A ternary switch model governing ERα ligand binding domain conformation

Abstract

The transcription factor estrogen receptor α is the primary driver of ER+ breast cancer progression and a target of multiple FDA-approved anticancer drugs. Ligand-dependent activity of ERα is determined by helix-12 conformation within the ligand binding domain. However, how helix-12 transitions from an unliganded (apo) state to active (estrogen-bound) or inactive (SERM/SERD-bound) states remains unresolved. Here, we present the crystal structure of an apo estrogen receptor α ligand binding domain from the teleost Melanotaenia fluviatilis, revealing a third distinct helix-12 conformation. Structural mass spectrometry and molecular dynamics simulations reveal that apo helix-12 is maintained in a stable and distinct conformation prior to ligand binding. Clashes between ligand and evolutionarily conserved residues L525, L536 and L540 displace helix-12, to promote activation or inactivation of the receptor. The crystal structure further reveals that breast cancer-associated mutations, Y537S and D538G, disrupt residue contacts critical for stabilising apo helix-12 conformation. We propose a model whereby helix-12 functions as a ternary molecular switch to determine receptor activity. These findings provide critical insights into the ligand-dependent and -independent regulation of estrogen receptor α and have significant implications for therapeutic intervention.

Estrogen receptor α is a primary driver of ER+ breast cancer and reproductive development. Here, the structure of the apo state is reported, providing a revised model for ligand-dependent and -independent regulation of receptor function.

Introduction

Oestrogen receptor α (ERα) is a ligand-responsive nuclear hormone receptor transcription factor that regulates physiological and cellular processes across vertebrates including reproduction, cellular proliferation and survival¹. In line with this, ERα signalling has been identified as a key driver of breast cancer development and progression, of which at least 70% are ER + ^2,3. Typically, binding of ligands to the ERα LBD determines the transcriptional outcome of the receptor through stabilising discrete conformational states of the C-terminal helix, H12 (Fig. 1a)⁴. The central model for ligand-dependent activation of nuclear receptors is that ligand binding switches H12 from a dynamic state to one that is stable^5,6. However, there is evidence that this is not the case for the ERα LBD. Hydrogen-deuterium exchange coupled mass-spectrometry (HDX-MS) analysis of the human ERα LBD shows that ligand binding shifts protein structure from a dynamic ensemble toward a stabilised state. Yet, no significant differences in the deuterium uptake of H12 occurs, supporting that H12 exists in a stable, discrete conformation prior to ligand binding^7–10. A clear understanding of the apo H12 conformation and how it is modulated holds enormous benefit for elucidating mechanisms of disease and pharmacological development, in addition to fundamental biology.

Fig. 1. X-ray crystal structure of the rfERα LBD reveals a distinct H12 conformation.

a Cartoon schematic illustrating that the conformation of H12 prior to ligand binding is unknown. Left: hERα LBD bound to E2 (PDB: 3UUD); right: hERα LBD bound to the SERM 4-hydroxytamoxifen (PDB: 3ERT). In the schematic, H12 is coloured pink while helices 3, 4, and 5, which form the AF2, are coloured orange. E2 and 4-hydroxytamoxifen are shown as sticks and coloured yellow. b Amino acid sequence alignment comparing hERα and rfERα LBDs. Ligand binding pocket (LBP) residues are shaded blue and activation function-2 (AF2) residues are shaded orange. c X-ray crystal structure of the apo rfERα LBD captured as a ‘head-to-head’ homodimer at 2.0 Å resolution (PDB: 9MDV). The structure reveals a distinct orientation of H12 (shown in pink) which assumes a vertical orientation between H3 and H11, resulting in partial obscurement of the AF2. d Molecular side chain interactions between H3, H11 and H12 in apo (left) and active E2-bound (right; PDB: 9D8Q) states, highlighting formation of the hydrophobic cluster. Residue side chains are shown as sticks and labelled. Highlighted in red are residues Y537 and D538 which are commonly mutated in ER+ breast cancer. e Top-down perspectives of the ligand binding pocket (LBP) of apo (left) and E2-bound (right) structures showing side-by-side comparisons of pocket structure. Important residue side chains are shown as sticks and labelled. Water molecules are shown as red spheres; E2 is shown as sticks and coloured yellow. f Alignment of LBP residues of apo and active E2-bound structures. Apo residues are coloured blue, while active E2-bound residues are coloured green. E2 is shown as sticks and coloured yellow.

In the context of receptor activation, agonists of ERα, such as the primary sex hormone estradiol (E2), bind to the ligand binding pocket (LBP) of the ERα LBD and induce a structural rearrangement of H12, forming the activation function-2 (AF2) surface along with H3 – H5 (Fig. 1a). Once formed, the AF2 allows specific recruitment of coactivator proteins to the ERα LBD via a conserved LxxLL motif, leading to activation of transcription¹¹. Additionally, it has been identified that mutations within H12 of the ERα LBD such as Y537S and D538G, which are prevalent in ER+ breast cancer, can circumvent the requirement of agonist binding to form the AF2. This leads to constitutive activation of the receptor, driving tumour development and disease, and in some instances endocrine resistance^2,9,12–15.

Due to the role of ERα in breast cancer, extensive research has been performed to identify ligands which disrupt formation of the AF2 and repress ERα signalling. This has led to the discovery of frontline therapies termed selective oestrogen receptor modulators/degraders (SERMs/SERDs), such as 4-hydroxytamoxifen, elacestrant and fulvestrant^10,16–20. While SERMs/SERDs bind within the LBP they induce a different structural rearrangement, positioning H12 in a stable conformation that remodels the AF2 to antagonise transcriptional activity via several mechanisms including corepressor recruitment, immobilisation within the nucleus, and degradation^4,21–25. Although the structure of the ERα LBD in these active and repressed states has been thoroughly characterised^4,26–29, there remains no structural information regarding the apo conformation of H12 in WT ERα that exists prior to ligand binding, or how H12 mutations such as Y537S and D538G escape this conformation to constitutively activate the receptor. Indeed, all present hypotheses for activation and inactivation of the receptor have been derived through retrospective analysis of ligand/peptide bound forms of the ERα LBD due to the historical challenges of apo state crystallisation.

The analysis of non-human protein homologues has proven valuable in understanding protein structure and function (e.g., telomerase reverse transcriptase TEN domain from the protozoan Tetrahymena thermophilia, or transcriptional machinery from yeast)^30–32. In a related study, we report the crystal structure of an orthologous ERα LBD bound to E2 and a human SRC coactivator peptide (PDB: 9D8Q) from the Australian teleost Melanotaenia fluviatilis (Murray River rainbowfish; rfERα)³³. The crystal structure exhibits conserved tertiary and quaternary structure, E2 binding mode and AF2-coactivator peptide interactions compared with the hERα LBD (Supplementary Fig. S1a–d). In addition, amino acid sequences constituting the LBP, AF2 and H11-H12 are near identical between the two orthologs, which share high sequence identity (63.75%) and sequence similarity (76.67%) across the entire LBD (residues 310–548) (Fig. 1b). Therefore, given the structural homology to the human ortholog and evolutionary conservation within regulatory regions, we aimed to use the rfERα LBD as a model for investigating ligand-dependent and -independent regulation.

Here, we report a high-resolution crystal structure of apo rfERα LBD, capturing a third distinct H12 conformation that has not been previously observed. Combined with extensive biophysical and functional validation, this structure reveals that H12 is stable in the apo conformation and physically modulated by ligands upon binding, which ultimately drive transcriptional outcome. Introducing breast cancer-associated mutations, Y537S and D538G, disrupts critical contacts required for stabilising the H11-H12 loop and confers constitutive activity. Our findings and comparative analysis allow us to propose a model where H12 functions as a ternary molecular switch that is modulated by ligand binding and destabilised by constitutively activating mutations, adopting three distinct states. Comparative structural and evolutionary analyses of ER and other nuclear receptors (NRs) challenge the notion of a universal activation mechanism. This study offers atomic-level insights into ligand-mediated modulation of a key transcription factor and the structural basis of cancer-driving mutations, with implications for therapeutic design.

Results and discussion

X-ray crystal structure of the apo rfERα LBD reveals a unique H12 conformation

We initially attempted to structurally characterise the apo wildtype human ERα LBD, which has been historically intractable to crystallisation. While a protocol to acquire pure, highly concentrated apo protein (>40 mg/mL) was developed, we were unable to obtain crystals. Instead, we successfully solved the crystal structure of the wildtype apo ERα LBD from an Australian teleost fish, Melanotaenia fluviatilis (Murray River rainbowfish; rfERα), previously studied by our laboratory³³, as a homodimer at 2.0 Å resolution (Fig. 1c and Table S1). Importantly, residues constituting the AF2 and H12 are highly conserved in hERα and rfERα orthologs. The crystal structure exhibits an overall global homology with the active E2-bound form (PDB: 9D8Q), with an RMSD (root mean square deviation) of 0.775 Å between monomers. However, H12 adopts a conformation uniquely distinct from that of the active state (Fig. 1d). In the active state, H12 rests perpendicular to H3 and H4 forming the lower half of the AF2 interface. In contrast, in the apo state, H12 is orientated vertically and wedges between H3 and H11, enclosing the LBP and partially masking the AF2 with its C-terminal end (Fig. 1d). Analysis of the interaction between H12 and H3/H11 reveals that residues L536 and L540 of the H12 LxxLL motif interact with M343 and T347 (H3), W383 (H5), and L525 (H11) to form a buried hydrophobic cluster. Externally, π-stacking between Y526 and Y537 and a salt bridge between K529 and D538 stabilise H12 in the vertical orientation. Together, these interactions shorten the H11-H12 loop by extending the α-helical structure of H12 and stabilise the apo conformation.

This distinct H12 conformation also results in significant changes within the LBP relative to the E2 bound receptor LBD (Fig. 1e and Supplementary Fig. 2a). For apo rfERα, the LBP of one monomer is occupied by three waters, and the other is occupied by a glycerol molecule, although no clear accessible solvent channels were observed (Supplementary Fig. 2b). In the apo crystal structure, H11 and H12 undergo substantial restructuring compared to the active conformation. The vertical orientation of apo H12 pries H3 and H11 apart, with the latter rotating outward into the solvent. This movement causes H524 (which forms a hydrogen bond with the second hydroxyl group of E2) to flip out of the LBP, with its position instead occupied by L525 (Fig. 1f). Structural alignment of the two states reveals that E2 would clash with L525, L536, and L540 of the apo receptor. However, the residues located on H5, the β-sheet, and H7 remain largely unchanged, except for E353 (H3) and R394 (H5), which form an ionic bond in the absence of ligand. The structural conservation of these residues between the two states suggests an important contribution to ligand binding by mediating recognition of the steroidal scaffold of the hormone. Taken together, these findings demonstrate that ligand binding physically modulates H12 conformation.

H12 maintains a third discrete conformation in apo rfERα LBD

To validate our apo crystal structure and suitability as a model for hERα, we conducted biophysical experiments to compare oligomeric state and overall solution structure of apo hERα and rfERα LBDs. Native mass-spectrometry (native-MS) analysis of apo rfERα and hERα LBDs revealed that both receptors predominantly form homodimers consistent with the crystal structure, with a minor population of monomers also detected (Fig. 2a). Orthogonal validation using inline size-exclusion chromatography small-angle X-ray scattering (SEC-SAXS) revealed statistically similar scattering profiles between the two orthologs (CorMap $P$ < 0.01; Bonferroni corrected³⁴), and SAXS molecular envelopes fitted to the crystal structure of the apo rfERα LBD demonstrated conservation of oligomeric state and homodimer structure in solution (Fig. 2b, c and Supplementary Fig. 3; Tables S2). These data confirm that the homodimeric quaternary structure observed in the crystal lattice is conserved in solution for both orthologs.

Fig. 2. Conformational landscape of the rfERα LBD.

a Native mass-spectrometry analysis of apo rfERα LBD (blue) and apo hERα LBD (grey) reveals the receptors predominantly form homodimers in the absence of oestrogen (E2). b Primary small-angle X-ray scattering (SAXS) profiles of the apo rfERα LBD (blue) and apo hERα LBD (grey) represented as log I(q) plot versus q plot, and c dimensionless Kratky plot demonstrating that protein samples are folded during SAXS data collection. d Consolidated deuterium-uptake percentage differences ($\%\Delta$D₂O) between apo and active E2-bound rfERα LBD obtained from hydrogen-deuterium exchange mass-spectrometry (HDX-MS) analysis mapped to the active E2-bound structure. The data show that H1 and H3 undergo significant stabilisation (>5% reduction in deuterium uptake) upon E2 binding, alongside H5, the β-sheet, H6, H7 and H11. Notably, deuterium uptake of H12 does not significantly differ between states. e Time-resolved percent D₂O uptake of representative H12 peptide (residues 541-553) for the apo rfERα LBD (blue) and when bound to E2 (green) at 10 s, 30 s, 60 s, 300 s, 900 s and 3600 s timepoints. All datapoints are shown and error bars represent the mean and standard deviation of at least two replicates. f Violin plot comparing the distribution of root-mean squared deviation (RMSD; in nm) of H12 (residues 537-546) between apo and active E2-bound states calculated from technical triplicate 5 µs all-atom molecular dynamics trajectories (n = 50,001 frames/trajectory). The violin plot box indicates IQR (25^th – 75^th percentile), centre marker is median, whiskers extend to 1.5x IQR. g As in (f) but comparing aggregate solvent accessible surface area (SASA; in nm²) of H12 residues. h Re-weighted free energy landscapes ($G/kT$) of apo (left) and active (right) trajectories derived from Markov state modelling analysis of the ensembles. The free energy values ($G/kT$) are projected onto the first two components of the time-structured independent component analysis (TICA). Approximate locations of metastable macrostates for apo (n = 3) and E2-bound ensembles (n = 4) are labelled numerically. Source data are provided as a Source Data file.

To investigate changes in tertiary dynamics upon E2 binding, we performed HDX-MS analysis of the rfERα LBD. Differential deuterium uptake revealed significant stabilisation (decrease in deuterium exchange) of H1, H3, H5, the β-sheet, H6, H7, and the base of H11 upon E2 binding the apo receptor (Fig. 2d and Supplementary Fig. S4). In contrast, H12 showed minimal change (<5%) in deuterium uptake/exchange (Fig. 2d, e), consistent with our structural data capturing H12 in two distinct, stable conformations representing the apo and E2-bound states of the rfERα LBD³³. These results align with previous HDX-MS analyses of the hERα LBD, which showed stabilisation of the same regions upon E2 binding, while the dynamics of H12 remained unchanged^7–10. All-atom molecular dynamics (MD) simulations of the rfERα LBD demonstrated that the RMSD and solvent-accessible surface area (SASA) of H12 were comparable between the apo and active E2-bound state (Fig. 2f, g). Moreover, a comparison of average Cα root-mean square fluctuation (RMSF) per residue agreed with the HDX-MS profile of the LBD, showing stabilisation of H1, H3, H5, β-sheet and H11 but no significant change in H12 dynamics (Supplementary Fig. S5a). Comparison of backbone RMSD across all residues, and H12 RMSD between apo rfERα LBD trajectories and apo hERα LBD trajectories (generated from a homology model), demonstrated that H12 remains relatively stable in the apo conformation with the LxxLL motif buried from solvent (Supplementary Fig. S6 and S7). The agreement between our HDX-MS and MD data is notable given the distinct timescales and methodologies. This is consistent with previous works detailing how atomistic simulations can afford mechanistic insight into the factors influencing deuterium exchange across biological timescales^35,36. For example, the lack of significant difference in deuterium exchange of H12 between states is supported by consistent RMSD and SASA values between apo and E2-bound simulations (Fig. 2e, g). These correlations would suggest that local changes in dynamics between states are predictive of the global stabilisation and solvent accessibility detected by HDX-MS.

Together, these data validate the apo crystal structure (show the agreement between the crystal structure and solution structure) and demonstrate that apo H12 exists in a third stable and discrete conformation, protecting buried residues from solvent exposure and regulating the accessibility of the AF2 surface to transcriptional coactivators. Our comparative biophysical and computational analyses, together with patterns of evolutionary and structural conservation, support the use of the rfERα LBD as a model for investigating regulatory mechanisms in human and other vertebrate orthologs.

Apo and E2-bound rfERα LBD traverse distinct conformational landscapes

Agonist binding to the apo ER LBD induces widespread changes in tertiary dynamics, with H1, H3, the β-sheet, and H11 undergoing the most significant stabilisation, as determined by HDX-MS analysis (Fig. 2d). Yet, how these collective changes occur remains unclear. To enhance our understanding of these changes, we performed additional E2-bound simulations of the rfERα LBD, and then analysed both apo and E2-bound trajectories (30 µs total), with Markov state modelling (MSM) to resolve discrete ensembles and structural perturbations^37,38. To reduce dimensionality of the trajectories, time-structured independent component analysis (TICA)³⁹ was performed using a set of pairwise distances between representative residues from structural elements correlating to results from the HDX-MS analysis, two H12 residues, and a stable reference residue. MSMs were then constructed with PyEMMA (see “Methods”) and coarse grained into a three-state model for the apo ensemble and a four-state model for the E2-bound ensemble using Perron-cluster cluster analysis (PCCA + ) (Supplementary Fig. S8 to S11)^37,40. Analysis of the MSMs revealed that the apo and E2-bound LBDs traverse between distinct conformational landscapes (Fig. 2h). In the apo ensemble, the receptor predominantly occupies one metastable state representing 84.0% of the stationary distribution and two minor states accounting for 5.8% and 10.2%, respectively (Supplementary Fig. S12). Mean first passage times (MFPTs) show that the transitions from the major to the minor states occur on the hundreds-of-microseconds timescale, whereas return transitions to the dominant state occur quickly, in less than 10 μs. In contrast, the E2-bound ensemble samples two major states accounting for 38.3% and 54.0% of the stationary distribution, and two minor states (7.3% and 0.4%). The transitions between these states occur over tens to hundreds of microseconds but reconvert to the two dominant states in relatively less time (Supplementary Fig. S12).

Comparative analysis of representative structures obtained from PCCA+ coarse graining⁴¹ demonstrates that the dynamic changes observed in the HDX-MS analysis are well represented across the metastable states. For example, H3 exhibits significant stabilisation upon E2 binding in the HDX-MS analysis (Fig. 2d). In the predominant apo metastable state, H3 hinges at its junction with H5 and flexes into the LBP, resulting in a bent helical conformation. In the two minor states H3 is also variable (Supplementary Fig. S12a). In contrast, in E2-bound metastable states, H3 is rigidified and locked into place by the steroidal scaffold of the hormone (Supplementary Fig. S12b). A potential contributor to the substantial reduction in deuterium uptake of H3 observed by HDX-MS is also reduced solvent accessibility within the LBP upon ligand binding, consistent with the presence of internal water molecules in the apo crystal structure which are displaced (Fig. 1e). RMSF analyses also indicate marked stabilisation of H3 upon E2 binding consistent with the reduced deuterium uptake and helical stabilisation. In the apo ensemble, the C-terminus of H1 and the H1-H3 loop (regions showing significant stabilisation observed by HDX-MS) adopt a heterogenous set of conformations. These appear to be associated with H3 and β-sheet flexibility, compared to the more stable E2-bound ensemble. Additionally, H6, H7 and H11 are more structured in the E2-bound ensemble as compared to the apo ensemble, where they frequently interact with the base of H3 (Supplementary Fig. S12b). Interestingly, the predominant difference between the two dominant metastable states within the E2-bound ensemble (3 and 4) lies in the configuration of H6 and the H11-H12 loop (Supplementary Fig. S12b). However, these contrasts do not affect the binding mode of E2 nor H12 conformation, indicating a degree of plasticity. Indeed, plasticity of the H11-H12 loop is also observed among crystal structures of the hERα LBD bound to E2 and chemically diverse agonist ligands in both wildtype and mutant forms^4,28,42,43. We propose that the observed reduction in deuterium uptake in the E2-bound ensemble reflects alleviation of steric hindrances involving H3 and reduced solvent accessibility within the LBP upon ligand binding. Importantly, in all metastable states H12 maintains a conformation consistent with the corresponding crystal structure. Together with the HDX-MS analysis and MD simulations, the MSMs and PCCA+ coarse-graining provide a deeper understanding of the conformational landscape across apo and E2-bound ensembles and provide further support that H12 assumes a stable conformation between transcriptional states.

Structural basis of constitutive activation

Our apo structural data provides an unprecedented opportunity to address longstanding questions regarding disease and ligand structure-activity relationships. In ER+ breast cancers, somatic mutations Y537S and D538G in H12 of the hERα LBD constitutively activate the receptor, driving uncontrolled cellular growth and tumour development^3,9,12,44. These mutations augment and confer receptor activation by introducing hydrogen bonding with D351 on H3 and improving H12 packing in the active conformation, respectively^9,28. The effects of these mutations on the apo receptor structure remain unclear. However, previously published HDX-MS analysis of Y537S and D538G mutant hERα LBD showed increased dynamics of the H11-H12 loop in the apo state compared to wildtype (specifically residues M528 to L540; an ~8-10% increase in deuterium uptake) suggesting that a destabilised H12 may more easily adopt a conformation amenable for coactivator binding in the absence of ligand⁹. In the apo conformation, Y537 forms a π-stacking interaction with Y526, while D538 establishes a strong ionic bond with K529 (Fig. 3a). Structurally, the Y537S mutation disrupts the π-stacking with Y526, while D538G severs the ionic bond with K529. Replicate MD simulations of apo mutant rfERα and hERα LBDs revealed notable destabilisation of residues spanning M528 to L540, consistent with the prior HDX-MS analysis⁹. In the simulations, these mutations disrupt residue interactions leading to destabilisation of the H12-loop and a shift in the RMSD distribution compared to wildtype (Fig. 3b). We introduced the Y537S and D538G mutations into the rfERα LBD to validate if constitutive activity is also conferred. Fluorescence polarisation coactivator recruitment assays with these mutants showed a substantial increase in coactivator peptide binding relative to the wildtype apo receptor, and equivalently for the hERα mutants (Fig. 3c), consistent with prior work⁹. Together, these findings provide direct structural evidence demonstrating that oncogenic mutations Y537S and D538G disrupt critical interactions, enabling escape of H12 from the apo conformation and allowing H12 to adopt the active conformation and constitutively activate transcription (Fig. 3d).

Fig. 3. Structural basis of H12 escape and ligand-dependent antagonism.

a Oncogenic mutations Y537S and D538G disrupt key interaction that stabilise H12 in the apo conformation. Close-up perspective of the wildtype apo rfERα LBD crystal structure (left), and computationally modelled Y537S (middle) and D538G (right) mutations. b RMSD density distributions of the H11-H12 loop (residues 528-540) for wildtype (blue), Y537S (orange) and D538G (purple) mutations, for apo rfERα (left) and apo hERα (right) LBDs, calculated from technical triplicate 1 µs all-atom molecular dynamics trajectories (n = 10,001 frames/trajectory). c In vitro fluorescence polarisation assays measuring coactivator recruitment for a human SRC-2 13-mer peptide probe (5-FAM–KHKILHRLLQDSS–COOH) for the rfERα (left) and hERα (right) LBDs with wildtype apo (blue), apo Y537S (orange), apo D538G (purple) and wildtype apo treated with 10 µM E2 (red). All datapoints are shown, reported as millipolarisation (mP) values, and error bars represent the mean and standard deviation from three experiments. d Cartoon schematic illustrating destabilisation of H12 in the apo conformation by Y537S and D538G mutations enabling escape to the active conformation in the absence of E2. e Crystallographic binding modes of the SERMs, 4-hydroxytamoxifen (PDB: 3ERT) and raloxifene (PDB: 1ERR), and SERDs, bazedoxifene (PDB: 6PSJ) and elacestrant (PDB: 7TE7), from hERα LBD crystal structures superimposed onto the apo rfERα LBD crystal structure. Residue side chains are shown as sticks and labelled. The corresponding chemical structure of each ligand is detailed below each structure. Source data are provided as a Source Data file.

SERMs and SERDs displace H12 from the apo conformation

We then investigated how frontline SERMs and SERDs affect the apo ERα LBD conformation upon binding. Previous analyses of ligand-dependent inactivation have been hampered by the absence of a genuine apo structure. Understanding the key molecular interactions facilitating inactivation of the receptor upon initial ligand binding holds significant value for elucidating mechanisms of action and pharmacological development. Superimposing the apo rfERα LBD with the inactive conformation hERα LBD bound to SERMs 4-hydroxytamoxifen and raloxifene, as well as SERDs bazedoxifene and elacestrant, revealed that disruption of the apo conformation is primarily mediated by steric clashes between the ligand’s side chain moieties and H12 (Fig. 3e). Similar to E2 activation, each antagonist directly clashes with residues L525, L536, L540 and M543, displacing H12 into the solvent. However, the bulky side chain moieties of each ligand (e.g., the dimethylaminoethyl group of 4-hydroxytamoxifen) prevent H12 from adopting either the apo or active conformations and instead force H12 to bind within the partially-formed AF2 groove^4,26. This mode of action is consistent with prior HDX-MS analyses of hERα bound to SERMs and SERDs^7–10. We also observed that, akin to E2 binding the apo receptor, SERMs and SERDs make key interactions with residues within the LBP that are unchanged between states via a core scaffold (Fig. 3e). These findings suggest that small molecules which can facilitate displacement of the apo H12 conformation and share chemical similarity to the core scaffold may possess SERM- and SERD-like behaviour.

A ternary molecular switch determines receptor function

Our findings reveal that H12 adopts a third discrete conformational state in the absence of ligand. We propose a model in which H12 functions as a ternary molecular switch that determines receptor activity (Fig. 4a). In this model, the receptor can exist in three distinct states where the output is determined by the chemical structure of the input ligand. In the first state (state 1), H12 adopts the apo conformation until a ligand binds to the receptor. If an agonist such as E2 binds, H12 transitions to the active conformation (state 2). Conversely, if a SERM (i.e., 4-hydroxytamoxifen) or SERD (i.e., elacestrant) binds, H12 adopts the inactive conformation (state 3). The key event that triggers conformational switching, regardless of ligand, are steric clashes with L525 (H11), L536 and L540 (H12), which would destabilise and displace H12 into solvent. During transition to state 2, repositioning of H3 and H11 prevents H12 reverting to state 1, leading to the shielding of hydrophobic residues and the completion of the AF2 interface. However, in the presence of Y537S and D538G mutations, apo H12 is destabilised by loss of key contacts. This destabilisation lowers the threshold to state 2, allowing the receptor to access the active conformation. As a result, the ternary switch is no longer strictly controlled by ligand binding but is biased toward activation, enabling constitutive transcriptional activity. In contrast, during the transition to state 3, the SERM/SERD chemical structure prevents H12 reverting to state 1 or state 2, forcing it to occupy the AF2 groove via its LxxLL motif, ultimately repressing transcriptional activity. The capacity of H12 to stably adopt multiple orientations likely stems from its unconstrained location at the C-terminus of the chain, in contrast to more restrained helices (i.e., H7 or H11) that are integral to domain structure. Thus, this structural flexibility enables H12 to be more easily modulated to adopt the three observed states and stably shield non-polar residues from solvent exposure.

Fig. 4. A ternary switch model describing ERα LBD conformational states and transcriptional activity.

a Cartoon diagram illustrating the ternary switch model. In the model, each state (1: apo; 2: active; 3: inactive) is shown as helices corresponding to H3, H4, H11 and H12 (in pink) for simplicity. Ligand binding to state 1 induces a conformational switch of H12 to either state 2 or state 3 but the outcome is dependent on the chemical structure of the ligand. However, in both outcomes the ligand stabilises H3 and clashes with the residues L525, L536 and L540 to displace H12 into solvent. In state 2, a requirement to shield hydrophobic residues from solvent, alongside repositioning of H3 and H11, stably anchors H12 in the active conformation thus forming the AF2. In state 3, the chemical structure of the SERM (shown is 4-hydrotamoxifen; OHT) or SERD prevents H12 from reverting to the apo conformation or adopting the active conformation, instead causing the helix to stably bind in the partially formed AF2 binding cleft via its LxxLL motif. The oncogenic mutations Y537S and D538G disrupt key interactions enabling escape from the apo conformation and spontaneous transition to state 2 without the need for agonist binding. b Top-down perspective showing ligand interactions within the LBP and H12 (in pink) for the LBDs of human PPARγ (PDB: 2PRG), TRβ (PDB: 3GWS), VDR (PDB: 1DB1), RXRα (PDB: 1FM9) and ERα (PDB: 1GWR). c Evolutionary conservation of residue positions calculated as Jensen-Shannon divergence of 11,547 ligand-dependent NR LBD sequences projected onto the crystal structure of active E2-bound hERα LBD (red = poorly conserved residue positions; blue = very conserved residue positions). d Sequence logo plots of the residue positions covering the base of H11 and the H11-H12 loop (top), and H12 (below) showing the sequence and chemical diversity of this region. Amino acids are coloured: basic = blue; acidic = red; aromatic = purple; polar = green; non-polar = black. Source data are provided as a Source Data file.

A limitation to this model is the absence of direct evidence from other ER LBDs depicting H12 in the observed apo conformation, given the lack of apo state crystallographic data from other species. Therefore, in order to conclusively validate the ternary switch model as universal, further experiments are required.

Ligand-dependent activation mechanisms are diverse among NRs

In the context of ER activation, the ternary switch model is discordant to most other NRs which describes a dynamic ‘disorder-order’ transition of H12 upon ligand binding^5,6. In the ‘disorder-order’ mechanism, H12 is considered intrinsically disordered or dynamic, projecting into solvent, and upon ligand binding is stabilised in the active conformation through direct contacts, thus forming the AF2 interface. This mechanism is well established for NRs like peroxisome-proliferator activated receptor γ (PPARγ), thyroid receptor β (TRβ), and the vitamin D receptor (VDR)^45–48. Comparative structural analysis of ligand-bound NR LBDs demonstrates that H12 stabilisation mechanisms vary significantly. For example, in PPARγ, TRβ and VDR LBDs the ligands rosiglitazone, T3 and vitamin D directly contact H12 through hydrogen bonding networks and van der Waals interactions, which aid in ‘trapping’ H12 in the active conformation (Fig. 4b). This is supported by HDX-MS analysis and nuclear magnetic resonance (NMR) of these receptors which show that H12 is significantly stabilised upon ligand binding, demonstrating a shift from either an unfolded or dynamic solvent-exposed ensemble to an ordered state^46,47,49. To our knowledge, no equivalent NMR data is available for the ERα LBD; however, our HDX-MS analysis of E2 binding to rfERα and previously reported analysis for hERα show non-significant differences in deuterium exchange of H12 upon agonist binding^7,9. These HDX-MS results are consistent with two distinct stable conformations of H12 observed between the apo and E2-bound crystal structures. This is further supported by the MD simulations which demonstrate comparable H12 RMSD and SASA between the two states, as well as PCCA+ coarse graining of the MSMs which showed H12 conformations consistent with crystal structures across metastable states. Taken together, the dynamics and structural differences between ER and PPARγ, TRβ and VDR strongly suggest that the initial configurations of H12 are determinant of the mechanism(s) of H12 modulation by agonist ligands. Additionally, these diverse and contrasting mechanisms highlight the intimate relationship between ligand chemical structure and H12 modulation to determine receptor activity.

An important distinction to draw between ERα and full agonist bound PPARγ, TRβ and VDR is that E2 does not contact H12 in the active conformation; H12 stabilisation is primarily mediated by hydrophobic packing and structural support from H11 following conformational change (Fig. 1d). Other NRs, such as retinoid-X receptor (RXR) and steroid hormone receptors such as androgen receptor (AR) and glucocorticoid receptor (GR), also do not contact H12 due to structural differences in H3, H11 and the H11-H12 loop (Fig. 4b). In these receptors, H12 shields hydrophobic residues in the active conformation like the ERα LBD. Furthermore, residues within this region are also historically important for conferring ligand specificities as NRs evolved⁵⁰. For instance, between ER and AR the substitutions T347N (N705), H524F (F876) and L525T (T877) modify residues that are essential for E2 specificity and activation, instead contribute toward androgen hormone specificity in AR⁵¹. To examine the diversity of this region, an evolutionary analysis of more than 11,500 ligand-sensitive NR LBD sequences was performed, revealing that residues within this region are poorly conserved between NRs and vary in chemical properties compared to those which form the AF2 binding cleft (Fig. 4c, d). Consistent with this, the conservation of residues within the AF2 binding cleft would preserve the binding of coregulators (via the LxxLL motif). However, in order to achieve stable interaction a functional H12 is necessary as mutation (or deletion) disrupts coactivation^52–55. However, to our knowledge, clinical mutations that affect H11 and H12 structure and dynamics in NRs (besides the constitutively activating ERα mutations) are uncommon. In combination with the reported experimental data discussed above, these analyses support the notion that ligand-activation mechanisms vary widely among nuclear receptors and likely coevolved alongside the emergence of ligand specificity to maintain regulatory functions. Further work is required to comprehensively deconvolute mechanisms of H12 modulation and dynamics upon ligand binding among other nuclear receptors.

In summary, we report the first crystal structure of an apo ER LBD, which was complemented by in-depth biophysical and computational analyses to reveal significant insights into the mechanisms of ligand-dependent and -independent regulation. Our findings present a model of receptor activation by hormones and inactivation by SERMs and SERDs used to treat ER+ breast cancer. The analysis revealed that both agonists and antagonist ligands disrupt apo H12 conformation through clashes with L525, L536 and L540, and coordinate with residues in the LBP via a similar core chemical scaffold. We also demonstrate that oncogenic mutations Y537S and D538G disrupt crucial contacts that stabilise the apo conformation of H12 providing a structural basis for their constitutive activating function in ER+ breast cancer. Taken together, this work uncovers insight into the regulation of a key transcription factor and serves as a foundation to inform therapeutic design.

Methods

Protein expression and purification

Construct design

The codon-optimised insert for the wildtype ligand binding domain (LBD) of Melanotaenia fluviatilis ERα (rfERα) was synthesised by Twist Bioscience (South San Francisco, CA, USA); codon optimised inserts for the human ERα (hERα) LBD, and also for Y537S and D538G variants of both orthologs, were synthesised by GenScript Biotech (Piscataway, NJ, USA). All codon-optimised inserts were cloned into a pET-11a expression vector with ampicillin resistance. The rfERα inserts contained an N-terminal non-removable hexahistidine tag, while the hERα inserts contained a removable N-terminal hexahistidine + SUMO solubility tag.

Expression of apo rfERα and hERα LBDs

Expression plasmids were transformed into either Escherichia coli strains BL21(DE3) (wildtype rfERα) or Lemo21(DE3) (all other constructs). Initially, 3 × 1 L conical flasks containing Luria broth (LB) supplemented with 100 µg/mL ampicillin were inoculated with 20 mL (1:50 dilution) of an overnight seed culture transformed with the expression plasmid of interest, and grown at 37 °C with shaking (200 rpm) for aeration. Upon the culture reaching an optical density of A_{600 nm} value of 0.8–1.0, expression of recombinant protein was induced by addition of IPTG to a final concentration of 500 µM at 16 °C for 20 h at 200 rpm.

Purification of apo rfERα LBDs

All purification steps described herein were performed at 4 °C or on ice. Following overnight expression, cells were harvested by centrifugation at 5000 × g for 10 min and each pellet was resuspended in 25 mL of ice-cold Buffer A (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 10 mM imidazole, 2 mM β-mercaptoethanol, and 0.1% Tween 20) and lysed by mechanical disruption. The lysates were clarified by centrifugation at 40,000 × g for 30 min, and then loaded onto a 5 mL HisTrap HP column (GE Life Sciences) connected to an NGC Medium-Pressure Liquid Chromatography System (Bio-Rad) at a flow rate of 4 mL/min, equilibrated with Buffer B (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 10 mM imidazole, 2 mM β-mercaptoethanol). The column was washed with 10 column volumes of Buffer B and then with 5 column volumes of 20% Buffer C (100% = 20 mM Tris-HCl pH 8.0, 500 mM NaCl, 250 mM imidazole, 2 mM β-mercaptoethanol) to remove weakly bound impurities. Bound rfERα LBD was eluted in two steps of 5 column volumes of 65% Buffer C and then 5 column volumes of 100% Buffer C. The elution fractions containing rfERα LBD were identified by SDS-PAGE analysis and diluted ~8-fold into no-salt Buffer D (20 mM Tris-HCl pH 8.0, 10% glycerol, 5 mM DTT). The diluted sample (~200 mL in total), was immediately loaded onto a 5 mL HiTrap Q HP ion exchange column (GE Life Sciences) equilibrated in Buffer E (20 mM Tris-HCl pH 8.0, 100 mM NaCl, 10% glycerol, 5 mM DTT) at a flow rate of 4 ml/min. The column was then washed with 5 column volumes of Buffer E and the bound protein was eluted over a 20-column volume linear gradient of 0–100% Buffer F (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 10% glycerol, 5 mM DTT). The bound rfERα LBD typically eluted at ~ 200 mM NaCl concentration. The elution fractions containing highly pure rfERα LBD were pooled and concentrated to ≥ 10 mg /mL using an Amicon Ultra centrifugal filter (10,000 MW cut-off, Sigma Aldrich) and flash-cooled in liquid nitrogen prior to storage at –80 °C. Protein concentrations were determined by measuring the absorbance at 280 nm and using the molar extinction coefficient and molecular weight calculated by ProtParam (https://web.expasy.org/protparam/).

Purification of apo hERα LBDs

The purification of hERα LBDs by nickel affinity and anion exchange chromatography were performed as described above for rfERα LBDs, with the exception that Buffer A was replaced with Buffer B, and Buffers D and E contained 2 mM β-mercaptoethanol in place of 5 mM DTT. Following the anion exchange chromatography step, elution fractions containing SUMO-hERα LBD were pooled and incubated with ~0.5 mg of recombinant SUMO protease overnight at 4 °C. The digested sample was then loaded onto a 5 mL HisTrap HP column (GE Life Sciences) equilibrated in Buffer E (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 10 mM imidazole, 2 mM β-mercaptoethanol). The digested hERα LBD was eluted by washing the column with 5 column volumes of 10% Buffer C, with the His-SUMO and SUMO protease staying bound to the column. Elution fractions containing highly pure hERα LBD were pooled and buffer exchanged 200-fold into Buffer E using an Amicon Ultra centrifugal filter (10,000 MW cut-off, Sigma Aldrich) and flash-cooled in liquid nitrogen prior to storage at –80 °C. The concentrations of hERα LBDs were determined using the same method as rfERα LBDs.

Fluorescence polarisation coactivator recruitment assays

The human SRC2-2 coactivator peptide probe (5-FAM–KHKILHRLLQDSS–COOH; 98% purity) was purchased from GenScript Biotech (Piscataway, NJ, USA)⁵⁶. For fluorescence polarisation assays a 200 µM E2 stock was prepared in 100% DMSO. The coactivator probe was prepared as a 10 mM stock in MQ H₂O and adjusted to ~pH 7 with dropwise addition of 2 M NaOH, with working stocks subsequently prepared at a final concentration of 100 µM in 10 mM Tris-HCl pH 7.5 and stored at –20 °C.

All fluorescence polarisation experiments were performed in triplicate as previously described with slight modification⁵⁷. Under apo conditions, samples (160 µL final volume) contained FP buffer (20 mM Tris-HCl pH 8.0, 100 mM NaCl, 0.05% Tween 20, 1 mM TCEP, 5% DMSO, 10% glycerol and 50 nM SRC2-2 probe), with the protein concentrations of each ER LBD variant varied. For WT hERα and rfERα LBDs, the experiment was also performed in the presence of 10 µM E2. Once prepared, the samples were incubated at 22 °C for 30 min, and then 150 µL of each sample transferred to individual wells of a non-binding flat-bottom black 96 well plate (Greiner Ref. 655900). Fluorescence polarisation was measured using a Pherastar FS microplate reader equipped with the FP 485 520 520 module, using default settings. Millipolarisation (mP) values were calculated using Eq.( 1) following subtraction of raw parallel and perpendicular emission values for a corresponding sample with the coactivator probe omitted.

$$mP=\frac{I_{parallel}-I_{perpendicular}}{I_{parallel}+I_{perpendicular}}\times 1000$$ 1

To investigate the association between the coactivator class probe with ER LBDs in the presence and absence of E2, the samples comprised FP Buffer and 8 different concentrations of ER LBD ranging between 0.3 and 5× the estimated dissociation constant (K_d). This was estimated by non-linear regression in GraphPad Prism using Eq.( 2), where ${mP}_{free}$ and ${mP}_{bound}$ are the fluorescence polarisation for free probe and saturated receptor respectively, $\left[L\right]$ is the total concentration of coactivator class probe, $\left[R\right]$ is the total concentration of ER LBD homologue and K_d is the dissociation constant.

$$mP={mP}_{free}+\left({mP}_{bound}-{mP}_{free}\right)\times \frac{\left(\left[L\right]+K_d+\left[R\right]\right)-\surd {\left(\left[L\right]+K_d+\left[R\right]\right)}^2-4\left[L\right]\left[R\right]}{2\left[L\right]}$$ 2

X-ray crystallography

Crystallisation experiments were performed using the sitting drop vapour diffusion method in 96-well Intelliplates (ArtRobbins) at 16 °C. Conditions promoting crystal formation were probed using the NR-LBD, PEG/Ion HT, Index HT, and Crystal Screen HT sparse matrix screens (Hampton Research). Reservoir volumes were 80 µL and crystals were formed in drops containing 1 µL 16 mg/mL apo rfERα LBD to 1 µL of reservoir solution. Diffracting crystals were obtained in the NR-LBD sparse matrix screen condition E5 (0.1 M NaCl, 0.1 M HEPES pH 7.0, and 225 PEG 2000 MME), appearing as large rods (~500 x 50 x 50 µm) after 1 week.

Data processing and refinement

Prior to data collection, crystals were cryoprotected by immersion in NVH oil and flash cooled in liquid nitrogen. Crystals were diffracted at 100 K at the MX1 beamline of the Australian Synchrotron⁵⁸. Images were indexed and integrated in XDS, with scaling and merging completed in Aimless^59,60. The phase problem was solved by molecular replacement in Phaser using Chain A of a previously determined structure of the rfERα LBD complexed with E2 and a human SRC2-2 coactivator-derived peptide (PDB: 9D8Q) as the search model, with ligands and solvent removed⁶¹. The resulting model underwent several rounds of refinement in phenix.refine and rebuilding in Coot, until the refinements converged^62,63. The final model displayed good overall geometry and contained no Ramachandran outliers. Data collection and refinement statistics are detailed in Table S1.

Native mass-spectrometry

Protein samples containing wildtype hERα or rfERα LBDs were diluted into 15 mL of ammonium bicarbonate (pH 7.0) and buffer exchanged five times using an Amicon Ultra centrifugal filter (10,000 MW cut-off, Merck Millipore) to a final concentration of 10 µM. Native mass-spectrometry (one technical replicate per sample) was performed on a Bruker Impact II HDMS Q-ToF mass spectrometer (Bruker Daltonics) with a nanoelectrospray ionisation source with the following instrument parameters: m/z range, 500-10,000; polarity, positive; capillary voltage, 2 kV; end plate offset, 500 V; and source temperature, 60 °C. Protein samples were electrosprayed from platinum-coated borosilicate glass capillaries (Harvard Apparatus, USA) prepared in-house. All spectra were acquired using the Compass Data Analysis software (v4.2 Bruker Daltonics) acquiring minimum of 50 scans per sample. Deconvolution of native mass spectra was performed using UniDec (v6.0.3 University of Arizona)⁶⁴ with the following parameters: m/z range, 1000-10,000; charge range, 1–50; sample mass every 1 Da; peak range, 1 Da; charge smooth width, 1; peak width, 0.85.

Size-exclusion chromatography coupled small-angle X-ray scattering

Small-angle X-ray scattering experiments were performed on the BioSAXS beamline on the Australian Synchrotron. Concentrated protein samples were diluted into buffer containing 20 mM Tris-HCl pH 7.5, 150 mM NaCl, 5% glycerol and 2 mM TCEP and buffer exchanged five times using an Amicon Ultra centrifugal filter (10,000 MW cut-off, Merck Millipore) to a final concentration of 5 mg/mL. The samples were loaded onto an equilibrated S200 5/150 GL column (Cytiva) using the Coflow autoloader⁶⁵ at a flow rate of 0.3 mL/min through a 3 mm UV flow cell (Knauer, Berlin, Germany). UV absorbances at 260 and 280 nm were measured immediately before presentation to the X-ray beam by an STS microspectrometer (Ocean Optics, Orlando, FL, USA). X-ray scattering was measured continuously during SEC at 295 K with a 1 s exposure time at 12.4 keV (λ = 1.00 Å) using a Pilatus3X 2 M detector positioned 2150 mm (1500 mm + 650 mm offset distance) from the sample capillary. This afforded a q-range of 0.0098 to 0.72027 Å^-1 following data reduction. Data was reduced by Fast Azimuthal Integration using Python and PyFAI with customised algorithms written for the BioSAXS beamline. Data were placed on an absolute scale using water in the measurement capillary as a standard and the nominal diameter of the capillary at the measurement position was 1.0 mm. The subtracted profiles were analysed using BioXTAS RAW (v2.3), and GNOM implemented in ATSAS (v4.0)^66,67. SAXS envelopes were calculated using the DAMMIN algorithm implemented in ATSAS (v4.0)⁶⁸ and fit to the high-resolution crystal structure of the apo rfERα LBD. The SAXS experimental data summary table can be found in Table S2.

Hydrogen-deuterium exchange detected by mass-spectrometry

Peptide identification

Differential HDX-MS experiments were conducted as previously described with several modifications⁶⁹. Peptides were identified using tandem MS (MS/MS) with an Orbitrap mass spectrometer (Q Exactive, ThermoFisher) over a 70-min time gradient. Product ion spectra were acquired in data-dependent mode with the top five most abundant ions selected for the product ion analysis per scan event. The MS/MS data files were submitted to Mascot (Matrix Science) for peptide identification. Peptides included in the HDX analysis peptide set had a MASCOT score greater than 20 and the MS/MS spectra were verified by manual inspection; mass tolerances for precursor ions were set to ±0.6 Da and ± 10 ppm for fragment ions. The MASCOT search was repeated against a decoy (reverse) sequence and ambiguous identifications were ruled out and not included in the HDX peptide set.

HDX-MS analysis

5 μL of 10 µM of apo rfERα LBD, or complexed with 5:1 molar ratio E2, was diluted into 20 μL D₂O buffer (20 mM Tris, pH 7.8, 150 mM NaCl, and 2 mM DTT) and incubated for various time points (0, 10, 60, 300, 900 and 3600 s) at 4 °C. The deuterium exchange was then slowed by mixing with 25 μL of cold (4 °C) 3 M urea, 50 mM TCEP, and 1% trifluoroacetic acid. Quenched samples were immediately injected into the HDX platform. Upon injection, samples were passed through an immobilised pepsin column (1 mm × 2 cm) at 50 μl min⁻¹ and the digested peptides were captured on a 1 mm × 1 cm C₈ trap column (Agilent) and desalted. Peptides were separated across a 1 mm × 5 cm C₁₈ column (1.9 μl Hypersil Gold, ThermoFisher) with a linear gradient of 4–40% CH₃CN and 0.3% formic acid, over 5 min. Sample handling, protein digestion and peptide separation were conducted at 4 °C. Mass spectrometric data were acquired using an Orbitrap mass spectrometer (Exactive, ThermoFisher). HDX analyses were performed in triplicate, with single preparations of each complex form. The intensity weighted mean m/z centroid value of each peptide envelope was calculated and subsequently converted into a percentage of deuterium incorporation (%D). This is accomplished determining the observed averages of the undeuterated control (t = 0 s) and fully deuterated spectra using the conventional formula described elsewhere⁷⁰. Statistical significance for the differential HDX data is determined by an unpaired t-test for each time point, a procedure that is integrated into the HDX Workbench software⁷¹. Corrections for back-exchange were made on the basis of an estimated 70% deuterium recovery, and accounting for the known 80% deuterium content of the deuterium exchange buffer. Each %D is averaged across all time points to determine an overall %D for each peptide. The HDX-MS experiment and data summary table can be found in Table S3.

Data rendering

The HDX data from all overlapping peptides were consolidated to individual amino acid values using a residue averaging approach. Briefly, for each residue, the deuterium incorporation values and peptide lengths from all overlapping peptides were assembled. A weighting function was applied in which shorter peptides were weighted more heavily and longer peptides were weighted less. Each of the weighted deuterium incorporation values were then averaged to produce a single value for each amino acid. The initial two residues of each peptide, as well as prolines, were omitted from the calculations. This approach is similar to that previously described⁷².

Molecular dynamics (MD) simulations

All-atom MD simulations for wildtype apo (PDB: 9MDV) and E2-bound (PDB: 9D8Q) rfERα, along with modelled Y537S and D538G variants, were performed using the same method outlined here, except for the duration of production simulations. A homology model of the apo hERα LBD (residues 310–548) was generated in ICM-Pro (Molsoft, LLC) using the apo rfERα LBD as a template^73,74; Y537S and D538G variants were generated using Coot⁶². MD simulations were performed using GROMACS with the Charmm36 forcefield^75–78. E2 topology was generated using SwissParam⁷⁹. Protein coordinates were placed in the centre of a dodecahedral box, situated at least 10 Å from the periodic edge boundary. The systems were solvated with the TIP3P water model⁸⁰ and Na⁺ ions were added to neutralise net charge of the system. Following system preparation, energy minimisation was performed using the steepest descent method for a maximum of 50,000 steps or until F_max reached <1000 kJ/mol/nm. Energy minimised systems were equilibrated using sequential 1 ns restrained simulations in the NVT and NPT ensembles at 310 K using Lincs constraint algorithm (hydrogen bonds), Particle Mesh Ewald electrostatics, Berendsen thermostat coupling, and for the NPT ensemble only, Parrinello-Rahman pressure coupling^81–84. In the apo conformation NVT and NPT simulations, restraints were applied to heavy protein and hydrogen atoms, and in the corresponding E2-bound simulations, restraints were also applied to E2 heavy atoms. Restraints were then released and production simulations performed in the NPT ensemble at 310 K using a 2-femtosecond timestep, with frames stored every 100 ps. Production simulations were performed in triplicate for at least 1 µs duration. Analyses were performed in Python using the MDTraj, NumPy, Matplotlib and Seaborn libraries^85–88. Comparisons between wildtype apo rfERα LBD and wildtype apo hERα LBD used the first 1 µs of the wildtype apo rfERα LBD trajectories. System parameters can be found in Table S4.

Markov modelling

To structurally analyse the conformational ensembles, Markov state modelling was employed; a powerful approach for studying kinetic data generated by MD trajectories, implemented with PyEMMA^37,40. The features used as input for time-structured independent component analysis (TICA)³⁹ consisted of pairwise distances between Cα atoms of ten residues located on secondary structure elements that are stabilised upon E2-binding (L319, M342, T347, A361, L402, L408, G415, M421, F425, L525), two H12 residues (L536, M543) and a reference residue that was stable in both states (W383). TICA was performed with a lag time of 7.5 ns for the apo state and 10 ns for the E2-bound state, retaining 46 and 34 dimensions, respectively, which correspond to 95% of the total kinetic variance. Microstate clustering was performed on the first two TICA components using the K-means algorithm⁸⁹, and different cluster sizes were explored to determine the optimal number microstate centres based on convergence of the implied timescale (ITS) plots (Supplementary Figs. S8–S9). MSMs were constructed using 75 microstates and a lag time of 25 ns for the apo ensemble, and 200 microstates and a lag time of 20 ns for the E2-bound ensemble. The number of metastable states for each MSM (three for apo and four for E2-bound) was determined by spectral gap analysis of the ITS as implemented in PyEMMA (http://emma-project.org/latest/index.html), stationary distributions and good separation in PCCA+ coarse graining⁴¹. The MSMs were statistically validated using the Chapman-Kolmogorov test⁴⁰. To compute mean first passage times transition path theory (TPT) was used^90,91, and twenty representative structures from each metastable state were obtained by sampling the stationary distribution of each MSM. Structural analysis was performed with PyMOL (Schrödinger, LLC).

Evolutionary sequence analysis

Primary amino acid sequences for ligand-activated NR LBDs were curated from the UniRef-100 and UniRef-90 databases with jackhmmer using the human sequence as a query⁹². Assembly of the multiple sequence alignment was performed in a three-step process. First, hits from jackhmmer queries of each receptor were aligned in MAFFT and edited to human residue positions in order to remove lineage-specific indels. Sequences with more than 30% gaps were removed. Following this, human NR LBD sequences were iteratively aligned with MAFFT to create a reference sequence profile which was used to merge the individual alignments using hmmalign software. To calculate the evolutionary conservation, the Jensen-Shannon divergence (JSD) of each residue position was calculated with SciPy, using the background ($q$) and position-specific ($p$) amino acid probability distributions⁹³. Sequence logos were created using WebLogo3 (https://weblogo.threeplusone.com/)⁹⁴.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualisation: D.P.M., J.L.P., J.B.B. Funding acquisition: J.B.B. Methodology: D.P.M., J.L.P. S.J.N., B.J., A.K.M. Investigation: D.P.M., J.L.P. S.J.N., B.J. Data analysis: D.P.M., J.L.P., S.J.N., B.J., B.D.P. Software: D.P.M. Visualisation: D.P.M. Writing—original draft:—D.P.M. and J.L.P. Writing—review and editing: D.P.M., J.L.P., S.J.N., B.J., A.K.M., B.D.P., P.G.R., J.B.B.

Peer review

Peer review information

Nature Communications thanks Mitsugu Araki, and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

Atomic coordinates for the X-ray crystal structure of the apo rfERα LBD are deposited in the Protein Data Bank (PDB) with the accession code 9MDV. Atomic coordinates for previously published structures can be accessed via accession codes: [10.2210/pdb1DB1/pdb], [10.2210/pdb1FM9/pdb], [10.2210/pdb1GWR/pdb], [10.2210/pdb2PRG/pdb], [10.2210/pdb3ERT/pdb], [10.2210/pdb3GWS/pdb], [10.2210/pdb3UUD/pdb], [10.2210/pdb9D8R/pdb]. SEC-SAXS data generated in this study are deposited in the Small Angle Scattering Biological Data Bank (SASBDB) under accession codes SASDWC7 (apo hERα LBD) and SASDWD7 (apo rfERα LBD). Data generated in this study are available as Source Data provided with this paper. Source data are provided with this paper.

Code availability

Code associated with the construction and analysis of Markov modelling can be found online at the BruningLab GitHub repository [https://github.com/BruningLab]. Analysis of molecular dynamics trajectories was performed using the openly available Python library, MDTraj [https://www.mdtraj.org/1.9.8.dev0/index.html].

Competing interests

B.D.P. is employed by Omics Informatics LLC, a company that provides HDX-MS analysis software (HDX Workbench) free to academics. The other authors have no competing interests to declare.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-65323-9.