Introduction

Top Abstract Introduction Materials and Methods Results and Discussion References

Interpreting drug-drug interactions at the molecular level may aid in the development of more effective and safer therapeutics. Because the CYP3A subfamily is probably the most important enzyme subfamily in terms of expression within liver and small intestine and involvement in metabolism of many drugs, focus on the regulation of expression of this subfamily is a priority. Recently, it was discovered that the pregnane X receptor (PXR¹) is a regulator of CYP3A transcription (Bertilsson et al., 1998; Blumberg et al., 1998; Kliewer et al., 1998) and is activated by most of the structurally diverse CYP3A inducers. This provided some degree of understanding for modulation of CYP3A expression in response to endo- and xenobiotics. The expansion of the role of PXR to regulation of expression of human multidrug resistance-associated protein-1 (which encodes for P-glycoprotein) and CYP2C8, enabling drugs to regulate their own metabolism and efflux, is beginning to be understood (Schuetz and Strom, 2001; Synold et al., 2001). The implications of PXR for drug development have resulted in some advocating the use of high-throughput assays to eliminate likely CYP3A inducers or alternatively progressing molecules through development with this caveat (Moore and Kliewer, 2000). It might be possible to avoid ligands for PXR since the ligand binding domain of human PXR is a large, flexible hydrophobic site, as described by X-ray crystallography (Watkins et al., 2001). Flexibility of the binding site probably allows promiscuity in accepting structurally diverse ligands (Table 1). To date, there have been several reports describing PXR binding as EC₅₀ values (Moore et al., 2000a,b; Staudinger et al., 2001). The present study uses these data to generate a pharmacophore that represents key features of ligands of the PXR binding site. This technique has been used previously (Ekins et al., 2000, 2001) because it represents an inexpensive, insightful approach to highlight the important binding features of ligands for enzymes and receptors. We have further tested this model with other PXR ligands and by aligning it within the X-ray structure.

	Materials and Methods

Top Abstract Introduction Materials and Methods Results and Discussion References

Modeling with Catalyst. The computational molecular-modeling studies were carried out using Silicon Graphics Octane and O2 workstations. Briefly, models were constructed using Catalyst version 4.5 (Accelrys, San Diego, CA) after importing the molecular structures described in the literature (Table 1; Moore et al., 2000a,b; Staudinger et al., 2001) and obtained from the ISIS MDDR-3D database (version 2000.2; MDL Information, Inc., San Leandro, CA) as the training set. Conformers of each ligand were generated as previously described for inhibition of cytochromes P450 (Ekins et al., 1999a, 2000, 2001) and used along with the reported EC₅₀ values to build 10 hypotheses (i.e., pharmacophores). The pharmacophore features considered for the model were hydrogen bond donors, hydrogen bond acceptors, and hydrophobic and ring aromatic features. After assessing all 10 hypotheses generated, the lowest energy cost hypothesis was used because this possessed features representative of all the hypotheses and had the lowest total cost.

Evaluating the Fit of the Pharmacophore to the X-Ray Structure of PXR. The Catalyst hypothesis was overlaid on each of the three SR12813 positions from the X-ray structure by minimizing the root mean square distance between each of the heavy atom positions of the X-ray (Watkins et al., 2001) and the Catalyst conformations. The root mean square distances were 2.246, 2.372, and 1.998 Å for positions 1, 2, and 3, respectively. Graphics of the overlays were created with WebLab ViewerPro (Accelrys).

Evaluating the PXR Binding of External Molecules. Twenty-eight molecules excluded from the training set and identified as PXR ligands by several groups were used to evaluate the pharmacophore. Conformations of these 28 structures were generated as described above and then fitted to the pharmacophore using the fast fit algorithm in Catalyst. Because these molecules were determined to activate/deactivate PXR but published EC₅₀ values are unavailable, a qualitative assessment was made of the predictions based on the relative activation of PXR.

	Results and Discussion

Top Abstract Introduction Materials and Methods Results and Discussion References

The 12 molecules with literature EC₅₀ data (Table 1) were used to generate a pharmacophore for PXR ligands using the Catalyst software. The model generated with these data consisted of one hydrogen bond acceptor and four hydrophobes and possessed an r² value correlation of observed and predicted EC₅₀ of 0.92. When molecules in the training set are fitted to the model, the most potent ligands, like hyperforin (Fig. 1A), fit all features well. Interestingly, the Catalyst model appears to accommodate large structures that were thought to alter the size of the binding site (Watkins et al., 2001); this can be seen in the distances between the hydrophobes and the hydrogen bond acceptor, ranging from 3.6 to 7.6 Å (Fig. 1B). The prediction of these features is in close agreement with the recently derived X-ray structure of the human PXR ligand binding domain (Watkins et al., 2001). This previous study suggested the binding site was largely hydrophobic, with some hydrogen bonding interactions with the cocrystallized ligand SR12813. This molecule aligns to the pharmacophore with a hydrogen bond through a P==O molecule feature (Fig. 1C). When this pharmacophore is aligned on the X-ray structure of SR12813 in complex with PXR, position 1 (as defined by Watkins et al., 2001) is the least consistent because the hydrogen bond acceptor feature on SR12813 (P==O) is not close to Ser²⁴⁷ (Fig. 1D). Positions 2 (Fig. 1E), and to a greater extent position 3 (Fig. 1F), seem to be consistent with the pharmacophore because the hydrogen bond acceptor feature is close to His⁴⁰⁷ (position 2) and Ser²⁴⁷ (position 3), described previously as hydrogen bonding regions within PXR (Watkins et al., 2001). The multiple hydrophobic features would also agree with the van der Waals contacts at multiple points (Watkins et al., 2001). One of the disadvantages of the pharmacophore approach is the perceived inability to model multiple binding modes in a single pharmacophore. EC₅₀ values can, however, be used to generate an "averaged" pharmacophore of the major features of the productive orientations of ligands in the binding site, as discussed previously (Ekins et al., 2001). In the case of PXR, it seems that we can describe at least two of the binding orientations of SR12813, namely positions 2 (Fig. 1E) and 3 (Fig. 1F).

View larger version (71K): [in this window] [in a new window]   Fig. 1.   Pharmacophores for PXR ligands generated using Catalyst and their overlay to the X-ray positions 1 to 3 derived for SR12813. A, hyperforin fitted to the PXR EC50 pharmacophore derived from 12 ligands. Cyan spheres represent hydrophobic features; green sphere represents a hydrogen bond acceptor with a vector in the direction of the putative hydrogen bond donor (larger sphere). Features of the hyperforin molecule of particular note include numerous oxygen (red) molecules. B, interfeature distances (angstroms) and angles (degrees) for the PXR pharmacophore; x-, y-, and z-coordinates for features are as follows. Hydrogen bond acceptor (point small sphere): 7.6, 7.4, 3.6; hydrogen bond acceptor vector (larger sphere): 7.46, 11.44, 3.98; hydrophobe: 7.53, 10.80, 1.05; hydrophobe: 7.99, 15.04, 10.69; hydrophobe: 13.75, 11.77, 7.94; hydrophobe: 4.20, 10.03, 4.62. C, SR12813 fitted to the PXR EC50 pharmacophore derived from 12 ligands. Cyan spheres represent hydrophobic features; the green sphere represents a hydrogen bond acceptor with a vector in the direction of the putative hydrogen bond donor (larger sphere). Features of the SR12813 molecule of particular note include numerous oxygen (red) and phosphorus (blue/green) molecules. D, Catalyst hypothesis and SR12813 conformation (in yellow) overlaid on X-ray position 1 of SR12813 (green). Important amino acids identified in (Watkins et al., 2001) are shown. E, Catalyst hypothesis and SR12813 conformation (in yellow) overlaid on X-ray position 2 of SR12813 (green). Important amino acids identified in (Watkins et al., 2001) are shown. E, Catalyst hypothesis and SR12813 conformation (in yellow) overlaid on X-ray position 3 of SR12813 (green). Important amino acids identified in (Watkins et al., 2001) are shown.

Activation data reported after exposure of PXR-transfected CV-1 or Caco-2 cells to a single concentration of test compound were used as an evaluation set for the pharmacophore generated in this study (Table 2). These 28 molecules were selected from a range of publications and represent some of the most diverse structures tested in vitro to date, many of which are PXR ligands. Some of these molecules when evaluated in multiple studies show a wide variability in -fold activation, for example pregnenolone, spironolactone, corticosterone, and phenobarbital. Ligand assay concentrations varied from 10 µM to 1 mM, further complicating any attempts at rank ordering these molecules (Table 2). Therefore, the 28 molecules were classified as potent (predicted EC₅₀ of <1 µM and observed >5-fold activation of PXR) and nonpotent PXR ligands (predicted EC₅₀ of >1 µM and observed <5-fold activation of PXR). The pharmacophore identifies potent PXR activators, such as troglitazone, dexamethasone-t-butylacetate, ecteinascidin, and nifedipine, differentiating them from weak PXR activators, such as quercetin, scopoletin, myricetin, and kaempferol. The model misclassifies the inactive taxotere probably because of the flexibility of this large molecule and the presence of multiple pharmacophore features. In general, the model seems to offer a useful approach for inferring whether a new molecule could be a PXR ligand in silico before verification in vitro. More experimental EC₅₀ data spanning a range of many orders of magnitude not included in the model building would be ideal to help validate the present model; but at present, this is not readily available. In addition, the pharmacophore does appear to be consistent with the crystal structure (Watkins et al., 2001), especially binding position 2 and 3 (Fig. 1, C, E, and F). The PXR pharmacophore contains hydrophobic and hydrogen bond acceptor features in a different arrangement to that previously identified as important in pharmacophores for CYP3A4 substrates and autoactivators (Ekins et al., 1999b) and CYP3A4 inhibitors (Ekins et al., 1999a). The apparent similarity between pharmacophore feature content and the dissimilarity of their arrangement in these different models is perhaps not surprising considering some, but not all, CYP3A4 substrates are also CYP3A inducers.

We have described a complimentary approach to using the protein structure for determining ligand binding potential with PXR that could assist discovery of further natural ligands by database searching. Modeling PXR data derived from other species should also help define the structural features that are unique and representative of subtle binding domain differences that result in species differences in CYP3A induction.