Structure-guided allosteric modulation of the delta opioid receptor

Abstract

Opioid analgesics remain essential for pain management but are associated with significant adverse effects, including respiratory depression, tolerance, and dependence. The δ-opioid receptor (δOR) represents a promising therapeutic target for developing safer opioid analgesics with reduced adverse effects compared to conventional μ-opioid receptor-targeting drugs. Positive allosteric modulators (PAMs) offer advantages over direct agonists by enhancing endogenous opioid signaling while preserving natural spatiotemporal activation patterns, potentially avoiding tolerance and dependence issues. Here, we present high-resolution cryo-EM structures of δOR complexed with the peptide agonist DADLE and the PAM MIPS3614, revealing a lipid-facing allosteric binding site formed by transmembrane helices 2, 3, and 4. MIPS3614 stabilizes the active receptor conformation through a critical hydrogen bond with residue N131^3.35 in the conserved sodium binding site, a key regulatory region controlling GPCR activation. Comprehensive mutagenesis, molecular dynamics simulations, and structure-activity relationships validate this proposed mechanism. Structure-guided optimization yields MIPS3983 with enhanced binding affinity and retained cooperativity. Our findings describe a molecular basis for δOR allosteric modulation and provide structural information relevant to the rational design of opioid therapeutics with improved safety profiles.

In this study, authors report cryo-EM structures of the delta opioid receptor bound to a positive allosteric modulator, showing it forms a hydrogen bond that disrupts a conserved sodium-binding site, stabilizing the active receptor and guiding drug design.

Introduction

Opioid analgesics are essential for pain management but are associated with significant adverse effects, including respiratory depression, gastrointestinal dysfunction, tolerance and dependence¹. With prescription-based drug dependence already beyond crisis point, the prevalence of these safety issues has contributed to the current opioid crisis, emphasising the need for therapeutic approaches with improved safety profiles.

Opioid receptors comprise a family of four G protein-coupled receptors (GPCRs): mu (μ), delta (δ), kappa (κ) and nociceptin/orphanin FQ peptide (NOP) receptors, which predominantly couple to inhibitory G_i proteins². While μ-opioid receptor (µOR) activation produces both the beneficial analgesia and detrimental side effects of conventional opioids^3,4, the δ-opioid receptor (δOR) has emerged as a promising alternative therapeutic target⁵. δOR agonists demonstrate efficacy in preclinical models of chronic pain, migraines and mood disorders with reduced respiratory depression and dependence liability compared to µOR agonists^6–10. Despite this promise, first-generation δOR agonists targeting the orthosteric binding site encountered development challenges due to on-target side effects, including seizures and tolerance development^11–13. While next-generation δOR agonists have successfully mitigated seizure risk, they still face significant hurdles, including the development of tolerance, insufficient clinical efficacy and off-target effects^10,14.

δOR agonists also represent promising therapeutic targets for gastrointestinal (GI) disorders. The enteric nervous system expresses functional δOR, and activation of the δOR-enkephalin axis provides important inhibitory control over intestinal motility and secretion¹⁵. This is evidenced clinically by racecadotril, an enkephalinase inhibitor that enhances endogenous opioid signalling to treat secretory diarrhoea¹⁶. The ability of positive allosteric modulators (PAMs) to selectively enhance endogenous opioid signalling when and where it occurs naturally makes them particularly attractive for GI applications, potentially providing therapeutic benefit for conditions like irritable bowel syndrome with diarrhoea (IBS-D) while preserving physiological control mechanisms and avoiding the adverse effects associated with sustained orthosteric receptor activation.

Despite these potential advantages across multiple therapeutic areas, a fundamental challenge for δOR agonists stems from their mechanism of action, which involves direct orthosteric site activation that drives receptor desensitisation and tolerance¹². Alternative approaches to safer opioid therapeutics have explored ligands with distinct pharmacological profiles¹⁷. These include receptor-targeted strategies such as partial agonists that exhibit reduced intrinsic efficacy¹⁸, biased agonists that selectively activate beneficial signalling pathways^19–21, allosteric modulators that target distinct allosteric sites to enhance endogenous ligand activity without directly activating the receptor^22,23, and bitopic ligands that bind to both orthosteric and allosteric sites, conferring both efficacy and selectivity to a single chemical entity^24,25. Additionally, indirect approaches include enkephalinase inhibitors that prevent degradation of endogenous opioid peptides and potentially avoid tolerance development²⁶.

PAMs targeting δOR represent a promising alternative therapeutic strategy^27–29. In response to pain, the body releases endogenous peptides such as enkephalins and endorphins, which activate opioid receptors to provide localised, natural pain relief. However, during severe or chronic pain, endogenous opioids are often insufficient, requiring administration of analgesics. Rather than directly activating the receptor, PAMs bind to distinct allosteric sites and selectively enhance the activity of endogenous opioid peptides, thereby preserving natural spatiotemporal signalling patterns while potentially circumventing some limitations inherent to orthosteric agonists³⁰. The first discovered δOR PAM, BMS-986187, potentiates enkephalin-mediated responses in a model system²². However, its poor physicochemical properties have limited its utility to peripheral in vivo studies, where it has shown potential in treating GI motility disorders¹⁵. In structural biology studies, the poor solubility of BMS-986187 likely prevented the determination of its allosteric site in X-ray crystallography studies³¹. We report a similar null finding when using BMS-986187 to determine a DADLE-bound δOR-G_i complex by cryo-EM. Fortunately, a medicinal chemistry campaign exploring the structure-activity relationships (SAR) of the xanthenedione scaffold of BMS-986187 has yielded MIPS3614, which exhibits substantially reduced cLogP values (cLogP 6.19 vs 4.75), while maintaining similar pharmacological properties to those of BMS-986187³².

In this study, we present a high-resolution cryo-EM structure of a δOR-DADLE-G_i complex co-bound to the PAM MIPS3614, revealing a lipid-facing allosteric binding site formed by residues on transmembrane helices 2, 3 and 4. Our structural analysis reveals that the carbonyl oxygen of MIPS3614 forms a hydrogen bond with residue N131^3.35 (superscript refers to Ballesteros-Weinstein numbering³³), which constitutes part of the conserved allosteric sodium ion binding site^34,35. Molecular dynamics simulations demonstrate that MIPS3614 stabilises an outward, lipid-facing rotamer conformation of N131^3.35 specifically associated with the receptor’s active state. We validate this mechanism through comprehensive mutagenesis and functional studies, demonstrating that removal of the carbonyl oxygen completely abolishes PAM activity and confirming the critical importance of the N131^3.35 interaction. We demonstrate that MIPS3614 effectively modulates δOR signalling in both recombinant systems and native tissues, providing proof-of-concept for therapeutic potential in GI disorders. Through structure-guided optimisation targeting additional binding site residues, we develop enhanced δOR PAMs, including MIPS3983, which exhibits improved binding affinity while maintaining cooperativity in recombinant systems. Collectively, our findings identify an allosteric binding site and define the mechanism by which these PAMs potentiate δOR agonist binding, providing information relevant to their rational design.

Results

Structure of δOR-DADLE

Several structures of the delta opioid receptor (δOR) have been determined in complex with small molecule agonists and antagonists, as well as exogenous selective peptides^{25,31,34,36–39}, yet the molecular basis of endogenous peptide binding and activation at δOR remains poorly characterised. We first determined the structure of δOR in complex with the agonist DADLE ([D-Ala², D-Leu⁵]-Enkephalin, Tyr-D-Ala-Gly-Phe-D-Leu). DADLE is a synthetic analogue of the endogenous peptide hormone Leu-Enkephalin (Tyr-Gly-Gly-Phe-Leu), both of which have similar pharmacological and selectivity profiles for the δOR⁴⁰.

To obtain structures of the δOR G protein complex, we used a chimeric miniature G protein (mG_si), similar to one previously used⁴¹, where the mG_si was fused to the carboxyl terminus of the receptor. This construct was co-expressed with Gβ₁ and Gγ₂ subunits, and complex formation was initiated with DADLE. The δOR-mG_si-DADLE complex was further stabilised by the addition of Nb35⁴² and purified in detergent by anti-FLAG affinity and size exclusion chromatography. Similar to previous studies³⁸, we observed anti-parallel oligomers of the δOR complex but were able to separate the monomeric complex for structure determination via size exclusion chromatography (Supplementary Fig. 1). The purified δOR-mG_si-DADLE complex was vitrified on gold grids and imaged using single-particle cryo-transmission electron microscopy (cryo-EM) on a Titan Krios microscope (Supplementary Table 1). We determined the cryo-EM structure of δOR-mG_si-DADLE to a global resolution of 2.0 Å (Fig. 1a–c, Supplementary Fig. 2 and Table 1), which was sufficient to model the majority of the backbone and side chains for the receptor, G protein, Nb35 and DADLE (Supplementary Fig. 3a).

Fig. 1. Overall structure of δOR-DADLE complex and comparison with related opioid receptor structures.

a Cryo-EM structure of the δOR-DADLE-mG_si complex at 2.0 Å resolution. The δOR is shown in teal, DADLE peptide in pink, mG_si in purple, Gβ₁ in dark blue, Gγ₂ in pink and Nb35 in grey. The inset shows a detailed view of DADLE bound in the orthosteric binding site with cryo-EM density contoured at 0.27. b Chemical structure of DADLE (Tyr-D-Ala-Gly-Phe-D-Leu), a synthetic analogue of the endogenous peptide Leu-Enkephalin. c Cryo-EM density map contoured at 0.27 showing the quality of density for DADLE and surrounding receptor residues, with the boxed region highlighting the peptide binding site. d Overlay comparison of δOR-DADLE (teal) with δOR-Deltorphin (orange, PDB: 8F7S) and µOR-Endomorphin (purple, PDB: 8F7R) showing structural similarities. e Extracellular view comparison of δOR-DADLE with δOR-Deltorphin and µOR-Endomorphin, highlighting differences in extracellular loop 3 (ECL3). f Intracellular view showing G protein coupling interfaces and conformational differences, particularly in intracellular loop 3 (ICL3) between the different opioid receptor-peptide complexes. g Cross-section view of DADLE binding in the orthosteric pocket showing the peptide’s extended conformation with the boxed region highlighting regions of the peptide binding site. Molecular interactions of h the N-terminal tyrosine (Y1) and D-alanine (DA2), i phenylalanine residue (F4) and j C-terminal D-leucine (DL5) of DADLE with the receptor.

Compared to previous structures of δOR-Deltorphin (PDB: 8F7S)³⁸, δOR-DADLE had a root mean square deviation (RMSD) of 1.04 Å for α-carbons across the receptor and 1.51 Å across the entire complex (excluding the dimer in 8F7S). δOR-DADLE was remarkably similar to μOR-Endomorphin (PDB: 8F7R)³⁸ with an RMSD of 0.76 Å for α-carbons across the receptor (1.34 Å across the complex). The extracellular pocket of δOR-DADLE was wider compared to δOR-Deltorphin, due to a more open conformation of ECL2 and ECL3. While the intracellular side showed similar receptor and G protein conformations, except for a change in conformation of ICL3 (Fig. 1d–f). G protein binding interfaces were similar, with only a modest rotation of the G protein. Some of the observed differences could be attributed to different G protein stabilisation strategies used, or to the neighbouring dimeric interfaces observed in previous complexes (Supplementary Fig. 4a–g). Similar to the previous cryo-EM structures, δOR-DADLE adopts a fully active conformation, with all activation motifs (toggle switch, PIF, NPxxY and DRY) displaying active conformations (Supplementary Fig. 4h–k).

The δOR-DADLE structure revealed that DADLE binds deep within the orthosteric binding site and occupies a similar location to that observed in previous structures of δOR-Deltorphin (Tyr-D-Ala-Phe-Glu-Val-Val-Gly) and μOR-Endomorphin (Tyr-Pro-Trp-Phe) (Fig. 1g–j)³⁸. The side chain of the N-terminal tyrosine (Y1), which is conserved across opioid peptides, sits in a hydrophobic pocket near TMs 3/5/6, surrounded by residues Y129^3.33, M132^3.36, V217^5.42 and H278^6.52, while the backbone nitrogen forms hydrogen bonds to D128^3.32 and Y308^7.43 (Fig. 1h). The side chain of F4 of DADLE extends towards a pocket composed of residues of TMs 2/3 and ECL1 and ECL2. Specifically, F4 forms hydrophobic interactions with L125^3.29, W114^ECL1 and C198^ECL2; this region of the peptide is further stabilised by a hydrogen bond from the backbone oxygen of G3 to K108^2.63 (Fig. 1i). The terminal D-leucine (DL5) extends towards TMs 3/4/5/6, and the side chain forms hydrophobic interactions with W284^6.58, and the carboxyl terminus of DADLE forms a salt bridge to K214^5.39 (Fig. 1j).

Molecular pharmacology of δOR PAMs

In a recent SAR study³², we discovered an analogue of BMS-986187, MIPS3614 (4), with reduced lipophilicity. MIPS3614 exhibited increased apparent solubility compared to BMS-986187, as evidenced by reduced turbidity when preparing 100 µM aqueous stock solutions (Supplementary Fig. 5a). Consistent with this observation, dynamic light scattering measurements showed baseline scattering for water and buffer controls, with MIPS3614 displaying profiles indistinguishable from controls under identical conditions. In contrast, BMS-986187 exhibited increased scattering and a detectable particle population, consistent with aggregation or precipitation and reduced apparent aqueous solubility relative to MIPS3614 (Supplementary Fig. 5b).

To understand the allosteric effects of BMS-986187 and MIPS3614 in different signalling pathways¹⁸, increasing concentrations of the allosteric ligand were co-added to increasing concentrations of DADLE in human embryonic kidney (HEK) 293 cells expressing human δOR (Fig. 2a–m). The operational model of allosterism⁴³ was fit to the data to estimate allosteric parameters of functional cooperativity (αβ), PAM affinity (pK_B) and efficacy of the PAMs (τ_B). In cAMP inhibition assays (Fig. 2b, h), both PAMs exhibited robust agonism due to high signal amplification, which made it difficult to fit an allosteric model to the data. In contrast, in less amplified assays, evidenced by significantly lower potencies of DADLE in these assays compared to cAMP, such as β-arrestin 2 (Fig. 2c, i), mG_si (Fig. 2d, j) and Nb33 (Fig. 2e, k) recruitment, both PAMs displayed less agonism and the data could be quantified with an allosteric model (Fig. 2f, l). Both PAMs exhibited similar levels of functional cooperativity (αβ = 5-7) and low levels of PAM agonism (τ_B = 0.1–1) across all recruitment assays. However, the functional binding affinity of MIPS3614 was approximately 10-fold lower (pK_B = 4.5–4.8) versus BMS-986187 (pK_B = 5.7–5.9), suggesting that the improvements made to PAM solubility were at the cost of reduced binding affinity. We confirmed that MIPS3614 also modulates the endogenous ligand Leu-Enkephalin, exhibiting comparable binding affinity (pK_B) and cooperativity (log αβ) with Leu-Enkephalin as with DADLE (Fig. 2l, m).

Fig. 2. Molecular pharmacology and functional characterisation of δOR positive allosteric modulators BMS-986187 and MIPS3614.

a Chemical structure of BMS-986187, a first-generation δOR PAM with limited solubility. b–f Pharmacological characterisation of BMS-986187 across multiple signalling pathways. Concentration-response curves showing DADLE-mediated responses with increasing concentrations of BMS-986187 (0–30 µM) in: b cAMP inhibition assay (n = 3), c β-arrestin 2 recruitment assay (n = 5), d mG_si recruitment assay (n = 5) and e Nb33 recruitment assay (n = 4). f Summary table of allosteric parameters from fitting the operational model of allosterism to (b–e), estimating functional cooperativity (αβ), PAM affinity (pK_B) and PAM efficacy (τ_B). g Chemical structure of MIPS3614, a second-generation δOR PAM with improved solubility compared to BMS-986187 through reduced cLogP. Key atoms are numbered. Equivalent characterisation of MIPS3614 using the same assay panel: h cAMP inhibition assay (n = 5), i β-arrestin 2 recruitment assay (n = 6), j mG_si recruitment assay (n = 5) and k Nb33 recruitment assay (n = 5). l MIPS3614 in a Nb33 recruitment assay with Leu-Enkephalin (Leu-Enk) as the agonist (n = 4). m A summary table of MIPS3614 allosteric parameters. n Validation of MIPS3614 allosteric effects in native tissue. Concentration-response curves for DADLE-mediated inhibition of electrically evoked contractions of the mouse colon with vehicle (n = 8), 1 µM MIPS3614 (n = 6), 10 µM MIPS3614 (n = 7), or 10 µM BMS-986187. Vehicle and 10 µM BMS-986187 were reproduced from Deo et al., for comparison³². o Quantification of DADLE potency changes from (n). Statistical significance was determined by one-way ANOVA (exact P values listed above bar graphs) with a Dunnett’s post-test using Vehicle for comparison. p Representative whole-organ colonic motility traces showing colonic motor complexes (CMCs) from n = 5 experiments under vehicle control (left) and 3 µM MIPS3614 treatment (right) at elevated intraluminal pressure. q Quantitative analysis of colonic motility with 3 µM MIPS3614 (n = 5). MIPS3614 significantly reduced CMC frequency without affecting velocity under conditions of elevated intraluminal pressure. Significance by unpaired two-tailed t-tests (*P = 0.0496). b–q Data are presented as mean ± SEM with n reporting biological replicates. Source data are provided as a Source Data file.

To assess if MIPS3614 had similar activity on the enteric nervous system (ENS) as BMS-986187^15,32, we examined its allosteric modulation on DADLE-mediated inhibition of electrically evoked contractions of the mouse colon (Fig. 2n). MIPS3614 at 1 μM and 10 μM significantly increased DADLE potency by 100-fold and 300-fold, respectively (Fig. 2o). Compared to BMS-986187, this corresponds to a 10-fold greater allosteric effect. We then examined 3 µM MIPS3614 in whole-organ colonic motility assays, measuring the frequency and velocity of colonic motor complexes (CMCs) (Fig. 2p). In contrast, when intraluminal pressure was elevated to increase CMC frequency, MIPS3614 significantly reduced this response and increased the time interval between contractions (Fig. 2q) without affecting CMC velocity. Collectively, these data suggest that, like BMS-986187, MIPS3614 can potentiate the effects of endogenous opioids released in response to increased intraluminal pressure.

Allosteric site of the delta opioid receptor

To determine structures of the δOR bound to allosteric ligands, the δOR-mG_si-DADLE complex protein was incubated with BMS-986187 or MIPS3614 prior to vitrification and cryo-EM data collection. We resolved 3D reconstructions of δOR-DADLE-BMS-986187 and δOR-DADLE-MIPS3614, both to a global resolution of 1.9 Å (Fig. 3a–c, Supplementary Figs. 6 and 7). However, compared to the δOR-mG_si-DADLE complex structure, we did not observe sufficient cryo-EM density to confidently model BMS-986187 (Fig. 3b), a finding consistent with prior attempts to determine the structure using X-ray crystallography³¹. Therefore, subsequent structural analysis was limited to δOR-DADLE and δOR-DADLE-MIPS3614 complexes. The cryo-EM maps for both structures were of high quality (Supplementary Fig. 3b). The δOR-DADLE, BMS-986187 and MIPS3614 structures were remarkably similar, with RMSDs of 0.10–0.45 Å for α-carbons of the entire complex (Fig. 3d–f).

Fig. 3. Structural comparison of δOR-DADLE complexes with and without positive allosteric modulators.

Consensus cryo-EM maps for the structures of a the δOR-DADLE complex, b the δOR-DADLE-BMS-986187 complex and c the δOR-DADLE-MIPS3614 complex contoured at 0.27, 0.45, and 0.30, respectively. δOR is coloured in shades of green, DADLE peptide in pink, mG_si in purple, Gβ₁ in dark blue, Gγ₂ in pink and Nb35 in grey. The inset shows cryo-EM density surrounding the allosteric site for MIPS3614 with cryo-EM density from the receptor-focused maps contoured at (a) 0.30, (b) 0.34 and (c) 0.35. d–f Comparison of the δOR structures from (a–c) with views from the d side, e extracellular surface and f intracellular surface.

Our structure of the δOR-DADLE-MIPS3614 revealed that the allosteric binding site is located on the extrahelical surface of TMs 2/3/4 (Fig. 4a–c). The cryo-EM density surrounding the ortho-cyclopropylbenzyl substituent was less defined, suggesting a degree of conformational heterogeneity in this region. Nonetheless, MIPS3614 could be unambiguously modelled (Fig. 4b). Notably, all compounds were synthesised as racemic mixtures; however, only the R-enantiomer clearly fitted the cryo-EM density (Supplementary Fig. 8a). Subsequently, chiral HPLC separation of the racemic mixture (MIPS3614) into 4a and 4b revealed that only one enantiomer had pharmacological activity, which we have defined as the R-enantiomer (4a) to match the structural data (Supplementary Fig. 8b).

Fig. 4. Structural characterisation and molecular validation of the MIPS3614 allosteric binding site.

a Overview of the δOR-DADLE-MIPS3614 complex showing the spatial relationship between the orthosteric binding site (DADLE in pink, dashed box) and the allosteric binding site for MIPS3614 in orange (dashed box). The allosteric site is located on the lipid-facing extrahelical surface of transmembrane helices 2, 3 and 4. b View of the MIPS3614 allosteric binding site with the receptor-focused cryo-EM map contoured at 0.35. c Detailed view of the MIPS3614 binding site showing key interacting residues from TM2, TM3 and TM4. d Molecular interaction diagram of MIPS3614 with δOR residues. Hydrogen bonds are denoted by black dashed lines, π-stacking interactions with green dashed lines, and hydrophobic interactions with solid red lines. e Mutational analysis of the MIPS3614 binding site using the Nb33 recruitment assay. Data show ligand-induced BRET responses. All mutant δOR cell lines were performed with n = 3 replicates except for WT (n = 7). f Effect of BMS-986187 at the M184^4.61F mutant (n = 3). Data were normalised to 10 µM DADLE response. g Heat map summarising the effects of mutations on DADLE maximum response (E_max) from (e). Blue indicates an E_max comparable to wild-type, while red indicates significantly reduced E_max values. E_max mean values are overlaid on the heat map. h Surface receptor expression quantified from ELISA. UT = Untransfected cells. Quad = S126^3.30A/Y130^3.34A/S177^4.54A/M184^4.61A. Data are from n = 3 experiments for all cell lines. i Heat map showing the allosteric modulation capacity (ΔpEC₅₀) of MIPS3614 across functional mutants. Values represent the change in DADLE potency in the presence of 30 µM MIPS3614. Red indicates loss of allosteric modulation, while blue indicates retained modulation. ΔpEC₅₀ mean values are overlaid on the heat map. e–i Data are presented as mean ± SEM with n reporting biological replicates. Statistical significance was determined by one-way ANOVA with a Dunnett’s post-test using WT for comparison (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). Mean ± SEM and P values for (g–h) are reported in Supplementary Table 2. Source data are provided as a Source Data file.

MIPS3614 adopts an extended binding pose with numerous molecular interactions (Fig. 4c, d). The tricyclic heterocyclic core is positioned at the centre of the transmembrane region, extending upward toward the extracellular surface. The core sits in a hydrophobic pocket formed by residues L97^2.52, A94^2.49, W173^4.50 and V181^4.58. Within this pocket, the tricyclic heterocyclic core forms a π-stacking interaction with Y130^3.34, which is further stabilised by two hydrogen bonds: one from the carbonyl oxygen in the 8-position to N131^3.35 and another from N1 to S126^3.30. The ortho-cyclopropylbenzyl group primarily interacts through hydrophobic residues, including L119^3.23, L120^3.24, A123^3.27, I127^3.31 and M184^4.61.

To validate the observed interactions, we performed a series of single-point mutations as well as a combined four-residue mutation and tested them in the Nb33 BRET assay. The maximal response (E_max) of DADLE was significantly reduced at multiple mutants, including Y130^3.34A, N131^3.35A, N131^3.35S, S177^4.54A, S177^4.54N and the quadruple mutant S126^3.30A/Y130^3.34A/S177^4.54A/M184^4.61A, compared to WT (Fig. 4e). The reduction in E_max correlated with reduced surface expression levels (Fig. 4g, h). Mutation of W173^4.50, which is conserved across nearly all class A GPCRs, to phenylalanine also resulted in reduced surface expression. In contrast, mutations at positions S126^3.30, V181^4.58 and M184^4.61 were well tolerated without affecting cell surface expression or the DADLE E_max response (Fig. 4g, h). For mutants that did not significantly reduce DADLE E_max, we calculated the change in DADLE EC₅₀ in the absence or presence of 30 μM of MIPS3614 (Fig. 4i). In WT δOR, 30 μM of MIPS3614 caused a 3-fold leftward shift in the DADLE potency, but this potentiation was reduced or absent in most mutants, indicating that residues within this pocket, particularly S126^3.30, N131^3.35 and M184^4.61, are critical for allosteric interactions with MIPS3614.

Mutation of the hydrogen-bonded S126^3.30 to alanine was well tolerated in terms of surface expression, with DADLE displaying a similar E_max to WT, but resulted in a significant reduction of the leftward potency shift of DADLE in the presence of 30 μM MIPS3614 (1.5-fold). Mutation of S126^3.30 to the bulky phenylalanine slightly reduced expression and DADLE E_max but completely abolished MIPS3614 modulation. We also mutated M184^4.61 to an alanine and phenylalanine, both of which reduced the modulation of MIPS3614. Although we were unsuccessful in observing clear cryo-EM density for BMS-986187, the M184^4.61F mutation had an even greater effect on its binding and modulation (Fig. 4f), indicating BMS-986187 binds the same allosteric site as MIPS3614.

Sequence alignment of δOR residues that interact with MIPS3614 with the corresponding μOR and κOR residues reveals the binding site is largely conserved across the opioid family (Supplementary Fig. 9). The majority of the site is fully conserved, including residues that form major interactions, such as S126^3.30 and N131^3.35, which form hydrogen bonds and Y130^3.34_, which forms π-stacking interactions. The differences between the opioid subtypes are primarily subtle changes to hydrophobic residues lining the pocket, such as L119^3.23I/V (μOR/κOR residue, respectively), A123^3.27I/I, V181^4.58L/I and M184^4.61M/I (Supplementary Fig. 9). The largest of these differences A123^3.27/I could potentially form steric clashes with the ortho-cyclopropylbenzyl group of MIPS3614, consistent with the A123^3.27I mutant resulting in a modest reduction in EC₅₀ shift by MIPS3614 (Fig. 4e).

To further assess the selectivity of MIPS3614, we tested its PAM activity against μOR and κOR as closely related opioid receptor subtypes, alongside SSTR2, which shares the conserved N^3.35 residue and M₄ mAChR as a distally related class A GPCR control. Consistent with previously published data in CHO cells demonstrating no PAM activity at μOR or κOR in cAMP and pERK1/2 assays³², MIPS3614 showed no significant PAM activity at κOR, M₄ mAChR or SSTR2 in a mini-Gsi recruitment assay, while modest PAM activity was observed at μOR (Supplementary Fig. 9d, e). Assessment using the conformationally-sensitive Nb33 biosensor revealed a less selective profile, with comparable PAM activity observed at both μOR and δOR (Supplementary Fig. 9f), likely reflecting different sensitivities between assay readouts and cell lines. For example, BMS-986187 was originally reported to show greater selectivity for δOR over μOR in functional assays⁴⁴, yet recent cryo-EM structures reveal it occupies the same conserved TM2-TM4 allosteric site at μOR as MIPS3614 occupies at δOR⁴⁵. Together, these findings suggest that apparent selectivity of allosteric modulators at opioid receptors is highly sensitive to assay system, signalling pathway and binding site conservation.

Molecular mechanism of positive allosteric modulation

Along with the A_2A adenosine receptor⁴⁶, the δOR and µOR have served as model systems for studying the allosteric effects of sodium ions on class A GPCRs^47,48. The sodium site is highly conserved across class A GPCRs³⁵, including the strongly conserved aspartate at position 2.50 (D^2.50) and an Na⁺ bound in this pocket stabilises the inactive receptor conformation. Upon activation, rearrangement of the TM helices leads to a collapse of the sodium pocket and expulsion of the ion, which likely is an essential part of the activation mechanism^35,49. This process may also be associated with protonation of D^2.50, disrupting the ionic interaction with the sodium and thereby facilitating receptor activation^49–51. At δOR, the sodium pocket includes N131^3.35, at a position that is typically (~70%) occupied by a hydrophobic residue in GPCRs³⁴. Comparison of N131^3.35 in the inactive (PDB: 4N6H) and active state DADLE- and DADLE-MIPS3614-bound structures reveals a change in the rotamer, resulting in a 5.5 Å movement of the side chain (Fig. 5a). In the inactive state, N131^3.35 points towards the sodium site and interacts directly with the sodium ion, whereas in the active state, the side chain rotates outward from the TM bundle. Notably, the carbonyl oxygen of MIPS3614 forms a hydrogen bond interaction with N131^3.35 in the cryo-EM structure, which may further stabilise the active state. Attempts to interrogate the contribution of N131^3.35 by mutagenesis to an alanine or a serine resulted in reduced E_max in response to DADLE and surface expression levels (Fig. 4e–h), consistent with prior attempts³⁴. However, mutation of N131^3.35 to an aspartic acid did not significantly affect DADLE E_max or expression but completely abolished the leftward potentiation of the DADLE response in the presence of MIPS3614 (Fig. 4i).

Fig. 5. Mechanism of allosteric modulation by disrupting the sodium binding site.

a Structural comparison of the conserved sodium binding site between inactive δOR (δOR:naltrindole, grey; PDB: 4N6H) and active MIPS3614-bound δOR (δOR:DADLE:MIPS3614, teal/orange). The overlay shows a 5.5 Å outward conformational change of N131^3.35 away from the sodium binding site (Na⁺ shown as a purple sphere) in the active state. Key sodium site residues D95^2.50 and S135^3.39 are labelled. The carbonyl oxygen from MIPS3614 forms a hydrogen bond with N131^3.35. b Occupancy of inward- or outward-pointing rotamers of N131^3.35 in MD simulations based on the MIPS3614-bound cryo-EM structure, in the presence of PAM, agonist, or Gα protein versus the apo state. c Frequency of D95^2.50-sodium distances in simulations of the inactive conformation: N131^3.35 inward (black-outlined bars, black mesh) and N131^3.35 restrained outward (blue bars, blue surface), with sodium near D128^3.32 in the orthosteric site. d pK_a of D95^2.50 calculated for DADLE-bound receptor simulations in the presence (purple bars) or absence (red-outlined bars) of MIPS3614. e Heat maps of TM2 and TM3 rotation angle distributions, measured as depicted on the cartoon receptor, from simulations based on the MIPS3614 cryo-EM structure, with and without PAM, agonist, Gα protein and D95^2.50 protonation. Light-grey lines facilitate comparison between panels. f Chemical structure of MIPS3878 (5), a deoxygenated analogue of MIPS3614; the red circle highlights the absent carbonyl oxygen in the tricyclic heterocyclic core. g Pharmacological characterisation of MIPS3878 across multiple signalling pathways showing complete loss of positive allosteric modulation. Data are mean ± SEM, n = 4 for all assays except cAMP inhibition (n = 5). h PAM competition assay using MIPS3614 as an allosteric agonist in a cAMP inhibition assay (see Fig. 2h), with 10 µM MIPS3878 added (n = 4). i PAM activity of MIPS3614 at δOR by Nb33 recruitment at 140 mM NaCl and 0 mM NaCl, where NaCl was replaced by equimolar choline chloride⁵⁵ (n = 4). Allosteric parameters were compared by two-sided unpaired t-test: pK_B (140 mM: 5.0 ± 0.1 vs 0 mM: 4.8 ± 0.2, P = 0.2113), log αβ (0.8 ± 0.1 vs 0.7 ± 0.1, P = 0.5788), log τ_B (−0.3 ± 0.1 vs −0.02 ± 0.04, P = 0.0068). Data are mean ± SEM with n reporting biological replicates. Source data are provided as a Source Data file.

To further investigate the influence of the allosteric modulator on the sodium binding site, we carried out molecular dynamics (MD) simulations of the MIPS3614-bound δOR conformation. We first compared the side chain conformation of N131^3.35 in simulations of the receptor bound to Gα protein, DADLE or MIPS3614 either individually or in combinations, as well as in the apo state (Fig. 5b). In agreement with our structural data, the N131^3.35 side chain almost exclusively remained in an outward conformation in the presence of the agonist, PAM and G protein. As expected, based on the inactive structure of the δOR, the side chain relaxed to the inward conformation in the simulation of the apo receptor. Notably, DADLE and MIPS3614, either individually or combined, stabilised the outward N131^3.35 conformation over the inward one. This shows that both agonist and PAM favour an active-like receptor conformation in the sodium pocket region.

To assess how the N131^3.35 orientation affects sodium binding, we performed MD simulations of the inactive receptor (PDB: 4N6H) in the orientation observed in the X-ray structure and with the asparagine side chain restrained in an outward conformation. The sodium ion present in the inactive structure remains bound at D95^2.50 in simulations with N131^3.35 unrestrained. In contrast, the outward rotation of the N131^3.35 side chain destabilises the ionic interaction (Fig. 5c), and the sodium ion instead moves towards D128^3.32 in the orthosteric pocket. This was further supported by MD simulations of the active-state δOR with a sodium ion placed close to D95^2.50, in which destabilization of the ion was also observed (Supplementary Fig. 10a). Furthermore, pK_a predictions for D95^2.50 obtained using PROPKA⁵² on simulation snapshots, show that MIPS3614 binding induces receptor conformations that increase the protonation of the side-chain carboxylate (Fig. 5d).

To identify potential large-scale motions in the receptor, principal component analysis (PCA) was applied to the simulations based on the MIPS3614 cryo-EM structure. Subsequently, the motions captured in the first principal components were analysed in the MD trajectories (Fig. 5e). This revealed that MIPS3614, DADLE, or Gα protein binding, as well as D95^2.50 protonation, each induce bending of TM3 that is consistent with available active and inactive δOR structures (Supplementary Fig. 10b). Moreover, this movement is associated with a concerted rotation of TM2 and TM3 compared to the apo state. In the simulation with the receptor bound to Gα protein, MIPS3614 and DADLE, the TM2/TM3 rotation leads to high occupancy of the outward conformation of N131^3.35 that enables hydrogen bonding to MIPS3614. The receptor conformations stabilised by the MIPS3614 hence contribute to displacing the sodium ion and favour protonation of D95^2.50, facilitating receptor activation.

To further validate the mechanism, we synthesised an analogue lacking the carbonyl oxygen on the tetrazoloquinazolinone core that forms the critical hydrogen bond to N131^3.35 (Fig. 5f). This deoxygenated analogue, MIPS3878 (5), was designed to test the functional importance of this specific interaction. MIPS3878 showed a complete loss of modulatory activity across multiple functional assays (Fig. 5g). In the Nb33 recruitment assay, MIPS3878 produced no shift in the EC₅₀ of DADLE, and similar results were observed in cAMP inhibition, β-arrestin 2 recruitment and mG_si coupling. To determine whether MIPS3878 was acting as a neutral allosteric ligand (NAL), binding to the allosteric site without modulating it, or had completely lost binding affinity, we performed competition studies. We assessed increasing concentrations of MIPS3614 in cAMP inhibition assays conducted in agonist mode (absence of orthosteric ligand), both with and without 10 μM MIPS3878 (Fig. 5h). In this experimental design, a neutral allosteric ligand (NAL) would function as an allosteric antagonist, competing with the response elicited by MIPS3614. However, MIPS3878 did not affect the response to MIPS3614 in cAMP assays at concentrations up to 10 µM, suggesting removal of the carbonyl oxygen reduces binding to the allosteric site.

Intriguingly, we observed a slight, but statistically significant concentration-dependent increase in the E_max of DADLE with increasing concentrations of MIPS3878 in the β-arrestin 2 assay (Fig. 5g). Prior mutation of residue N131^3.35 to alanine at δOR resulted in constitutive β-arrestin activation³⁴. We speculate that the loss of the hydrogen bond with N131^3.35 may confer biased allosteric properties to MIPS3878 allowing it to weakly bind to the MIPS3614 allosteric site, potentially acting as a pathway-selective positive allosteric modulator favouring β-arrestin 2 activation over G protein coupling. This contrasts with BMS-986187, which is biased towards G protein signalling relative to β-arrestin 2 activation⁵³, and highlights the importance of Na⁺ binding site in controlling the active conformation of class A GPCRs⁵⁴.

To further probe the relationship between Na⁺ and MIPS3614, we examined PAM activity in the absence of extracellular Na⁺_, replaced by equimolar choline chloride⁵⁵. PAM activity was preserved under Na⁺-free conditions (Fig. 5i), consistent with our structural data demonstrating that MIPS3614 occupies the allosteric site in the active receptor conformation in the absence of Na⁺. A modest but concentration-dependent enhancement of PAM-agonist activity was observed at 0 mM NaCl compared to 140 mM NaCl, while the DADLE concentration-response curve remained unchanged, suggesting the effect is specific to the interaction between MIPS3614 and the Na⁺ site rather than reflecting broader changes in receptor signalling or G protein coupling efficiency.

Structure-guided optimisation of δOR allosteric modulators

The initial structure-activity relationship (SAR) studies for the BMS-based scaffold focused on modifications to the substituent in the 9-position⁵⁶, while a subsequent investigation examined modifications to the tricyclic heterocyclic core³². Based on the binding pose of MIPS3614 in our δOR structure, we hypothesised that extending substituents from the 4-position of the heterocyclic core would enhance interactions with residues within the allosteric pocket, particularly S177^4.54.

A new series of compounds was synthesised (Figs. 6a, 6a–i) and evaluated for their ability to potentiate DADLE response in the Nb33 recruitment assay (Fig. 6b, c). The new analogues generally demonstrated 10-fold improved binding affinity (pK_B) compared to MIPS3614 (Fig. 6b). However, compounds containing polar substituents, specifically the ethanol and acetic acid groups in 6c and 6e, showed significantly reduced functional cooperativity and binding affinity. This loss of activity was attributed to unfavourable polar interactions within the predominantly hydrophobic allosteric pocket. Based on these results, we focused subsequent studies on MIPS3983 (6i), which contains an ethyl propanoate substituent and demonstrated improved binding affinity (pK_B = 5.4 compared to 4.6 for MIPS3614) while retaining a substantial cooperativity factor of 12. The ethyl propanoate group appeared to provide an optimal balance between enhanced binding through additional interactions and maintenance of appropriate physicochemical properties for the binding site. We chose MIPS3983 for structural studies because of the larger size of the propanoate group compared to appendages of compounds 6a, 6b, 6d and 6g, which had comparable pK_B and αβ values to MIPS3983.

Fig. 6. Structure-guided optimisation of δOR positive allosteric modulators.

a Chemical scaffold showing the tricyclic heterocyclic core structure with variable R-group substituents at the secondary amine position (dashed circle). Nine analogues were synthesised (6a-i) to explore structure-activity relationships and enhance interactions with the allosteric binding pocket, particularly targeting residue S177^4.54. b Heat maps summarising the pharmacological properties of the new analogues derived from concentration response curves in the Nb33 recruitment assay (shown in c). Top panel shows binding affinity (pK_B) values, and the bottom panel shows functional cooperativity (αβ) values with mean values overlaid on the heat map. N.D. not determined. Statistical significance was determined by one-way ANOVA with a Dunnett’s post-test using MIPS3614 for comparison (*P = 0.0304; ****P < 0.0001). Data are from n = 7 experiments for MIPS3614, n = 6 for 6a, n = 4 for 6b, n = 7 for 6c, n = 5 for 6d, n = 5 for 6e, n = 6 for 6f, n = 5 for 6g, n = 4 for 6h, n = 5 for 6i. Data for (c) are presented as mean ± SEM with n numbers reporting biological replicates. d Cryo-EM structure of the δOR-DADLE-MIPS3983 complex showing improved cryo-EM density in the allosteric binding site. The inset displays cryo-EM density from the receptor-focused map at a contour of 0.15. e Detailed view of MIPS3983 binding interactions in the allosteric pocket. Compared to MIPS3614, the tricyclic heterocyclic core of MIPS3983 appears to rotate in a manner that allows hydrogen bonding to both S126^3.30 and S177^4.54. f Molecular interaction diagram of MIPS3983 with δOR residues. Hydrogen bonds are denoted by black dashed lines, π-stacking interactions with green dashed lines, and hydrophobic interactions with solid red lines. g Comparison of density maps across different δOR structures highlighting potential lipid or cholesterol binding sites near the MIPS3614 allosteric site. Source data are provided as a Source Data file.

Using the same methodology employed for MIPS3614, we determined a cryo-EM structure of the human δOR in complex with DADLE and MIPS3983 (Fig. 6d, Supplementary Fig. 11 and Table 1). MIPS3983 was bound to the same allosteric site as MIPS3614, with notably improved interpretability of the cryo-EM density around the tricyclic heterocyclic core and the ortho-cyclopropylbenzyl group. This enhanced density definition likely reflects the improved binding affinity of MIPS3983. The additional propanoate substituent showed more ambiguous density, suggesting either conformational flexibility or potential steric interactions with surrounding detergent or cholesterol molecules observed near the binding site. Nevertheless, the propanoate could be modelled into the cryo-EM density extending below the primary allosteric binding pocket. The tricyclic heterocyclic core and 9-(4-((ortho-cyclopropylbenzyl)oxy)phenyl) groups of MIPS3983 formed the same key interactions observed with MIPS3614. Additionally, MIPS3983 established a new hydrogen bond between S177^4.54 and the nitrogen atom of the heterocyclic core, confirming our design hypothesis (Fig. 6d). The propanoate substituent extended downward from the heterocyclic core, forming additional hydrophobic interactions with W173^4.50 and V174^4.51.

These structural and functional studies demonstrate that the allosteric pocket can accommodate additional substituents that enhance binding affinity through new productive interactions. The success of MIPS3983 validates the structure-based design approach and provides a foundation for further optimisation of this allosteric modulator scaffold. The improved affinity and retained cooperativity of MIPS3983 make it a valuable tool compound for studying δOR allosteric modulation.

Discussion

Multiple approaches have been explored in the search for safer opioid analgesics, including partial agonists that provide submaximal receptor activation⁵⁷, biased agonists that preferentially activate G protein over β-arrestin pathways¹⁹, and allosteric modulators that enhance endogenous ligand activity³⁰. Our structural and functional characterisation of δOR PAMs shows how allosteric modulation is achieved at this receptor, and identifies features that could be exploited therapeutically.

Although we could not confidently model BMS-986187 binding due to insufficient cryo-EM density, we successfully determined the allosteric binding site using MIPS3614, a more soluble analogue with similar pharmacological properties. The shared loss of allosteric activity at the M184^4.61F mutant suggests that both compounds bind the same site. BMS-986187 binding to the same allosteric site as MIPS3614 is further supported by a cryo-EM structure of BMS-986187 bound to the δOR and µOR (Supplementary Fig. 12a) in a publication released during manuscript revision⁵⁸. Our structural determination corrects previous computational predictions of the BMS-986187 binding site, which had relied on molecular docking and limited mutagenesis data⁵⁹. The experimentally determined binding site reveals interactions and binding mechanisms that were not anticipated from earlier modelling efforts, highlighting the critical value of high-resolution structural data for understanding allosteric modulation. Our own selectivity data, together with recent structural and pharmacological studies further demonstrate that apparent subtype selectivity at this conserved TM2-TM4 allosteric site is highly sensitive to the assay system and signalling pathways, underscoring both the challenges and opportunities for developing selective PAMs at this pocket^45,58.

The δOR allosteric binding site was located in a lipid-facing pocket formed by residues from transmembrane helices 2, 3 and 4. Lipid-facing allosteric sites are emerging as a common feature of GPCR-positive allosteric modulators, though with distinct binding mechanisms⁶⁰. The CB₁ receptor PAM ZCZ011 binds at a similar transmembrane location⁶¹ but utilises purely hydrophobic interactions without engaging the sodium binding site (Supplementary Fig. 12b–e). In contrast, the μOR-selective PAM BMS-986122 occupies a different lipid-facing pocket 15 Å away⁶², formed by TMs 3, 4 and 5 (Supplementary Fig. 12f–i). The selectivity of BMS-986122 for μOR over δOR likely stems from a steric clash with the larger methionine residue at position 3.45 in δOR versus threonine in μOR.

Our identification of the δOR PAM binding site reveals a mechanism whereby allosteric modulation is achieved through direct stabilisation of the sodium binding site in its collapsed, active-state conformation³⁵. Structural comparison with the µOR-BMS-986122 structure⁴⁴ reveals that δOR and µOR PAM binding sites are spatially distinct, yet conformationally linked through their shared connection to the Na⁺ binding site. Prior pharmacological findings demonstrated that BMS-986122 and BMS-986187 binding produces a rightward shift in agonist vs NaCl concentration response curves^44,63, consistent with both PAMs modulating receptor activity through destabilising the inactive Na⁺-bound conformation, albeit from different binding sites. This effect of PAM binding on NaCl concentration response curves is likely common to class A PAMs that stabilise the active conformation of GPCRs. Consistent with this mechanism, cell-based signalling assays in the absence of extracellular Na⁺ demonstrated preserved PAM activity for MIPS3614, with a modest enhancement of PAM-agonist activity, confirming that MIPS3614 acts independently of Na⁺ occupancy at the allosteric site (Fig. 5i).

Our MD simulations show that the interaction between MIPS3614 and N131^3.35 contributes to the displacement of the sodium ion and favours the protonation of D95^2.50, thereby facilitating receptor activation. The simulations reveal concerted motions of TM2 and TM3 that impact the sodium pocket architecture. These conformational rearrangements alter the local environment surrounding the conserved D95^2.50, shifting its pK_a toward higher values and thus favouring protonation. This protonation event further destabilises sodium binding and promotes ion expulsion, creating a positive feedback mechanism that stabilises the receptor’s active conformation. The coupling between MIPS3614 binding, N131^3.35 reorientation, and sodium site disruption represents a key mechanistic link between allosteric modulation and receptor activation.

The high resolution of our cryo-EM maps (Fig. 6g) enabled identification of cryo-EM density consistent with lipid or sterol binding sites both above and below the δOR allosteric pocket, highlighting that MIPS3614 binds at a membrane-lipid interface. We did not model specific lipids or sterols into our maps due to potential ambiguity in distinguishing between different membrane components and detergent molecules; however, the presence of this density demonstrates that the allosteric pocket architecture is partly defined by membrane components^60,64. These observations were consistent with the MD simulations, which showed that there were several stable lipid molecules surrounding MIPS3614 (Supplementary Fig. 13), confirming that lipid interactions are a stable and defining feature of this binding site.

The membrane-embedded nature of the δOR allosteric binding site has important implications for medicinal chemistry, particularly in explaining the structure-activity relationships observed in our campaigns and others. SAR studies of the BMS-986187 tolyl group showed that polar heterocycles and larger aromatic rings reduced PAM activity⁵⁶, suggesting size constraints imposed by nearby lipids and the importance of maintaining hydrophobic character for favourable membrane interactions. Similarly, MIPS3614 analogues with bulky polar modifications to the tetrazoloquinazolinone secondary amine (6c/6e) showed reduced activity, likely due to steric clashes or unfavourable polar interactions with lipids positioned below the binding site. These findings demonstrate that membrane lipids are integral components of the allosteric binding site and must be considered in structure-based drug design efforts targeting this pocket^60,64.

Our findings may also inform the elaboration of these modulators to exploit features of the receptor beyond the allosteric pocket itself. The identification of lipid-binding sites adjacent to this pocket suggests that tethering allosteric modulators to membrane-targeting moieties could enhance binding affinity and selectivity. This approach has been applied to other GPCRs, such as neurokinin 1 receptor antagonists linked to lipid groups that show enhanced potency, affinity and duration of action⁶⁵. Additionally, given recent findings that δOR activation from endosomes provides sustained pain inhibition⁷, lipid-conjugated PAMs might facilitate subcellular targeting to specific membrane compartments, potentially extending therapeutic effects.

From a therapeutic perspective, δOR PAMs offer several potential advantages over direct agonists, including preservation of endogenous signalling patterns, which could reduce the propensity for tolerance development²⁷. Although this study supports the use of δOR PAMs in GI disorders²⁹, future studies will be necessary to validate the potential of δOR PAMs for treating pain. Our structural and mechanistic data provide a starting point for structure-based drug design of improved δOR PAMs. Medicinal chemistry efforts can target the subtle sequence differences identified between opioid receptor subtypes or exploit the lipid-binding regions to develop next-generation compounds with enhanced selectivity and therapeutic profiles. Realising this potential will require more comprehensive selectivity profiling of MIPS3614 and future analogues across the opioid receptor family using orthogonal assay systems including radioligand binding, functional signalling and biophysical approaches across multiple cell lines and species.

Methods

Ethics statement

All procedures involving animals were approved by the Monash Institute of Pharmaceutical Sciences Animal Ethics Committee (approval 40039) and adhered to the ARRIVE guidelines.

Structure

Expression and purification of Nb35

Nb35 was expressed using an autoinduction method in BL21 (DE3) Rosetta Escherichia coli^66,67. Briefly, cells were transformed and grown at 37 °C under antibiotic selection with 100 μg/mL carbenicillin and 35 μg/mL chloramphenicol in autoinduction media (50 mM Phosphate buffer, pH 7.2, 2% tryptone, 0.5% yeast extract, 0.5% NaCl, 0.6% glycerol, 0.05% glucose and 0.2% lactose). At an OD₆₀₀ of 0.7, the cells were transferred to 20 °C for approximately 16 h before being harvested by centrifugation. Cell pellets were stored at −80 °C until purification. Nb35 was purified using Ni-affinity chromatography⁶⁸.

Expression and purification of δOR-mG_si-DADLE complex

The full-length human delta opioid receptor gene was modified to include an N-terminal hemagglutinin (HA) signal sequence and FLAG-tag, and a C-terminal 8xHis tag, GGGS linker, 3 C protease cleavage site and mini-G_si (mG_si)^69,70. The resulting FLAG-δOR-mG_si construct was cloned into the pFastBac vector (Invitrogen) for insect cell expression.

FLAG-δOR-mG_si and Gβ₁γ₂ were co-expressed in Tni insect cells (Expression Systems), using the Bac-to-bac baculovirus expression system (Invitrogen). The Tni cells were grown to a density of 4 × 10⁶ cells per ml in ESF 291 serum-free media (Expression Systems) and infected with a 4:1 ratio of FLAG-δOR-mG_si:Gβ₁γ₂ baculoviruses. Cells were incubated for 48–60 h at 27 °C, before harvesting by centrifugation (6000 × g, 20 min, 4 °C) and cell pellets stored at −80 °C.

For the purification of δOR-mG_si-DADLE complex, the cell pellet was thawed and resuspended in lysis buffer (20 mM HEPES, pH 7.4, 5 mM MgCl₂) with protease inhibitors (0.2 mM PMSF, 5 µg/mL leupeptin, 5 µg/mL soybean trypsin inhibitor and 1 mM benzamidine hydrochloride), 500 U benzonase and 1 μM DADLE peptide. The resuspension was stirred at room temperature for 15 min to allow for lysis before centrifugation at 30,000 × g for 20 min at 4 °C. The pellet was resuspended in solubilisation buffer (20 mM HEPES, pH 7.4, 100 mM NaCl, 5 mM MgCl₂, 5 mM CaCl₂, 0.5% LMNG, 0.03% CHS) with protease inhibitors (0.2 mM PMSF, 5 µg/mL leupeptin, 5 µg/mL soybean trypsin inhibitor and 1 mM benzamidine hydrochloride), 500 U benzonase. The resuspended pellet was homogenised with a Dounce and complex formation initiated with the addition of 1 μM DADLE peptide, 1 mg Nb35 and 25 mU/mL apyrase. The complex was incubated with stirring for 2 h at 4 °C before centrifugation at 30,000× g for 30 min at 4 °C to remove insoluble material. The complex was then bound to M1 anti-FLAG-affinity resin by gravity flow before washing with wash buffer (20 mM HEPES, pH 7.4, 100 mM NaCl, 5 mM MgCl₂, 5 mM CaCl₂, 0.01% LMNG, 0.0006% CHS and 1 μM DADLE). The complex was eluted with elution buffer (20 mM HEPES, pH 7.4, 100 mM NaCl, 5 mM MgCl₂, 0.01% LMNG, 0.0006% CHS, 1 μM DADLE, 10 mM EGTA and 0.1 mg/ml FLAG peptide) before concentration in an Amicon Ultra-15 100 kDa molecular mass cut-off centrifugal filter unit (Millipore). The complex was further purified by size exclusion chromatography on a Superdex 200 Increase 10/300 GL (Cytiva) in 20 mM HEPES, pH 7.4, 100 mM NaCl, 5 mM MgCl₂, 0.01% LMNG, 0.0006% CHS and 1 μM DADLE. Final fractions were assessed by SDS-PAGE, pooled, concentrated to 10 mg/mL and flash-frozen in liquid nitrogen before storage at -80 °C

Grid preparation and imaging

For the allosteric complexes, the sample was thawed and the allosteric ligand was added (100 μM for BMS-986187, 200 μM for MIPS3614, 100 μM for MIPS3983) and incubated overnight on ice at 4 °C. No allosteric ligand was added for the δOR-mG_si-DADLE complex structure. 3 µL of sample was applied to glow-discharged (15 mA, 180 s) UltrAufoil R1.2/1.3 300 mesh holey grid (Quantifoil) and was frozen in liquid ethane using a Vitrobot mark IV (Thermo Fisher Scientific) at 100% humidity and 4 °C with a blot time of 2 s and blot force of 10–14. Data were collected using a G1 Titan Krios microscope (Thermo Fisher Scientific) equipped with S-FEG, a BioQuantum energy filter and K3 detector (Gatan). The Krios was operated at an accelerating voltage of 300 kV with a 50 μm C2 aperture, 100 μm objective aperture inserted, and zero-loss filtering (10 eV slit width), at 105 kX magnification in nanoprobe EFTEM mode. Data were collected using aberration-free image shift (AFIS) with Thermo Fisher EPU software.

Cryo-EM data processing

All datasets were collected at 0.82 Å/pix and processed using a standardised workflow. Movies were motion corrected using UCSF MotionCor2⁷¹ and CTF parameters estimated using GCTF⁷². Movies for δOR-mG_si-DADLE-MIPS3983 were motion corrected and output in float16 using Relion v5.0^73,74 and CTF parameters were estimated using CTFFIND 4.1⁷⁵. Particles were picked using the general model of crYOLO⁷⁶ and extracted with RELION v3.1^73,74 (v5.0 for δOR-mG_si-DADLE-MIPS3983) at 60 pix, 4.92 Å/pix. Following import to cryoSPARC v4.1.2⁷⁷ for 2D classification and Ab-Initio model generation, heterogeneous refinement was performed with 2 initial models and a junk trap (1 initial model for δOR-mG_si-DADLE-MIPS3983). Full-sized particles were reextracted at 360 pix, 0.82 Å/pix and subjected to subsequent rounds of 2D classification and heterogeneous refinement in cryoSPARC v4.1.2. Selected particles underwent Bayesian Polishing in RELION and final non-uniform refinement with CTF refinement. For δOR-mG_si-DADLE-MIPS3983, an additional 3D classification step without alignment was performed (6 classes, T = 10) before final refinement. Dataset-specific details can be found in Supplementary Figs. 2, 6, 7 and 11.

Model building and refinement

An initial δOR template model was generated from the cryo-EM structure of the µ opioid receptor (PDB: 6DDE)⁷⁸. An initial model for the miniG_si was generated from the structure of CCK1:mG_sqi complex (PDB: 7MBY) and the Gβ₁Gγ₂, and Nb35 were taken from the CCK1:DNG_s complex (PDB: 7MBX)⁴¹. Models were placed into the cryo-EM maps using UCSF ChimeraX⁷⁹ and rigid-body-fit using PHENIX⁸⁰. Models were refined with iterative rounds of manual model building in Coot⁸¹ and ISOLDE⁸², and real-space refinement in PHENIX. Restraints for all ligands were generated using the GRADE web server (https://grade.globalphasing.org). Model validation was performed with MolProbity⁸³ and the wwPDB validation server⁸⁴. Figures were generated using UCSF ChimeraX⁷⁹ and PyMOL (Schrödinger).

Computational chemistry

To study ligand binding to the lipid-facing allosteric pocket, we performed MD simulations with all-atom representation of the receptor, G protein, ligand, native-like membrane and solvent. MD simulations were performed using Gromacs 2024.2⁸⁵ patched with Plumed 2.9.2. Amber03, GAFF2 and Slipids force fields were used for protein, small molecule ligands and membrane, respectively^86–89. The simulation box (Supplementary Table 3) with the receptor and an asymmetric membrane was built using CHARMM-GUI⁹⁰. The cryo-EM structure reported in this manuscript was used for the active-state receptor simulations, and the 4N6H PDB structure was used for the inactive-state calculations. Receptor termini were capped with N-methyl and acetate caps. The membrane was composed of nine lipid types and included cholesterol, sphingomyelin, phospholipid headgroups: phosphatidylcholine, phosphatidylethanolamine, phosphatidylserine, phosphatidylglycerol and palmitoyl, oleyl and linoleoyl lipid tails. The simulation box was filled with TIP3P water and 0.15 M NaCl. Ligand partial charges were calculated with PyRED⁹¹, and ligand topologies were generated using ACPYPE⁹² (Supplementary Files). The target volume was achieved with 5 ns NPT simulation with the Berendsen barostat, followed by 300 ns box pre-equilibration in the NVT ensemble with position restraints of 10,000 kJ/mol nm² imposed on the protein. After the membrane pre-equilibration, the simulation boxes with specified ligand sets were prepared and equilibrated. Insertion of MIPS3614 required removal of the overlapping lipids and was followed by an additional 250 ns NVT equilibration with maintained restraints on protein atoms. Energy minimisation of each system was then performed until the maximal force was less than 100.0 kJ/mol/nm. This was followed by 500 ps of equilibration run in the NVT ensemble and 1 ns in NPT ensemble with Parrinello-Rahman barostat and position restraints of 1000 kJ/mol nm² applied to the protein main chain and Cα atoms, respectively. The dimensions of the simulation boxes were approximately 83 Å × 83 Å × 141 Å with Gα protein and 83 Å × 83 Å × 107 Å without Gα protein. The Nose-Hoover thermostat was used for temperature coupling at 309 K, and the Parrinello-Rahman barostat was used for semiisotropic pressure coupling at 1 bar for the production runs. Long-range electrostatics were treated with the particle mesh Ewald method⁹³. The LINCS algorithm⁹⁴ was used to constrain the hydrogen-heavy atom bonds, and a time step 2 fs was used. All-atom unbiased MD production simulations were run for 1 μs in six replicas per complex, and trajectory frames were saved every 400 ps. Final coordinates of the ligand-receptor complexes are provided in Supplementary Files. MIPS3614 stability in the allosteric pocket was assessed with RMSD calculations. The allosteric modulator remained close to the binding mode observed experimentally in a vast majority of the simulation replicas (Supplementary Fig. 10c). To perform PCA of all complexes in a common subspace, trajectories were truncated to the receptor heavy atoms using gmx trjconv, concatenated and fitted using the Cα atoms. PCA of the resultant trajectory was then performed with the Gromacs tools gmx covar and gmx anaeig. Trajectory analyses were performed with Gromacs tools, the open-source, community-developed PLUMED library (version 2.9.2)⁸⁶, and VolMap tool. PROPKA was used to calculate pK_a values⁵². Plots and heat maps were generated using Gnuplot⁵². Molecular visualisations were created with PyMOL⁹⁵ and VMD⁹⁶.

Enzyme-linked immunosorbent assay (ELISA)

Flp-In-HEK293 cells were seeded at 50,000 cells per well into poly-D-lysine–coated 48-well plates, and 4-6 h later transfected with 100 ng/well of FLAG-hδOR WT or mutants using a 1:6 ratio of DNA:PEI in 150 mM NaCl. Cells were fixed with 3.7% (v/v) paraformaldehyde in tris-buffered saline (TBS) for 30 min. For receptor expression, cells were permeabilized by 30-min incubation with 0.5% (v/v) NP-40 in TBS. Cells were then incubated in blocking buffer [1% (w/v) skim milk powder in 0.1 M NaHCO₃ for 4 h at room temperature and incubated with mouse M2 anti-FLAG antibody (Sigma, #F1804, 1:2000, 2 h at 4 °C). After washing three times with TBS, cells were incubated with anti-mouse IgG, HRP-linked antibody (Cell Signalling Technology, Cat#7076, 1:2000) for 1 h at room temperature. Cells were washed and stained using TMB substrate (Sigma-Aldrich). Absorbance at 450 nm was measured using an EnVision Multilabel Reader (PerkinElmer).

Functional assays

For cAMP inhibition, FlpIn HEK293 cells stably expressing human δOR were seeded in 10 cm dishes and transfected with 5 μg of cAMP sensor using YFP-Epac-RLuc (CAMYEL) BRET sensor. For Nb33, mG_si, or β-arrestin 2 recruitment assays, FlpIn HEK293 cells were seeded in 10 cm dishes and transiently transfected with 1 μg of donor (human δOR-RLuc8 WT or mutant) and 4 μg of acceptor (nanobody33-Venus, miniG_si-Venus or β-arrestin2-YFP) using linear polyethylenimine (Polysciences, Warrington, PA, USA) at a DNA:PEI ratio of 1:6 in 150 mM NaCl. Twenty-four h post-transfection, 30,000 cells/well were seeded in poly-D-lysine-coated 96-well white CulturPlates (PerkinElmer, Glen Waverly, VIC, Australia) and incubated for twenty-four h at 37 °C and 5% CO₂. For selectivity experiments, FlpIn HEK293 cells were seeded in poly-D-lysine-coated 96-well white CulturPlates (PerkinElmer, Glen Waverly, VIC, Australia) and transiently transfected with 10 ng of donor (human δOR-RLuc8, human μOR-RLuc8, human κOR-RLuc8, human M4-RLuc8 or human SSTR2-NLuc) and 40 ng of acceptor (miniG_si-Venus) using linear polyethylenimine (Polysciences, Warrington, PA, USA) at a DNA:PEI ratio of 1:6 in 150 mM NaCl. Forty-eight hours post-transfection, the cells were washed and then incubated with Hanks’ balanced salt buffer containing 10 mM HEPES, pH 7.4, for 30 min at 37 °C. To investigate the effects of removing extracellular Na⁺ using the Nb33 assay, cells were instead washed and incubated with assay buffer (20 mM HEPES, 1.26 mM calcium chloride, 5.33 mM potassium chloride, 0.44 mM potassium phosphate (monobasic), 0.30 mM potassium phosphate (dibasic), 0.5 mM magnesium chloride, 0.41 mM magnesium sulphate and 5.60 mM glucose) containing either 140 mM NaCl or 140 mM ChCl⁵⁵. Increasing concentrations of DADLE in the absence or presence of increasing concentrations of each allosteric modulator were added and incubated for 5 min. Cells were then incubated for 7 min with a final concentration of 5 μM coelenterazine H (Nanolight Technology, AZ) and for cAMP inhibition also 30 μM forskolin (Sigma-Aldrich, Castle Hill, NSW, Australia). BRET was measured on a LUMIstar Omega instrument (BMG LabTech, Mornington, VIC, Australia) at 12 min post agonist addition using 475 ± 30 nm/535 ± 30 nm filters. For kinetic reads, cells were incubated for 5 min with a final concentration of 5 µM coelenterazine H (Nanolight Technology, AZ). Kinetic BRET signals were then measured on a LUMIstar Omega instrument (BMG LabTech, Mornington, VIC, Australia) using 475 ± 30 nm/535 ± 30 nm filters in 1-min intervals, where a baseline reading was taken for the first 5 min. Increasing concentrations of orthosteric ligand in the absence or presence of allosteric modulator were added and readings were taken for an additional 20 min. To determine BRET signal, the ratio of light emitted at 535 ± 30 nm by YFP or Venus over the light emitted at 475 ± 30 nm by Renilla Luciferase was quantified.

Animals

C57Bl/6 J mice (6–8 weeks, male) were obtained from the Monash Animal Research Platform. Mice were housed under a 12 h light/dark cycle, with controlled temperature (24 °C) and humidity, and free access to food and water. All procedures involving animals were approved by the Monash Institute of Pharmaceutical Sciences Animal Ethics Committee (approval 40039) and adhered to the ARRIVE guidelines. Mice were humanely killed by cervical dislocation, and the distal colon was removed and placed in modified Krebs buffer (in mM: 118 NaCl, 4.7 KCl, 1.1 MgSO₄•7H₂O, 1.18 KH₂PO₄, 25 NaHCO₃, 11.6 glucose and 2.5 CaCl₂).

Tissue contraction assay

Segments of distal colon were placed under a resting tension of 0.5 g and equilibrated for 30 min before use. Neurogenic contractions of the circular muscle were evoked by transmural electrical field stimulation (EFS; 0.5 ms duration, 3 pulses/s, 60 V). Once baseline responses to EFS were established, increasing concentrations of Leu-Enk (final concentration 1 pM-10 µM) were added to the organ baths, along with MIPS3614 (1 or 10 µM), BMS-986187 (10 µM) or vehicle (DMSO). EFS was repeated 3 times per test concentration. The peak amplitude of each EFS-evoked contraction was normalised to the baseline response in the absence of test compounds. To confirm tissue viability, carbamoylcholine (carbachol, 10 μM) was applied after each experiment⁹⁷.

Spatiotemporal mapping

Propagating contractions were assessed in the intact colon by video mapping¹⁵. After an equilibration period, motor patterns were recorded with elevated intraluminal pressure for 20 min. The effects of MIPS3614 on the number, velocity and interval between pressure-evoked colonic motor complexes (CMCs) were assessed relative to vehicle control.

Data analysis

GraphPad Prism software (v. 9.0-v11.0) was used to analyse all functional data and for statistical analysis. Data points are presented as mean ± standard error of the mean (s.e.m.) based on at least three biologically independent experiments in duplicate. Concentration-response curves were fitted using the three-parameter log(agonist) vs. response equation. pEC₅₀ values were extracted from the curve fit of each individual experiment. E_max was calculated by subtracting the Bottom from the Top parameter from the curve fit of each individual experiment. Statistical analysis of ∆pEC₅₀ and E_max values was performed using one-way analysis of variance (ANOVA) with Dunnett’s multiple comparison–corrected post hoc test against δOR WT, with a p-value of <0.05 considered significant.

When possible, the data were also fit to the operational model of allosterism⁴³ to determine equilibrium dissociation constants of agonist and allosteric modulator are represented by K_A and K_B, respectively. The efficacy of orthosteric and allosteric ligands was respectively denoted by τ_A and τ_B. αβ represents the cooperativity factor with α denoting the cooperativity between agonist and allosteric modulator at the level of binding (binding cooperativity), and β representing the cooperativity between agonist and allosteric modulator at the level of efficacy (efficacy cooperativity; or scaling factor). pK_B and log αβ values were extracted from the global fit of the grouped data of at least three biologically independent experiments. Statistical analysis of pK_B and log αβ values was performed using one-way analysis of variance (ANOVA) with Dunnett’s multiple comparison–corrected post hoc test against MIPS3614, with a p-value of <0.05 considered significant.

Chemistry

Provided in Supplementary Methods section of the Supplementary Information.

Reporting summary

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

Author contributions

A.B.G., P.J.S., C.V., D.M.T. designed the overall project. J.I.M., O.D. designed, expressed and purified protein samples. J.I.M., H.V. performed sample vitrification and cryo-EM imaging. J.I.M. and D.M.T. processed the EM data and generated and analysed atomic models. M.D.N., V.P., N.B., A.B.G. performed pharmacology experiments in recombinant systems. S.A. performed pharmacology experiments in native tissues and whole organ experiments. O.D., P.J.S., B.C., M.J. designed compounds. O.D. synthesised and characterised compounds. D.B. performed and analysed molecular dynamics simulations. M.D.N., V.P., N.B., A.B.G., S.A., D.P.P., S.E.C., C.V., D.M.T. analysed pharmacology data. D.P.P., S.E.C., A.C., B.C., M.J., J.C., A.B.G., P.J.S., C.V., D.M.T. provided project supervision. J.I.M., M.D.N., O.D., A.B.G., D.M.T. wrote the manuscript with contributions from all authors.

Peer review

Peer review information

Nature Communications thanks Susruta Majumdar, Nokomis Ramos-Gonzalez, Bryan Roth, Aida Shahraki, and Hu Zhang for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was funded by a National Health and Medical Research Council (NHMRC) Investigator Grant 1196951 (D.M.T.) and 2026533 (A.B.G), an NHMRC Ideas Grant 2021675 (D.P.P., S.E.C., C.V., J.I.M.), an NHMRC Program Grant 1150083 (A.C.), an Australian Research Council (ARC) DECRA DE240100931 (A.B.G), a Sven and Lilly Lawski Foundation stipend N2024-0043 (D.B.), and a Swedish Research Council Grant 2021-4186 (J.C.). The MD simulations were enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council through grant agreement no. 2022-06725. The work was supported by the Monash University Ramaciotti Centre for Cryo-Electron Microscopy, and the Monash eResearch capabilities, including M3 High-performance computing.

Data availability

Atomic coordinates for δOR-DADLE-mGsi-Gβ₁γ₂-Nb35, δOR-DADLE-MIPS3614-mGsi-Gβ₁γ₂-Nb35, and δOR-DADLE-MIPS3983-mGsi-Gβ₁γ₂-Nb35 were deposited in the Protein Data Bank (PDB) with the accession codes 9YDP, 9YDQ, and 9YDR, respectively. Cryo-EM maps were deposited in the Electron Microscopy Data Bank under the following accession codes: EMD-72825 (DADLE consensus map), EMD-72828 (DADLE receptor focused map), EMD-72826 (DADLE-MIPS3614 consensus map), EMD-72829 (DADLE-MIPS3614 receptor focused map), EMD-72827 (DADLE-MIPS3983 consensus map), EMD-72830 (DADLE-MIPS3983 receptor focused map), EMD-72831 (DOR-BMS-986187 consensus map), and EMD-72832 (DADLE-BMS-986187 receptor focused map). The source data underlying Fig. 2, Fig. 4, Fig. 5, Fig. 6, Supplementary Fig. 8b, 9 are provided as a Source Data File. MD trajectories, together with initial and final simulation coordinates, have been deposited at Zenodo (https://doi.org/10.5281/zenodo.21219908). Source data are provided with this paper.

Competing interests

A.C. is a co-founder and holds equity in Septerna Inc. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76730-x.