Molecular communication within and between proteins forms the basis for biological signaling. Allostery involves coupling of ligand binding at one site with a conformational or dynamic change at a distant site, thereby affecting binding at that site. Identifying the transmission pathways between sites and understanding their dynamic nature is of fundamental interest and potentially of high relevance for drug design, where distinct ligands may drive differentiating effects downstream of the signaling events. Here, we identify and explore the allosteric communication network within the glucocorticoid receptor (GR) ligand-binding domain (LBD).
GR is a nuclear hormone receptor with essential roles in metabolism and resolution of inflammation (1). GR agonists are efficacious for treatment of inflammatory, allergic, and immunological disorders (2, 3), but it remains a longstanding challenge to design differentiating ligands that separate the anti-inflammatory efficacy from side effects such as diabetes, muscle wasting, and osteoporosis (4). GR comprises the N-terminal domain, the DNA binding domain, and the C-terminal LBD. Ligand-free GR predominantly resides in a chaperone complex in the cytoplasm. Ligand binding leads to partial release of chaperone proteins and nuclear translocation. Within the nucleus, GR binds to specific DNA sequences and promotes coregulator assembly and transcriptional regulation (1).
Ligands bind to the fully occluded binding pocket at the center of the LBD (Fig. 1A) (5–7) and modulate the structure of the receptor (8, 9). The receptor conformation, in turn, directs the DNA interaction pattern and guides the protein composition of the enhancer or promotor complexes to drive a specific genomic response (1, 10). Allosteric effects between the intrinsically disordered N-terminal domain and the DNA binding domain have been investigated in detail by Hilser and coworkers (11), showing how binding to individual domains shapes the ensemble of states and tunes the output between activation and repression. However, less is known about ligand-driven allosteric effects and how these might influence signaling.
The LBD harbors two regions involved in binding coregulator proteins: the activation function 2 (AF-2), located at the interface between helices 3, 4, and 12 (1, 5), and the transcriptional activation function tau2, which has been mapped to helix 1 (1, 12). Crystal structures of GR LBD in complex with different ligands and coregulator peptides have shown that the receptor can adopt different conformations (5–7, 9). In particular, the position of helix 12 is of key importance for receptor activity across the nuclear hormone receptor family (13–17). Recent reports on one other nuclear hormone receptor, the peroxisome proliferator–activated receptor γ (PPARγ), have demonstrated that helix 12 and the coregulator-binding surface adopt an ensemble of conformations, the relative populations of which respond to ligand binding (18–20). This indicates that conformational flexibility forms an integral part of receptor activation and leads to the hypothesis that ligand-driven shifts within conformational ensembles explain how different ligands can promote different levels of receptor activation (graded response) and tissue-specific effects (biased signaling).
Recent developments in nuclear magnetic resonance (NMR) have made it possible to detect weakly populated conformational states and determine their lifetimes (21, 22). High-resolution NMR studies of GR LBD have previously been hampered by poor spectrum quality and sample aggregation, but we recently reported near-complete backbone assignments for wild-type GR LBD (GRwt) and the stabilized F602S mutant (GRF602S) in the ligand- and coregulator-bound state (23). Building on that work, here we identify the structural pathways of allosteric transmission, including a dynamic coupling between the ligand-binding pocket and the coregulator interfaces AF-2 and tau2. We further describe how activation is manifested through the dynamic equilibrium of alternative helix 12 conformations and how this responds to binding various ligands and coregulator peptides. Combining NMR with surface plasmon resonance (SPR), we unravel the free energies of allostery between ligand-binding, coregulator-binding, and helix 12 activation and quantify the allosteric response to variations in coregulator identity and ligand structure. We demonstrate that the allosteric transmission pathway is preserved among ligands exhibiting subtle variations in structure, whereas different classes of ligands with distinct pharmacophores apparently rewire the allosteric transmission pathway, opening the possibility to design novel ligands that exert specific downstream effects.
Conformational dynamics link the ligand-binding pocket and AF-2 interface
To study the structural and dynamical consequences of ligand and coregulator binding to the LBD, we introduced 13CHD2-labeled methyl groups in methionine residues, which have favorable NMR properties (24) and are distributed throughout the LBD, including the ligand-binding pocket and the AF-2 coregulator-binding site (Fig. 1A). The 13 LBD methionines have easily distinguishable methyl signals in the 1H-13C heteronuclear single-quantum coherence (HSQC) spectrum, which we assigned by site-specific methionine to leucine replacements (23). As shown in Fig. 1 (B to D), chemical shift changes of the methionine methyl groups sensitively report on structural and dynamical differences between various ligand complexes. For example, the cyclohexyl substituent of desisobutyryl-ciclesonide (dibC) extends into a region between helices 3 and 7 (Fig. 1C), where it pushes on M560 and M639, causing significant changes in the chemical shifts of these residues (Fig. 1B). In the same region, the methyl group in the C16 position of dexamethasone (dex) changes the environment of M560 relative to that in the cortisol-bound receptor, leading to further differences in the chemical shift for this residue among the different ligand-bound forms. In contrast, the ligands are relatively conserved at the other end of the ligand-binding pocket (Fig. 1D) and the chemical shift of M601 remains in a similar position for all three complexes. However, we note that the chemical shift of M604 is distinct for the complex with cortisol, which, unlike dex and dibC, lacks a conjugated double bond at the C1-C2 position (Fig. 1D). Last, compared to the methionines described above, the chemical shift of M752 shows only minor changes in response to the structural differences of the three ligands (Fig. 1B), in keeping with its remote location at the AF-2 interface (Fig. 1A). However, these subtle chemical shift changes provide critical information on gradual shifts in the relative populations of alternative helix 12 conformations, as explained further below.
We next investigated how these ligand-specific effects might translate into altered conformational dynamics of the LBD. We performed 13C Carr-Purcell-Meiboom-Gill (CPMG) relaxation dispersion experiments (24) to directly monitor dynamic exchange between conformations in GRwt and GRF602S. Initially, we studied GRF602S in complex with dex and a peptide from the TIF2 coregulator (GRF602S-dex-TIF2). M752, located in the N-terminal part of helix 12 and interacting with the bound TIF2 peptide, shows a marked relaxation dispersion profile indicating exchange dynamics on the millisecond time scale between alternative conformations (Fig. 2A). The results are similar for GRwt-dex-TIF2 (see Fig. 3A), but the data quality is higher for GRF602S, presumably because of its greater stability and solubility (5). M565, located in helix 3 near helix 12, also shows relaxation dispersion in the GRF602S-dex-TIF2 complex, albeit of much lower amplitude (Fig. 2A). Furthermore, M601, located in the ligand-binding pocket, also exhibits enhanced relaxation through line broadening, resulting in low signal intensity (Fig. 1B). Other methionine side chains do not show signs of conformational exchange. For example, there is no significant conformational exchange for M593, which is located in helix 4 and in contact with TIF2, suggesting that this part of the AF-2 motif adopts a single conformation in the GR-dex-TIF2 complex.
To probe further the conformational exchange dynamics, we performed backbone 15N CPMG dispersion experiments (21, 25) on GRwt and GRF602S in complex with dex and the TIF2 peptide. The relaxation data provide unequivocal evidence of conformational exchange in GRwt-dex-TIF2, but broad NMR lines and poor signal-to-noise ratios leave few residues amenable to detailed analysis by CPMG experiments. The situation is improved for GRF602S-dex-TIF2, which yielded significant relaxation dispersions for 14 residues (fig. S1). The backbone data reveal conformational exchange for several additional residues surrounding the methionines showing relaxation dispersion or line broadening (M752, M565, and M601; described above). Figure 2B highlights all residues exhibiting conformational exchange on the GR LBD structure. E751, L753, and I756 in the N-terminal half of helix 12 exhibits conformational exchange, whereas the C-terminal half, which is more remote from the ligand-binding site and the bound peptide, is unaffected. E751 is in direct contact with the TIF2 peptide. I756, together with M752, forms part of the AF-2 pocket where the canonical LxxLL coregulator motif binds. L753, on the other hand, forms part of the ligand-binding pocket where it contacts W600 in helix 5. W600, in turn, is centrally placed and is adjacent to both M601 and I756. Both the indole amino group of W600 and the methyl of M601 exhibit weak exchange signals.
Together, the results reveal a tight network of dynamic interactions bridging from the ligand to the N-terminal end of helix 12 (Fig. 2B). This network delineates a pathway for dynamic allosteric communication between the ligand-binding pocket and the AF-2 interface.
Conformational dynamics link the coregulator-binding sites AF-2 and tau2
S534, V538, E542, and L544 in helix 1 all show 15N relaxation dispersion (fig. S1) and belongs to the coregulator-binding site tau2 (residues 529 to 556). Several exchanging residues at the C-terminal end of helix 3 connect this segment to the AF-2 interface (Fig. 2B). A key residue for this pathway appears to be K579, which interacts directly with the TIF2 peptide and, together with E755, forms a charge clamp that orients the coregulator motif (5). Next to K579, residues A580 and I581 also exhibit 15N relaxation dispersion. In addition, K576 and the side chain of W577 exhibit 15N exchange broadening, one helix turn below. Together, these residues form a second pathway of dynamic coupling, linking AF-2 to tau2, which suggests that coregulator binding at one of the sites might allosterically affect the conformation of the other site and thereby its coregulator affinity. These results thus provide an atomic-resolution explanation for the previous observation that the tau2-binding coregulator Hic-5 acts to stabilize higher-order GR-coregulator complexes (1, 26, 27).
GR exhibits a rapid equilibrium between alternative conformations
To quantify the exchange dynamics, we fitted a minimal, two-state exchange process simultaneously to the 13C and 15N dispersion profiles of GRF602S-dex-TIF2. The fitted parameters comprise the exchange rate, kex = 1147 ± 82 s−1; the relative population of the minor (high-energy) state, pm = 19 ± 7% (pm = 1 − pM, where pM is the major ground state); and the residue-specific chemical shift differences between the exchanging conformational states, Δδ(15N) = |δm − δM| = 0.5 to 1.2 parts per million (ppm) (fig. S2). Fitting the GRwt-dex-TIF2 data with kex and pm fixed to values similar to those obtained for GRF602S-dex-TIF2 yields wild-type Δδ values in good agreement with the results for the mutant (fig. S2), strongly indicating that the exchanging conformations are highly similar in the two protein variants; this result is expected given the remote location of the site of mutation from the AF-2 site.
The present finding that the N-terminal end of helix 12 undergoes dynamic exchange between alternative conformations in the ternary complex goes beyond previous observations from crystal structures of GR, where the helix conformation varies among different ligand- and coregulator-bound complexes. Because helix 12 forms a substantial part of the AF-2 surface, the alternative conformations can likely be distinguished by additional coregulator proteins and trigger downstream rearrangements leading to differential transcriptional activity. To this extent, the alternative conformations detected here might exhibit different levels of activation. On the basis of these observations, we formed the hypothesis that dynamic exchange between alternative conformations and their relative populations constitute a continuous dial for transcriptional activation that is tuned by the detailed structure of each individual ligand.
The dynamic equilibrium of helix 12 conformations responds to ligand and coregulator
To test this hypothesis, we next compared the 13C CPMG relaxation dispersion results for M752 of GRwt in complex with five different steroid ligands [dex, cortisol, prednisolone (pred), AZ938, and dibC; Fig. 3] and two peptide fragments from the coregulator proteins TIF2 and PRGC1. The different GRwt-ligand-TIF2 complexes show subtle variation in their dispersion profiles (Fig. 3, left), indicating that the bound ligands affect the exchange dynamics of M752, to gradually shift the relative populations of helix 12 conformations. The limited data, with M752 being the single probe, are not sufficiently precise to unequivocally quantify differences among these complexes in their exchange parameters. However, assuming that the exchanging helix 12 conformations—and, consequently, the 13C chemical shift difference between them—are identical in all complexes, the magnitude of the dispersion step (i.e., the difference between the highest and lowest relaxation rates) in Fig. 3 should increase with the value of pm. Thus, it appears that the complexes with cortisol, AZ938, and dibC (Fig. 3, B, D, and E) have lower pm values than those with dex and pred (Fig. 3, A and C). This conclusion is also supported by the 13C chemical shifts observed for M752 in the 1H-13C HSQC spectra of the GRwt-ligand-TIF2 complexes (Fig. 1B), which scale as expected with pm (δobs = Δδ pm + δM; table S1). Similar to our observations, changes in the relative populations of alternative receptor conformations were recently observed in a study using a chemical probe attached to helix 12 of PPARγ (18, 19).
Upon substituting the TIF2 peptide for a PRGC1 peptide in the GRwt-ligand-coregulator complexes, we observe that M752 no longer shows relaxation dispersion (Fig. 3), suggesting that a single conformation of helix 12 dominates in all of the GRwt-ligand-PRGC1 complexes or possibly that the chemical shift is identical in the major and minor conformations. However, the former interpretation is supported by the result that M565 also does not show any dispersion in GRwt-dex-PRGC1, contrary to the case in the TIF2 complex (Fig. 2A). These data indicate that different coregulators with unique amino acid sequences vary in their ability to drive ligand-specific responses and that the thermodynamic coupling is distinct (see further below).
Agonist efficacy in functional assays follows the population of helix 12 minor conformation
To investigate whether the observed population shift of helix 12 conformations might correlate with the functional response of GR, we measured agonist efficacy in a transactivation reporter gene assay (Fig. 4) (28). The top effect relative to dex (100%) varies from 77% for dibC and cortisol, 86% for pred, to 90% for AZ938. Despite being a partial agonist, dibC binds GR with high potency. Conversely, dex and pred have relatively high efficacies but weaker potencies for GR. These results underscore that efficacy does not directly relate to potency. However, the functional efficacy matches qualitatively with the pm values from the NMR relaxation experiments, where the GRwt complexes with dibC and cortisol both appear to have lower pm than the compounds with higher efficacy. The cortivazol analog AZ938 forms a notable exception with high functional efficacy combined with a low pm value. AZ938 is distinct from the other steroid ligands in that the A-ring keto-moiety has been replaced with a fluorophenyl diazole (Fig. 3D). Structural studies have shown that this motif induces a sizeable rearrangement of the ligand binding site, specifically affecting the conformations of Q570 and R611 such that a new pocket opens up between helices 3 and 5 (29). These structural changes may alter the allosteric communication pathway to the AF-2 site and affect the active helix 12 conformation, as also suggested by the chemical shift of M752 (table S1). On the basis of the results for the remaining four congeneric steroid ligands, we suggest that the minor conformation of helix 12 might represent the active receptor, in keeping with the hypothesis formulated above.
Coregulator-binding affinities show significant variation among ligand-LBD complexes
To get insights into the thermodynamic coupling between ligand and coregulator binding to GR-LBD, we used SPR to measure the kinetic on- and off-rate constants, kon and koff, which yield the equilibrium association constant, Ka = kon/koff. We measured these parameters for both PRGC1 and TIF2 peptide binding to the five different GR-ligand complexes, using both GRwt and GRF602S (Fig. 5 and fig. S3). The results show that kon, koff, and Ka vary among the different complexes. The kinetic rate constants determined by SPR also indicate that the off-rates of the coregulator peptides are too slow to give rise to the NMR relaxation dispersion profiles, thereby providing independent evidence that the latter are caused by conformational exchange within the ligand- and coregulator-bound states rather than exchange between free and bound states.
In comparing Ka among the different ligand-coactivator pairs, we find that both GR constructs have higher affinity for PRGC1 than for TIF2, irrespective of which ligand is bound. In addition, GRF602S exhibits higher affinities than GRwt in every case (table S2). The variation in Ka among the different GR-ligand complexes tends to be greater for TIF2 than for PRGC1. Furthermore, the variation is greater in GRwt than in GRF602S for both coregulators.
Ligand structure tunes the allosteric free energy and can affect the transmission pathway
Allosteric coupling between the ligand- and coregulator-binding sites has been observed previously for several nuclear hormone receptors (8, 30, 31) but has rarely been quantified. To interpret in a unified and quantitative manner the combined experimental results from SPR and NMR in terms of allosteric couplings, we used a three-site statistical thermodynamic model for ligand (L) and coregulator (P) binding and transcriptional regulation (A) involving the conformational change of helix 12. This model involves three states for each site, corresponding to two unbound states, which are either binding incompetent or binding competent, and the bound state. To interpret the present data, it suffices to include only a subset of all possible states (Fig. 6A; see fig. S4 and Supplementary Text for a full description of the model). The probability of each state is a function of the free energies of forming the binding-competent conformation at a given site (ΔGi), binding to a site (Δgbi), and allosteric coupling or cooperativity (Δgij) between sites (32–34). It has been shown that changes in allosteric effects can arise from population shifts within the ensemble of states as a consequence of changes in ΔGi without any changes in Δgij (34). Here, we focus on isolating the free energies of site-to-site couplings, Δgij.
It is informative to calculate the ratio of binding affinities for pairs of different ternary complexes, because it reveals how allosteric couplings between sites vary with ligand and coregulator peptide identity. The free energy of binding a coregulator peptide to a GR-ligand complex is given by ΔG = ΔGP + ΔgLP + ΔgLA + ΔgbP. It follows that the difference in ΔG between two different GR-ligand complexes is ΔΔG = −RT ln[Ka(Lm, Pi)/Ka(Ln, Pi)] = ΔΔgLP + ΔΔgLA. However, the minor variation in pm (i.e., the relative population of active helix 12 conformations detected by NMR) between the different ternary complexes indicates that the variation ΔΔgLA contributes relatively little to variations in coregulator-binding affinity (fig. S5), which instead is dominated by ΔΔgLP. Thus, the SPR results can be interpreted to extract differences between ligands in their strength of allosteric coupling to coregulator binding, ΔΔgLP.
Comparing complexes with different ligands but the same coregulator, the affinity ratios Ka(Lm,Pi)/Ka(Ln,Pi) cover the range 2.3 to 11.7 for GRwt-ligand-TIF2 and 1.2 to 4.0 for GRwt-ligand-PRGC1 (table S2) when using as reference the GRwt-cortisol-coregulator complex, which has the weakest affinity in both cases. Figure 6B shows a comparison of the relative allosteric coupling between the L and P sites, ΔΔgLP, for TIF2 and PRGC1 across the series of five ligands. The cooperativity ΔgLP varies significantly (by factors of 2 to 3) with the identity of both ligand and coregulator, although the standard errors are relatively large for PRGC1. Nonetheless, the datasets for the two coregulators are significantly correlated. This result suggests that the mechanism of allosteric transmission between the ligand-binding pocket and the AF-2 interface is highly related for the two coregulators.
GRF602S shows higher affinity for coregulators, compared to GRwt (Fig. 5 and fig. S3), but in this case, the ratios Ka(Lm,Pi)/Ka(Ln,Pi) are considerably lower for both TIF2 and PRGC1, with the maximum ratio being less than 3 (table S2). This result is explained by stabilization of the binding-competent conformation of site P, ΔGP(GRF602S) < ΔGP(GRwt) (see fig. S5), in agreement with previous observations that the F602S mutation increases hormone responsiveness (8, 35). However, the mutation does not seem to have any greater impact on the cooperativities between sites, ΔgLP or ΔgPA (fig. S5). Figure 6C shows the relative allostery, ΔΔgLP, in GRF602S, which can be directly compared with the corresponding data for GRwt (Fig. 6B). The data for the four complexes with ligands containing the cyclic ketone moiety (cortisol, pred, dex, and dibC) are highly correlated and fall on a straight line. This result reinforces the notion that for congeneric ligands, coregulator binding involves the same allosteric pathway between the ligand pocket and the AF-2 interface irrespective of the coregulator identity. Furthermore, the rank order among these complexes is preserved between GRwt and GRF602S, indicating that this conclusion holds for both protein variants, as also indicated by the CPMG relaxation dispersion data (Fig. 2 and fig. S2). The divergent result for AZ938 is likely because of its distinct structure, which results in conformational rearrangements in the AZ938 complex (described above) and apparently leads to a modified communication pathway from the ligand-binding pocket to the AF-2 site and an alternative helix 12 conformation (table S1). The results presented in Fig. 6 (B and C) thus reveal how these ligand-induced perturbations affect the free energy of allosteric coupling, ΔgLP.
Differential allosteric coupling of helix 12 activation with ligand or coregulator binding
On the basis of the results from the CPMG relaxation dispersion experiments and the functional assay, we tentatively interpret the major and minor conformations of helix 12 as transcriptionally “inactive” and “active,” respectively, in line with conclusions reached for PPARγ (18–20). In the framework of the statistical thermodynamic model, the population ratio pm/pM depends on ΔGA, ΔgLA, and ΔgPA, but the variation in pm/pM among GRwt-ligand-TIF2 complexes depends only on ΔgLA. Thus, the NMR experiments probe the relative allosteric effect on the transcriptionally active conformation elicited by different ligands. Qualitatively, the NMR data show that ΔgLA depends on the ligand identity in the GRwt-ligand-TIF2 complexes (Fig. 3), similar to our conclusions regarding ΔgLP. A semiquantitative interpretation indicates that pm/pM varies between 0.1 for dibC and cortisol to 0.2 for dex and pred, which corresponds to ΔΔgLA ≤ 0.7 RT (fig. S7). Therefore, while the GR-dibC complex has the highest affinity for coregulator peptides and exerts the greatest cooperativity between the ligand- and peptide-binding sites, ΔgLP (Fig. 6, B and C), it displays weaker cooperativity between ligand binding and receptor activation, ΔgLA.
Next, comparing differences in pm/pM between TIF2- and PRGC1-bound ternary complexes, we isolate ΔgPA. The lack of detectable conformational exchange in the GRwt-ligand-PRGC1 complexes (Fig. 3) suggests that a single conformation dominates in this case, indicating that PRGC1 shows either much stronger or weaker allostery, ΔgPA, compared to TIF2. Calculations suggest that PRGC1 has a stronger cooperativity ΔgPA than TIF2 by a factor of at least 2.5 (fig. S8); this result is similar to what is found for ΔgLP, as discussed above (fig. S6). In the GRwt-ligand-PRGC1 complexes, any ligand-dependent variation in ΔgLA is unobservable because of the high level of activation. The marked difference between TIF2- and PRGC1-bound GR-LBD in their conformational dynamics highlights how the specific amino acid sequence of the coregulator can modulate the receptor’s response to ligand binding and thereby affect downstream pathways.
Our results provide a unique view of the allosteric transmission pathway through which ligands modify the conformational dynamics of the receptor. While recognizing that NMR chemical shifts do not necessarily respond to conformational changes, potentially leaving “blind” spots in our analysis, we believe that uniform 15N labeling provides a sufficient number of probes to offer a satisfactory view of exchange dynamics throughout the LBD. The CPMG relaxation dispersion data directly visualize a network of residues undergoing concerted dynamics that couples the ligand-binding site to the AF-2 and tau2 sites and helix 12 activation. Residues in the binding pocket connect the ligand to the N-terminal end of helix 12 and the C-terminal end of helix 3 and onward to helix 1, providing the means for the receptor to tailor the coregulator interactions in response to the ligand pharmacophore. The network identified here shows partial agreement with previous predictions derived from a statistical analysis of coevolution across the nuclear hormone receptor family (36) but differs in several respects from conclusions reached from mutagenesis and modeling studies (8, 37) in that our data reveal a contiguous network that does not seem to include helix 11 as the primary conduit of communication. It is possible that these differences reflect the detailed variation in structural homology among nuclear hormone receptors.
We used the conformational dynamics of the M752 methyl group as the principal indicator of receptor activation in 10 different ternary complexes. While the use of a single probe limits the quantitative interpretation of the dynamic data alone, the full picture emerging from the combination of 13C relaxation dispersion and chemical shift data, SPR, functional assays, and statistical thermodynamic modeling is consistent. Receptor activation is strongly dependent on the coregulator identity, and our model reveals how this, in turn, arises from ligand- and coregulator-dependent cooperativity between sites. Because coregulator expression profiles vary among cell types, this result indicates that the potential to achieve ligand-specific effects will vary between different tissues.
The mechanistic and quantitative model presented here describes how binding of various ligands and coregulators differentially affect receptor allostery to shift the dynamic ensemble between alternative conformations with different regulatory properties. Our results complement previous reports on allostery between the N-terminal domain and the DNA binding domain of GR, which have revealed how allostery varies among isoforms or mutational variants that destabilize individual domains and generate population shifts within the ensemble, leading to either activation or repression of transcription (11). The observed variation in the population of active conformations amount to regulation by a continuous dial rather than a binary structural switch, a phenomenon explaining key concepts such as graded response and biased signaling.
Our work further shows that the allosteric pathway is maintained for a given class of ligands with common pharmacophore, although subtle changes in ligand structure lead to variation in the allosteric free energy. By contrast, substantial changes in ligand structure can affect the interaction network within the protein, thereby rewiring the allosteric transmission pathway. Thus, ligands with different chemical scaffolds and pharmacophores may create new opportunities to influence the structural ensembles and communication pathways in unique ways that will be important to consider when developing mechanistically differentiating GR ligands (3, 13, 17, 28, 38).
MATERIALS AND METHODS
Expression and purification GR LBD
Construct design, expression, and purification of wild-type GR LBD (residues T529 to K777) and the mutant F602S were carried out as described (23). Expression of samples labeled with 12CβD212CγD213 CHD2 methionine (24), synthesized in-house as described below, was carried out using the PASM-5052 autoinduction medium (39).
Synthesis of 12CβD212CγD213CHD2 methionine
(S)-2-(tert-Butoxycarbonylamino)-4-(((R)-3-(tert-butoxycarbonylamino)-3-(1,1,2,2 2H4)-carboxypropyl)disulfanyl)-(3,3,4,4,2H4)-butanoic acid. dl-Homocystine-(3,3,3′3′,4,4,4′4′2H8) (1.00 g, 3.62 mmol) was dissolved in 10% sodium carbonate (24.0 ml) and dioxane (21.0 ml) at 0°C. Di-tert-butyl dicarbonate (1.74 g, 7.96 mmol) was added portionwise, and the reaction mixture was allowed to reach room temperature and stirred for 18 hours. The pH was adjusted to 4 with 10% citric acid and extracted with ethyl acetate (3 × 50.0 ml). The organic layers were combined, dried (MgSO4), and concentrated under reduced pressure. This resulted in 1.72 g (quantitative yield) (S)-2-(tert-butoxycarbonylamino)-4-(((R)-3-(tert-butoxycarbonylamino)-3-(1,1,2,2,2H4)-carboxypropyl)disulfanyl)-(3,3,4,4, 2H4)-butanoic acid as a white solid. [500 MHz, dimethyl sulfoxide (DMSO)] δ 1.38 (18 H, s), 3.98 (2 H, d), 7.15 (2 H, d), and 12.55 (2 H, bs). Liquid chromatography–mass spectrometry (LC-MS): mass/charge ratio (m/z) 477 [M+H]+.
2-(tert-Butoxycarbonylamino)-4-(13C, 2H2)-(methylthio)-(3,3,4,4,2H4)-butanoic acid. (S)-2-(tert-Butoxycarbonylamino)-4-(((R)-3-(tert-butoxycarbonylamino)-3-(1,1,2,2,2H4)-carboxypropyl)disulfanyl)-(3,3,4,4,2H4)-butanoic acid (1.52 g, 3.20 mmol) was dissolved in 7% ammonium bicarbonate (12.0 ml) and DMSO (4.00 ml). The mixture was degassed, and (2S,3S)-1,4-dimercaptobutane-2,3-diol (0.540 g, 3.52 mmol) was added under inert atmosphere. The reaction was stirred at room temperature for 0.5 hours. Iodomethane-(13C, 2H2) (0.800 ml, 12.8 mmol) was added dropwise, and the reaction mixture was stirred for 2.5 hours. Dichloromethane (25.0 ml) was added to the reaction, and the organic layer was washed twice with 0.5 M hydrochloric acid (20.0 ml). The organic phase was dried (MgSO4) and concentrated under reduced pressure. The compound was purified by preparative high-performance liquid chromatography (HPLC) on a Kromasil C8 column [10 μm 250 × 50 inside diameter (ID) mm] using a gradient of 15 to 55% acetonitrile in water/acetonitrile/formic acid 95/5/0.2 buffer over 25 min with a flow of 100 ml/min and detection by ultraviolet at 210 nm. This resulted in 0.370 g (45%) of 2-(tert-butoxycarbonylamino)-4-(13C,2H2)-(methylthio)-(3,3,4,4,2H4)-butanoic acid as a white solid. (500 MHz, DMSO) δ 1.38 (9 H, s), 1.99 (1 H, d), 3.98 (1 H, d), 7.11 (1 H, d), and 12.48 (1 H, bs). LC-MS: m/z 255 [M-H]−.
2-amino-4-(13C, 2H2)-(methylthio)-(3,3,4,4,2H4)-butanoic acid TFA salt. 2-(tert-Butoxycarbonylamino)-4-(13C, 2H2)-(methylthio)-(3,3,4,4,2H4)-butanoic acid (0.340 g, 1.31 mmol) was dissolved in dichloromethane (1.00 ml) in a 25.0-ml flask. 2,2,2-Trifluoroacetic acid (TFA) (0.800 ml, 10.8 mmol) was added dropwise to the mixture at room temperature under inert atmosphere. The reaction was stirred for 2 hours. Water (4.00 ml) was added to the mixture and lyophilized, which was repeated two times, and resulted in 0.210 g (41%) of 2-amino-4-(13C, 2H2)-(methylthio)-(3,3,4,4,2H4)-butanoic acid as a TFA salt and a clear oil that solidified over time. (500 MHz, DMSO) δ 2.01 (1H, d), 3.99 (1 H, s), and 8.27 (2 H, bs).
NMR sample preparation
The protein was transferred at low concentration (<0.03 mM) to a 20 mM phosphate buffer [10% (v/v) D2O] (pH 7.5) supplemented with 1% CHAPS, 2 mM dithiothreitol (DTT), 50 μM dex, 0.02% NaN3, and protease inhibitor using a PD10 desalting column (GE Healthcare). After adding a suitable coactivator peptide, corresponding to a fragment of nuclear receptor coactivator 2 (KENALLRYLLDKDD; TIF2) or peroxisome proliferator–activated receptor γ coactivator 1-α (PPQEAEEPSLLKKLLLAPANT; PRGC1), the sample was concentrated to 0.3 mM. The exchange of LBD-bound dex for other ligands was conducted in the absence of stabilizing CHAPS and coactivator peptide. Excess ligand was dissolved in 50 mM tris (pH 9) containing 10% DMSO (to increase ligand solubility), and dex-bound LBD was exchanged twice into this buffer using a PD10 desalting column. Last, the protein was transferred to the aforementioned DMSO-free phosphate buffer containing CHAPS.
NMR resonance assignment
Assignment of the 1H/13C/15N backbone and 1H/13C methionine methyl resonances has been reported previously for GRwt and GRF602S (23).
Relaxation dispersion measurements
Transverse relaxation-optimized spectroscopy (TROSY)–based relaxation-compensated 15N CPMG experiments (25, 40, 41) were implemented using interleaved acquisition of various CPMG frequencies before sampling the indirect 15N dimension. Data were acquired at a temperature of 298 K and static magnetic field strengths of 14.1 and 18.8 T. 13C CPMG experiments on 12CβD212CγD213CHD2 methionine-labeled samples were performed as described (24) at a temperature of 298 K. Experiments on GRwt with TIF2 were acquired at static magnetic field strengths of 14.1 and 20.0 T, whereas experiments on GRwt with PRGC1 were acquired at 18.8 T, and experiments on GRF602S with TIF2 were acquired at 14.1 and 18.8 T. Relaxation dispersion data were acquired using a linear two-point approximation of the exponential decay (42) at CPMG frequencies (νCP) of 0 (reference), 67, 100, 133, 200, 267, 300, 333, 400, 500, 600, 800, and 1000 Hz, with duplicates acquired at 0, 200, 400, and 600 Hz. Spectra were processed using Topspin 3.4 (Bruker) and analyzed using NMRView (43) or Sparky (44). The dependence of the transverse relaxation rate constant (R2) on νCP was fitted to a two-state exchange process using MATLAB (MathWorks Inc.) to yield the exchange rate (kex), as well as the populations (pi) of the exchanging states and the chemical shift differences (Δδ) between these, as described (21, 45). Error estimates were based on Monte Carlo simulations using 1000 samples (46).
Surface plasmon resonance
All SPR measurements were run on Biacore 3000 (GE Healthcare) using as running buffer 10 mM Hepes, 50 mM NaCl, 0.5 mM Tris(2-carboxyethyl)phosphine (TCEP), 0.05% Tween 20 (pH 7.6). To avoid mass transport limitations that may obscure the kinetic analysis/fitting, biotinylated coregulator PRGC1 or TIF2 was immobilized on a two-dimensional carboxylated dextran chip (Xantec) at 270 and 380 response units (RU), respectively, using covalently immobilized NeutrAvidin (2200 RU) as a linker.
GR LBD was incubated with respective ligand at 50 μM and injected for 2 min at 400, 200, and 100 nM in running buffer. After each injection, the surface was regenerated with running buffer supplemented with 0.05% SDS. For all binding curves, the association and dissociation phases were fitted using a typical Langmuir 1:1 interaction model to extract the association (kon) and dissociation rate constant (koff), which was further used to estimate the binding affinity constant according to Ka = kon/koff. As a primary validation of the data, the affinities extracted from kinetic fitting were compared with the equilibrium response (i.e., binding coverage) for the different ligands, which were in excellent agreement.
GR reporter gene assay
To identify compounds showing agonist activity at human GR, we used a semi-automated reporter assay based on GRE-lacZ transfected ChagGo-K-1 cells. The induction of lacZ gene expression through the binding of ligand-bound GR to GRE in the promoter of lacZ is measured as the up-regulation of the β-galactosidase activity through a color reaction (change in absorbance).
Stably transfected cryopreserved cells were suspended in RPMI 1640 (Gibco) with 10% fetal bovine serum (Hyclone), 1× nonessential amino acids (Gibco, 100×), and 1 mM sodium pyruvate (Gibco). The cells were then seeded, 50,000 cells per well in 200 μl in a 96-well plate (Costar 3595), and incubated for 24 hours at 37°C, 5% CO2, and 95% humidity. Compounds were serially diluted, 1:3 dilutions, to give a 10-point concentration response curve. Compounds and controls were then added to the 96-well cell plate in a volume of 1 μl with a 1 or 0.2 mM start concentration, which gives a final assay start concentration of 5 and 1 μM, respectively. As 100% control, dex was used at a final concentration of 1 μM, and as 0% control, DMSO (0.5%) was used. Incubation at 37°C continued for another 24 hours. Cell medium was removed, and the plate was washed once with phosphate-buffered saline (PBS), and 50 μl of 0.1% Triton X-100 (Chemicon) was added to each well and incubated at room temperature for 20 min, followed by addition of 40 μl of reaction mix per well containing 2.5 mM MgCl2 (Sigma), 0.8% (v/v) β-mercaptoethanol (Sigma), and ortho-nitrophenyl-β-galactoside (2.2 mg/ml; Sigma) in 0.1 M sodium phosphate buffer (pH 7.5), and incubated for 1 hour at 37°C. The reaction was stopped by adding 100 μl per well of stop buffer, 300 mM glycine (Sigma), and 15 mM EDTA (Sigma), pH adjusted to 11.3. Absorbance was measured at 420 nm in PHERAstar Plus3. Raw data output was analyzed in Genedata Screener software, a dose response curve was generated using a four-parameter logistic fit, and the EC50 (median effective concentration) values were calculated.