Sequence/Structure Alignment for NR-LBDs Used to Construct VDR-LBD Model I (14-Oct-99)

                 H1                   H2                                       H3                         H4

      |        H5              S1       S2   H6               H7                    H8

                 H9                               H10       |     H11               H12

HOMOLOGY MODEL OF VDR-LBD

X-ray crystallographic structures have been published for several nuclear receptor ligand-binding domains (NR-LBDs): (a) unliganded RXR (Bourguet et al., 1995), (b) liganded RAR (Renaud et al., 1995), (c) liganded TR (Wagner et al., 1995), (d) liganded ER (Brzozowski et al., 1997; Tanenbaum et al., 1998), (e) liganded PR (Williams & Sigler, 1998), and (f) liganded PPAR (Nolte et al., 1998). Sequence-homology comparisons indicate that TR would make the best template for homology-extension modeling of VDR, but the coordinates of this structure have never been released [not available in the Protein Data Bank (www.rcsb.org) as of 10/18/99]. We therefore selected PR as our principal template, but modeled two short segments on the corresponding parts of RAR (see below). We also excised 60 residues of the long insert in the VDR-LBD which projects outward between helices H2 and H3; secondary-structure predictions for this sequence are weak and there are no good matches to it in the PDB. Hence any attempt to model it would be arbitrary, and its position in the protein structure is such that it cannot play a significant role in ligand binding, coactivator binding or receptor dimerization (assuming that these functions are embodied in portions of VDR that are homolgous to those of other members of the NR family).

A substantial literature covers the topic of the conformation(s) accessible to 1,25-(OH)2-D3 and we thoroughly reviewed it before selecting a starting structure for the ligand. For the A-ring we selected the chair form with the C-1 hydroxyl group in an axial position (which makes the C-3 hydroxyl group equatorial). We tested both cis- and trans-conformations of the seco-B ring, normally preferring the latter. A standard steroid conformation was employed for the C,D-ring moiety.

Figure 1 depicts the sequence alignment used for model I; the legend describes its construction. Alignments were generated **before** deletion of the large inserted loop (positions 162-221). Note that a multi-alignment has much higher information content and reliability than a two-sequence comparison and our final manual adjustment took all six sequences into account.

We used various modules of the Insignt II software package from MSI to carry out the modeling calculation, first superposing all of the known NR-LBD structures to define the positions of helices and the shape of the ligand pocket with bound ligand. Residues 125-141 were chosen for H1, which is therefore longer than H1 of the PR template. Standard parameters were used to extend this alpha helix. The rest of the structure was built with the Insight II Homology Module using the coordinates of PR as a template and choosing the loops manually from the automatic selections generated by the software. In two parts of the sequence (residues 143-161 and 283-293) we used RAR coordinates in place of PR to give a "hybridcomplex" template. Considerations of accomodating the H2/H3 loop in an unstrained structure and positioning the beta-hairpin at the inner end of the binding pocket so as to account for affinity labeling and mutation data (see above) dictated the selection of the RAR segments in the model. All residue positions in the "hybridcomplex" were then mutated to the corresponding VDR amino acid sidechains.

We used the Builder Module of Insight II to construct the ligand molecule and then placed it in an initial docking position in the model based on the superposition of the ligands in the known liganded LBD structures and the result of affinity labeling studies with the 3-bromoacetate ester. Using the Docking Module we further explored ligand conformations and manually adjusted the 1,25-(OH)2-D3 position with reference to the bond and electrostatic energy terms. Next the liganded hybridcomplex was exported to XPLOR for further refinement. H-bond partners for the hydroxyl groups on the A ring were chosen on the basis of their position vis-a-vis the ligand as well as mutational data. Constraints were imposed so as to force the hydroxyl group on C-3 to form an H-bond to the sidechain of serine-237 and to force the C-1 hydroxyl to accept an H-bond from the guanidinium group of arginine-274 and donate one to the hydroxyl group of serine-275. Repeated rounds of minimization were conducted alternately with and without the H-bond constraint terms. The final structure generated was robust, changing very little when subjected to molecular dynamics. Note that throughout this procedure the sidechain of 1,25-(OH)2-D3 was **not** constrained to any particular position, but simply allowed to seek a minimum-energy position.