The κ-opioid receptor (KOR) represents a highly desirable therapeutic target for treating not only pain but also addiction and affective disorders1. However, the development of KOR

analgesics has been hindered by the associated hallucinogenic side effects2. The initiation of KOR signalling requires the Gi/o-family proteins including the conventional (Gi1, Gi2, Gi3, GoA

and GoB) and nonconventional (Gz and Gg) subtypes. How hallucinogens exert their actions through KOR and how KOR determines G-protein subtype selectivity are not well understood. Here we

determined the active-state structures of KOR in a complex with multiple G-protein heterotrimers—Gi1, GoA, Gz and Gg—using cryo-electron microscopy. The KOR–G-protein complexes are bound to

agonist binding at KOR. These results provide insights into the actions of opioids and G-protein-coupling specificity at KOR and establish a foundation to examine the therapeutic potential

of pathway-selective agonists of KOR.

Opioid receptors are G-protein-coupled receptors (GPCRs) that have important roles in pain sensation. Almost all clinically used opioids act through the μ-opioid receptor (MOR). However,

their use is associated with severe side effects, including a high potential for abuse, addiction and death due to respiratory depression in overdose3. The magnitude of these problems has

led to a search for opioid alternatives for the treatment of pain and related conditions4. The activation of opioid receptors recruits downstream effectors, including heterotrimeric G

proteins (including Gα, Gβ and Gγ subunits) and β-arrestins. Specifically, opioid receptors primarily couple to the Gαi/o family (Gi1, Gi2, Gi3, GoA, GoB, Gz and gustducin (Gg)) (Extended

Data Fig. 1a). Some of these subtypes can mediate non-overlapping signalling pathways depending on the GPCR involved5,6,7,8. Whether signalling through individual pathways has redundant

roles or separately drives the therapeutic efficacy and side effects of opioids remains mostly unclear.

KOR is a highly desirable therapeutic target for treating not only pain but also addiction and affective disorders. KORs have gained increasing attention owing to their unique analgesic

activity—they are predominantly expressed in pain-related neurons, and drugs that target KOR do not lead to addiction or cause death due to overdose as observed for MOR agonists1. The lack

of rewarding/euphorigenic effects initially encouraged the development of KOR-agonist drugs as non-addictive analgesics9. Potent and selective KOR agonists have been developed, and these

agonists produce effective peripheral and central analgesia. However, mood disorders such as dysphoria and psychotomimesis have been frequently observed as side effects of KOR agonists,

which has limited their therapeutic application2. Here we determined the atomic structures of KOR in complex with different G-protein transducers and hallucinogenic ligands to help to

elucidate the actions of opioids and the molecular basis for Gαi/o subtype selectivity.

Although many efforts have been dedicated to the structural and molecular basis underlying the differences between G-protein and arrestin signalling, the roles of individual G-protein

subtypes and the molecular determinants of subtype selectivity remain largely unclear. Sequence alignment of the seven Gi/o subtypes suggests that they could be further grouped into four

subclasses on the basis of sequence identity (Gi1, Gi2 and Gi3; GoA and GoB; and Gz and Gg) (Extended Data Fig. 1b). To further understand the role of KOR–G-protein coupling and signalling,

we determined the structures of KOR in complexes with four representative Gi/o subtypes (Gi1, GoA, Gz and Gg) at nominal resolutions of 2.71 Å, 2.82 Å, 2.65 Å, and 2.61 Å, respectively,

using single-particle cryo-electron microscopy (cryo-EM; Fig. 1a, Supplementary Fig. 1 and Extended Data Table 1). In particular, KOR–Gi1 and KOR–GoA are bound to a psychotropic salvinorin

analogue, methoxymethyl-salvinorin B (momSalB)10. However, cryo-EM experiments of KOR–Gz or KOR–Gg bound to momSalB yielded only low-resolution reconstructions (resolution of around 4.5–5 Å)

that prevented the delineation of detailed molecular interactions. Thus, we leveraged another highly potent KOR agonist, GR89,696 (ref. 11), to obtain high-resolution structures of KOR–Gz

and KOR–Gg.

a, Cartoon representations of KOR–G-protein complexes. Structures of KOR–Gi1 and KOR–GoA are bound to momSalB. Structures of KOR–Gz and Gg are bound to GR89,696. b, Structural alignment of

the four Gα subunits. Distances of movement from the N terminus are labelled.

The high-resolution maps of the four structures enabled unambiguous modelling of the agonist-bound heterotrimeric complexes (Supplementary Fig. 2). The overall differences between the four

structures are subtle (root mean square deviations (r.m.s.d.) of 0.5 Å), with the exception of the Gα subunit in each complex (Fig. 1b). G-protein interactions with the receptor are

canonically driven by the α5 and the N-terminal (αN) helices of the Gα subunit. The overlay of the four different G-protein subtypes showed that they adopt similar conformations in the α5

helix but differ in the extent of movement in the αN helix (Fig. 1b). In particular, relative to Gi1, both GoA and Gz exhibit a 6 Å displacement in the αN helix, whereas Gg has a smaller 2 Å

displacement. Notably, alignments of the MOR–Gi1 structure12 with KOR–Gi1 indicate that the αN helix of Gi1 in MOR displays a position that is distinct from that of KOR–Gi1, whereas the α5

helix shows an orientation and interaction pattern similar to those in KOR (Extended Data Fig. 1c).

The overall structures of KOR in the Gi1/oA/z/g-bound states are similar to the previously reported nanobody-stabilized active conformation (KOR–Nb39)13 (r.m.s.d., 0.8 Å) (Extended Data Fig.

1d). Notably, a comparison of these two structures reveals that the intracellular end of transmembrane helix 6 (TM6) in the KOR–Gi1 protein complex moves 2 Å closer to TM7. Nb39 stabilizing

a different receptor conformation is further supported by its positive allosteric ability to enhance agonist binding affinity (Extended Data Fig. 1e). Another feature unique to

G-protein-bound KOR is the presence of a well-defined intracellular loop 3 (ICL3) conformation that is absent in the Nb39-stabilized KOR, presumably due to its inherent flexibility (Extended

Data Fig. 1d). Similar differences have also been captured between MOR–Gi112 or β2AR–Gs14 and their corresponding nanobody-stabilized active states15,16, which further corroborate that a

nanobody can stabilize a conformational state that mimics but is not identical to the G-protein-coupled state.

KORs have a prominent role in the modulation of human perception. Salvinorins, such as salvinorin A (SalA)17,18, are a group of naturally occurring hallucinogens with dissociative effects

elicited by activating the central KORs. momSalB is a semi-synthetic analogue of SalA and displays similar in vivo pharmacology compared to SalA19,20. GR89,696 is a potent and long-lasting

KOR agonist that produces antinociception and dysphoria but with unknown hallucinogenic properties21. Different binding poses of momSalB and GR89,696 were observed in the KOR orthosteric

pocket. This is consistent with their divergent chemical structures—GR89,696 is an alkaloid (containing basic nitrogen atoms) and momSalB is a terpenoid (lacking basic nitrogen atoms) (Fig.

1a). The pyrrolidine nitrogen atom in GR89,696, as well as many other ligands including KOR’s endogenous dynorphin ligands22, is essential for the binding to KOR and enables the ligand to

act as a hydrogen-bond (H-bond) donor and forms a salt bridge with the carboxylate side chain of Asp1383.32 in the binding pocket (where the superscript values indicate Ballesteros–Weinstein

numbering for GPCRs23) (Fig. 2a). As salvinorin ligands (such as momSalB) lack the basic nitrogen atom, there are no attractive electrostatic interactions observed between the salvinorins

and Asp1383.32. Indeed, neither D1383.32A nor D1383.32N (the mutation in KOR DREADD24) showed detrimental effects in the binding affinity or agonistic potency of SalA, whereas both mutants

abolished the interaction with endogenous dynorphin ligands24,25,26. The mutation D1383.32N resulted in a significant loss of potency in U50,488 and GR89,696, but had minimal effects on

momSalB (Fig. 2b). The side chain of Asp1383.32 pointing to the methoxymethyl group of momSalB also explains an interesting observation that D1383.32N could further enhance the binding

affinity and potency of SalA and salvinorin B (SalB)24, probably due to the switch from the unfavourable acceptor–acceptor interaction to attraction resulting from the new H-bond

interactions between the side chain of mutated asparagine and methoxy oxygen of the ligand.

a, The binding poses of momSalB and GR89,696 in their respective complex structures. The salt bridge or H-bond interactions in Gz- and Gg-coupled structures are shown as black dashed lines.

This salt bridge or H-bond interaction is absent in momSalB-bound KOR. b, The highly conserved anchoring residue Asp1383.32 has a different role in momSalB, GR89,696 or U50,488-mediated KOR

activation. Data are normalized to the percentage of the reference agonist U50,488. Data are grouped data ± s.e.m. of n = 3 biological replicates. The full quantification parameters for this

experiment are provided in Supplementary Table 1. c, Specific residues in the orthosteric pockets that interact with momSalB or GR89,696. Note that the I1353.29L mutation was included in

KOR structure constructs to increase the expression level. d, Mutagenesis screening of binding-pocket residues using G-protein-mediated cAMP inhibition assays. The effect on the potency of

momSalB or GR89,696 was quantified on the basis of the log[EC50] values. Data are log[EC50] ± s.e.m. of n = 3 biological replicates. Statistical significance for each mutant was calculated

using one-way analysis of variance (ANOVA) with Dunnett’s multiple-comparison test compared with the wild type (WT); *P  GoA > Gz > Gg) (Fig. 4g). This is consistent with measurements of the

binding affinity of different G proteins (Gi1 > GoA > Gz > Gg) at KOR. These intrinsic differences in G proteins, including the binding affinity and coupling efficiency, add pharmacological

evidence to determinants for G-protein-subtype selectivity. Our structural observations from different G proteins in complex with KOR show that the Gi/o family subtypes share a highly

conserved mechanism in interacting with KOR, but that each maintains pharmacological differences. Considering that many GPCRs can couple to different G-protein families, such as the β2AR

coupling with both Gs and Gi/o (refs. 38,54), whether β2AR displays differential binding affinities with Gs and Gi/o may help to explain its G-protein coupling specificity.

Furthermore, the allosteric activity of G proteins can be regulated by GDPs or GTPs that decouple G proteins from the receptor. It is known that nucleotide-specific conformations exist

between nucleotide-free G proteins and GDP- or GTP-bound G proteins. When comparing the crystal structure of the uncoupled GDP-bound Gi1 heterotrimer55 and nucleotide-free (KOR–Gi1, GoA, Gz

and Gg) heterotrimers (Extended Data Fig. 11a), several conformational displacements are noted (Extended Data Fig. 11b). The activated receptor engages the C terminus of the α5 helix of

Gαi1, which undergoes an upward helical extension (8.6 Å) into the receptor core (Extended Data Fig. 11c) compared with the uncoupled G protein structure. The insertion of the α5 helix into

the transmembrane helical bundle of the receptor has the following two consequences. First, the loop connecting the α5 helix and β6 strand moves outward 5 Å. Second, the movement of the α5

helix disrupts the original hydrophobic interactions between the α5 and α1 helices, leading to a displacement of the P loop. Both the P loop translocation and loss of coordination with GDPs

are necessary for GDP release44,45. In agreement with the ternary complex model, our saturation binding data show that GDPs or GTPs act as negative regulators of agonist binding kinetics in

the presence of G proteins. One limitation of this study is that we used an in vitro overexpressed system with engineered receptors and G proteins to measure ligand activity, which cannot be

extended to in vivo without further experiments.

In summary, we have elucidated the molecular interaction details between highly conserved Gi/o subtypes and KOR using cryo-EM-derived atomic models. We have also examined the structural

determinants of ligand selectivity and efficacy at KOR. Using structural pharmacology analysis, we revealed the intrinsic differences between these previously under-represented Gi/o subtypes

and demonstrated that subtype selectivity is probably a combinational result of receptor conformational dynamics, the binding affinity of G proteins and cooperativity between agonist

binding and G-protein coupling. Such findings are important both in understanding GPCR-mediated signalling and in the generation of new research tools and therapeutics based on the potential

of G-protein-selective agonists.

For the human KOR, we used a construct the same as the previously determined active-state KOR13. In brief, the construct (1) lacks N-terminal residues 1–53; (2) lacks C-terminal residues

359–380; (3) contains Met1–Leu106 of the thermostabilized apocytochrome b562 RIL (BRIL) from E. coli (M7W, H102I, R106L) in place of receptor N terminus residues Met1–His53. This N-terminal

Bril will be removed using a PreScission cleavage site in the end. The single chain Fab scFv16 has the same sequence as previously reported56. A 6×His tag was added to the C-terminal scFv16

sequence with a PreScission cleavage site inserted between. For the G-protein heterotrimers, individual G-protein constructs (Gi1, GoA, Gz and Gg) were engineered (labelled as dominant

negative, DN)47 for the binding of scFv16, and then subcloned into a designed vector that co-expresses the Gβ1 and Gγ2. Further modifications were made to enable a stable complex between

KOR, G-protein heterotrimer and scFv16. Specifically, Gi1(DN) includes S47N, E245A, G203A and A326S. GoA(DN) includes C3S, S47N, G204A, E246A, A326S and M249K. For Gz(DN), the N-terminal

sequence was replaced with the Gi2 sequence to allow for better interaction with scFv16; other mutations include S47N, G204A, E246A, R249K, N262D and A327S. For Gg(DN), the N-terminal

sequence was replaced with the Gi2 sequence; other mutations include S47A, G203A, E245A, H248K, T261D, A326S and N251D.

The Bac-to-Bac Baculovirus Expression System was applied to generate high-quality recombinant baculovirus (>10−9 viral particles per ml) for protein expression (KOR, G-protein heterotrimers

and scFv16). For the expression of KOR–G–scFv16 protein complex, each heterotrimeric G protein, including Gα (Gi1, GoA, Gz or Gg), Gβ1 and Gγ2 was coexpressed with KOR and scFv16,

respectively, by infection of Spodoptera frugiperda Sf9 cells at a cell density of 2.5 × 106 cells per ml in ESF921 medium (Expression System) with the P1 baculovirus at a multiplicity of

infection (MOI) ratio of 2:2:0.5. Cells were collected by centrifugation (125 rpm at 27 °C) for 48 h after infection, washed with HN buffer (25 mM HEPES pH 7.4, 100 mM NaCl), and stored at

−80 °C for future purification.

The compounds used in this study—(−)-U50,488 (0496) and GR89,696 (1483)—were purchased from Tocris. momSalB was synthesized by a method described previously57. After purification by silica

gel column chromatography, momSalB was a single spot on TLC (silica, 20% ethyl acetate, dichloromethane) with an Rf of 0.49. An NMR spectrum of momSalB was collected to confirm the chemical

identity (Supplementary Fig. 6), which is consistent with the expected spectrum reported previously58.

We thawed the cell pellet and incubated it in buffer containing 20 mM HEPES pH 7.5, 50 mM NaCl, 1 mM MgCl2, 2.5 units Apyrase (NEB), 10 μM agonist (final concentration) and protease

inhibitors (500 mM AEBSF, 1 mM E-64, 1 mM leupeptin, 150 nM aprotinin) for 1.5 h at room temperature. We then collected the membrane by centrifugation at 25,000 rpm for 30 min at 4 °C. The

membrane was solubilized in buffer (40 mM HEPES pH 7.5, 100 mM NaCl, 5% (w/v) glycerol, 0.6% (w/v) lauryl maltose neopentyl glycol (LMNG), 0.06% (w/v) cholesteryl hemisuccinate (CHS), 10 μM

agonist and protease inhibitors) with 200 μg scFv16 in the cold room. After 5 h, the supernatant was collected by centrifugation at 30,000 rpm for 30 min at 4 °C and incubated with 1 ml

TALON IMAC resin (Clontech) and 20 mM imidazole overnight in the cold room. The next day, the resin was collected and washed with 10 ml buffer containing 20 mM HEPES pH 7.5, 100 mM NaCl, 30 

mM imidazole, 0.01% (w/v) LMNG, 0.001% (w/v) CHS, 5% glycerol and 5 μM agonist. The protein was then eluted with the same buffer supplemented with 300 mM imidazole, concentrated and further

purified by size-exclusion chromatography on the Superdex 200 increase 10/300 column (GE healthcare), which was pre-equilibrated with 20 mM HEPES pH 7.5, 100 mM NaCl, 100 μM TCEP, 0.00075%

(w/v) LMNG, 0.00025% (w/v) glyco-diosgenin (GDN) and 0.00075% (w/v) CHS, 1 μM agonist. Peak fractions were collected, concentrated and incubated with PNGase F (NEB), PreScission protease

(GenScript) to remove the potential glycosylation and N-terminal His–BRIL, respectively, and 100 μg scFv16 at 4 °C overnight. The next day, cleaved His–BRIL and protein, uncleaved protein

and proteases were separated by the same procedure as described above. Peak fractions were concentrated to 3–5 mg ml−1 for electron microscopy analysis. Four KOR–G-protein–scFv16 complexes

were purified according to the same procedure except that different agonists were used.

The scFv16 protein was expressed by infection of Sf9 cells at a cell density of 2.5 × 106 cells per ml in ESF921 medium (Expression System) with the P1 baculovirus at an MOI of 2. After 96 

h, the cell culture medium containing secreted scFv16 protein was collected by centrifugation at 4,000 rpm for 15 min. The pH of the supernatant was adjusted to 7.5 by addition of Tris-base

power. Chelating agents were quenched by the addition of 1 mM nickel chloride and 5 mM calcium chloride and incubation with stirring for 1 h at room temperature and 5 h in the cold room. We

removed the precipitates by centrifugation and the resultant supernatant was further cleaned with 0.45 μm filter paper, and incubated with 2 ml Ni-NTA resin and 10 mM imidazole overnight in

the cold room. The Ni-NTA resin was washed the next day with 20 ml buffer (20 mM HEPES pH 7.5, 100 mM NaCl, 0.00075% (w/v) LMNG, 0.000075% (w/v) CHS, 0.00025% (w/v) GDN, 20 mM imidazole).

The protein was eluted with the same buffer supplemented with 300 mM imidazole, concentrated and further purified on the Superdex 200 increase 10/300 column. Monomeric fractions were pooled,

concentrated, flash-frozen in liquid nitrogen and stored at −80 °C until future use.

The expression of heterotrimeric G protein was achieved by infection of Sf9 cells at a cell density of 2.5 × 106 cells per ml in ESF921 medium (Expression System) with the P1 baculovirus at

an MOI of 2. After 48 h, cells were collected and lysed in buffer containing 200 mM NaCl, 40 mM HEPES pH 7.5, 0.2% Triton X-100, 5% glycerol, 3 mM β-me and protease inhibitors. The

supernatant was isolated by centrifugation at 40,000 rpm for 50 min and incubated with 1 ml Ni-NTA resin and 20 mM imidazole overnight at 4 °C. The resin was collected the next day and

washed with 20 ml buffer containing 100 mM NaCl, 20 mM HEPES pH 7.5, 5% glycerol, 20 mM imidazole and 3 mM β-me. The protein was then eluted with elution buffer (300 mM NaCl, 20 mM HEPES pH 

7.5, 5% glycerol, 3 mM β-me and 300 mM imidazole), concentrated and further purified on the Superdex 200 increase 10/300 column, which was pre-equilibrated with buffer the same as the

elution buffer except without the imidazole. The peak fractions were concentrated, flash-frozen in liquid nitrogen and stored at −80 °C for future binding assays.

The purified samples (3–4 μl) were applied to glow-discharged 300-mesh Au grids (Quantifoil R1.2/1.3) individually and vitrified using a Vitrobot mark IV (Thermo Fisher Scientific). Cryo-EM

imaging was performed on the Talos Artica system operated at 200 kV at a nominal magnification of ×45,000 using a Gatan K3 direct electron detector at a physical pixel size of 0.88 Å. Each

stack video was recorded for 2 to 2.7 s in 60 frames at a dose rate of about 15 e− px−1 s−1, leading to a total exposure dose indicated in Extended Data Table 1. Videos were collected

automatically with SerialEM59 using an optimized multishot array procedure60.

Dose-fractioned image stacks were processed for beam-induced motion correction followed by contrast transfer function estimation. Particles were selected using Blob particle picker,

extracted from the micrograph and then used for 2D classification and 3D classification followed by non-uniform refinement. All of these steps were performed in cryoSPARC61,62.

Maps from cryoSPARC were used for map building, refinement and subsequent structural interpretation. The dominant-negative Gi1 trimer model and scFv16 model were adapted from the cryo-EM

structure of the MRGPRX2–Gi1 complex (Protein Data Bank (PDB): 7S8M)63. GoA, Gz and Gg trimer models were built from the Gi1 trimer model, followed by mutating the non-conserved residues

back to the wild-type GoA, Gz and Gg. The receptor KOR model was taken from the active-state KOR–Nb39 structure (PDB: 6B73)13. The receptor, G proteins and scFv16 were docked into the

cryo-EM map using Chimera64. The complex models (KOR–G-protein–scFv16) were manually built in Coot65, followed by several rounds of real-space refinement using Phenix66. The model statistics

were validated using Molprobity67. Structural figures were prepared using Chimera or PyMol (https://pymol.org/2/).

For the KOR–Gαi-mediated cAMP inhibition assay, HEK293T (ATCC CRL-11268) cells were co-transfected with human KOR or various mutants along with a split-luciferase-based cAMP biosensor

(GloSensor, Promega) at a 1:1 ratio (KOR:GloSensor). After 16 h, the transfected cells were plated into poly-l-lysine-coated 96-well white clear-bottom cell culture plates with DMEM + 1%

dialysed FBS at a density of 40,000–50,000 cells per 200 μl per well and incubated at 37 °C with 5% CO2 overnight. The next day, 3× drug solutions were prepared in fresh drug buffer (20 mM

HEPES, 1× HBSS, 0.3% bovine serum albumin (BSA), pH 7.4). The plates were decanted the next day and received 40 μl per well of drug buffer (20 mM HEPES, 1× HBSS, pH 7.4) followed by addition

of 20 μl of 3× drug solutions for 15 min in the dark at room temperature. Cells then received 20 μl luciferin (4 mM final concentration) supplemented with isoproterenol (300 nM final

concentration), stimulating the production of endogenous cAMP through β2 adrenergic Gs activation, and incubated in the dark at room temperature. After 15 min, luminescence intensity was

quantified using the Mithras LB 940 multimode microplate reader (Berthold Technologies). Data were plotted as a function of drug concentration, normalized to percentage U50,488 stimulation,

and analysed using log (agonist) versus response in GraphPad Prism (v.9.3.1).

To measure the agonist-stimulated G-protein (wild type and mutants) activation by KOR and various mutants, a BRET2-based cell assay was used. Specifically, four plasmids (KOR, Gα, Gβ, Gγ)

were used, in which each Gα is tagged with a luciferase (Rluc8) and Gγ is tagged with an N-terminal GFP. Specifically, the Gαi1/Gβ3/Gγ9, GαoA/Gβ3/Gγ8, Gαz/Gβ3/Gγ1 and Gαg/Gβ3/Gγ1

combinations were used for BRET2 Gi1, GoA, Gz and Gg experiments, respectively. Detailed information of the GFP-Gγ and Gα-Rluc8 constructs was described previously30. HEK293T cells were then

transfected with the four plasmids (KOR, Gα-Rluc8, Gβ, Gγ–GFP) using a 1:5:5:5 DNA ratio of receptor:Gα-RLuc8:Gβ:Gγ-GFP2 (100 ng receptor, 500 ng Gα–RLuc8, Gβ and Gγ–GFP2 for 10 cm dishes).

Transit 2020 (Mirus Biosciences) was used to complex the DNA at a ratio of 2 μl Transit per μg DNA in Opti-MEM (Gibco-Thermo Fisher Scientific). Then, 16 h after transfection, cells were

plated in poly-l-lysine-coated 96-well white clear-bottom plates in plating medium (DMEM + 1% dialysed FBS) at a density of 40,000–50,000 cells in 200 μl per well and incubated overnight.

The next day, the plates were decanted and washed once with 60 μl drug buffer (20 mM HEPES, 1× HBSS, pH 7.4) and then 60 μl drug buffer containing coelenterazine 400a (Nanolight Technology)

at a final concentration of 5 μM was added to each well. After 5 min for substrate diffusion, 30 μl 3× drug solutions in fresh drug buffer (20 mM HEPES, 1× HBSS, 0.3% BSA, pH 7.4) was added

to each well and incubated for an additional 5 min. Finally, the plates were read on the Mithras LB 940 multimode microplate reader (Berthold Technologies) with 400 nm (RLuc8-coelenterazine

400a) and 510 nm (GFP2) emission filters for 1 s per well. The GFP to Rluc8 ratio was calculated, plotted as a function of drug concentration, normalized to percentage U50,488 stimulation

and analysed using log (agonist) vs response in GraphPad Prism (v.9.3.1).

Saturation binding assays were performed using the construct BRIL-wt-KOR(54–368) reconstituted into nanodiscs comprised of KOR, spMSP1D1 and lipid mixture (POPC:POPE:POPG = 3:1:1) at a molar

ratio of 1:3:100. Binding assays were set-up in 96-well plates in standard binding buffer (50 mM Tris-HCl, 10 mM MgCl2, 0.1 mM EDTA, 0.1% BSA, pH 7.4) at room temperature. Saturation

binding assays with 0.1–20 nM 3H-U69,593 in the standard binding buffer were performed to determine the equilibrium dissociation constant (Kd) and Bmax. To determine the effects of G

proteins on 3H-U69,593 binding, each G protein (final concentration 1 μM) was incubated with 3H-U69,593 and homogenous membrane fractions for 3.5 h at room temperature. Data were analysed

using GraphPad Prism (v.9.3.1) using a one-site model.

For the competitive binding assay, 3H-JDTic (0.68 nM), homogenous membrane fractions expressing KOR and 3× GR89,696 solutions were incubated in 96-well plates in standard binding buffer in

the absence or presence of four G proteins in various concentrations (final concentration: 1,900 nM, 190 nM, 19 nM, 1.9 nM, 0 nM) for 3.5 h at room temperature in the dark, and then

terminated by rapid vacuum filtration onto chilled 0.3% PEI-soaked GF/A filters followed by three quick washes with cold wash buffer (50 mM Tris-HCl, pH 7.4) and read. Results (with or

without normalization) were analysed using GraphPad Prism (v.9.3.1) using one-site or allosteric IC50 shift models.

The cell-surface expression levels of wild-type KOR and its mutants were measured using an enzyme-linked immunosorbent assay (ELISA). In brief, HEK293T (ATCC CRL-11268) cells were

transiently transfected with wild-type KOR and KOR mutant DNA at the same quantity. After 24 h, cells were plated in poly-l-lysine-coated 96-well white clear-bottom plates in plating medium

(DMEM + 1% dialysed FBS) at a density of 40,000–50,000 cells in 200 μl per well and incubated overnight. The next day, plates were decanted and fixed with 4% (w/v) paraformaldehyde for 10 

min at room temperature. Cells were then washed twice with 1× phosphate-buffered saline (PBS) (pH 7.4) and blocked by 1× PBS containing 0.5% (w/v) non-fat milk for at least 30 min at room

temperature followed by incubation with anti-Flag (M2)–horseradish peroxidase-conjugated antibodies (Sigma-Aldrich, A8592) diluted 1:20,000 in the same buffer for 1 h at room temperature.

After washing three times with 1× PBS, 1-Step Ultra-TMB ELISA substrate (Thermo Fisher Scientific, 34028) was added to the plates and the plates were incubated at 37 °C for 15–30 min and

terminated by addition of 1 M sulfuric acid (H2SO4) stop solution. Finally, the plates were read at a wavelength of 450 nm using the BioTek Luminescence reader. The data were analysed using

GraphPad Prism (v.9.3.1).

To measure the expression levels of four wild-type G proteins and their mutants, HEK293T (ATCC CRL-11268) cells were transiently transfected with the same quantity of wild-type and mutant G

proteins DNA. After 16 h, cells were plated in poly-l-lysine-coated 96-well white clear-bottom plates in plating medium (DMEM + 1% dialysed FBS) at a density of 40,000–50,000 cells in 200 μl

per well and incubated overnight. The next day, the plates were decanted and washed once with 60 μl drug buffer (20 mM HEPES, 1× HBSS, pH 7.4), then 60 μl drug buffer containing

coelenterazine 400a (Nanolight Technology) at a final concentration of 5 μM was added to each well. After 5 min for substrate diffusion, plates were read in a Mithras LB 940 multimode

microplate reader (Berthold Technologies) with 400 nm (RLuc8-coelenterazine 400a) and 510 nm (GFP2) emission filters for 1 s per well. The Rluc8 values represented the G-protein expression

levels and were plotted in the GraphPad Prism (v.9.3.1).

Analysis of GTPase activity of G proteins (Gi1, GoA, Gz, Gg) was performed by using a modified protocol of the GTPase-Glo assay (Promega). G proteins were serially (1:1) diluted into various

concentrations with a buffer of 300 mM NaCl, 20 mM HEPES pH 7.5 and 1 mM DTT, and 5 μl was dispensed into each well of a 384-well plate. The reaction was initiated by adding 5 μl 1 μM GTP

solution to 5 μl G proteins. After incubation for 90 min at room temperature, 10 μl reconstituted GTPase-Glo reagent was added to the sample and incubated for 30 min at room temperature.

Luminescence was measured after addition of 20 μl detection reagent and incubation for 10 min at room temperature using the Mithras LB 940 multimode microplate reader (Berthold

Technologies). The data were analysed using GraphPad Prism (v.9.3.1).

The Gromacs simulation engine (v.2020.3)68 was used to run all molecular dynamics simulations under the Charmm36 force-field topologies and parameters69,70. Charmm force-field parameters and

topologies for the ligands momSalB and GR89,696 were generated using Charmm-GUI’s Ligand Reader & Modeller tool70. The loop grafting and optimization for modelling missing side chains and

loops was performed in the ICM-Pro (v.3.9-2b) molecular modelling and drug discovery suite (Molsoft)71. The structurally conserved helix-8 (Hx8) amphipathic helical motifs in KOR were

modelled using human antagonist-bound KOR (PDB: 4DJH)26 as the template structure. The lobe in Gi1, GoA, Gz and Gg proteins was modelled using a human agonist-bound CB2–Gi structure (PDB:

6PT0)72. Structure regularization and torsion profile scanning were performed using ICMFF force field73. The GR89,696-bound structures of KOR complexes with Gz and Gg proteins as well as

momSalB-bound KOR with Gi1 and GoA proteins were then uploaded to the Charmm-GUI webserver69, where the starting membrane coordinates were determined by the PPM74 server using the Charmm-GUI

interface. The complexes were then embedded in a lipid bilayer composed of 1,2-dipalmitoyl-sn-glycero-3-phosphatidylcholine (DPPC), 1,2-dioleoyl-sn-glycero-3-phosphatidylcholine (DOPC) and

cholesterol (CHL1) following the recommended ratio of 0.55:0.15:0.30, respectively75. The GR89,696-bound KOR complex with Gz contained 330 DPPC, 90 DOPC and 180 CHL1 lipids, 64,400 water

molecules, and 178 sodium and 176 chloride ions. The GR89,696-bound KOR complex with Gg contained 330 DPPC, 90 DOPC and 180 CHL1 lipids, 64,227 water molecules, and 184 sodium and 175

chloride ions. The momSalB-bound KOR complex with Gi1 contained 220 DPPC, 60 DOPC and 120 CHL1 lipids, 43,172 water molecules, 124 sodium and 116 chloride ions. The momSalB-bound KOR complex

with GoA contained 220 DPPC, 60 DOPC and 120 CHL1 lipids, 41,663 water molecules, and 126 sodium and 113 chloride ions. All of the systems were first processed for 50,000 steps of initial

energy minimizations, then 60 ns of equilibration, followed by production runs of up to 750 ns for the KOR–Gg based system and 550 ns for the rest (Gi1, GoA and Gz-bound KOR systems). The

simulations were carried out on GPU clusters at the University of Southern California’s High-Performance Computing Center. The temperature of 310 K and v-rescale thermostat algorithm were

used during the production run76. The analyses of molecular dynamics trajectories were performed using the GROMACS software package68.

For BRET2 and cAMP-inhibition assays, in the case of more than two groups, log-transformed EC50 values were first analysed using one-way ANOVA. If significant, the Dunnett’s

multiple-comparison test was used to compare each mutant with the wild-type one, and the Tukey’s multiple-comparison test was used to compare log-transformed EC50 values between each group.

In the case of two groups, log-transformed EC50 values were analysed using unpaired two-tailed Student’s t-tests to compare each mutant with a wild-type receptor. For the cell-surface

expression studies, the optical density at 450 nm values of each mutant were normalized to the wild-type KOR receptor (normalized as 100%), and the resultant values were then first analysed

using one-way ANOVA. If significant, a Dunnett’s multiple-comparison test was used to compare each mutant with the wild-type receptor. For G-protein expression studies, the Rluc values of

each mutant were normalized to the wild-type G protein (normalized as 100%), and the resultant values were then first analysed using one-way ANOVA. If significant, a Dunnett’s

multiple-comparison test was used to compare each mutant with the wild-type G protein. For radioligand binding and GTP turnover assays, data were analysed using unpaired two-tailed Student’s

t-tests. In one-way ANOVA and unpaired two-tailed Student’s t-test analysis, the significance threshold was set at α = 0.05. Asterisks denote statistical significance; *P