Enantiomer-specific activities of an LRH-1 and SF-1 dual agonist

Chirality is an important consideration in drug development: it can influence recognition of the intended target, pharmacokinetics, and off-target effects. Here, we investigate how chirality affects the activity and mechanism of action of RJW100, a racemic agonist of the nuclear receptors liver receptor homolog-1 (LRH-1) and steroidogenic factor-1 (SF-1). LRH-1 and SF-1 modulators are highly sought as treatments for metabolic and neoplastic diseases, and RJW100 has one of the few scaffolds shown to activate them. However, enantiomer-specific effects on receptor activation are poorly understood. We show that the enantiomers have similar binding affinities, but RR-RJW100 stabilizes both receptors and is 46% more active than SS-RJW100 in LRH-1 luciferase reporter assays. We present an LRH-1 crystal structure that illuminates striking mechanistic differences: SS-RJW100 adopts multiple configurations in the pocket and fails to make an interaction critical for activation by RR-RJW100. In molecular dynamics simulations, SS-RJW100 attenuates intramolecular signalling important for coregulator recruitment, consistent with previous observations that it weakly recruits coregulators in vitro. These studies provide a rationale for pursuing enantiomerically pure RJW100 derivatives: they establish RR-RJW100 as the stronger LRH-1 agonist and identify a potential for optimizing the SS-RJW100 scaffold for antagonist design.

LRH-1 and SF-1 have very similar, lipophilic ligand binding pockets that have been challenging to target with synthetic molecules. The native ligands of these receptors are likely phospholipids (PL), as they have been crystallized with a variety of PL in the binding pockets. Unlike typical NR hormones, PL do not make polar interactions deep in the binding pocket, and they partially protrude into the solvent [22][23][24][25][26][27] . This unusual binding mode has made it difficult to predict types of synthetic molecules able to bind these receptors and modulate transcriptional activity. The compound RJW100 (Fig. 1a) is one of the few reported chemical scaffolds shown to activate these receptors and has been used by several groups to probe LRH-1 and SF-1 biology 16,[27][28][29][30][31][32][33] .
RJW100 is a racemic mixture of two enantiomers that contain hexahydropentalene cores, two bulky styrene substituents, a hydrophobic n-hexyl substituent, and a hydroxyl group capable of making polar interactions (Fig. 1a). Differences in stereochemistry are found at positions 1 (the hydroxyl group), 3a (the external styrene), and 6a (Fig. 1a). Due to the cis-fused structure of the hexahydropentalene scaffold, positions 3a and 6a adopt the same absolute stereochemistry for RJW100: one enantiomer has S stereochemistry at both 3a and 6a, and the other has R stereochemistry in these positions. For simplicity, the RJW100 enantiomers will be hereafter called RR-RJW100 and SS-RJW100 for the stereochemistry at positions 1 and 3a/6a, respectively. Because of the bicyclic nature and differences in the stereochemistry, the hexahydropentalene cores of the enantiomers fold in opposite directions, thus resulting in very different three-dimensional structures. HPLC traces for RR-RJW100 generated by enantioselective synthesis (black curve) and for the racemate (blue curve). Separation by HPLC resulted in enantiopure RR-RJW100 and SS-RJW100 (both 100% ee). (c) Enantioselective synthesis of RR-RJW100. The (R)-enantiomer of the alcohol 1 was obtained from the racemate by kinetic resolution using a crotylation reaction with Novozyme 435, an immobilized Candida antartica lipase as the catalyst, which is known to be selective for the (S)-enantiomer of alcohols with a similar structure 54 . The reaction was stopped at around 50% conversion. The recovered enantioenriched alcohol (R)-1 (95:5 enantiomer ratio) protected as the tert-butyldimethylsilyl ether and converted to RR-RJW100 (95:5 enantiomer ratio) using the published procedure for the racemate 33,34 . The enantio-enriched RR-RJW100 was used to identify the two HPLC peaks from the racemate and was not used for experiments with LRH-1/ SF-1. (d) CD traces of the RJW100 enantiomers used in these studies.
Binding affinities of RJW100 enantiomers. Previous coregulator recruitment studies indicate that both RJW100 enantiomers bind LRH-1 and SF-1 34 , but ligand binding affinities have not been measured. To determine binding affinities, we utilized a fluorescence polarization-based, equilibrium ligand binding assay in which a fluorescein-labelled probe is displaced by unlabelled competing ligands 39 . Differences in probe affinity for LRH-1 and SF-1 were calculated prior to the competition assays and are accounted for in the equation for fitting the competition data (details in the methods section and our previous publication) 39 . Both enantiomers outcompete the probe, indicating that they bind the receptors in the binding pockets (Fig. 2a,b). The RR-RJW100 enantiomer has an approximately threefold higher affinity for LRH-1 than SS-RJW100 (K i = 0.4 µM versus 1.2 µM, Fig. 2a). Comparison of log K i values calculated from three independent experiments show that the difference in affinity is statistically significant (p = 0.01 by two-tailed, paired t-test). However, magnitude of the affinity difference is fairly small and may not be biologically important. For SF-1, the binding affinity of RR-RJW100 trends higher than SS-RJW100, but the difference does not reach statistical significance (K i = 13 µM for RR-RJW100 versus 30 µM for SS-RJW100, p = 0.2, Fig. 2b). Therefore, both enantiomers directly bind LRH-1 and SF-1 and do not exhibit dramatic differences in affinity.
Differing effects of RJW100 enantiomers on receptor thermostability and transcriptional activity. Although binding affinities are not dramatically different, the RJW100 enantiomers have distinct effects on protein stability and function. We used differential scanning fluorimetry (DSF) to probe the thermostability of purified receptor-ligand complexes, quantified by the temperature at which the complexes are 50% unfolded (T m ). Proteins in solution are continuously sampling conformational states, and ligand-dependent changes in T m provide insights into the energetics of receptor conformations recognized or induced by differ-  www.nature.com/scientificreports/ ent ligands. Data are presented as change in T m relative to PL-bound receptors. As previously observed 35 , RR-RJW100 stabilizes LRH-1 relative to a PL ligand (Fig. 3a). The RR enantiomer is similarly stabilizing to SF-1, increasing the T m of the complex by around 4 °C (Fig. 3b). In contrast, SS-RJW100 does not significantly stabilize LRH-1 or SF-1 (Fig. 3a,b). Differences in T m induced by the enantiomers are unrelated to differences in binding affinities, since DSF experiments utilized saturating ligand concentrations (20-fold molar excess). Mutagenesis experiments support the idea that these energetic differences arise from different ligand binding modes. One of the primary mechanisms used by RJW100 to bind and activate LRH-1 is a water-mediated hydrogen bond made by the ligand hydroxyl group with residue T352 35 , and we tested the effects of disrupting this interaction in DSF experiments. LRH-1 stabilization by RR-RJW100 is completely dependent on residue T352, but the T m of the LRH-1-SS complex is unaffected by a T352V mutation (Fig. 3c). This indicates a differential reliance on the T352 interaction for binding LRH-1. The T m differences induced by the enantiomers suggest that they recognize different conformational states, which could impact receptor activation. Indeed, while both enantiomers similarly activate SF-1, RR-RJW100 is a much stronger activator of LRH-1 than SS-RJW100 in luciferase reporter assays (Fig. 4a,b). LRH-1 activation by SS-RJW100 is attenuated at all doses tested and is reduced by as much as 46% compared to RR-RJW100 (Fig. 4b).
The weaker activity is associated with an inability to form a productive interaction with residue T352. Consistent with DSF studies (Fig. 3c) and our previous work 35 , RR-RJW100 is unable to activate a T352V LRH-1 mutant. However, this mutation does not affect activation by SS-RJW100 (Fig. 4c). SS-RJW100 activity also does not depend on an interaction with residue H390 like the closely-related analogue, GSK8470 40 . Our previous work demonstrated that GSK8470 cannot activate H390A-LRH-1 35 , but we find here that SS-RJW100-driven LRH-1 Results are presented as the difference in T m relative to protein bound to phospholipid. Significance of T m differences was determined by two-tailed, paired Student's t-tests from parallel experiments. *, p < 0.05. (c) DSF with wild-type (WT) or mutant (T352V) LRH-1 demonstrates the differential role of residue T352 in LRH-1 stabilization by RJW100 enantiomers. Each bar is the mean ± SEM for three experiments conducted in triplicate. Statistical significance was determined by twoway ANOVA followed by Sidak's multiple comparisons test. *p < 0.05. . Luciferase reporter assays measuring transcriptional activity of full-length SF-1 (a) or LRH-1 (b) in Hela cells. Each point represents the mean fold change versus DMSO-treated cells ± SEM for three experiments each conducted in triplicate. Significance of differences in activity was determined by two-way ANOVA, followed by Sidak's multiple comparisons test. *p < 0.05 versus SS-RJW100. (c) Mechanistic differences LRH-1 activation by SS-RJW100 compared to related molecules. Wildtype (WT) or mutant LRH-1 was overexpressed in Hela cells prior to treatment with 30 µM RR-or SS-RJW100. Each bar indicates mean fold change (± SEM) in LRH-1 activity after agonist treatment for 24 h relative to DMSO-treated cells. The experiment was conducted three times in triplicate. Fold change is calculated relative to DMSO-treated cells overexpressing WT LRH-1. Significance was determined by two-way ANOVA followed by Sidak's multiple comparisons test. *p < 0.05 versus DMSO-treated cells for each group of bars (WT, mutant, or pCI empty vector. The A349F mutation introduces a bulky amino acid that blocks the binding pocket and was used as a negative control. www.nature.com/scientificreports/ activity is unaffected by this mutation (Fig. 4c). Together, these results demonstrate that the chirality of RJW100 affects its ability to bind and activate LRH-1 but has a lesser impact on SF-1. Mutagenesis studies demonstrate that SS-RJW100 likely utilizes different mechanisms to activate the receptor than other agonists with the same core scaffold.
Crystal Structure of SS-RJW100 bound to LRH-1. To understand how RR-and SS-RJW100 differentially affect LRH-1 activity, we determined the 1.7 Å X-ray crystal structure of SS-RJW100 bound to the LRH-1 ligand binding domain (LBD) ( Table 1, Fig. 5a). The overall protein conformation is very similar to the published RR-RJW100-LRH-1 structure (PBD ID 5L11) 35 , with an average root mean square deviation (RMSD) of 0.1 Å. Strong electron density in the binding pocket indicates that the ligand is bound (Fig. 5b); however, modelling its precise position was challenging. Unlike in the RR-RJW100-LRH-1 structure, there is no clear indention in the density that indicates the puckering of the bicyclic core (see Supplementary Fig. S1 online). In addition, no single conformation of the ligand is sufficient to account for the electron density in the binding pocket (Fig. 5b). Modelling the ligand in two conformations with partial occupancies provides a better fit to the density and improves ligand B factors (values that reflect vibrational motion). With either one or two molecules in the model, B factors are higher for the ligand than for protein atoms in the structure, suggesting that the ligand is relatively mobile. However, the ratio of ligand/ protein B factors is 2.1 ± 0.4 with one molecule versus 1.63 ± 0.07 with two molecules modelled in the pocket, a significant improvement (p = 6 × 10 -12 by two-tailed, unpaired Student's t-test). Refinement of the structure indicates that the two SS-RJW100 conformers are present with nearly equal occupancies (52% occupancy for conformer 1 and 48% occupancy for conformer 2). The binding modes of the two SS-RJW100 molecules are very different than observed for RR-RJW100 and for the related agonist GSK8470 (seen through superposition of the structure with PDB 5L11 and PDB 3PLZ, respectively) (Fig. 5c). The two SS-RJW100 molecules are also positioned very differently from each other: the hydroxyl groups point in opposite directions (Fig. 5c). The ligand makes several hydrophobic interactions (see Supplementary Fig. S1 online), but neither orientation places the hydroxyl group near the water network coordinating residue T352 like RR-RJW100 (Fig. 5d). This explains why T352V mutation does not affect LRH-1 activation by SS-RJW100 (Fig. 4c). Instead, the SS-RJW100 hydroxyl group is within hydrogen bonding distance of LRH-1 residue M345 (conformation 1, Fig. 5d) or the backbone carbonyl of S383 (conformation 2, see Supplementary Fig. S1 online). However, the complete absence of density surrounding the hydroxyl groups for both conformers (Fig. 5b) suggests that these interactions are unstable. www.nature.com/scientificreports/ In contrast to the hydroxyl group, the ligand density is very strong surrounding the phenyl rings from the external and internal styrene moieties (positions 3a and 4, respectively, Fig. 1a). Interestingly, the phenyl rings of the two conformers overlap despite the differences in the overall orientations of the molecules and the different sizes of the groups at positions 3a and 4 (Fig. 5b). They also overlap precisely with the phenyl rings from RR-RJW100 (Fig. 5c). This observation implies an important role for the styrene moieties in orienting the agonist in the pocket that is maintained even when it results in a nonideal positioning of the ligand hydroxyl group.
The "frustrated" SS-RJW100 ligand fails to form stable interactions in the LRH-1 binding pocket. Because of the difficulty placing SS-RJW100 in the structure, we sought to improve our model using Ensemble Refinement, a tool in the Phenix software suite 41 . This tool refines crystallographic data using a combination of molecular dynamics simulations and translation-libration-screw (TLS) to model local vibrational disorder and global disorder, respectively. This generates an ensemble of models that fits the crystallographic www.nature.com/scientificreports/ data better than a single model 41 . Superposition of the 35 models in the ensemble shows more motion in solventexposed regions, particularly near the bottom of the molecule, while the internal part of the structure is more stable (Fig. 6a). This is not surprising and is very similar to an LRH-1-RR-RJW100 model generated through ensemble refinement (Fig. 6b). However, there is a dramatic difference with the ligands: SS-RJW100 adopts many very different conformations within the binding pocket in the composite model, while RR-RJW100 exhibits a single preferred conformation (Fig. 6a,b, inset). As a result, there is a marked difference in the stability of the hydrogen bond made by the hydroxyl group of each enantiomer (with LRH-1 residue M345 for SS-RJW100 or with the water molecule coordinating residue T352 for RR-RJW100). SS-RJW100 makes an OH-mediated hydrogen bond in only five out of the 35 models (14%), while RR-RJW100 maintains the hydrogen bond in 95% of the models (p < 0.0001, by two-tailed, unpaired Student's t-test Fig. 6c, left). Interestingly, the interaction made . The input models for ensemble refinement were PDB 6VC2 (from this publication) with one conformation of SS-RJW100 modelled in the pocket and PDB 5L11, respectively 35 . The ensembles are represented as cartoon "putty, " in which the size of the cartoon is proportional to residue B factors. Insets depict close-up views of the ligand conformations in each model of the ensemble. (c) Left, Stability of hydrogen bonds made by the RJW100 hydroxyl groups was determined by measuring the distances between the hydroxyl groups and the closest binding interaction in each model of the ensembles. For SS-RJW100, we measured the distance between OH and M345 (sulphur atom), and for RR-RJW100 we measured the distance between the OH and the water atom coordinating residue T352. Grey shaded area indicates distances in hydrogen-bonding range. Right, Stability of π-π stacking with residue H390 was measured in each frame. Grey shaded area indicates distances in bonding range. (d) The intact water network contains four molecules, but this network is disrupted in the SS-RJW100 ensemble. The pie chart summarizes the number of water molecules in each model in the ensemble. www.nature.com/scientificreports/ by the SS-RJW100 styrene with LRH-1 residue H390 is much more stable: edge-to-face π-π stacking occurs in 88% of the models (Fig. 6c, right). However, this is less stable than the same interaction by RR-RJW100, which persists in 100% of the models (Fig. 6d, p = 0.009 by two-tailed, unpaired Student's t-test). Finally, SS-RJW100 has a destabilizing effect on the water network that coordinates a small group of polar residues in the LRH-1 binding pocket (including T352). This network contains four water molecules, and destabilization of the network is associated with lower activity of LRH-1 agonists 35 . In the LRH-SS-RJW100 ensemble, the water network is intact for only 35% of the models versus 79% of the models when the RR enantiomer is bound (Fig. 6d).
To explore ligand dynamics more rigorously, we conducted 1 microsecond molecular dynamics simulations (MDS) with the enantiomer-bound LRH-1 structures. All trajectories had stable RMSDs of less than 1.9 Å (see Supplementary Fig. S2 online). Over the course of the simulation, hydrogen bonding by the RR-RJW100 hydroxyl group with the water molecule coordinating T352 is maintained for 58.2% of the time. Hydrogen bonding of SS-RJW100 with either M345 (with the ligand starting in conformation 1) or S383 (starting in conformation 2) is less stable, persisting for 16.2% and 23.7% of the simulations, respectively. We also measured the stability of edge-to-face π-π stacking with H390, defined as having an angle between the ring planes between 60 and 120° and distance within 5.5 Å between rings 35,42 . The interaction is maintained for 62.0% of the simulation with RR-RJW100, 56.1% of the time for SS-RJW100 (conformer 1), and 48.7% of the time for SS-RJW100 (conformer 2). These results are consistent with the analyses from Ensemble Refinement. Therefore, the reduced activity of SS-RJW100 is associated with an inability to form stable, directional interactions in the binding pocket.

SS-RJW100 diminishes LRH-1 allosteric activation networks. Activation of LRH-1 by ligands
involves an allosteric signalling network between the ligand binding pocket and the activation function surface (AFS), the site of coregulator binding. A key part of this allosteric network is the helix 6/ β-sheet surface (AF-B) near the ligand binding pocket (Fig. 7a). The AF-B is utilized by both synthetic and phospholipid agonists to modulate coregulator associations via allosteric communication with the AFS in a mechanism involving correlated motion between these two distant sites 22,24,36 . To understand how two ligands with similar binding affinities differentially affect LRH-1 transcriptional activity, we investigated whether chirality affects receptor conformation and allostery in MDS. First, we examined ligand-dependent differences in receptor conformation that might impact allosteric signalling. A comparison of root mean square fluctuation (RMSF) of Cα atoms in each complex over the course of the simulations shows generally higher fluctuations in the presence of either SS-RJW100 conformer versus RR-RJW100 (Fig. 7B). This is consistent with the lower thermostability of the LRH-1-SS-RJW100 complex described in Fig. 3. Notably, the AF-B is the most prominent region destabilized by both SS-RJW100 conformers (Fig. 7b). Since the AF-B to AFS communication axis involves correlated motion between these two sites, the marked difference in RMSF in AF-B suggests the potential that this communication is altered when SS-RJW100 is bound.
To quantify the allosteric communication between the AF-B and AFS in the LRH-1-RJW100 complexes, we constructed dynamical networks and conducted suboptimal path analysis. Communication between distant sites in a protein occurs through correlated motion of residues along paths between them. These paths of communication can follow thousands of possible routes, but the paths containing the strongest correlated motion (the optimal path and a subset of suboptimal paths) convey the most information. In suboptimal path analysis, each Cα is considered a "node, " and the nodes are connected to each other by "edges. " The length of an edge is inversely proportional to the correlated motion of the two nodes it connects, such that shorter edges indicate more correlation 24,43 . The sum of edge lengths along a path therefore indicates the strength of the path. A histogram comparing the lengths of the 1000 shortest suboptimal paths between the AF-B and AFS for the three complexes reveals that stronger communication occurs when RR-RJW100 is bound compared to either conformer of SS-RJW100 (Fig. 7c). This evidence of stronger communication to the site of coregulator binding is consistent with the higher E max of Tif2 binding to LRH-1 in the presence of RR-RJW100 previously reported 34 . Reduced ligand-driven communication to the AFS provides an explanation for the lower levels of transcriptional activity induced by the SS enantiomer (Fig. 4c) and its reduced ability to recruit the Tif2 coactivator 34 .

Potential of the SS-RJW100 binding pose for antagonist design. Design of LRH-1 antagonists has
been particularly challenging, and we sought to gain insights from the SS-RJW100 binding mode that could aid future antagonist design. Nuclear receptor antagonists often disrupt the (AFS) through physical displacement of the activation function helix (AF-H), thereby interfering with coregulator recruitment 44 . A classic example of this is 4-hydroxytamoxifen (4-OHT), an antagonist of the oestrogen receptor and oestrogen related receptors (Fig. 8a,b). The 4-OHT mechanism of action is illustrated in Fig. 8a, which shows the superposition of two structures of oestrogen related receptor gamma (ERRγ) in the active state or bound to 4-OHT (from PBD 2GP7 and 2GPU, respectively) 45 . The clash of 4-OHT with residue F450 of ERRγ causes displacement of the AF-H (Fig. 8a). To explore the potential for the SS-RJW100 scaffold as an antagonist, we examined its proximity to the LRH-1 AF-H. Both dominant binding modes of SS-RJW100 orient it such that portions of the molecule are within 6 Å of residue L532 on the LRH-1 AF-H, which is in the analogous position to ERRγ F450. Conformer 2 is slightly closer, with position 2 of the molecule 5.5 Å from L532 (Fig. 8a). Therefore, the addition of bulk at position 2 could result in AF-H displacement. To illustrate this point, we docked a previously-described RJW100 analogue named "30-endo" 33 into the LRH-1-SS-RJW100 structure, using the enantiomer with the same ring pucker as SS-RJW100 (Fig. 8b,c). The acetoxy group at position 2 of 30-endo follows a trajectory toward the AF-H and is 2.9 Å away from L532 (Fig. 8c). Incorporating additional bulk into the group at position 2 could increase this clash and may result in an effective LRH-1 antagonist. www.nature.com/scientificreports/

Discussion
Chirality is important for drug design: one enantiomer is often responsible for on-target activity, while the other could have no effect, on-target beneficial effects, or deleterious effects 38 . In these studies, we uncovered striking functional differences between the RJW100 enantiomers. Compared to RR-RJW100, SS-RJW100 destabilizes LRH-1 and is a weaker activator in luciferase reporter assays (Figs. 2, 3, 4). The reduced activity is associated with unstable hydrogen bonding by the ligand and impaired allosteric signalling between the AF-B and the AFS (Figs. 5, 6, 7). These findings provide a rationale for pursuing enantiomerically pure versions of RJW100 and its derivatives, suggesting that improved LRH-1 activity could be achieved with R stereochemistry at the 3a position for molecules with this scaffold. Curiously, the differences between the enantiomers are much less pronounced for SF-1 than LRH-1. SS-RJW100 destabilizes SF-1 relative to RR-RJW100 in DSF assays, but it does not significantly affect binding affinity or transcriptional activity. The reason that SF-1 is less sensitive is unclear. SF-1 has a very similar binding pocket to LRH-1, with 53% sequence identity overall and 61% sequence identity in the LBD. Understanding the reason SF-1 is less sensitive than LRH-1 to RJW100 chirality will likely require a structure of SF-1 bound to a synthetic molecule, which is not currently available.
In addition to delineating mechanisms that cause differential activity of the RJW100 enantiomers, these studies provide insights that could guide future drug design. Development of LRH-1/ SF-1 modulators has been challenging, in part because natural ligands of these receptors (PL) interact with the receptors in an unusual manner compared to many other NR ligands [22][23][24][25][26][27] . The capacity for the deep part of the LRH-1/ SF-1 pockets to accommodate synthetic ligands is not fully understood nor apparent from observing PL-bound structures. Structural and mechanistic studies with small molecule agonists have established the importance of two deep pocket interactions in positioning ligands with the same 3a stereochemistry as RR-RJW100. The first is the hydrogen bond with a small polar patch in the deep binding pocket near residue T352, and the second is π-π stacking with www.nature.com/scientificreports/ residue H390 18,33,35,36 . With RR-RJW100, the configuration of the π-π stacking and hydrogen-bonding sites is complementary to the shape of the pocket, allowing an extremely stable binding mode. Understanding this has been vital for LRH-1 agonist development, enabling the discovery of potent and efficacious new lead agonists with a robust structure-activity relationship 18,37,39 . Likewise, the novel binding poses of SS-RJW100 present opportunities to target other regions of the binding pocket. This could be especially important for development of LRH-1 antagonists, which has lagged behind agonist discovery. There are no published crystal structures of LRH-1 bound to an antagonist, and the current leading antagonist binds with similar affinity to SS-RJW100 39,46 .
Notably, both conformations of SS-RJW100 in the structure prevent an interaction with T352, a critical driver of LRH-1 activation. This property is favourable for an antagonist, particularly in combination with other modifications that improve affinity or enhance receptor inactivation. While the instability of SS-RJW100 may be viewed as an impediment to rational design, several factors highlight the potential of this stereochemistry. The strong electron density surrounding the styrene rings, together with dynamical analyses, suggest these regions are stable relative to the rest of the molecule and are likely driving positioning of the ligand (Figs. 5, 6, 7). Stabilizing a desired conformation of the ligand could be possible through modifications to molecule that increase binding site compatibility. For example, one could envision adding bulk to the bicyclic core to stabilize conformation 2 and dispalce the AF-H, a common strategy in NR antagonist design (Fig. 8).
Together, these studies have illuminated key differences in the function of RJW100 enantiomers. Insights gained from this work further the understanding of how LRH-1 and SF-1 are regulated by synthetic ligands and will aid future endeavours in drug discovery for these receptors.
After stirring for 10 min at 0 °C the ice bath was removed, and the stirring continued for 30 min at RT before quenching with H 2 O (20 mL). The whole mixture was poured onto H 2 O (100 mL) and the products extracted with dichloromethane (2 × 75 mL). The combined organic phases were washed with H 2 O (2 × 100 mL) and brine (100 mL), dried over MgSO 4 , filtered and concentrated in vacuo to give the crude material as a yellow oil. Purification by flash chromatography column on Al 2 O 3 (basic, grade III) with hexane as the eluent provided (R)-(7-phenylhept-1-en-6-yn-3-yloxy)(tert-butyl)dimethylsilane ((R)-2) as a pale yellow oil (0.193 g, 67%) whose spectral data was consistent with those obtained for racemic product 33 . Chiral HPLC showed that the compound so formed corresponded to the slower running of the two enantiomers and retained the 95:5 enantiomeric ratio of the starting enyne (Fig. 1b). Following DLPC exchange, protein was re-purified by size exclusion chromatography into assay buffer (150 mM NaCl, 20 mM Tris-HCl (pH 7.4), and 5% glycerol). Mutant LRH-1 (T352V) was purified using the same conditions as wild-type LRH-1. SF-1 LBD (residues 218-461) in the pMSC7 vector was transformed in E. Coli strain BL21 (pLysS) and grown in LB at 37 °C to an OD 600 of 0.6. Expression was induced with 0.5 mM IPTG, followed by overnight growth at 18 °C. SF-1 was purified as described above for LRH-1 except for the addition of 5 mM tris(2-carboxyethyl)phosphine (TCEP) to the nickel affinity chromatography buffers and that it was not exchanged with DLPC prior to re-purification by size exclusion chromatography. The vector for His-tagged Tobacco Etch Virus (TEV) protease was a gift from John Tesmer (University of Texas at Austin). The pMSC7 vector was provided by John Sondek (University of North Carolina at Chapel Hill).
Mutagenesis. Point mutations to LRH-1 in the pCI vector or in the pMSC7 vector have been described previously 35 .
Fluorescence polarization (FP) binding assay. Assays utilized a fluorescein-labelled probe (6 N-FAM) that is displaced by unlabelled competitors 39 . Competition assays were conducted in opaque, black 384 assay plates using a combination of 10 nM 6 N-FAM with 5 nM LRH-1 ligand binding domain (LBD) or 30 nM 6 N-FAM with 25 nM SF-1 LBD (these concentrations were determined through optimization steps in our previous work) 39 . Competitor ligand concentrations ranged from 2 -11 to 2 -4 M. A constant amount of DMSO was present in each well (6.7% v/v). Assays were conducted three times in quadruplicate. FP was measured on a BioTek Neo plate reader. GraphPad Prism (v.8) was used to analyze the data using a one-site, fit K i curve, which fits the K i of the unlabelled competing ligand using the EC 50 of the competition curve, the concentration of the fluorescent probe, and the K d of the probe for each receptor, as follows: www.nature.com/scientificreports/ Significance of differences in K i values from parallel experiments was determined using two-tailed, paired t-tests, with p < 0.05 considered statistically significant. Differential scanning fluorimetry. Purified protein (0.2 mg/ ml, in an assay buffer of 20 mM Tris HCl pH 7.5, 150 mM NaCl, and 5% glycerol) was combined with RJW100 enantiomers (100 µM) overnight at 4 °C. SYPRO orange dye (Thermo Fisher) was added to the complexes the next day, at a final dilution of 1:1000. Complexes were heated at a rate of 0.5 °C/min on a StepOne Plus thermocycler (Applied Biosystems), with excitation/ emissions wavelengths of 488 nm/603 nm. Background was subtracted from reactions containing agonists and dye but no protein. Three replicates were performed in triplicate, using two separate protein preparations. T m was calculated using the Bolzman equation (GraphPad Prism, v.8). Significance of differences in T m shifts were determined using two-tailed, paired t-tests, with p < 0.05 being considered statistically significant.
Cell culture. Hela cells were purchased from and authenticated by Atlantic Type Culture Collection. Cells were cultured in phenol red-free MEMα medium (Thermo Fisher) supplemented with 10% charcoal-stripped foetal bovine serum. Cells were maintained under standard culture conditions. Reporter gene assays. Hela cells were plated in white-walled, clear-bottomed 96-well plates. Cells were seeded at densities of 7500 cells/ well, yielding a confluency of ~ 70-80%. The following day, cells were transfected with the FugeneHD transfection reagent (Promega) at ratios of 5 µl Fugene: 2 µg DNA. Cells in each well were transfected with 5 ng full-length LRH-1 in a pCI vector (wild-type or with point mutations as described in the figure legends) and two reporter plasmids: (1) containing a portion of the SHP promoter cloned upstream of firefly luciferase in the pGL3 basic vector (20 ng/well) and (2) a constitutive Renilla luciferase vector under the control of the CMV promoter, used as a control for transfection efficiency (1 ng/well). Cells transfected with the pCI empty vector and reporter plasmids were used as negative controls. Cells were treated with agonists 24 h after transfection at concentrations indicated in the figure legends. The DualGlo kit (Promega) was used to quantify luminescence signal 24 h after treatment on a BioTek Neo plate reader. SF-1 luciferase experiments were conducted identically, except that full-length SF-1 (in a pcDNA3.1 vector) was overexpressed instead of LRH-1 (with empty pcDNA3.1 as the negative control).
Crystallography. LRH-1 LBD was purified by nickel affinity chromatography as described above. The His-tag was cleaved from its fusion partner using tobacco etch virus (TEV) protease overnight with dialysis at 4 °C into imidazole-free buffer A. Cleaved protein was incubated with SS-RJW100 overnight at 4 °C (fivefold molar excess). The protein-ligand complex was re-purified by size exclusion into a buffer of 100 mM ammonium acetate (pH 7.5) 100 mM sodium chloride, 1 mM EDTA, 2 mM DTT, and 2 mM 3-[(3-cholamidopropyl) dimethylammonio]-1-propanesulfonate (CHAPS). Following size exclusion chromatography, additional compound was added to ensure saturation (fivefold molar excess) and combined with a peptide corresponding to the NR Box 3 of the Tif2 coactivator with the sequence H 3 N-KENALLRYLLDKDDT-CO 2 (RS Synthesis). This complex was concentrated to 6 mg/ml and used for crystallization. Crystals were generated by hanging drop vapor diffusion at 18 °C, using microseeding to aid nucleation. Seeds were derived from crystals of racemic RJW100 bound to LRH-1 (generated as previously described) 35 . Drops contained protein:crystallant:seeds at a 1:1:1 ratio. The crystallant was 0.05 M sodium acetate, pH 4.6 and 5-11% PEG4000.
Crystal data collection and structure refinement. Crystals were flash-frozen in liquid nitrogen, using a cryoprotectant of crystallant in 30% v/v glycerol. Data were collected remotely at the Argonne National Laboratory, South East Regional Collaborative Access Team, Beamline 22ID. Data were processed with HKL2000 47 and phased by molecular replacement in Phenix 48 , using PBD 5L11 as the search model. Ensemble refinement was performed using the Phenix program "Ensemble Refine" 41 . Phenix version 1.16-3549 was used with the Ensemble Refine package: https ://www.pheni x-onlin e.org/.

Molecular dynamics simulations.
All molecular dynamics simulations were performed for 1000 ns with Amber18 49 as previously described 50 . Three models of the LRH-1 ligand binding domain were used: (1) with RR-RJW100 (PDB 5L11), (2) with SS-RJW100 (conformer 1 from PBD 6VC2), or (3) with SS-RJW100 (conformer 2 from PDB 6VC2). Complexes were solvated in an octahedral box of TIP3P water with 10 Å buffer supplemented with Na + and Clions at 150 mM. Dynamic Network analyses were carried out by selecting Cα atoms in the protein as nodes for network construction. A pair of nodes was connected by edges if they maintained a distance of < 4.5 Å for at least 75% of the simulation. Carma and NetworkView plugin in Visual Molecular Dynamics (VMD) 51 were utilized for producing the dynamic network 52 . The edge distances were derived from pairwise correlations as a measure of communication within the network. Suboptimal paths between the AF-B and the AFS sites were identified using the Floyd-Warshall algorithm 53 and analysed by the subopt program in the VMD NetworkView plugin 51 . Cα atoms of residues 410 and 534 were used as source and sink nodes. Molecular modelling. An enantiomer of the "30-endo" ligand ((2S,3aR,6aS)-5-hexyl-6-phenyl-6a-(1phenylvinyl)-1,2,3,3a,4,6a-hexahydropentalen-2-yl acetate) 33 was modelled into the SS-RJW100-LRH-1 structure using Maestro v. 12.1.013 (Schrodinger) by editing conformation 2 of SS-RJW100 and performing energy minimization.