One way of understanding life at molecular level is to obtain the three-dimensional (3D) structures of the molecules. Such structural views represent a static picture in the life of a molecule giving a sneak preview what it might be in reality. For the understanding of the functional implication of vitamin D (VD) signaling it is also important to look at various structural complexes of the vitamin D receptor (VDR), which may outline its possible dynamics and mechanics. VDR is able to perform the ligand-, DNA- and protein recognition, and interaction in a multi-task manner thus can be viewed as molecular machine which will regulate gene expression with the combination/sum of all these particular functions.

Before the year 2000 people in the VD field were only guessing how may the 3D structure of VDR look like. Some implications were coming from already known crystal structures e.g., receptors for retinoids such as retinoid X receptor (RXR) (Bourguet et al., 1995) and retinoic acid receptor (RAR) (Renaud et al., 1995). Not until exactly 14 years ago the structure of VDR-1α,25-dihydroxyvitamin D₃ (1,25D₃) complex has been solved (Rochel et al., 2000) and a long journey started in understanding the binding of various VDR analogs and the structure-based analog design. Within the first 5 years more structures have been solved (Tocchini-Valentini et al., 2001, 2004; Eelen et al., 2005) but the long awaited answer to explain the differences in physiological effects for various ligands was one of the biggest disappointment that crystal structures provided since no substantial distinction could be made for the conformation of the active VDR-ligand complexes.

Compared to the beginning of the last decade a huge number of X-ray crystal structures are available for VDR. In detail, there are VDR LBDs from three different species H. sapiens (34), R. norvegicus (40) and D. rerio (13) and four DBD-DNA complexes from H. sapiens. The basic information about these complexes is summarized in Table 1. This data makes also possible to analyze orthologous molecules with reflection to functional and structural differences. However, to understand this aspect well it would be beneficial to have more data from numerous organisms. In this review evolutionary aspects and species-specific difference will be not discussed in depth. More space will be given to general domain organization, binding mode of natural ligands and recognition of DNA by VDR. Some data coming from molecular dynamics (MD) simulations will be also discussed since this approach represents a compromise in obtaining 3D structural models and have been proven to be well aligned with the wet lab data. At last, recent data from small-angle X-ray scattering (SAXS), cryo-electron microscopy (cryo-EM) and H/D exchange (HDX) experiments will be discussed with some perspectives highlighted.

From functional and structural organization point of view VDR is formed by DNA-binding domain (DBD:24–89 aa; domain C), ligand-binding domain (LBD:126–427 aa; domain E), a connective hinge (domain D) between them and a short A/B domain located at the N-terminus. Compare to RXR it completely lacks the F domain, which is the very last part of the LBD after helix 12. The LBD is formed by a three layer anti-parallel α-helical sandwich (Figures 1A,B highlighted in green, blue and red). Based of the particular structure it contains all together 11–13 α-helices (Li et al., 2003) (Figure 2A). The internal structure of the LBDs of the respective nuclear receptors (NRs) shows a high similarity with specialized diversity based on functional properties of the particular receptor. In VDR, the LBD is responsible for active ligand recognition and interaction with partnering proteins such as coregulators, and RXR to form the functionally active RXR-VDR heterodimer. In particular, helices 3, 4, and 10–12 are involve in the interaction with protein partners. Interestingly, to date all solved VDR crystal structures show very ubiquitous and conserved organization of the of overall structural fold not reflecting the divergent nature of the bound natural or synthetic ligands. What are in fact the differences in structures that reflect various physiological effects of the particular ligands? Allegedly there will not be a simple answer to this question since we may face the limitation on the complexity of the available ligand-VDR structures or have to allow the possibility that the subtle differences between the structures may cause a “butterfly effect” that have more extreme consequences in terms of VDR signaling than initially thought. By all means there are important differences in the metabolism of various synthetic ligands and a possible unique coactivator recruitment may also play its role. However, none of these possibilities can be fully explored using the available VDR crystal structures. Nevertheless, what we may agree on is that all VDR crystal structures show agonistic conformation, surprisingly even in case of antagonists, that is canonically represented by a closed conformation of the helix 12 providing a docking platform for the recruitment of coactivators. This may be due to the shifted equilibrium that drives VDR for closed helix 12 with minimal energy conformation. In addition, the VDR structures do tell us the binding mode, anchoring points and subtle changes in the position of residues that may be effectively used for de novo design of superagonist such as AMCR277A (PDBID:3CS4) (Hourai et al., 2008). The frequently mentioned subtle changes that are characteristic for the ligand-binding pocket (LBP) may be further analyzed and can explain some of the binding differences between various ligands in correlation to their functional and biological properties.

The “lower part” of the LBDs of all ligand-activated NRs contains a LBP which volume size range between 400 and 1400 ų (Figures 2A,B light-brown mesh) (Nagy and Schwabe, 2004). It is not quite different in case of VDR where LBP primarily serves for effective recognition of various natural ligands such as 1,25D₃ and its metabolites or bile acids. In addition, this is one of the most important parts to modulate VDR's activity via various synthetic compounds. The VDR pocket can be placed in the middle range of the volume scale showing rather high dynamic plasticity toward various ligands.

The first VDR crystal structure confirmed the conserved contact or anchoring points for the interaction of VDR with 1,25D₃ (Rochel et al., 2000) (Figure 2B). The residues involved in the positioning of the 1,25D₃ in the LBP are Y143 (helix 1) and S278 (helix 5) that contact the 1,25D₃1α-hydroxyl (OH) group, S237 (helix 3) and R274 (helix 5) contact 3β-OH, and H305 (loop 6–7) and H397 (helix 11) interact with 25-OH. In most of the cases if these anchoring point are disrupted a decrease in the activation potential of the ligand can be observed. Besides these residues the interior surface of the VDR LBP is formed of about 40 mostly non-polar amino acids. An interesting approach is to look how the LBP is changed upon ligand-binding especially its volume or how much volume (%) is occupied by the ligand. A rather straightforward example is the comparison of the 1,25D₃ and MC1288 LBPs. Although the volumes of these ligands are highly comparable 434 ų and 427 ų (probe radius 1.9 Å), respectively, the volumes of the LBP show slight differences 776 ų and 643 ų (same probe radius) leading to ratio of 56 and 66% of ligand occupancy of the LBP volume (Molnár et al., 2006). When a ligand occupancy is higher compared to 1,25D₃ then it increases the stability of VDR. This factor holds true for ligands that show high structural similarity and binding mode to 1,25D₃. It seems to some extend that the ratio of the ligand to LBP volume can be a good descriptor of the ligands' activation potential. In addition, the actual shape of the cavity also reflects differences in the binding of various ligands which is illustrated in Figure 3. There are slight differences in the shapes of the LBP depending on the bound ligand. The red arrow shows the part where the shape is conserved well. Topologically this is the place where the 1,25D₃ A-ring is located e.g., helices 1, 3, and 5. The more plastic part is the one where the 1,25D₃ side chain is found with helices 6, 7, and 11. This region shows the highest variation between the ligand-bound structures and is indicated with green arrow Figure 3. In addition, small changes can be also detected for instance the part highlighted with green circle. Very interesting in the difference between the two Gemini structures 2HCD and 4IA1, where the largest difference for the two side chains is that “C17–20 threo 20S—Gemini” has the hydrogens in its methyl groups substituted with deuterium. Interestingly, in this case the pocket shows more compact shape in the region with the double side chains and the volume of the location highlighted with green circle is decreased, which may be due to modification of the A-ring. Another interesting question is the maximum volume to which the VDR LBP can be stretched. MD simulations showed that by docking a Gemini with fluorinated methyls groups, (CF3)₂-Gemini, the LBP could be expanded by 1/3 of the 1,25D₃ LBP volume. In proportion to this, the compound's volume in the pocket is also increased about 30%.

An appealing view opens up when a comparison of the VDR structure with one of its closest relative pregnane X receptor (PXR) is made (Watkins et al., 2001). As for VDR there are plenty of PXR crystal structures available and it can extend its LBP to very large volume (~1400 ų) to accommodate various compounds. The published VDR structures lack the insertion domain (Δ166–216 aa), but even without it the LBP as discussed earlier is as large as 700–800 ų with MD simulations showing that it can expand beyond 1000ų, which is already comparable to PXR's LBP. This may suggest that VDR is able to accommodate variety of other compounds in addition to 1,25D₃ such as LCA/3-keto-LCA (Makishima et al., 2002). The later have been crystallized with rVDR and will be discussed in this review. The indications for binding additional compounds besides 1,25D₃ are coming also from P. marinus (sea lamprey) where despite of lack of the calcified skeleton and teeth it may serve as a xenobiotic activator for detoxification by regulating P450 enzymes (Whitfield et al., 2003; Krasowski et al., 2005). However, it is yet to be determined whether there are existing other ligands that bind to VDR. One part of the LBD, which may allow the binding of these compounds is the insertion domain. Although its clear functional role has not been identified and it seems that it is not directly required for the binding of the 1,25D₃ (Rochel et al., 2001), it may play some other roles. A mutation C190W was reported in patients that results in loss of 1,25D₃ binding (Malloy et al., 1999), though this may be due to the disruption of the VDR structure by introducing a large bulky tryptophan residue. Secondly, the homologous part of PXR (142–431 aa) especially the occurrence of the two β-strands and the associated coiled regions are responsible for the expandability of the PXR's LBP thus an analogous role cannot be out ruled in case of VDR as well. There has been also proposed that an alternative pocket, which can be considered as an enlargement of the original pocket, is formed in the VDR that extends toward the helix 2/β-sheet region of the LBD (Mizwicki et al., 2004). Especially this can be observed with covalently locked 1,25D₃-derived compounds such as 1,25(OH)₂-lumisterol, which has been showed by in silico docking studies.

Due to the space limitation this review cannot address and discuss all the VDR-ligand complexes, for some more details see text below or recent reviews (Carlberg and Molnár, 2012; Carlberg et al., 2012), but some of the structures with natural ligands will be discussed in more details. One of them is 3-epi-1α,25-dihydroxyvitamin D₃ (3-epi-1,25D₃) a 1,25D₃ metabolite that has been shown to exhibit tissue specific activities comparable to 1,25D₃ (Norman et al., 1993; Reddy et al., 2000). The structural analysis showed a binding mode very similar to that of 1,25D₃ with interesting compensation for the lacking S278-3α-OH hydrogen bond for the epimer using water mediated contacts Figure 4 (Molnár et al., 2011). Interestingly, the same water channel is present in the 1,25D₃ complex and was observed with other complexes as well (Tocchini-Valentini et al., 2001; Hourai et al., 2006).

A widely accepted fact is that precursor 25-hydroxyvitamin D₃ (25D₃) and its metabolite 24R,25-hydroxyvitamin D₃ (24,25D₃) does not posses significant biological activities. However, it has been shown that in Cyp27b1^−/− cells, that are unable to produce actively 1,25D₃, the VD signaling may be primary mediated via 25D₃ (Lou et al., 2010). It stays a matter of discussion that under physiological condition how much of the VD signaling is mediated via 1,25D₃ vs. 25D₃. 24,25D₃ shows only weak potency of influencing VD signaling at concentration 500nM although an enhancement for human osteoblast differentiation has been shown at concentration 1 μM (van Driel et al., 2006). Docking and subsequent MD simulations have been done to see the binding mode of these compounds and it has been confirmed that the position of residues mediating the anchoring hydrogen bonds are conserved with the exception of R274 which is located further than 3.5 Å from 24,25D₃ Figure 5. Another residue that is unable to make a binding contribution is S237, but its position in the pocket is maintained. The void created by the missing R274 increases the LBP and lowers the occupancy factor for this compound. The analysis of the simple binding mode confers the activity range from 1,25D₃ > 25D₃ > 24,25D₃. This also shows in general the importance of the 1α-OH group for the potency of the VDR agonist.

Recently, the crystal structures for another group of natural ligands have been solved. From the identification of secondary bile acids as VDR agonist (Makishima et al., 2002) the interesting question remained how these compounds bind to VDR. From the structural data it is evident that the litocholic acid (LCA) and 3-keto litocholic acid (3kLCA) are located in the opposite orientation than 1,25D₃ (Figure 6). The 24-carboxyl group faces the β-turns of VDR, the β-region of the steroid backbone the helix 6–7/11 region and the A-ring is in the direction of helix 12. The S274 (hVDR S278) and Y143 hydrogen bonds are conserved in all structures (Figure 6). The difference for this part of the ligand-binding comes from the water mediated contacts with both R270 (hVDR R274) and S233 (hVDR S237). These contacts seem to be weaker compare to 1,25D₃ but not so weak as in case of 25D₃ and 24,25D₃, where in fact they are missing due to lack of 1α-OH group. The opposite part of the bile acids shows also weaker hydrogen bonding network than for the previously discussed VD metabolites including 1,25D₃. The structural comparison between the two bile acids shows a less complex, more straightforward binding mode for 3kLCA with only one water molecule involved. Whereas, for LCA there are no direct contacts with H301 (hVDR H305) and H393 (hVDR H397). All these contacts are mediated thorough waters. This also may explain the lower VDR activation potential of 3kLCA compared to LCA.

I would seems that the evolutionary design of VDR reached by recognizing 1,25D₃ its perfection. All the anchoring points show effective hydrogen bonding and by looking at the limited data for natural ligands any deviation from 1,25D₃ will result in the use of alternative bridging contacts such water molecules. This shows a rather limited adaptability of VDR in effectively recognizing its natural ligands, yet larger than we envisioned a decade ago. One must keep in mind that the activation of VDR with one or another metabolite will largely depend on the local cellular concentration of this compounds leading for instance for favored activation via bile acids instead of classical 1,25D₃ binding.

VDR belongs to the class of zinc finger transcriptional factors with DBD that consists of a highly conserved 66 aa residue core (Khorasanizadeh and Rastinejad, 2001) and an adjacent C-terminal extension. The conserved core has two zinc fingers where one contains fours cystein residues per atom of zinc (Figure 7). This feature allows VDR to effectively recognize and bind hormone response elements (REs) - termed VDREs. VDREs are typically made up of two hexameric half-sites whose consensus sequence is 5′-RGKTCA-3′, where R = A or G and K = G or T. The half-sites may be arranged in various orientation most commonly forming a direct repeat with three neutral base pairs separating the half-sites (DR3) (Umesono et al., 1991). The unliganded VDR can occupy its REs also as a homodimer (Carlberg et al., 1993). Upon binding of ligand, VDR interacts with RXR and forms a heterodimeric complex that binds to VDREs with 5′-prime bound RXR.

The data for structural view on VDR-DNA recognition is scarce. To date we have only four structures published where three represent VDR homodimers on direct repeat 3 (DR3) from mouse osteopontin (mSPP) (CACGGTTCACGAGGTTCA), rat osteocalcin promoter (rOC)(CACGGGTGAATGAGGACA) and a canonical DR3 element (cDR3) (CACAGGTCACGAAGGTCA) (Table 1) (Shaffer and Gewirth, 2002). The last structure represents the DBDs of the heterodimeric RXR-VDR on canonical DR3 element (cDR3) (TTAGGTCACGAAGGTCAA) (Table 1) (Shaffer and Gewirth, 2004). The 66 to 70 aa of the DBD are structurally coordinated by two zinc atoms that create a structure (Figure 7), in which one short α-helix is interacting directly with the major groove of the DNA (Härd et al., 1990; Lee et al., 1993). The VDR homodimers show asymmetric head to tail arrangements. The experimentally determined range of affinities of DR3s used for crystallographic studies are mSPP > cDR3 > rOC with mSPP supporting both VDR homodimer and RXR-VDR heterodimer binding, cDR3 requiring 10x higher VDR homodimer protein levels and weak heterodimer binding, whereas rOC is unable to bind VDR homodimers and has a very weak heterodimer binding ability (Freedman and Towers, 1991; Nishikawa et al., 1993; Toell et al., 2000). The change at third position, a purine to pyrimidine, of the consensus half sites AGGTCA allows the additional water mediated hydrogen contact of E42 with the DNA, which increase the stability of the mSPP-VDR complex. In case of rOC the reason for diminished VDR homodimer binding is because there is a G at position five of the upstream half-site GGGTGA, where supposedly RXR is bound in case of heterodimer binding. In high affinity half-sites in this location there is a C in the first strand but a complementary G in the second DNA strand. The interaction, which involves hydrogen bonds is between the R50 of VDR and the G from the second complementary strand. In rOC upstream half-site instead of G there is a C in the second strand, which is not a hydrogen bond acceptor thus R50 cannot interact with it. There is also some agreement between the strength of the homodimer binding and a sum of all existing DNA-VDR contacts within 3.5 Å calculated with the ncont program of the CCP4 suite (Winn et al., 2011). The results show 88, 85, and 83 contacts for rOC, cDR3, and mSPP, respectively. Interesting is also the interacting surface ratio between the two VDR homodimer molecules in percentage (5′upstream: 3′downstream) 38.52:61.48% for rOC, 46.21: 53.8% for the cDR3 and a reversed ratio of 57.3:42.7% for mSPP. This would suggest that for strong VDR homodimer binding there is a more contribution from the upstream half-site, but a clear conclusion cannot be reached based on this limited data set. In the RXR-VDR crystal structure there are two asymmetric units where the full complex shows an orientation RXR to VDR for unit 2 and a reversed orientation VDR to RXR for unit 1 with RXR bound on the downstream part of the VDRE. This may be due to the stabilization contacts between the adjacent VDR molecules where the hinge of one VDR molecule is stabilized with the DBD of the second VDR molecule. The general organization of the unit 2 is depicted also on the Figure 7 bottom. The two zinc atoms (light blue color) with the respective cysteins are visualized (the second heterodimer unit is missing from the representation). The overall number of contacts is only 65, which is much lower than in case of VDR homodimers discussed earlier. This indicates that the binding of the RXR-DBD heterodimer to cDR3s is not optimal and for an effective binding a certain point of VDRE degeneration is needed. Moreover, the contribution of the monomers to the binding is rather interesting. The ratio between interacting monomeric surfaces is 56.22:44.26% (RXR-VDR) suggesting a higher contribution of RXR to the binding Figure 7 top (see also the H/D exchange experiments discussed below). This is most likely due to the reverse orientation of the RXR-VDR on the cDR3. The surfaces of the interacting atoms are visualized in different color DNA (red), RXR (blue) and VDR (green). It is to be noted that compared to DNA-protein interaction there are hardly any interactions between the protein monomers. None for RXR-VDR, for VDR homodimers there are two for mSPP and rOC, and five on cDR3. This is quite in agreement with full length receptor studies which suggest that most of the heterodimerisation is contributed from hinges and LBDs.

The main aim of this review was to collect and discuss structural data that is related to vitamin D signaling. The structural data for various isolated domains (LBD and DBD) show their organization on atomic level. This data is sufficient for understanding the particular ligand- and DNA-protein interactions but fails to provide spatial information on the mutual orientation of the domains of RXR-VDR on its natural promoters. It also fails to highlight inter-domain communication after DNA-, ligand- or cofactor-binding. However, they are irreplaceable tools for structure-based drug design and mechanistic view of the VDR action. The dissemination of the information derived from structural data and in silico models may help to understand how VDR works in its natural settings and provides a space for the intervention in various diseases. The recent SAXS, Cryo-EM and H/D exchange studies with full length RXR-VDR complexes show more complex views on VDR function and provide first tools for the integration of structural information with genomic, epi-genomic, transcriptional and functional data.

There are numerous scientific questions connected to VDR that may be interesting to answer on structural level. The list is not complete but here are some of them: