Structural studies unravel the active conformation of apo RORγt nuclear receptor and a common inverse agonism of two diverse classes of RORγt inhibitors

The nuclear receptor retinoid acid receptor-related orphan receptor γt (RORγt) is a master regulator of the Th17/IL-17 pathway that plays crucial roles in the pathogenesis of autoimmunity. RORγt has recently emerged as a highly promising target for treatment of a number of autoimmune diseases. Through high-throughput screening, we previously identified several classes of inverse agonists for RORγt. Here, we report the crystal structures for the ligand-binding domain of RORγt in both apo and ligand-bound states. We show that apo RORγt adopts an active conformation capable of recruiting coactivator peptides and present a detailed analysis of the structural determinants that stabilize helix 12 (H12) of RORγt in the active state in the absence of a ligand. The structures of ligand-bound RORγt reveal that binding of the inverse agonists disrupts critical interactions that stabilize H12. This destabilizing effect is supported by ab initio calculations and experimentally by a normalized crystallographic B-factor analysis. Of note, the H12 destabilization in the active state shifts the conformational equilibrium of RORγt toward an inactive state, which underlies the molecular mechanism of action for the inverse agonists reported here. Our findings highlight that nuclear receptor structure and function are dictated by a dynamic conformational equilibrium and that subtle changes in ligand structures can shift this equilibrium in opposite directions, leading to a functional switch from agonists to inverse agonists.

The Th17 lineage of T helper cells plays an essential role in protective immunity against a variety of bacteria such as Mycobacteria tuberculosis and Staphylococcus aureus and pathogenic fungi such as Candida albicans (1). Individuals with genetic defects in the Th17 pathway are susceptible to recurrent bacterial infections and often develop unrelenting chronic mucocutaneous candidiasis (2,3). However, Th17 cells, which produce the eponymous IL-17A (often simply referred to as  and other proinflammatory cytokines, including IL-17F, IL-21, IL-22, and granulocyte-macrophage colony-stimulating factor (GM-CSF), are also potent inducers of multiple autoimmune diseases in animal models (4 -8) and are strongly implicated by human genetic studies in the pathogenesis of most common human autoimmune diseases, including psoriasis, psoriatic arthritis, Crohn's disease, ankylosing spondylitis, rheumatoid arthritis, and multiple sclerosis (9 -13). Given the prominent roles of the Th17/IL-17 pathway in autoimmunity, therapeutic interventions targeting this pathway have been intensely pursued. Two monoclonal antibodies (mAb) targeting IL-17A, secukinumab (Cosentyx) and ixekizumab (Taltz), have been approved recently for the treatment of moderate to severe plaque psoriasis. Secukinumab is also approved for psoriatic arthritis and ankylosing spondylitis (14). In phase 3 clinical trials, both secukinumab and ixekizumab have demonstrated remarkable efficacy for treatment of psoriasis with ϳ90% of patients achieving a 75% reduction in psoriasis area and severity index (PASI75) and ϳ40% achieving a PASI100 response with complete clearing of skin lesions (15). Similar efficacy has been achieved by brodalumab (Siliq), a mAb targeting the IL-17 receptor IL-17RA, which was recently approved by the United States Food and Drug Administration to treat adults with moderate to severe plaque psoriasis (16). Other mAbs, such as guselkumab, tildrakizumab, risankizumab (BI 655066), and AMG 139, that target the p19 subunit of IL-23, a critical cytokine for pathogenic Th17 lineage commitment and expansion, also show excellent efficacy for psoriasis (15). Collectively, the success of these biologics has strongly validated clinically IL-17, IL-17R, and IL-23, all key players in the Th17 pathway, as valuable therapeutic targets for autoimmune diseases.
The nuclear receptor (NR) 2 retinoid acid receptor-related orphan receptor ␥t (ROR␥t) is a master transcription factor of Th17 cells, being both necessary and sufficient for IL-17 expres-sion, and it is essential in promoting Th17 cell differentiation while suppressing the Th1 program (17,18). ROR␥t is also required for the production of IL-17 from other cell types, including ␥␦ T cells, invariant natural killer T cells, and group 3 innate lymphoid cells, and is necessary for Th17 cells to produce other proinflammatory cytokines such as IL-22, GM-CSF, and the IL-23R (19). Therefore, small-molecule modulators against ROR␥t are also a highly attractive therapeutic modality that has the potential for meaningful pharmacological differentiation from the specific anti-IL-17 or anti-IL-23 mAbs. Many small-molecule modulators of ROR␥t have been reported in the literature in recent years, and two compounds have reached phase 2 clinical trials for treatment of psoriasis (20 -22).
There are three members in the ROR subfamily of human NRs as follows: ROR␣ (RORA, NR1F1), ROR␤ (RORB, NR1F2), and ROR␥ (RORC, NR1F3). The members of the subfamily share about 50% sequence identity in their ligand-binding domains (LBD). Two isoforms of ROR␥ exist: the canonical ROR␥ (ROR␥1) and ROR␥t (ROR␥2), which lacks the first 21 N-terminal amino acids due to alternative promoter usage. The ROR␥ isoform is expressed in most tissues and is involved in many physiological functions (23). In contrast, ROR␥t is solely expressed in lymphoid lineage cells of the immune system, consistent with its essential role in the development of lymph nodes as well as Th17 cells (23). Despite the word "orphan" in ROR␥'s name, recent studies with sterol auxotroph cells have convincingly shown that sterol lipids, including certain cholesterol biosynthetic intermediates and oxysterols, are physiological ligands for ROR␥ (24 -26). An important link between lipid metabolism and regulation of Th17 pathogenicity was established recently by the discovery that CD5 antigen-like (CD5L) acts as a negative regulator that alters the balance of lipid saturation and directly affects the availability of sterol ligands for ROR␥t (27).
Crystal structures of ROR␥ LBD in complex with hydroxycholesterols and various synthetic ligands have been reported (28 -36). But to our knowledge, no apo structure of ROR␥ LBD has yet been published. Here, we report the crystal structures of ROR␥t LBD in both apo and ligand-bound states with two novel classes of synthetic ligands. The apo ROR␥t structure adopts a predominantly active conformation, which is supported by the NMR experiments in solution. Density functional theory (DFT) calculations were used to elucidate the energetics of the structural determinants underlying the active conformation of ROR␥t. A common theme for the structural mechanism of action of the two distinct classes of inverse agonists was revealed from analyses of the respective complex structures and normalized crystallographic B-factors. We emphasize that the dynamic conformational equilibrium is a fundamental attribute to understand NR structure and function. Indeed, subtle changes in ligand structures can shift the equilibrium in opposite directions and lead to a functional switch from an agonist to an inverse agonist.

Production and crystallization of SRC2 peptide-tethered ROR␥t-LBD
Structural studies of NRs are often carried out in the presence of carefully chosen cofactor peptides that bind and stabi-lize the receptors, as is the case in the first crystallographic study of the ROR␥t-LBD (28). When examining the published crystal structure (PDB code 3L0L), we noticed that the C␣-C␣ distance is only 8 Å between the first resolved N-terminal residue of the steroid receptor coactivator-2 (SRC2) peptide and the C-terminal residue (Ser-507) of the ROR␥t-LBD. Modeling suggested that it might be possible to covalently tether an SRC2 peptide to the C terminus of ROR␥t-LBD via a simple -GGGlinker. The tri-glycine linker provides the maximal conformational flexibility to preserve the native interactions between the tethered partners with minimal interference. The tethered system is thermodynamically more stable than the untethered complex due to the reduction of macroscopic translational entropy through tethering (37,38). We therefore produced the His 6 -ROR␥t-LBD(260 -507) -GGG-EKHKILHRLLQDS (SRC2 peptide) construct and expressed it in Escherichia coli. The chimeric protein expressed well in a soluble form and was straightforward to purify using standard affinity and size-exclusion chromatography (see under "Experimental procedures"). Similar approaches have been used in the study of the PXR-LBD-SRC-1p complex (39), as well as MHC-II-peptide complexes and other protein-protein interactions (40,41). Without the complication of achieving the right stoichiometry for the peptide/receptor mixture, the apo crystals of ROR␥t-LBD with tethered SRC2 peptide could be reproducibly obtained by sitting or hanging drop vapor diffusion at room temperature (see under "Experimental procedures"). In most cases, co-structures with ligands were obtained by soaking of compounds into the apo ROR␥t-LBD crystals. Very occasionally, co-crystallization, where a compound was pre-incubated with the protein solution prior to crystallization, was also used to solve a co-complex structure. This is the case for compound 2 (Table 2), for which co-structures were obtained through both soaking and co-crystallization.

Apo ROR␥t-LBD structure and the structural determinants for its active conformation
The apo ROR␥t-LBD structure was solved in the space group P4 1 2 1 2 with one polypeptide chain per asymmetric unit. The statistics for data collection and refinement are summarized in Table 1. The apo ROR␥t-LBD structure is very similar to that of the 25-hydroxycholesterol (25-HC)-bound ROR␥t (28) (PDB code 3L0L), with r.m.s.d. of backbone and all heavy atoms at 0.54 and 1.20 Å between the two structures, respectively ( Fig.  1). Similar to 25-HC-bound ROR␥t, the apo structure assumes an active conformation, with H12 as an integral part of the AF-2 surface that captures the SRC2 coactivator peptide (Fig. 1A). The -GGG-linker residues are in a flexible loop conformation with weak electron densities and elevated B-factor values. There are no specific interactions between the glycine linker and the rest of the protein. Notably, the apo ROR␥t contains a large ligand-binding pocket, with a cavity volume of 940 Å 3 (Fig.  1A). As is typical for nuclear receptors, the ligand-binding pocket is predominantly hydrophobic (575 Å 3 , 61% of total cavity volume). In contrast, the cavity volumes for hydrophilic, basic, and acidic compositions are 161 Å 3 (17%), 178 Å 3 (19%), and 26 Å 3 (3%), respectively. Interestingly, almost the entire 25-HC can fit comfortably in the ligand-binding pocket of apo ROR␥t; only some adaptive side-chain movements from His-323, Leu-324, and Mer-365 are required to accommodate the B-ring of 25-HC (Fig. 1B). In fact, the His-323 side-chain conformation switch constitutes the largest structural difference between the apo and 25-HC bound ROR␥t. It has been shown recently that a number of cholesterol precursors, such as lanosterol and desmosterol as well as oxysterols, are the natural endogenous ligands for ROR␥t (24 -26). Our structure of apo ROR␥t suggests that the ligand-binding pocket may be preformed to readily accommodate such ligands.
The question then arises as to how ROR␥t keeps H12 in the active conformation and maintains a large cavity in its interior in the absence of a ligand. One key structural element is the His-479 -Tyr-502-Phe-506 triplet. Tyr-502 and Phe-506 reside on the same face of the two-turn H12, forming close interactions with His-479 on H11 (Fig. 1C). The most prominent of the interactions is the H-bond between the phenol of Tyr-502 and the imidazole N ⑀2 of His-479. Here, the N ⑀2 of His-479 acts most likely as an H-bond acceptor, and thus His-479 is in the neutral (uncharged) form. His-479 also engages in an edge-to-face aromatic packing interaction with Phe-506. Furthermore, a favorable aromatic ring packing exists between Tyr-502 and Phe-506. The side-chain rotamer of His-479 is further stabilized through a water-mediated H-bond network linking its N ␦1 to the backbone carbonyl oxygen of Gln-475. To more quantitatively assess the significance of these interactions, an ab initio calculation of the interaction energy among the side chains of the HYF triplet was carried out using the DFT method (see under "Experimental procedures"). The DFT calculation used the Truhlar hybrid meta exchange-correlation functional M06-2X with a 6 -32G** basis set, which is wellsuited for calculating non-covalent interactions such as H-bonding and -stacking that are dominated by dispersion forces (42,43). The resulting interaction energy for the HYF triplet is Ϫ12.90 kcal/mol. This is a very significant energy that provides the anchoring force to stabilize H12 of ROR␥t in the active conformation in the absence of a ligand. The HYF triplet is conserved among the ␣/␤/␥ isoforms of the ROR subfamily but is not found in any other NRs. For example, the corresponding residues in PPAR␥ are His-449 -Leu-469 -Tyr-473 with no H-bonding among them. The interaction energy is only Ϫ3.00 kcal/mol when PPAR␥ is in the active conformation and is negligible when PPAR␥ is in the inactive conformation (interaction energies calculated using the same DFT method, with PPAR␥ coordinates from PDB code 1PRG) (see supplemental Fig. S1). Because of the lack of a stabilizing triplet in PPAR␥, H12 and indeed the entire ligand-binding pocket of apo PPAR␥ exhibit significant conformational mobility as revealed by NMR (44) and fluorescence anisotropy studies (45). The crystal structure of apo PPAR␥ also showed elevated crystallographic B-factor values for the H12 and ligand-binding pocket residues, and H12 was in fact captured in both active and inactive states (46).
There is another unique structural element in ROR, namely the H11Ј helix. Again, H11Ј is present in all three isoforms of the ROR subfamily but is not found in most of the other NRs. As illustrated in Fig. 1D, H11Ј and H12 together bury 322 Å 2 of solvent-accessible surface area, which is 29% of H12's total surface area. Therefore, a notable helical packing interaction exists between H11Ј and H12. Helical packing in general is an important stabilizing force in protein folding and in protein-protein interactions (47,48). Here, the H11Ј-H12 packing contributes to stabilize H12 in the active conformation. Interestingly, H12 and the SRC2 peptide together bury 509 Å 2 of solvent-accessible surface area (Fig. 1D), which amounts to 46% and 32% of total surface area for H12 and the SRC2 peptide, respectively, and it underscores the critical role H12 plays in recruiting coactivators. In summary, ROR␥t utilizes a unique HYF triplet to provide significant interaction energy to anchor H12 in the active conformation, which is further stabilized by helical packing with a unique H11Ј element.

ROR␥t-LBD is capable of recruiting coactivator peptide in the absence of a ligand
A hallmark of NR in the active state is its ability to recruit coactivators to initiate transcription. To investigate whether apo ROR␥t is competent to bind coactivators, we carried out solution NMR studies using 13 C/ 15 N-labeled samples from a ROR␥t(259 -518) construct as well as the crystallography construct ROR␥t(260 -507)-G 3 -SRC2. The ROR␥t(259 -518) construct encompasses the entire ROR␥t ligand-binding domain, without C-terminal truncation beyond H12 and without tethering of a coactivator peptide. As shown in Fig. 2, titration of the SRC2 peptide into the 13 C/ 15 N-labeled ROR␥t(259 -518) protein resulted in distinctive chemical shift perturbations to the backbone amide peaks in the 15 N-TROSY spectra ( Fig. 2A) as well as to the methyl peaks in the 13 C-HSQC spectra (Fig. 2B) of ROR␥t(259 -518). These results clearly demonstrate that the native ROR␥t-LBD in solution is capable of binding to a coactivator peptide, such as the SRC2 peptide used here, in the absence of a ligand. Furthermore, the backbone 15 N-TROSY spectrum of the SRC2-bound ROR␥t(259 -518) bears strong resemblance to that of the ROR␥t(260 -507)-G 3 -SRC2 (Fig.  2C). Some of the differences between the two spectra with proton chemical shifts centered around 8 ppm in Fig. 2C are likely due to the difference of C-terminal sequences after residue 507. The resemblance is even more striking for the methyl resonances in the 13 C-HSQC spectra (Fig. 2D and compare with B), with the perturbed methyl peaks from ROR␥t(259 -518) shifting to the exact positions of the corresponding methyl peaks from ROR␥t(260 -507)-G 3 -SRC2. Therefore, the SRC2-tethered ROR␥t-LBD largely recapitulates the native ROR␥t-LBD in the presence of the SRC2 peptide in solution. It should be noted from qualitative assessments of the NMR spectra that the native apo ROR␥t-LBD exhibits more conformational flexibil- ity in solution than the SRC2-tethered apo ROR␥t-LBD. There is also evidence that the C-terminal residues beyond H12 (i.e. residues 508 -518) in the native ROR␥t-LBD are largely disordered both in solution and in crystal forms. 3

Binding mode of two diverse classes of ROR␥t inverse agonists and the structural mechanism of action (sMOA)
Two classes of ROR␥t inverse agonists were discovered through a high-throughput screening campaign using a ROR␥t gene reporter assay (see under "Experimental procedures"). Two representative compounds from the two classes are shown in Table 2. These compounds had sub-micromolar binding affinity to ROR␥t in a fluorescence polarization (FP) competition assay using a probe with similar potency in a cell-based ROR␥t reporter gene assay. More importantly, these compounds also demonstrated Th17 primary cell activity, inhibiting the production of IL-17. A 2.04-Å resolution co-complex structure of compound 1 with ROR␥t-LBD was obtained through soaking of the com-

Table 2
Inverse agonists of ROR␥t reported in this study pound in the apo ROR␥t crystals. As shown in Fig. 3A, compound 1 binds across the entire cavity of the ligand-binding pocket, starting from one end with the pyridine nitrogen making a strong H-bond with Arg-367, the urea carbonyl in the middle forming a water-mediated H-bond with the backbone carbonyl of Phe-377, and the isobutylene moiety extending into a highly hydrophobic region lined with aliphatic and aromatic residues, and ending with the phenyl amide interacting with the critical HYF triplet. Most notably, the phenyl ring of compound 1 protrudes into the His-479 -Tyr-502 pair, disrupting the H-bond  Fig. 1A and key ligand-binding site residues highlighted as sticks. The simulated annealing omit difference maps were calculated with the entire compound molecules omitted and plotted in blue at 2.5 level for compounds 1 and 2 and at 3.0 level for compound 3. The apo structure of ROR␥t is shown in pale green as in Fig. 1A.

Apo and ligand-bound ROR␥t structures
between them. His-479 is forced to adopt a different side-chain conformation, which is stabilized through the formation of two new H-bonds with the amide linker of compound 1 and the backbone carbonyl of Leu-475, respectively (Fig. 3B). The two aromatic residues on H12, Tyr-502 and Phe-506, however, lose their close interactions with His-479. In this case, the interaction energy of the HYF triplet is reduced to only Ϫ1.4 kcal/mol as calculated using the same DFT protocol, which amounts to a loss of 11.5 kcal/mol stabilizing energy for H12 compared with the apo structure. Therefore, the disruption by compound 1 of the critical H-bond between His-479 and Tyr-502, with a concurrent loss of the aromatic packing between His-479 and Phe-506, results in the severe destabilization of H12 in its active conformation.
Compound 2 is structurally very different from compound 1. To understand how compound 2 binds to ROR␥t and what is the structural basis for its mechanism of action, we also solved its co-complex structure with ROR␥t-LBD. Both soaking and co-crystallization approaches were attempted, and both yielded co-structures that were highly similar to each other; therefore, only the structure from co-crystallization is reported here. Despite having completely unrelated chemical structures, compounds 1 and 2 occupy essentially the same space in the ligand-binding pocket of ROR␥t (Fig. 3C). The isobutyl group of compound 2 aligns with the pyridyl ring of compound 1, although the former does not engage Arg-367 in an H-bond. The methyl piperidine moiety of compound 2 fills the same hydrophobic sub-pocket that accepts the isobutylene moiety of compound 1. A very similar water-mediated H-bond is observed between compound 2 and the backbone carbonyl of Phe-377. Most interestingly, the chlorobenzyl group of compound 2 also intercalates into the His-479 -Tyr-502 pair, forcing a side-chain conformation switch for His-479 and breaking up the His-479 -Tyr-502 H-bond (Fig. 3D). The interaction energy of the HYF triplet is lost almost completely at Ϫ1.1 kcal/mol based on the DFT calculation. The loss of favorable interactions for His-479 is partly compensated for by the formation of new H-bonds with the amide linker of compound 2 and the backbone carbonyl of Leu-475, in a way reminiscent of the case with compound 1. Therefore, it appears that we have uncovered a common mechanism by which diverse chemical scaffolds may exert inverse agonist activities against ROR␥t through direct disruption of the tightly interacting HYF triplet thereby destabilizing the active conformation of ROR␥t.

Differential B-factor analysis
The crystallographic B-factor, also called atomic displacement parameter, is a measure of atomic displacement from its equilibrium or mean position, and it captures structural flexibility as well as positional variations due to thermal vibrations. Not surprisingly, B-factors have been widely exploited to probe protein flexibility (49), thermal stability (50), enzyme-active sites (51), and more recently to provide an integrated description of protein dynamics by combining with order parameters derived from solution NMR studies (52). Likewise, we reasoned that the destabilization of ROR␥t's H12 due to the binding of compounds 1 and 2 should be manifested by elevated B-factors in the H12 region of the respective complex structures. Typically, B-factor analyses of proteins are performed to compare different regions of the same protein structure. Here, we want to compare B-factors in the same regions of the protein from two different structures. To allow for a meaningful comparison of B-factors between the H12 region of interest in the bound state and the corresponding apo state, we first introduce a modified B-factor as shown in Equation 1, where ͗B͘ is the median B-factor, calculated separately for backbone and side-chain atoms, respectively. BЈ can be viewed as a normalized B-factor. It is not influenced by variation of overall B-factors between different crystals. BЈ is dimensionless. A negative (positive) BЈ value can be interpreted as indicative of higher (lower) rigidity and stability than the average of the structure. To identify regions of altered flexibility or stability in the compound-bound complex versus the corresponding apo state, we examine Equation 2, An atom with positive B diff means it has elevated flexibility and is less stable compared with its apo state. Conversely, an atom with negative B diff means it is further rigidified and more stable than in the apo state. In Fig. 4, the B diff values are visualized with color ramps from blue (negative) to white (0) to red (positive) for H11 and H12 of ROR␥t in complex with compounds 1 (Fig.  4A) and 2 (Fig. 4B), respectively. As already suggested by the DFT calculations, the destabilization of H12 and the HYF triplet by the ligands is reflected in the B diff plots by red colorations of H12.
In addition to showing how ligands can destabilize H12, we surmised that it should also be possible that ligands can further stabilize H12 in the active conformation. To this end, we studied compound 3, which was synthesized as an analog of compound 1 following a typical structure-activity relationship (SAR) optimization strategy. Compound 3 is structurally very similar to compound 1 except having an ether linker in place of an amide in compound 1. Compound 3 showed good binding to ROR␥t (FP K d ϭ 0.19 M), but surprisingly it did not register an inhibitory activity in the reporter gene assay (IC 50 Ͼ3.3 M). To understand its unexpected lack of inverse agonist activity, we solved a 2.34 Å resolution co-complex structure of compound 3 with ROR␥t-LBD through soaking of the compound in the apo ROR␥t crystals (Fig. 3E). Surprisingly, compound 3 does not disrupt the tightly interacting HYF triplet. Instead, the phenyl ether of compound 3 points toward Trp-317 and forms a tightly packed aromatic cluster with the HYF triplet and Trp-317 together, thereby further stabilizing ROR␥t in the active conformation (Fig. 3F). This stabilization effect is confirmed by the differential B-factor analysis shown in Fig. 4C. Because the inverse agonist activities of compound 1 and 2 are due to their destabilization effect on the active receptor conformation of ROR␥t as discussed above, it is expected that stabilization of the active conformation by compound 3 would elicit the opposite effect on the receptor. In other words, compound 3 should be an agonist of ROR␥t. A fluorescence polarization (FP)-based coactivator recruitment assay using a fluorescein-labeled coactivator peptide (D22) containing the LXXLL motif was developed to test this. As shown in Fig. 5, compound 3 indeed behaves as an agonist by enhancing the recruitment of D22 with an EC 50 of 0.54 M.

Discussion
Nuclear receptors are ligand-regulated transcription factors that orchestrate the assembly of coregulatory complexes (coactivators and corepressors) to modulate transcription through chromatin-remodeling activities (53). The role of a high affinity endogenous ligand is "classically" viewed as a trigger of a molecular switch whereby the binding of the ligand induces a conformational change involving H12 that switches the receptor from an inactive or repressed state to an active state competent for recruitment of coactivators and subsequent initiation of transcription (54). This conceptually appealing but overly simplistic view of NR activation is still frequently cited in the literature, even though mounting evidence has accumulated revealing NR proteins are highly dynamic and versatile. For instance, REV-ERB␣ (NR1D1) and REV-ERB␤ (NR1D2) do not even have a H12 and use heme as a ligand (55,56); NR activities are spatially and temporally regulated (57) and involve multiple structural elements with specific higherorder structures at play (58,59). NRs are also known to be regulated by ligand-independent mechanisms such as posttranslational modifications (60,61). For example, ROR␥t has been shown to be modulated by both acetylation (62) and ubiquitination (63) with direct functional effects on Th17 biology. In this study, we showed that the apo ROR␥t LBD can adopt an active conformation with the capability to recruit a coactivator peptide. We believe that this active conformation is not influenced by the triple-glycine linker because these glycines do not make any specific contacts with the rest of the protein in the crystal structure and are highly flexible with very weak electron densities and high B-factors. In full support of this notion, our solution NMR studies with the ROR␥t(259 -518) construct demonstrated that the non-tethered native ROR␥t LBD is capable of directly binding to the SRC2 coactivator peptide in the absence of a ligand, and the NMR spectra of SRC2-bound ROR␥t LBD closely resemble those of the SRC2-tethered ROR␥t-LBD, suggesting both having similar conformations in solution.
A number of NRs, such as estrogen-related receptor ␥ (ESRRG) (NR3B3), nerve growth factor IB (NGFIB) (NR4A1), nuclear receptor-related 1 (NURR1) (NR4A2), and liver receptor homolog-1 (LRH-1) (NR5A2), are capable of adopting active conformations independent of ligands (64). The crystal structure of apo LRH-1 (65) in particular shows an active conformation with a voluminous but empty cavity (820 Å) in the ligand-binding pocket, very similar to the case with apo ROR␥t revealed here. The specific structural elements stabilizing apo LRH-1 in the active conformation have been attributed to an extended H2 that acts as an extra layer with optimal helical packing against H3 to hold AF2 helices, including H12, in the activated state (65). In this study, we have carried out detailed structural and computational analyses to demonstrate that the tightly interacting His-479 -Tyr-502-Phe-506 triplet is the primary structural element responsible for anchoring ROR␥t in the active conformation, which is further strengthened by additional helical packing with the extra H11Ј helix. Both the HYF triplet and the H11Ј are structural features unique to the ROR subfamily of NRs. Proteins exist not in a single fixed state but rather as a dynamic ensemble in the biologically relevant environment. The apo crystal structure of ROR␥t LBD reported here is only a snapshot of a continuum of conformations sampled by ROR␥t in solution. The snapshot revealed by crystallography represents a low energy state having a dominant population. Under physiological conditions, ROR␥t should also be able to sample other lower population (higher energy) states, including inactive ones. In fact, our NMR studies with the ROR␥t(259 -518) construct have confirmed that ROR␥t can indeed bind directly to a corepressor peptide derived from the silencing mediator of retinoid and thyroid hormone receptors-2 (SMRT2) in solution in the absence of a ligand (supplemental Fig. S2). Therefore, we have shown that ROR␥t has the conformational elasticity to bind either a coactivator (active conformation) or a corepressor (inactive conformation) in solution independent of ligands. It would be reasonable to posit that in general the presence of ligands or coregulators can alter the populations of various distinct states an NR may assume and shift the conformational equilibrium toward further activation or inactivation of the receptor. Under this dynamic equilibrium paradigm of NR modulation, the abundance of specific ligands, coactivators, or corepressors would dictate the activation state of an NR in a particular cellular context.
The sMOA revealed by the crystallographic studies of the two novel classes of ROR␥t inverse agonists dovetails with the dynamic equilibrium paradigm. Despite their high structural diversity, the two classes of inverse agonists share a common sMOA; they both disrupt the critical HYF triplet anchor thereby destabilizing the active conformation of H12 and shifting ROR␥t toward inactivation. Conversely, an agonist of ROR␥t should do the opposite. This is the case with compound 3, which forms an extended aromatic cluster together with the HYF triplet and further stabilizes H12 in the active conformation. The dynamic picture of H12 mobility and stability in response to ligand binding is supported by the differential B-factor analysis as well as the ab initio calculations of the interaction energies employed in this study. Recently, cholesterol biosynthetic intermediates have been identified as natural ligands for ROR␥t. We believe that these endogenous ligands work similarly to compound 3 in that they also bind and further stabilize the active conformation of ROR␥t. Under physiological conditions, the actions of endogenous ligands are likely necessary to achieve sustained activation of ROR␥t due to other factors that can tilt the conformational equilibrium of the receptor toward repressed states.
Finally, it is interesting to note that subtle changes in the ligand structure can result in diametrically opposite functional responses of a nuclear receptor, as demonstrated by the two structurally very similar urea compounds 1 and 3 reported here. This phenomenon can be understood most naturally from examining the dynamic equilibrium of receptor states, as compounds 1 and 3 shift the conformational equilibrium of ROR␥t in opposite directions. Similar results have been reported by Rene et al. (35), who showed a class of benzylsulfonamides as full inverse agonists of ROR␥t with phenylsulfonamide analogs showing agonistic activities. The underlying sMOA described by the Genentech group is identical to what we have observed here, in that the phenylsulfonamides stabilize the active conformation of ROR␥t, and the benzylsulfonamides destabilize it. The subtleties that minimal variation of ligand structures can lead to dramatically different functional activities are not limited to NRs but are in fact well known in the GPCR field (66). Complex dynamic states are involved in the ligand regulation as well as signal transduction of GPCRs (67,68). The dynamic equilibrium paradigm appears to be a unifying framework applicable for understanding the subtle structure-activity relationships often encountered in receptor drug discovery.

Crystallization, data collection, and structure determination
Initial screens of crystallization conditions were carried out with Crystal Screens I and II from Hampton Research, and NR-LBD HT-96 screening kit from Molecular Dimensions (MD1-34), using sitting drop vapor diffusion on a Mosquito instrument at 20°C. Drops containing 0.1 l of protein mixed with 0.1 l of well solution were equilibrated against 80 l of well solution. Follow-up optimizations were performed manually using hanging drop vapor diffusion with drops containing 1 l of protein mixed with 1 l of well solution equilibrated against 500 l of well solution. Diffraction-quality apo or ligand-bound ROR␥t crystals were obtained with 2-8% (w/v) PEG4000, 0.6 M NaCl, and 0.1 M PIPES, pH 7.0. Crystals were flash-frozen in liquid nitrogen with cryoprotectant prepared using the corresponding reservoir condition supplemented with 20% glycerol. Protein-ligand co-crystals were obtained either from soaking of the compound into apo ROR␥t crystals or from co-crystallization of the pre-formed protein-ligand complex solution. For co-crystallization, a compound was added to the protein stock solution with a protein-ligand molar ratio of about 1:5. The complex solution was incubated at room temperature for 40 min and then centrifuged at 14,000 rpm for 35 min prior to crystallization experiments. For soaking experiments, 3 mM compound prepared in DMSO stock solution was added in a drop containing the apo ROR␥t crystals and soaked at room temperature for 6 h. The soaked crystals were then flash-frozen in liquid nitrogen using the crystallization solution supplemented with 25% PEG400 as a cryoprotectant. X-ray diffraction data were collected at beamlines X06SA or X06DA (for co-crystals with compound 1) of the Swiss Light Source using Pilatus 6M and 2M-F detectors, respectively. Diffraction data were processed using either d*TREK or XDS (69,70). The initial structure was determined by molecular replacement with Phaser (71) as implemented in the PHENIX software suite (72) using the structure of ROR␥t LBD in complex with 22(R)-hydroxycholesterol (PDB code 3L0J) as a search model. Ligand geometry restraints were generated using Corina (Molecular Networks GmbH). Multiple rounds of positional and isotropic B-factor refinement using phenix.refine followed by manual rebuilding using Coot (73) were carried out for each structure. The quality of the final model was evaluated using MolProbity (74). Structural figures were generated using PyMOL (Schrödinger, LLC).

NMR spectroscopy
All NMR data were acquired at 30°C on an 800-MHz Bruker AvanceIII spectrometer equipped with a triple resonance ( 1 H/ 13

Fluorescence polarization competitive binding assays
The fluorescence polarization (FP) measurements were conducted using an Envision plate reader (PerkinElmer Life Sciences) using the ROR␥t-SRC2 construct produced in-house (see above), and a fluorescent TAMRA (Thermo Fisher Scientific, Inc.) probe synthesized with an in-house compound binding to ROR␥t. Compound was diluted in assay buffer (20 mM Tris, pH 7.5, 150 mM NaCl, and 0.05% CHAPS), and 10 l of 20 M compound was mixed with 10 l of 2 M ROR␥t-SRC2 in the same buffer and incubated for 30 min at room temperature. The TAMRA probe was then added to the mixture at a final concentration of 15 nM. After 30 min of incubation at room temperature, the FP signal was measured (excitation wavelength 531 nm; emission wavelength 595 nm), and the K d values were determined as described (78).

ROR␥t-Gal4 reporter gene assay
HEK293T cells were co-transfected with a plasmid pBIND containing the chimera of the DNA-binding domain of the yeast Gal4 protein and the ligand-binding domain of human ROR␥t (Gal4DBD-hROR␥t LBD), along with the luciferase reporter plasmid pGL4.31 (luc2P/GAL4UAS/Hygro, Promega). The positive control was co-transiently transfected with both plasmids, and the negative control had only the pGL4.31 promoter sequence. Assays were assembled in 384-well Greiner plates where transiently transfected cells and test compound at varying concentrations were incubated at 37°C and 5% CO 2 for 20 -24 h. The next day, assay plates were taken out and equilibrated at room temperature for 20 -30 min. Bright-Glo TM luciferase assay system (Promega) was used to detect luciferase production. After addition of Bright-Glo detection reagent, the plates were incubated at room temperature for 20 min. The plates were read on an Envision plate reader to measure luminescence signal. The relative light unit (RLU) signal was converted to percent of control value relative to control and blank wells.

IL-17 production in Th17 primary cell assay
Frozen CD4 ϩ T cells (AllCells) were thawed and resuspended in X-VIVO media (Lonza) at a cell density of 1 ϫ 10 6 cells/ml. Skewing cytokines were added to media at the final concentrations of 30 ng/ml IL-23, 10 ng/ml IL-1␤, 10 ng/ml IL-6, 2 ng/ml IL-2, 4 ng/ml TGF␤1, 5 g/ml IL-4, and 5 g/ml IFN␥ and mixed with activated beads (Miltenyi Biotec catalogue no. 130-091-441) at 1 bead/cell. The cells were incubated under the stimulatory conditions for 72 h at 37°C. Skewed cells were spun and resuspended in Iscove's media (Invitrogen catalogue no. 12440) with a cell density of 1.11 ϫ 10 6 cells/ml. Cells were seeded at 90 l/well into Corning black/clear TC-treated plates (Corning catalogue no. 3603) to give 100,000 cells per well. 10 l of medium-diluted compounds was added, and plates were incubated for 2 h at 37°C and 5% CO 2 . 50 l of cytokine/bead mixture was added to all wells except blank wells, which received media only. Cell plates were incubated at 37°C and 5% CO 2 for 48 h. Afterward, plates were spun, and 50 l of supernatant from each well was collected and transferred to MSD Vplex IL-17A assay plate (Meso Scale Diagnostics, LLC) for detection of IL-17 expression, following the manufacturer's protocol. IC 50 values were obtained by fitting 10-point concentration-response data to a four-parameter logistic equation in ActivityBase (ID Business Solutions Ltd.).

Coactivator recruitment assay
An FP-based assay was developed for coactivator recruitment. The fluorescence polarization measurements were conducted on an Envision (PerkinElmer Life Sciences) using a ROR␥t-GST construct produced in-house. The probe is a fluorescein-labeled coactivator peptide (D22) from Invitrogen containing the LXXLL motif and optimized for binding to ROR␥t ligand-binding domain. 10 l of compounds diluted in 20 mM Tris, pH 7.5, 150 mM NaCl, and 0.05% CHAPS were mixed to 10 l of 2 M ROR␥t-GST in the same buffer and incubated for 30 min at room temperature. The fluorescein D22 probe was then added to the mixture at a final concentration of 15 nM. After 30 min of incubation at room temperature, the FP signal was mea-sured (excitation wavelength 480 nm; emission wavelength 535 nm).

DFT calculation
Interaction energies between His-479, Tyr-502, and Phe-506 of ROR␥t LBD were calculated using the DFT routine implemented in Jaguar (79). A hybrid functional for non-covalent complexation energies, M06-2X (80), was applied together with a 6 -31G** basis set to calculate DFT energies. A self-consistent reaction field method using a Poisson Boltzmann solver (81) was applied as continuum solvation model with a dielectric constant of ⑀ ϭ 4. Interaction energies were calculated as the difference between the DFT energy of the residue triplet with atomic coordinates of His-479, Tyr-502, and Phe-506 taken from the crystal structures reported here and three DFT energies associated with the individual residues calculated separately. To preserve the coordinates found in the crystal structures, only single point energies were calculated. The interaction energies for corresponding PPAR␥ residues were calculated based on the same DFT method, with PPAR␥ coordinates taken from PDB code 1PRG.