Structural insights into μ-opioid receptor activation

Nanobody stabilized structure of the μOR

The active states of ligand-activated GPCRs are likely unstable, even when bound to full agonists^20-23. However, the active conformation can be stabilized by interactions between a receptor and its cognate G protein. This stabilization is reflected in a higher affinity for agonists when GPCRs are in complex with their cognate G protein²⁴. In the case of the μOR, the affinity for the morphinan agonist BU72 is enhanced by 47 fold when coupled to the G protein G_i (Fig. 1b, c). Efforts to obtain a structure of activated μOR in complex with G_i have thus far not been successful. As an alternative, we have previously utilized camelid single-domain antibody fragments, nanobodies, as G protein-mimetics to stabilize the active conformation of the β₂AR and M2R for structural study^12,13,17. For the β₂AR, the conformation of the receptor obtained in complex with the G_s mimetic nanobody 80 (Nb80) was nearly identical to that in the β₂AR-G_s complex²⁵ (RMSD 0.61 Å).

To generate G protein-mimetic nanobodies for the μOR, llamas were immunized with purified μOR bound to the peptide agonist DMT¹-DALDA²⁶ and reconstituted into phospholipid vesicles¹². We examined the ability of selected nanobodies to stabilize the high-affinity state for μOR agonists. Purified μOR was reconstituted into high-density lipoprotein (HDL) particles and agonist competition assays were performed in the presence or absence of nanobodies (Fig. 1b). In presence of 5 μM nanobody 39 (Nb39), the affinity of the potent morphinan agonist BU72²⁷ increases from 470 pM to 16 pM (Fig. 1b). BU72 has a dissociation half-life of 140 minutes in the presence of Nb39 (Extended Data Fig. 1b). Nb39 also enhances the affinity of μOR agonists DAMGO and endomorphin-2, indicating that the effect is not limited to morphinan agonists (Extended Data Fig. 1a).

Crystals of the μOR bound to BU72 and Nb39 were obtained in a monoolein lipidic mesophase²⁸ and a complete data set to 2.1 Å was obtained by merging diffraction data from four crystals (Extended Data Table 1). Nb39 binds to the intracellular surface of μOR (Fig. 1d, Extended Data Fig. 2), and mediates the majority of packing interactions between lipidic layers and adjacent Nb39-μOR complexes in the crystal lattice. There are no packing interactions involving the extracellular surface of the receptor (Extended Data Fig. 1c). We observe a limited parallel dimeric packing interaction between μOR molecules involving the extracellular end of transmembrane helix 1 (TM1), TM2, the first extracellular loop (ECL1), and helix 8 with a buried surface area of 460 Ų (Fig. 1e). A similar interface between TM1, TM2, and helix 8 was also observed in the inactive structure of μOR with a slightly larger buried surface area of 615 Ų (Fig. 1e). The inactive structure also identified a more extensive parallel dimer interaction involving TM5 and TM6 with a buried surface area 1460 Ų (Fig. 1f). This interaction involving TM5 and TM6 would not be compatible with the conformational changes we observe in the active state (Fig. 1f). It is important to note that the physiologic relevance of these interfaces remains unclear.

Structural differences in the extracellular surface between inactive and active μOR are relatively small (Fig. 2a) with the exception of the proximal N-terminus, as discussed below. Conformational changes at the cytoplasmic surface of the μOR observed upon activation are similar to those observed for the β₂AR, M2R and rhodopsin, with large outward movement of TM6 and a smaller inward movement of TM5 and TM7 (Fig. 2a, Extended Data Fig. 3). The conserved E/DRY motif at the intracellular end of TM3 plays a role in maintaining GPCRs in the inactive state. In rhodopsin, an ionic interaction between R135^3.50 and E247^6.30 in TM6 stabilizes TM6 in an inactive conformation (superscript numbers follow the Ballesteros-Weinstein numbering method for GPCRs²⁹). While there is no acidic amino acid at the end of TM6 of μOR that can form a similar salt bridge with R165^3.50, R165^3.50 can form a hydrogen bond with T279^6.34 (Fig. 2b). In the active states of the μOR and metarhodopsin II, R^3.50 forms a hydrogen bond with Y^5.58, stabilizing the inward movement of TM5 (Fig. 2b). R^3.50 and Y^5.58 assume a similar orientation in the M2R and the β₂AR-G_s complex; however, they are not close enough to form a hydrogen bond (Extended Data Fig. 3a).

Agonist binding pocket

The morphinan scaffold of BU72 binds to the activated μOR in a similar orientation to that observed previously for the irreversible antagonist β-funaltrexamine (β-FNA) at the inactive μOR (Fig. 3a, Extended Data Fig. 4). Figure 3b shows that the overall structural differences in the orthosteric binding pockets of active and inactive μOR are relatively subtle. The majority of interactions between BU72 and active μOR are hydrophobic or aromatic in nature, with the exception of two conserved polar interactions (Fig. 3c, Extended Data Fig. 4). As observed previously in the inactive structures of the μOR, δOR and κOR³⁰, the phenolic hydroxyl of BU72 engages in a water-mediated interaction with H297^6.52. In the active state, this network is more extended, and involves Y148^3.33 and the backbone carbonyl of K233^5.39. An ionic interaction between the morphinan tertiary amine and D147^3.32 is seen in both the active μOR bound to BU72 and in inactive μOR bound to β-FNA (Fig. 3c).

We also observe an unexpected interaction between BU72 and the amino-terminus of the μOR, which forms a lid over the ligand-binding pocket (Fig. 3d and Extended Data Fig. 5). In particular, the amino-terminal residue H54 is positioned 2.6 Å from secondary amine of BU72. It is of interest that binding of BU72, as well as other structurally unrelated peptide agonists like DAMGO and DMT¹-DALDA, lead to conformational changes in the truncated amino terminus that can be detected by nuclear magnetic resonance (NMR) spectroscopy (Sounier et al., companion manuscript). While interactions between the amino-terminus and the transmembrane core may indeed be important for μOR function³¹, the specific interaction observed here is unlikely to be physiologically relevant as H54 is not highly conserved and the H54A mutation does not alter the affinity of BU72 in μOR with the full length amino-terminus (μOR wild-type K_i = 21 pM and μOR H54A K_i = 30 pM in HEK293 membranes). Additionally, using a bioluminescence resonance energy transfer assay³², we observe no significant difference in the EC₅₀ of G_i activation by BU72 between the wild-type receptor and the H54A mutant (EC₅₀ for wild-type of 79±17 pM and H54A mutant of 67±20 pM).

While the morphinan core of BU72 is unambiguously placed within the electron density, there are two unexplained features of the ligand. We observe a strong positive electron density (F_o-F_c signal) between the H54 side chain and the BU72 amine (Extended Data Fig. 5c). Attempts to identify the source of this density, including mass spectrometry for an alternative ligand structure and the presence an anomalous signal for a coordinated metal were unsuccessful (Extended Data Fig. 4). Another unexpected finding is the near-planar geometry of the sp³ hybridized carbon of BU72 to which the pendant phenyl group is attached (Extended Data Fig. 4e-h). This geometry could be explained by a double bond between this carbon and the adjacent nitrogen. We observed a minor fraction of such a compound by mass spectrometry in our preparation of BU72. However, this compound was not observed in mass spectrometry of the μOR-ligand-Nb39 complex used for crystallography. Ultimately, we modeled a higher-energy conformation of BU72 within the observed electron density (see Extended Data Fig. 4 for further discussion).

In the active μOR, there is a water-filled cavity lined by polar and aromatic side chains that extends off of the intracellular end of the ligand-binding pocket (Fig. 3e). While a similar cavity is observed in inactive μOR, the active-state cavity is larger and completely contiguous with the morphinan-binding site. Substitution of morphinans with a cyclopropylmethyl at the tertiary amine generally results in a ligand with antagonist activity⁷. As an example, BU74, which only differs from BU72 in having a cyclopropylmethyl substituent on the tertiary amine (Fig. 3f) is an antagonist at the μOR³³. While BU74 can be docked into the active state μOR structure with the cyclopropylmethyl within the polar cavity, there are potential clashes with Y326^7.43 in TM7 and W293^6.48 in TM6. Moreover, the cyclopropylmethyl substituent would displace one or more water molecules that form part of a polar network described in more detail below. Consistent with these observations, MD simulations reveal that the antagonist BU74 is unstable in the position occupied by the agonist BU72 and rapidly shifts away from this initial pose (Extended Data Fig. 6).

Propagation of conformational changes

The structural difference in the orthosteric ligand-binding pocket between active and inactive μOR is relatively subtle (Fig. 3b). It is difficult to identify specific interactions between the structurally rigid agonist and the receptor as being key for μOR activation. This stands in contrast to activation of the β₂AR and M2R, where specific polar interactions between the receptor and the smaller, more flexible agonists contribute to structural changes associated with activation. These polar interactions stabilize a 2 Å inward movement of TM5 in the β₂AR and a 2 Å inward movement of TM6 in the M2R.

While agonist stabilized changes in the binding pocket differ for the μOR, β₂AR, and M2R, the overall structural changes observed at the G protein interface are very similar (Extended Data Fig. 3). The mechanism by which agonist-stabilized changes propagate to the cytoplasmic surface appears to be more similar between μOR and the β₂AR compared to the M2R (Fig. 4a). For both the β₂AR and the μOR, there is a rearrangement of the packing of a triad of conserved amino acids F^6.44, P^5.50 and I^3.40 (which we term the conserved core triad) that lie just below the binding pocket (Fig 4a). This rearrangement is associated with a counterclockwise rotation (when viewed from the extracellular surface) and outward movement of the cytoplasmic end of TM6. Thus, the subtle agonist-stabilized rearrangement in the conserved core triad may initiate the cascade of structural changes involved in activation of the μOR and β₂AR. The role of the conserved core triad appears to differ in the M2R, where the smaller V111^3.40 forms a weaker packing interaction with F396^6.44 and P198^5.50. As a result, the triad in both inactive and active states of the M2R appear similar to the active states of the β₂AR (Fig. 4a).

Given the similarity in allosteric propagation between μOR and β₂AR, we sought to identify how BU72 stabilizes the active conformation of the conserved core triad. In the β₂AR, the rearrangement of the conserved core triad can be attributed to an agonist stabilized inward movement of TM5^12,13. For the μOR, however, there are no specific interactions between the receptor and BU72 that stabilize the inward movement of TM5. Instead, we identify a set of interactions that together appear to stabilize the rearrangement of the conserved core triad in the μOR.

Both β-FNA and BU72 share a common morphinan scaffold. This common scaffold, however, is positioned differently in the inactive and active structures. In both states, residues I296^6.51 and V300^6.55 in TM6 and W318^7.35 and I322^7.39 in TM7 form a common hydrophobic surface for binding the morphinan ligands (spheres, Fig. 4b,d). While this surface is similar in both inactive and active states, differences in the position of the rigid morphinan agonist and antagonist result in differences in the positions of key ligand substituents relative to specific residues in TM3 and TM6 that are coupled to the conserved core triad of F289^6.44, P244^5.50 and I155^3.40 (Fig. 4b,d).

In the active μOR structure, the morphinan scaffold of BU72 adopts a pose that is sterically incompatible with the inactive position of TM3. As a result, the residues of TM3 that interact with the tertiary amine (D147^3.32) and methyl substituent (M151^3.36) of the agonist shift 1.5 Å toward TM2 relative to its position in the inactive structure (Fig. 4b). We utilized MD simulations to assess whether the agonist favors this displacement in TM3. In simulations of activated μOR with the agonist removed, the previously ligand-contacting residues on TM3 quickly relax toward their inactive positions (Fig. 4c), often without global structural change of the receptor. These motions are tightly coupled to motion of the conserved core triad residue I155^3.40 towards its inactive position, causing it to reposition relative to F289^6.44 (Fig. 4b,c). Notably, the position of D147^3.32 is also coupled to the rotameric state of N150^3.35, which coordinates an allosteric sodium ion in the inactive state and forms a hydrogen bond with a backbone carbonyl of I146^3.31 in the active state (Fig. 4b,c).

Another link between the ligand binding pocket and the conserved core triad may be mediated via TM6 through W293^6.48 (Fig. 4d,e). In the active μOR structure, the aromatic group of BU72 is positioned only slightly closer (0.6 Å) to W293^6.48 as compared to the same aromatic group of β-FNA in the inactive state. However, MD simulations suggest that agonists stabilize W293^6.48 in the rotamer observed in the active-state crystal structure. In simulations of the active state, removal of BU72 results in a change in the favored rotamer of W293^6.48 (Fig. 4e). Conversely, we assessed whether an agonist bound to the inactive state would stabilize the side chain of W293^6.48 in the rotamer associated with the active state. Here, we simulated the inactive state structure with the antagonist β-FNA replaced with the agonist β-fuoxymorphamine (β-FOA), which differs from β-FNA solely in a methyl substituent at the morphinan tertiary amine³⁴ (Extended Data Fig. 7a). In these simulations, the pose of β-FOA tends to shift towards that observed for BU72 and the TM3 residues D147^3.32 and N150^3.35 shift towards their active-state positions; in concert, the side chain of W293^6.48 shifts towards its active position (Extended Data Fig. 7b,c). W293^6.48 is spatially juxtaposed to the conserved core triad residue F289^6.44, and the conformation of W293^6.48 may therefore serve as an important link between the ligand-binding pocket and the triad (Fig. 4d).

Role of polar network in GPCR activation

In addition to rearrangement of the conserved core triad, comparison of inactive and active structures of μOR reveals an extensive network of polar interactions between the orthosteric binding pocket and the G protein-coupling interface that must rearrange upon activation (Fig. 5a). The high-resolution electron density maps of the active μOR allow us to detect more ordered water molecules within this polar network than have been observed in other active-state GPCR structures reported to date. To provide a better comparison of this polar network, we examined the high-resolution inactive structure of the highly homologous δOR¹¹. The inactive structure of δOR at a resolution of 1.8 Å identifies more water molecules within this polar network than the 2.8 Å inactive structure of μOR and therefore highlights the contribution of many hydrogen bonds in stabilizing both inactive and active states of opioid receptors (Fig. 5b). These hydrogen bonds represent many low energy molecular switches that have to be broken and reformed in a concerted manner to achieve the active conformation. The polar network interactions for inactive state μOR are likely to be identical to the inactive state of δOR because the specific amino acids involved in this network and their side chain conformations are identical between the two homologous receptors.

Many of the residues in the polar network are conserved (Extended Data Fig. 8a, b, e and f), suggesting that the polar network may also be conserved and play a similar role in activation of other Family A GPCRs. While there are fewer water molecules observed in the active state structure of the β₂AR, the ones that are resolved are in the same positions as those observed in the μOR. Even though the resolution in the active β₂AR and M2R structures is not sufficient to observe as many ordered water molecules, the positions of the conserved side chains lining the polar core of the β₂AR and M2R are nearly identical to those of the μOR (Extended Data Fig. 8f), suggesting that they are stabilized by a similar hydrogen bonding network.

In the high-resolution inactive state structure of the δOR, there is an ordered sodium ion that is adjacent to the conserved core triad and is coordinated by the side chains of D^2.50, N^3.35 and S^3.39, as well as by W^6.48 through a water molecule (Fig. 5b,c). A sodium ion at a similar coordination site has been observed in inactive structures of the protease activated receptor subtype 1 (PAR1)³⁵, the adenosine A_2A receptor (A_2AR)³⁶ and the β₁ adrenergic receptor (β₁AR)³⁷. For many GPCRs, including the μOR, agonist binding affinity and/or G protein activation is allosterically inhibited by sodium³⁸. Consistent with allosteric stabilization of the inactive state by sodium, we do not observe a sodium ion in the active state structure of the μOR, and the residues that formed the sodium-binding site rearrange to preclude sodium binding in the active state (Fig. 5c).

The polar network ends at the cytoplasmic surface with a web of interactions involving the conserved NPxxY sequence (Fig. 5d). Upon receptor activation, the NPxxY motif in TM7 moves inward towards TM5 where N332^7.49 and Y336^7.53 participate in a hydrogen bond network involving Y252^5.58 in TM5, the backbone carbonyls of L158^3.43 in TM3 and V285^6.40 in TM6, and three ordered water molecules (Fig. 5d). A similar hydrogen bond network was previously observed in the structure of metarhodopsin II¹⁹. Similar water-mediated hydrogen bonds are likely present in active β₂AR and M2R as the side chains of key residues in this region occupy similar positions (Extended Data Fig. 3b).

It is interesting to speculate on the role of this polar network in the energetics of μOR activation. NMR studies reveal that allosteric coupling of the agonist-binding pocket and the G protein-binding interface in the μOR is relatively weak, and structural changes in TM6 are observed only in the presence of a G protein-mimetic nanobody (Sounier et al., companion manuscript). A similar observation has been made for the β₂AR^20-22,39. Using both double electron-electron resonance (DEER) spectroscopy and NMR spectroscopy we observed that in the presence of a saturating concentration of the catecholamine agonist isoproterenol only 20% of β₂AR has TM6 in an active-like conformation²⁴. This stands in contrast to the more efficient coupling between the orthosteric binding pocket and TM6 in rhodopsin, as reflected in the ability to crystallize rhodopsin in an active state without a G protein or G protein mimetic nanobody⁴⁰, and biophysical studies that reveal a stronger allosteric coupling between the orthosteric binding pocket and TM6 in rhodopsin⁴¹. Comparison of the inactive state structures of rhodopsin and the δOR reveals that the δOR has a more extensive polar network on the cytoplasmic side of the ligand binding pocket (Extended Data Fig. 9). This is particularly notable when considering the network of hydrogen bonds that maintain TM6 in the inactive conformation (Extended Data Fig. 9). As noted above, a similar polar network likely stabilizes the inactive states of the μOR and β₂AR (Extended Data Fig. 8). The less extensive polar network stabilizing the inactive state of rhodopsin is compensated for by the covalent inverse agonist, 11-cis-retinal. This balance of non-covalent polar interactions and a covalent ligand imparts rhodopsin with virtually no basal activity, but the ability to rapidly and efficiently respond to photoisomerization of retinal. In contrast, the polar network in the μOR and β₂AR help to maintain unliganded receptors in an inactive state, but the need to disrupt this network makes activation of these receptors energetically less efficient.

Expression and purification of μOR

Full length Mus musculus μOR bearing an amino-terminal Flag epitope tag and a carboxy-terminal 6xHis tag was expressed in Sf9 insect cells using the BestBac baculovirus system (Expression Systems). To facilitate removal of flexible amino- and carboxy-terminal regions, a tobacco etch virus protease recognition sequence was inserted after residue 51 and a rhinovirus 3C protease recognition sequence was inserted before residue 359. Insect cells were infected with baculovirus encoding μOR at a density of 4×10⁶ cells/ml for 48-60 hours at 27° C. Receptor was solubilized and purified in a final buffer comprised of 25 mM HEPES pH 7.4, 100 mM NaCl, 0.01% MNG (Anatrace), 0.001% cholesterol hemisuccinate (CHS) as previously described⁹.

Llama immunization and selection of μOR-binding nanobodies

Purified μOR bound to the antagonist naloxone and with the amino- and carboxy- termini cleaved was incubated with an excess of the agonist DMT¹-DALDA (NIDA Drug Supply Program) and further purified by size exclusion chromatography in a buffer comprised of 20 mM HEPES pH 7.5,100 mM NaCl, 0.1% dodecylmaltoside (DDM, Anatrace), 0.03% CHAPS, 0.01% CHS, and 1 μM DMT¹-DALDA. The resulting agonist bound receptor was reconstituted into phospholipid vesicles composed of DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine, Avanti Polar Lipids) and Lipid A in a 10:1 (w:w) ratio at a final receptor concentration of 1.3 mg/ml. The resulting reconstituted receptor was flash frozen in liquid nitrogen in 100 μg aliquots for llama immunization.

One llama was immunized over a period of 6 weeks with 0.3 mg of liposome-reconstituted μOR purified as described above and bound to the agonist DMT¹-DALDA. A phage display library of nanobodies was prepared from peripheral blood lymphocytes as previously described⁴². μOR-binding nanobodies were identified by selecting phage that bound liposome-reconstituted μOR in the presence or absence of agonist. Ten clones from three families were enriched during rounds of selection. One family, which includes Nb39, bind the intracellular surface and function as G protein mimetics. Another family of nanobodies, including clone Nb35, was identified that competes directly with orthosteric antagonists of the μOR and binds at the extracellular surface.

Expression and purification of nanobodies

Nanobodies bearing a carboxy-terminal 6xHis tag were expressed in the periplasm of Escherichia coli strain WK6. Cultures were grown to an OD₆₀₀ of 1.0 at 37°C in Terrific Broth medium containing 0.1% glucose, 2 mM MgCl₂, and 50 μg/mL ampicillin and induced with 0.5 mM isopropyl-β-D-thiogalactoside (IPTG). Cells were harvested after overnight growth at 25°C and incubated in a buffer containing 200 mM Tris pH 8.0, 0.5 mM EDTA, 500 mM sucrose, 0.5 mg/mL lysozyme for 1 hour at 25°C. Bacteria were osmotically lysed by rapid dilution in a 4× volume of water. The periplasmic fraction was isolated by centrifugation of cell debris, and was supplemented with NaCl to a final concentration of 150 mM as well as imidazole to a final concentration of 25 mM. Nanobodies were purified from the periplasmic fraction by nickel affinity chromatography, and were subsequently purified by size exclusion chromatography in a buffer comprised of 25 mM HEPES pH 7.5 and 100 mM NaCl. Peak fractions were pooled and concentrated to approximately 5 mM.

Radioligand characterization of BU72 and Nb39

Radioligand competition assays were performed using purified μOR reconstituted into high-density lipoprotein (HDL) particles comprised of the lipids POPC and POPG (Avanti Polar Lipids) in a 3:2 molar ratio as previously described^43,44. Heterotrimeric G_i was prepared by co-expressing human G_iα1, human Gβ1 and Gγ2 subunits in Hi5 insect cells using baculoviruses encoding the individual subunits. The G protein was purified as previously described²⁵.

For competition binding experiments, a mixture of 0.02 nM μOR and 2.3 nM ³H-diprenorphine (³H-DPN, Perkin Elmer) was incubated with varying concentrations of agonist in a binding buffer comprised of 25 mM HEPES pH 7.4, 100 mM NaCl, and 0.1% BSA. Experiments were also performed with either 5 μM Nb39 or 500 nM G_i. Binding reactions were incubated for 4 hours at 25°C. Free radioligand was separated from bound radioligand by rapid filtration onto a Whatman GF/B filter pretreated with 0.1% polyethylenimine with the aid of a 48-well harvester (Brandel). Radioligand activity was measured by liquid scintillation counting. Competition binding data were fit to a one-site model for μOR alone and μOR with Nb39 and a two-site model for μOR incubated with G_i using GraphPad Prism 6.0.

Dissociation studies for BU72 were performed using the method of Motulsky and Mahan⁴⁵. ³H-DPN was diluted in an assay buffer comprised of 20 mM Tris, pH 7.4, 150 mM NaCl, and 0.05% BSA containing μOR in HDL particles with either vehicle or different concentrations of BU72 alone or in the presence of Nb39. Binding reactions were incubated at 25°C in the dark and nonspecific binding was determined in the presence of 10 μM naloxone. Aliquots of this binding reaction were removed at specified time points over the course of 2 or 8 hours and filtered through Whatman GF/C filters with the aid of a Brandel harvester. As above, radioligand activity was measured by liquid scintillation counting. Dissociation rates for BU72 were determined by fitting data in the ‘kinetics of competitive binding’ program in GraphPad Prism 6.02. For K1 and K2, rates of ³H-DPN association and dissociation were determined through independent studies following the same method as above.

Purification and crystallization of μOR-BU72-Nb39 complex

Initial crystals of a μOR-BU72-Nb complex diffracted to low resolution. In an effort to improve the specific activity of purified μOR, we utilized an extracellular binding nanobody, Nb35 during the purification of the receptor. Nb35 binds in the orthosteric pocket of μOR with a K_i of 12 nM. Purified ligand-free μOR, with the amino- and carboxy- termini cleaved was mixed with a 2.5× excess of Nb35, and the complex was isolated by Ni-NTA chromatography. Functional μOR in complex with Nb35 was eluted in a buffer containing 25 mM HEPES pH 7.5, 100 mM NaCl, 0.01% MNG, 0.001% CHS, and 250 mM imidazole. The agonist BU72 and Nb39 were then added in excess. In the presence of Nb39, BU72 has exceptionally high affinity for the μOR and displaces Nb35, resulting in the desired μOR-BU72-Nb39 complex. The μOR-BU72-Nb39 complex was isolated by size exclusion chromatography and concentrated to approximately 50 mg/mL for crystallization trials.

For crystallization, the purified μOR-BU72-Nb39 complex was reconstituted into a 10:1 (w:w) mixture of monoolein and cholesterol (Sigma). The protein solution and lipid were mixed in a 1:1.5 ratio (w:w) and the lipidic cubic phase was attained using the two-syringe method⁴⁶. 30 nL of the resulting mesophase was dispensed onto 96-well glass plates and overlaid with 500 nL of precipitant solution using a Gryphon LCP robot (Art Robbins Instruments). Crystals grew in a precipitant comprised of 15-25% PEG300, 100 mM HEPES pH 7.0-7.5, 1% 1,2,3-heptanetriol, 0.5-1.0% Polypropylene glycol P 400 (Hampton Research) and 100-300 mM (NH₄)₂HPO₄. Crystals were observed after 2 days and reached full size in 1 week. Crystals were harvested with mesh grid loops (MiTeGen) and flash frozen in liquid nitrogen. Loops were screened at the Advanced Photon Source GM/CA beamlines 23ID-B and 23ID-D. Diffraction data were collected at a wavelength of 1.033 Å using a 10 μm beam with 5 fold attenuation and exposed for 0.5-1 s. An oscillation width of 0.1-0.5 degrees was used and diffraction images from 4 crystals were merged to create the final data set.

Structure determination and refinement

Diffraction images were indexed, scaled, and merged by XDS⁴⁷ with statistics summarized in Extended Data Table 1. The structure was determined by molecular replacement in Phaser⁴⁸ with initial search models including inactive μOR (PBD ID: 4DKL) and Nb80 (PDB ID: 3P0G). The structural model was iteratively rebuilt in Coot⁴⁹ and refined using Phenix⁵⁰. Final refinement statistics are summarized in Extended Data Table 1. Molprobity⁵¹ was used for structure analysis and indicated that 96.3% of residues where within favored Ramachandran regions with 3.7% in allowed regions. No residues were identified as Ramachandran outliers. Figures were prepared using MacPyMol (Schrödinger).

System setup for molecular dynamics simulations

Simulations of the μ opioid receptor (μOR) were based on both the antagonist-bound inactive-state crystal structure (PDB ID: 4DKL) and the agonist-bound active-state crystal structure described in this manuscript. Coordinates were prepared by first removing all non-ligand and non-receptor molecules except for the cholesterol neighboring TM7 and for crystallographic water molecules near the receptor. For inactive μOR simulations, the T4 lysozyme was removed and acetyl and methylamide capping groups were placed on R263^ICL3 and E270^ICL3. For active μOR simulations, the nanobody was removed. In both cases, Prime (Schrödinger, Inc.) was used to model missing side-chains, and capping groups were then added to the N- and C- termini of the receptor. Histidine residues were simulated as the neutral Nε tautomer. Other titratable residues were simulated in their dominant protonation state at pH 7 except for D114^2.50, which was charged in inactive simulations and neutral in active simulations. A sodium ion was placed adjacent to D114^2.50 in inactive simulations.

The μOR was simulated in seven distinct conditions. These include: (1) The unliganded, inactive μOR, prepared by deleting the covalently bound, co-crystallized ligand, β-FNA, and adding a proton in its place to K233^5.39; (2) The inactive μOR with the co-crystallized ligand β-FNA; (3) The inactive μOR with agonist β-FOA (which does not bind covalently); (4) The unliganded, active μOR, prepared by deleting the co-crystallized ligand, BU72; (5) The active μOR with the co-crystallized ligand BU72; (6) The active μOR with the co-crystallized ligand BU72, with the N-terminal peptide deleted; (7) The active μOR with the antagonist BU74, with the N-terminal peptide deleted. Simulations of the active μOR without N-terminal peptide were prepared by deleting residues 52 through 64 of the receptor. Simulations with β-FOA were prepared by docking β-FOA to the crystallographic pose of β-FNA. Simulations with BU74 were prepared by docking BU74 to the crystallographic pose of BU72 and rotating the torsion angle of the methylcyclopropyl group to agree with that of β-FNA's methylcyclopropyl group.

We performed three to six simulations per condition (Supplementary Section). Simulations in a given condition were initiated from identical structures, but with initial atom velocities assigned independently and randomly.

It should be noted that in all liganded simulations, including those with β-FNA, β-FOA, BU72, and BU74, the ligand's tertiary amine nitrogen was protonated and therefore the ligand was simulated as a cation. This is necessary for the ligand to form the conserved salt bridge with neighboring D147^3.32.

Each of the resulting prepared μOR receptor structures was then aligned to the Orientations of Proteins in Membranes (OPM)⁵² entry for the inactive μOR using MacPyMOL (Schrödinger). The μOR was modified with disulfide bridges and inserted into a hydrated, equilibrated palmitoyloleoylphosphatidylcholine (POPC) bilayer using the CHARMM-GUI interface^53-56. Sodium and chloride ions were added to neutralize the system, reaching a final concentration of approximately 150 mM. All simulations contained one μOR receptor embedded in a lipid bilayer with 160 POPC molecules.

Simulation protocol and analysis

Each simulation was performed on two GPUs using the CUDA version of PMEMD (Particle Mesh Ewald Molecular Dynamics) in Amber14^57-59 with 2.5 fs time steps. Simulations were heated from 0 K to 100 K in the NVT ensemble and then from 100 K to 310 K in the NPT ensemble, both with 10.0 kcal mol⁻¹ Å⁻² harmonic restraints applied to the protein and to the lipids. Initial velocities were assigned randomly at the first heating step with Langevin dynamics. Subsequently, simulations were equilibrated in the NPT ensemble at 310 K (controlled with a Langevin thermostat) with pressure of 1 bar (controlled with anisotropic Berendsen weak-coupling barostat), with harmonic restraints on all protein atoms tapered off 1.0 kcal mol⁻¹ Å⁻² in a stepwise fashion every 2.0 ns starting at 5.0 kcal mol⁻¹ Å⁻² to 0.0 kcal mol⁻¹ Å⁻², for a total of 12.0 ns of equilibration. Bond lengths to hydrogen atoms were constrained using SHAKE. Production simulations were performed in the NPT ensemble at 310 K and 1 bar, using a Langevin thermostat for temperature coupling and a Monte Carlo barostat for pressure coupling, and were initiated from the final snapshot of the corresponding equilibration simulation. Nonbonded interactions were cutoff at 9.0 Å, and long-range electrostatic interactions were computed using the particle mesh Ewald (PME) method, with an Ewald coefficient β of approximately 0.31 Å⁻¹ and B-spline interpolation of order 4. The FFT grid size was 84 × 84 × 96 for all simulations. We performed a total of 27 simulations, which are summarized in the Supplementary Section.

We used the CHARMM36 parameter set for protein molecules, lipid molecules, and salt ions, and the CHARMM TIP3P model for water; protein parameters incorporated CMAP terms^60-63. These parameters were assigned using the ParmEd implementation of Chamber, provided with AmberTools14. Ligand parameters were based on the results from the CHARMM ParamChem web server, version 0.9.7.1⁶⁴, and incorporated specific modifications to partial charges. When ParamChem reported large penalties (errors) for estimated partial charges, better estimates were determined by submitting appropriately chosen fragments to ParamChem (see below).

Parameterization of β-FNA, the co-crystallized ligand in the inactive μOR structure, required additional steps due to the added complexity of its covalent bond with K233^5.39 on the receptor. To assign parameters to the rest of β-FNA, the molecule was fragmented as shown in “Supplementary Section on Parameterization” and then re-assembled. In addition, a molecule consisting of K233^5.39 bonded to the portion of β-FNA on the extracellular side of its morphinan scaffold was submitted to ParamChem, and the results were used to modify charges near the bond between the β-FNA carbon and Lys side-chain nitrogen. A modified version of the ParmEd program was used to form a bond between β-FNA and K233^5.39, to parameterize that and neighboring bonds, and to modify neighboring partial charges according to the derived parameters. The related ligand β-FOA (which does not bind covalently to the μOR) was divided into the same three fragments as β-FNA for parameterization, with the following modifications: the bond between the ligand and K233^5.39 was cleaved, a C=C double bond was formed between the two carbons adjacent to this bond, and the methylcyclopropyl group was replaced with a methyl group. The co-crystallized ligand of the active μOR, BU72, was similarly parameterized by splitting it into fragments, individually uploading the fragments to the ParamChem web server, and assembling the results together. The fragments can be found in the Supplementary Section. The ligand, BU74, was parameterized identically to BU72, except that its tertiary amine methyl group was replaced with a methylcyclopropyl group.

Quantitative analysis of trajectories was conducted in both cpptraj, an Amber14 software package, and VMD⁶⁵. Plots were smoothed with a moving average, with a symmetric triangular smoothing window. The window length was 20 ns, except during the first 10 ns of each simulation, when the smoothing window extended from time zero to double the time at which the data was recorded. Plots were rendered in R with the ggplot2 package as well as in Python with the Matplotlib package.

To compute movement of specified atoms from the active position towards the inactive position, as plotted in Fig. 4C, the active and inactive crystal structures and each simulation frame were aligned using the Cα atoms of residues 80–95, 102–13, and 181–205 (TMs 1, 2, and 4). For each specified atom, the atom position in simulation was projected onto the line connecting the atom positions in the active and inactive crystal structures, and the distance of the projected point from the position in the active crystal structure position was determined.