5, and PM2 were generously provided by SA Kliewer, JT Moore, and LB Moore (GlaxoSmithKline, Research Triangle Park, NC, USA). The expression vectors were either full-length receptors (i.e., containing both a DBD and LBD; human, mouse, rat, rabbit, and chicken PXRs) or GAL4/PXR chimeras that contain only the LBD of the PXR receptor (BXRα, BXRβ, Xenopus tropicalis PXR, and zebrafish PXR). For the full-length expression vectors, the reporter plasmid was CYP3A4-PXRE-Luc, a construct that contains a promoter element from CYP3A4 (recognized by PXR DBDs) driving luciferase expression. For the GAL4/LBD expression constructs, the reporter plasmid was tk-UAS-Luc, which contains GAL4 DNA binding elements driving luciferase expression. The following transfection ratios of reporter, receptor, and β-galactosidase plasmids were used (ng/well): human, mouse, rabbit, and rat PXRs – 25/2.7/20; chicken PXR – 10/1/20; Xenopus tropicalis PXR, zebrafish PXR, and Ciona VDR/PXR – 75/50/20.

The PXR activation assay was a luciferase-based reporter assay 1745. On day 1, 30,000 cells/well were seeded onto 96-well white opaque plates (Corning-Costar, Corning, NY, USA). On day 2, cells were transfected using the calcium phosphate precipitation method with expression vector or 'empty' control vector and luciferase reporter plasmid. On day 3, the cells were washed and then incubated with medium containing charcoal-dextran treated fetal bovine serum (Hyclone, Logan, UT, USA) and drugs or vehicle. On day 4, the cells were washed and the medium replaced with serum-free medium. Cells were washed with Hanks' buffered salt solution and then exposed to 150 μL lysis buffer (Reporter Lysis Buffer, Promega). Separate aliquots were taken for measurement of β-galactosidase activity (Promega) and luciferase activity (Promega Steady-Glo luciferase assay).

Activation of receptor by ligand was compared to receptor exposed to identical conditions without ligand ('vehicle control'). In general, dimethyl sulfoxide (Sigma) was used as vehicle and was adjusted to be 0.5% (v/v) in all wells. A control was also run with transfection of 'empty' vector (i.e., lacking the receptor cDNA) and reporter vector to control for activation of reporter vector by endogenous receptor(s). In experiments with a variety of activators, activation by endogenous receptors was not seen.

To facilitate more reliable cross-species comparisons, complete concentration-response curves for ligands were determined in the same microplate as determination of response to a maximal activator. This allows for determination of relative efficacy, ε defined as the maximal response to test ligand divided by maximal response to a reference maximal activator (note than ε can exceed 1). The following maximal activators and their concentrations were as follows: human PXR – 10 μM rifampicin; mouse and rat PXRs – 10 μM pregnenolone 16α-carbonitrile; rabbit PXR – 50 μM 5β-pregnane-3,20-dione; chicken PXR – 20 μM nifedipine; Xenopus tropicalis PXR – n-propyl-p-hydroxybenzoate 50 μM; zebrafish PXR – 20 μM 5α-androstan-3α-ol; and Ciona VDR/PXR – 20 μM 6-formylindolo-[3,2-b]carbazole. All comparisons to maximal activators were done within the same microplate. Luciferase data were normalized to the internal β-galactosidase control and represent means ± SD of the assays. Concentration-response curves were fitted using Kaleidagraph software (Synergy Software, Reading, PA, USA). In combining data from multiple experiments, the pooled variance was calculated by the formula s_pooled = {[(n₁-1)s₁² + (n₂-1)s₂² + ... + (n_k-1)s_k²]/[N-k]}^-1/2, where there are N total data points among k groups, with n replicates in the i^th group.

To test for cytotoxicity, two assays that have been well-validated in HepG2 cells were used: 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) reduction and alamar blue reduction. Both assays sensitively measure the ability of viable cells to metabolize the parent compound to a metabolite that can be detected by spectrophotometry or fluorometry 79. HepG2 cells were seeded at a density of 20,000 cells/well (100 μL per well) into clear 96-well microplates (for the MTT assay) or black, opaque 96-well plates (for the alamar blue assay) and grown for 24 hours. The next day, 100 μL solutions of drug concentrations or vehicle controls in cell growth medium at twice the intended final concentration were added to the cells (final volume 200 μL). The cells were again incubated for 24 hr. For the MTT assays, MTT (In vitro toxicology assay kit, MTT-based; Sigma, St. Louis, MO, USA) was dissolved at 5 mg/mL in warm cell growth medium. 20 μL of this solution was added to the cells (total volume 220 μL), and the plates incubated for another 4 hrs. After incubation, the supernatant was removed and 50 μL of solubilization buffer provided in the Sigma kit with 0.5% DMSO was added. DMSO was added to ensure total solubility of the formazan crystals. Plates were shaken for 2 min, and the absorbance recorded at 590 nm. The percent viability was expressed as absorbance in the presence of test compound as a percentage of that in the vehicle control (with subtraction of background absorbance).

For the alamar blue assays, alamar blue stock solution (Biosource International; Camarillo, CA, USA) was diluted 1:1 with cell growth medium and 50 μL of this was added to each well, yielding a final concentration of 10% alamar blue (total volume 250 μL). The plates were exposed to an excitation wavelength of 530 nm, and the emission at 590 nm was recorded to determine whether any of the test drug concentrations fluoresce at the emission wavelength. Plates were returned to the incubator for 5 hr and the fluorescence was measured again. The percent viability was expressed as fluorescence counts in the presence of test compound as a percentage of that in the vehicle control (with subtraction of background fluorescence). Drug concentrations that cause > 30% loss of cell viability in the MTT assay or > 15% loss of cell viability in the alamar blue assay were not used in the determination of concentration-response curves for activation of PXRs.

Pharmacophore modelling was performed as described previously 4554. Briefly, computational molecular modeling studies were carried out using Discovery Studio 1.7 Catalyst™ (Accelrys, San Diego, CA) running on a Sony Vaio with Intel Centrino processor. Pharmacophore models attempt to describe the arrangement of key features that are important for biological activity. Briefly, the Catalyst™ models were employed to generate hypotheses. Molecules were imported from sdf files, the 3-D molecular structures were produced using up to 255 conformers with the Best conformer generation method, allowing a maximum energy difference of 20 kcal/mol. Hypogen PXR pharmacophores for each species were generated with Catalyst™ using the 16 molecules in Table 4. Molecules highlighted in bold type were used for training as they are common to all species – molecules with no effect were given the arbitrary EC₅₀ value of 10,000 μM (10 mM).

Ten hypotheses were generated using these conformers for each of the molecules and the EC₅₀ values, after selection of the following features: hydrophobic, hydrogen bond acceptor, and hydrogen bond donor with up to 4 excluded volumes. After assessing all ten generated hypotheses, the hypothesis with the lowest energy cost was selected for further analysis as this possessed features representative of all the hypotheses and had the lowest total cost. The quality of the structure activity correlation between the estimated and observed activity values was estimated by means of an r value. Additionally 6-formylindolo-[3,2-b]carbazole was aligned with carbamazepine and n-butyl-p-aminobenzoate with the HIPHOP alignment to ascertain the pharmacophore for Ciona VDR/PXR.

The following sequences were used for phylogenetic analysis and ancestral sequence reconstruction (some links are from the Ensembl database 80): human VDR [GenBank: NM_00376], chimpanzee VDR [Ensembl:ENSPTRT00000009010], rhesus monkey VDR [Ensembl:ENSMMUT00000009414], cow VDR [Ensembl:ENSBTAT00000021832], dog VDR [Ensembl:ENCAFT00000014497], mouse VDR [GenBank: NM_008504], rat VDR [GenBank: NM_009504], chicken VDR [GenBank: AF011356], Japanese quail VDR [GenBank: U12641], Xenopus laevis VDR [GenBank: U91849], Atlantic salmon (Salmo salar) VDR [GenBank: AJ780914], fugu VDR [Ensembl:NEWSINFRUT00000138841], bastard halibut (Paralichthys olivaceus) VDR [GenBank: AB037674], zebrafish VDR [GenBank: AF164512], medaka VDR [Ensembl:ENSORLT00000001311], stickleback fish (Gastrosteus aculeatus) VDR [Ensembl:ENSGACT00000006308], sea lamprey VDR [GenBank: AY249863], Ciona intestinalis VDR/PXR [GenBank: BR000137], human CAR [GenBank: NM_005122], chimpanzee CAR [ENSPTRT00000002884], rhesus CAR [GenBank: AY116212], cow CAR [Ensembl:ENSBTAT00000012145], dog CAR [Ensembl:ENSCAFT00000020528], pig CAR [GenBank: AB214979], Baikal seal (Phoca sibirica) CAR [GenBank: AB109553], Northern fur seal (Callorhinus ursinus) CAR [GenBank: AB109554], mouse CAR [GenBank: NM_009803], rat CAR [GenBank: NM_022941], pig CAR [GenBank: AB214979], opossum CAR [Ensembl:ENSMODT00000006393], human PXR [GenBank: AF061056], chimpanzee PXR [ENSPTRT00000028510], rhesus monkey PXR [GenBank: AF454671], cow PXR [Ensembl:ENSBTAT00000026059], mouse PXR [AF031814], rat PXR [GenBank: NM_052980], rabbit PXR [GenBank: AF188476], opossum PXR [Ensembl:ENSMODT00000023109], chicken PXR [GenBank: AF276753], Xenopus laevis PXRα [GenBank: BC041187], Xenopus laevis PXRβ [GenBank: AF305201], Xenopus tropicalis PXR [Ensembl:ENSXETT00000039109], fugu PXR [Ensembl:NEWSINFRUT00000171584], medaka PXR [Ensembl:ENSORLT00000022473], Tetraodon nigriviridis PXR [Ensembl:GSTENT00026021001], zebrafish PXR [GenBank: AF454674, GenBank: AF502918], domestic silkworm (Bombyx mori) ecdysone receptor [GenBank: L35266], ixotid tick (Amblyomma americanum) ecdysone receptor [GenBank: AF020187], and purple sea urchin (Strongylocentrotus purpuratus) liver X receptor [GenBank: XM_774904]. Sequences were aligned using ClustalW 81 and Tcoffee software 82 and manually adjusted as needed.

Phylogeny was inferred by maximum likelihood using PHYML software 8384, assuming a WAG protein model and a 4-category discrete gamma distribution of among-site rate variation. To estimate support, 100 bootstrap replicates were analyzed. Ancestral protein sequences of AncR1, AncR2, and AncR3 (see nodes in Figure 4) were inferred by maximum likelihood using PAML 3.15 software 8586 on the maximum likelihood phylogeny of 49 amino acid sequences of extant VDRs, PXRs, and CARs (see Figure 4). For ancestral reconstruction, the JTT+G model (supported with 100% posterior probability in the Bayesian analysis) was assumed. Residues that interacted closely with ligands in published X-ray crystallographic structures of human VDR 585960, rat VDR 6162, zebrafish VDR 63, human PXR 293031323334, human CAR 6465, and mouse CAR 66 were identified and designated in Figure 5 and Additional files 7 and 8 as 'ligand-binding residues'.

Intrinsic disorder prediction of protein sequences were performed using the PONDR VL3H algorithm 6887. The disorder calculations for each amino acid residue are available as Additional file 10.