Insights into the binding mode and mechanism of action of some atypical retinoids as ligands of the small heterodimer partner (SHP)

Homology modelling

Sequence similarity searches were used to identify a pool of nuclear receptors sharing homology relationships with the ligand binding domain of SHP and endowed with different conformational features. The selected nuclear receptors are reported in Table 1. Not surprisingly, the best hit of the similarity search resulted in the recently disclosed ligand binding domain (LBD) of DAX-1 (pdb code: 3F5C, E-value = 2.8e-23). Alike SHP, DAX-1 belongs to the NR0B group of the nuclear receptor superfamily. During embryogenesis, DAX-1 is involved in cell differentiation in testes and adrenal tissues and plays a key role in maintaining the pluripotent state of embryonic stem cells [34–36] The conformation of helix 12 (H12) of DAX-1 is in an antagonistic state, according to the receptor that acts as global repressor of many nuclear receptors [33]. Noteworthy, helix 10 (H10) is collapsed into the canonical binding site slightly lowering the druggability of this binding pocket as evidenced by the site score (site score = 0.98). This feature should apparently prevent the binding of ligands to DAX-1.

The chicken ovalbumin upstream promoter transcription factor-2 (COUP-TF2, pdb code: 3CJW, E-value = 1.0e-16) is the second receptor in the rank of E-values. It is one of the three functionally non redundant receptors belonging to the CUP family of orphan nuclear receptors. These nuclear receptors are differently involved in fetal development and in metabolic homeostasis of adult organisms [37]. From a structural point of view, the ligand binding domain of COUP-TF2 shares similar features with DAX-1, namely the antagonistic conformation of H12 and the collapse of H10 into the canonical binding site [38].

The next receptor in the rank is the Ultraspiracle protein (USP, pdb code: 1Z5X; E-value = 5.3e-8; pdb code: 1G2N; E-value = 3.1e-8). Ultraspiracle protein (USP) is the ortholog of the vertebrate retinoid X receptor (RXR) in insects and mediates the heterodimerization of several nuclear receptors like RXR does in vertebrates [39]. It is worth pointing out that two different conformational states of this receptor have been solved by X-ray crystallography [39, 40].

Although in either USP structures H12 adopts the antagonistic conformation and H10 collapses into the canonical ligand binding site (similarly to DAX-1 and COUP-TF2), there are two additional features in the structure of 1G2N. Firstly, the loop connecting helix H1 to helix H3 (L1-3) folds around helices H3, H11 and H12 [41]. This particular fold locks the H12 in a resting state and prevents any of its conformational movements. Remarkably, the loop L1-3 is partly or not at all solved in the structures of DAX-1, COUP-II and USP-1Z5X, thus evidencing its high flexibility. Secondly, the crystal structure of USP-1G2N shows a druggable allosteric binding site filled by a phospholipids molecule (site score = 1.23). Very interestingly, this seems to be not an artifact of the crystallization procedure since the same allosteric binding site has been found as functional in other nuclear receptors such as the steroidogenic factor-1 (SF-1) and LHR-1 [42, 43]. In previous works, we already used the 1G2N structure of USP as template to model the structural features of SHP and its interaction with compounds 1 and 2 [22, 32].

Thus, while entries a–c may resemble an antagonistic conformation of SHP with a collapsed canonical binding site, entry d may be a suitable template to construct an antagonistic conformational model of SHP endowed with an allosteric binding site.

In the bottom part of the E-value ranked proteins, we selected the canonical agonist and antagonist conformations of the retinoid X receptor-α (RXRα, pdb code: 1MV9; E-value = 7.0e-7; pdb code: 1DKF; E-value = 6.2e-7). This selection was motivated by the fact that the known small molecule modulators of SHP belong to the chemical class of synthetic retinoids and some members of this class may bind to the canonical binding site of nuclear retinoic acid receptors (RARs) or RXRs as well. Accordingly, we deemed advisable to not exclude at this stage the hypothesis that the molecular mechanism of SHP agonists and antagonists could comply with the classic principles for modulation of nuclear receptors [44]. Thus, entries e and f were instrumental to build two conformational models of SHP with a druggable canonical binding site and H12 in the agonistic and antagonistic states, respectively.

A multiple alignment between the sequences of SHP and the above receptors was generated (Fig. 2) and used to construct four 3D conformational models of the receptor: (a) SHP_DAX-1 model (H12-antagonist, H10-collapsed) as representative of the conformational features observed in 3F5C (DAX-1), 3CJW (COUP-TF2) and 1Z5X (USP); (b) SHP_USP (H12-antagonist, allosteric site) as example of the conformational features observed in 1G2N (USP); (c) SHP_RXRα (H12-agonist, canonical site) as representative of the conformational features observed in 1MV9 (RXRα); (d) SHP_RXRα* (H12-antagonist, canonical site) as example of the conformational features observed in 1DKF (RXRα). The resulting 3D models of SHP were geometrically validated and their folding was checked using the Verify3D server (see supporting information).

Docking experiments

In this part of the study, we used the homology models of SHP (models a–d, Table 1) to perform docking experiments of agonist (3Cl-AHPC, 2), competitive antagonist (3A-AHPC, 3) and an additional atypical retinoid molecule (ST1926, 4) whose structure is closely related to 3Cl-AHPC (2), albeit no data is available about its binding to SHP [45, 46]. Concerning ST1926 (4), in particular, the idea was to assess whether it could also bind and modulate the receptor by adopting a molecular mechanism similar to 3Cl-AHPC (2). Docking experiments were carried out into both the canonical and allosteric binding sites of SHP. Briefly, the canonical binding site refers to the hydrophobic cleft that is almost conserved in the ligand binding domain (LBD) of all nuclear receptors and is linked to the stabilization of the activation function on helix 12 (H12). The allosteric pocket refers to the phospholipids binding site that has been unexpectedly discovered in some crystal structures of nuclear receptors (e.g. USP, LRH-1, SF-1). Noteworthy, a growing number of experimental evidences suggest a modulatory role of this site upon phospholipids binding at least in LRH-1 and SF-1 [42, 43, 47, 48]. The canonical and allosteric sites share a common volume, forming together an “L” shaped cavity (Fig. 3a). The “L” shaped cavity is wedged between the Loop108-138 and the helix H12, with positively charged residues being located at the bottom of each arm.

From docking experiments, two alternative binding poses for each compound in either the canonical or allosteric binding sites of models a–d of SHP were selected for further studies (Table 2). In all of these complexes, the adamantane moiety of compounds 2–4 occupies the wedge of the “L” shaped cavity, whereas the carboxylic group alternatively points to the positively charged residues at the bottom of each arm of the cavity (Fig. 3b).

Molecular dynamic simulations

Since it was not possible to uniquely identify the most energetic favored solutions on the basis of the scoring function of docking experiments (IFD score, Table 2), we performed 10 ns MD simulations of the complexes in full water solvated box. The following tasks, in particular, were the goals of this last part of the work: (a) to optimize the binding energies of compounds 2–4 taking into account any possible ligand induced fit effect of the receptor and, thereby, aiding the identification of their most energetic favored solutions into a specific binding cleft and conformation of SHP; (b) to design mutagenesis experiments; (c) to understand the molecular mechanisms responsible of the agonistic and antagonistic modulation of the receptor by the atypical retinoid molecules.

Concerning the first task, from the inspection of the average and standard deviations of the interaction energies of compounds 2–4 with the receptor (Table 3), it is found that they show a stable and energetic favored solution when docked into the allosteric site of the conformational model of SHP_USP. In addition, the agonist (3Cl-AHPC, 2) displays an alternative favored binding pose into the canonical cleft of the conformational model SHP_DAX-1, as evidenced by the slight difference of interaction energy from the selected solution of SHP_USP (ΔE = 5.6 kcal/mol). Figures 4 and 5 show the interactions of compounds 2–4 into the conformational models of SHP according to the solutions selected from MD simulations.

On the basis of these results, a number of residues was selected for mutagenesis experiments including Arg34, Trp92, Phe96, Gln134, Arg138, Trp148 and Arg238 (Fig. 2). These experiments are detailed in the next section.

The final task of molecular dynamic simulations was to shed light on the molecular mechanisms responsible of the agonistic and antagonistic modulation of SHP. To this end, we investigated the effect of the agonist (2) and antagonist (3) on the stability of the three nuclear receptor boxes and helix H12 in the conformational model SHP_USP. Indeed, being in agreement with the competitive profile of the agonist (2) and antagonist (3) as well as mutagenesis experiments (see below), it is only the conformational model SHP_USP that is able to host compounds 2–3 in a common binding site, namely the allosteric site.

Table 4 reports the averages and the standard deviations of RMSD values calculated on the backbone atoms of the receptor and, more specifically, on the backbone atoms of the selected structural motifs. From the inspection of the table, it is found that 3Cl-AHPC (2) and (3A-AHPC, 3) induce a similar pattern of stabilization (RMSD<3.0) and destabilization (RMSD>3.0) of the NR boxes, with the NR boxes 1, 3 being stabilized and NR box-2 being destabilized. Remarkably, compounds 2 and 3 show a different stabilization degree of helix H12 that makes sense in view of their pharmacological profile. Thus, while the agonist (2) destabilizes the antagonistic conformation of H12, the antagonist (3) stabilizes it in SHP.

ST1926 (4), a compound not yet tested in SHP functional assays, shows a behavior akin to the agonist 3Cl-AHPC (2), with the destabilization of the antagonistic conformation of H12. However, conversely to compounds 2 and 3, it shows a higher stabilization of the NR boxes 1, 3 and a lower destabilization of the NR box 2.

Mutational studies

To verify the presence of a functional binding site in SHP, both the allosteric and canonical binding sites hypothesized were studied using preliminary mutagenesis experiments. One or more amino acids which were thought to be significantly interacting with the ligand in these hypothesized binding sites were mutated. All the mutants were studied for their functional changes using a reporter assay in which SHP was used to repress the HNF4α induced luciferase expression driven by ApoCIII promoter (Table 5; Fig. 6).

In case of the allosteric binding site (hypothesis A), a double SHP mutant with Arg238 and Arg138 replaced by acidic Asp residue was constructed in order to invert the electrostatic profile of the receptor at the entrance of the allosteric site, thereby making repulsive the interaction with the carboxylic group of a putative ligand. Two additional single mutants of SHP, with Arg238 and Arg138 replaced by alanine residue, were also made to understand the specific contribution of the single residue to SHP function. Finally, SHP mutants Trp92Ala, Phe96Ala and Trp148Ala were studied to test the involvement of these aromatic amino acids in the binding of a putative SHP ligand. Two mutants, SHP(Arg238Asp + Arg138Asp) and SHP(Phe96Ala) completely lost their function in repressing the HNF4a induced luciferase activity on pApoCIII-Luc while the other mutants did not show any significant changes in their function in comparison to the wtSHP protein.

On the other hand, the canonical binding site (hypothesis B) was verified by constructing two other SHP mutants: SHP(Arg34Ala) and a swap double mutant SHP(Arg34Gln + Gln134Arg). None of these two mutants showed any difference in their repression function as compared to the wild type SHP protein.

Homology modelling

The sequence of human SHP (Entrez code: NP_068804) was submitted to a BLAST [56] search versus the RCSB protein (PDB) database [57] using the WU-BLAST server [58]. The resulting top ranked proteins are shown in Table 1. SiteMap algorithm, as implemented in the Schrodinger software package [59], was used to locate binding sites into each of the resulting structures, assigning a site score (S) [60]. This score is calibrated so that a satisfactory binding site is endowed with S ≥ 1.0.

After this analysis, four structures (pdb codes: 3F5C; 1G2N; 1MV9; 1DKF) were identified as suitable templates to build different homology models of SHP on the basis of conformational criteria and the site score. Despite its slightly low site score (=0.98), DAX-1 (pdb code: 3F5C) was included in the template structures because of its closest sequence similarity relationship to SHP. Multiple sequence alignments and the construction of SHP homology models were carried out using Prime v. 2.1. The multiple alignments were carefully checked to avoid gap insertion where conserved secondary structure motifs were present. At this regard, it should be mentioned that the large loop insertion of SHP (residues 110–138) was modeled as previously reported [22]. For each of the generated model, the side chains of residues were refined identifying the low energy conformations using a rotamer library as implemented in Prime v. 2.1. Finally, the homology models were submitted to the protein preparation refinement protocol as implemented in the Schrodinger package, consisting in the optimization of the hydrogen bond network and a restrained minimization (maximum RMSD allowed 0.30 Å) using the OPLS-2001 force field. The statistical and geometrical validations of the final models were carried using Molprobity and Verify 3D servers [61–63].

Docking studies

Ligands 2–4 were prepared with the ligand preparation tool implemented in the Schrödinger Suite 2009. In particular, we considered for each compound different tautomers and ionization states at the physiological pH of 7.0. Docking experiments into the four conformational models of SHP were carried out using Glide 5.5 and the extra precision (XP) protocol. During these experiments, the induced fit protocol of Glide was employed to take into account any possible ligand induced conformational rearrangement of the binding site. Two solutions for each ligand, namely the top scored one and its most diverse solution among the first ten ranked solutions, were stored for the analysis of the binding modes. In particular, the calculation of the RMSD of heavy atoms was used to identify the second solution of docking experiments.

Molecular dynamics

The complexes resulting from docking experiments (Table 2) were solvated in an orthorhombic box using TIP3P water molecules [64]. Sodium and chlorine atoms were also placed to counter-balance the net charges of the receptor. All the molecular dynamic simulations (MDs) were performed using Desmond 2.2 software [65, 66] and the force field OPLS-2005 as implemented in the Schrödinger Suite 2009. In particular, the simulation protocol was constituted of the Desmond default relaxation steps, 2 ns equilibration phase and a final production phase of 8 ns. In particular, Desmond default relaxation steps comprised: (a) the energetic minimization of the system over a maximum of 2000 steps, with the first 10 steps using the steepest descendent and the convergence criterion being set to 50 kcal/mol Å⁻¹, harmonic restrains placed on the solute atoms (force constant = 50.0 kcal/mol); (b) an analogous energetic refinement of the system without restrains; (c) MDs of 12 ps at temperature 10°K with harmonic restrains on the solute heavy atoms (force constant = 50.0 kcal/mol), NVT Berendesen ensemble; (d) MDs of 12 ps at temperature 10°K retaining the harmonic restrains, NPT Berendesen ensemble; (e) MDs of 24 ps heating phase to temperature 300°K retaining the harmonic restrains, NPT Berendesen ensemble; (f) MDs of 24 ps at temperature 300°K without harmonic restrains, NPT Berendesen ensemble. Both equilibration and productions phases were run using the NPT Berendesen ensemble at temperature 300°K. During all the simulations, a time step of 2 fs was used. Other not specified parameters were set to the default values. The atomic coordinates of each system were saved every 4.8 ps. The occupancy of ligand-receptor hydrogen bond interactions was calculated in the production phase of the simulations (last 8 ns). Schematic interactions shown in Fig. 5 were drawn using MOE [67].

Mutagenesis experiments

Mutations were introduced in Human SHP gene by PCR based site directed mutagenesis using quick change site directed mutagenesis kit (Stratagene, CA, USA). Two complementary oligonucleotide sets were designed as PCR primers to introduce each mutation whose sequences were listed in the Table s2. Reaction mixtures were set up according to the manufacturer’s instructions using 50 ng of template DNA (hSHP cloned in pcDNA3) and 125 ng of primers. Cyclic parameters: denaturing at 95 °C for 30 s, followed by 18 cycles at 95 °C for 30 s, 55 °C for 1 min and 68 °C for 7 min. The reaction mixture was digested with Dpn1 for 1 h at 37 °C and finally transformed into XL1 blue super competent cells. Sequences of all constructs were confirmed using DNA sequencing.

Luciferase reporter assay

Cos7 cells were maintained in Dulbecco’s minimal essential medium in the presence of 10% FBS. pApoCIII-Luc and pcDNA3-HNF4α were kindly provided by Dr. Frances Sladek while human SHP was cloned into pcDNA3. Different hSHP mutants were constructed as described earlier. For luciferase assays, cells were plated in 24 well plates 1 day before transfection and transfection was carried out using lipofectamine 2000 (Promega, Madison, USA). Total DNA in each transfection was adjusted by adding appropriate amounts of pcDNA3 empty vector. Approximately 48 h after transfection, cells were harvested and luciferase activities were measured using luciferase assay systems (Promega) according to manufacturer’s instructions. The luciferase readings were normalized against β-galactosidase activity as an internal control. The transfection experiments were repeated thrice, each time carried out in triplicates.