Main

Heterotrimeric G proteins transduce information to intracellular partners by modulating GTP binding and hydrolysis3. Through their interaction with G-protein-coupled receptors (GPCRs) and effectors, G proteins provide the transducer function that is necessary for the conveyance of extracellular information4,5. Heterotrimeric G proteins consist of an α subunit bound to a β and γ subunit dimer; they remain a trimer while the α subunit is bound to GDP6. Receptors provide the transmembrane conduit for a signal between the extracellular agonist and the intracellular G-protein transducer7. Specifically, GPCRs undergo a conformational change that acts to catalyse a reaction between the receptor and the Gα protein8,9,10. This interaction shifts the affinity for Gα binding to GDP to conditions that favour GDP release and GTP binding3. Thus, the receptor acts as a GEF and this reaction is considered to be primarily unidirectional11 (Fig. 1a). However, there have been observations that the GTP loading function of the receptor is reversible—that is, the receptor may facilitate the release of GTP from Gα. Early examples of this reversible interaction used nonhydrolysable forms of GTP such as GTPγS3,12, wherein the dissociation of radiolabelled GTPγS could be observed upon agonist binding to the receptor. One such study examined the kinetics of the release of 35S-GTPγS in cells expressing the mu opioid receptor (MOR) and found that the rate of release of nucleotide was increased as a function of a single saturating concentration of agonist and that partial and full agonists maintained their rank order efficacy in both exchange reactions13 (35S-GTPγS binding and 35S-GTPγS release).

Fig. 1: GPCRs induce both GTP binding and GTP release from Gα proteins and the process is agonist-mediated.
figure 1

a, Schematic of the proposed model, showing the conventional pathway of GDP-to-GTP exchange (left) and the expanded model to allow for both GTP and GDP release (right) as detailed in the linked Article14. A, agonist; R, receptor (asterisks indicate different active states); Ka, affinity constant; G protein; Gapo, unbound G protein (blue); GGDP, GDP-bound G protein (green); GGTP, GTP-bound G protein (red); α1 and α2, active state affinities. b, DAMGO-stimulated binding and release in CHO-MOR cells presented as raw data (in disintegrations per minute (dpm)) and the normalization to baseline and maximum response. c–g, Normalized binding and release with indicated agonists in CHO-K1 cells expressing MOR (c; n = 3 binding, 3 release), KOR (d; n =3 binding, 5 release), 5-HT1AR (e; n = 8 binding, 7 release), M2R (f; n = 3 binding, 3 release) and SST2R (g; n = 4 binding, 4 release). The raw data are presented in Extended Data Fig. 1. MPE, maximum possible effect. b–g, Data are mean ± s.e.m. and potency is presented as mean with 95% confidence interval. h, Comparison of potency in 35S-GTPγS binding versus release by unpaired, two-tailed t-test for each receptor comparing the individual potency (pEC50, where EC50 is half-maximal effective concentration) values measured per experiment. Data are mean with 95% confidence interval. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; NS, not significant (P > 0.05).

Source Data

Full size image

Here we investigate the significance of the release mechanism as a function of agonist concentration and how it can influence drug responsiveness in vivo. In a linked Article14, we present extensive pharmacological and biochemical characterizations of the release reaction, which we summarize in the three-state coupling model (Fig. 1a). In the study, we show that the GTP-release function of the receptor adheres to the pharmacological principles that pertain to the GTP-binding function of the receptor. In summary, the release function is dependent on agonist concentration and can be reversed by antagonists, and competitive interactions are preserved between orthosteric agonists and antagonists. Moreover, we demonstrate that the effect is due to activation of the receptor population and not merely a function of receptor occupancy. We provide experimental evidence and a functional state model that establishes that the efficacy and potency of an agonist to promote GTP release can differ from its efficacy and potency to induce GTP binding. Therefore, an agonist may demonstrate selectivity for affecting the equilibrium of the functional active state of the G protein and can show a preference for one state over the other. We also provide evidence that an agonist may have a different rank order potency and efficacy for the two states of the exchange function of a GPCR.

In the present study we show that agonists can induce both GTP binding and GTP release from Gα in a concentration-dependent manner, and that this can be observed for several different GPCRs (Fig. 1). GTP binding is assessed using a conventional method that entails incubating isolated cell membranes in the presence of 35S-GTPγS and increasing concentrations of agonists15,16. To observe GTP release, we use a ‘pulse-chase’ paradigm, which entails first loading the membrane preparations with 35S-GTPγS. Since many GPCRs are negatively regulated by sodium ions, removal of sodium allows for the constitutive activation of all sensitive receptors and the subsequent loading of 35S-GTPγS binding to G proteins. After the pulse, the chase entails dilution of the membranes and inclusion of an excess of unlabelled (cold) GTPγS in the presence of sodium (see Methods).

The two reactions are compared in Fig. 1b using membranes prepared from cells overexpressing mouse MOR. The data are presented as radioactivity counts for both the binding and the release assay; to facilitate comparison of the potencies, the data are also normalized to the baseline (0%) and the highest concentration used in each response (100%), and the curve is inverted for the release function. For the MOR, the potency of DAMGO ([D-Ala2, N-MePhe4, Gly-ol]-enkephalin), an enkephalin analogue, is conserved for both assays and the same is true for met-enkephalin (Fig. 1c). The exchange effect can also be observed for dynorphin A (1–17) at the kappa opioid receptor (KOR), serotonin (5-HT) at the serotonin 1A receptor (5-HT1AR), carbachol at the muscarinic 2 receptor (M2R) and somatostatin-14 (SST-14) at the somatostatin 2 receptor (SST2R) (Fig. 1d–g). Notably, agonist potencies at MOR and M2R are conserved in the two states, whereas dynorphin, serotonin and somatostatin are significantly more potent at their cognate receptors for promoting the release of GTP (Fig. 1h; individual curves are shown in Extended Data Fig. 1).

There are multiple clinically relevant opioid agonists that span a broad range of pharmacological characteristics (including partial agonists and biased agonists); therefore, we used these tool compounds to determine whether the release function and binding function could be dissociated at the MOR. In addition, we tested two new compounds, which were selected on the basis of their scaffold variation from biased MOR agonists introduced by our laboratory (the SR series—for example, SR-17018) and for their characteristics as full agonists that are less potent than morphine in cellular assays. The latter consideration was based on a desire to not introduce more potent opioid agonists to the scientific literature. For each drug, DAMGO was assayed in parallel to serve as a reference, since DAMGO maintains the same potency in both responses and serves to define the maximum efficacy in both assays in this cell line. Not unexpectedly, several agonists perform similarly to DAMGO, preserving the potency in both responses; however, some agonists show a differential preference for potency (Fig. 2a) and/or efficacy (Fig. 2b) for one state over the other (see Extended Data Fig. 2 for curves and Extended Data Table 1 for parameters). Since the two effects were measured in parallel with DAMGO, we also determined the difference in transduction efficiencies (ΔΔlogR; Fig. 2c and Extended Data Table 1) for each agonist in the release assay and the binding assay. This representation permits normalization between responses to directly compare agonist activity17. The two new agonists show significant gains in the release function, having nearly a hundred-fold gain in selectivity for the release active state, as measured by the difference in transduction efficiencies; we have named these compounds muzepan1 and muzepan2 (Fig. 2d), as they are mu opioid receptor-acting compounds containing an ‘azepane’ ring.

Fig. 2: Opioid agonists exhibit differential preferences for GTP binding and release in CHO-MOR cell membranes.
figure 2

a–c, Comparisons of the mean of the individual 35S-GTPγS binding and release: potencies (a), maximum efficacies (Emax normalized to DAMGO (100% versus baseline (0%)) (b), and difference in transduction efficiencies (ΔΔlogR, relative to DAMGO) (c).Data are mean with 95% confidence interval. Unpaired t-test was used for comparing binding and release parameters for each compound. Extended Data Fig. 2 shows concentration–response curves and Extended Data Table 1 presents parameters and number of individual replicates (n ≥ 3). d, Chemical structures of muzepan1 and muzepan2 with binding affinities (pKi with s.e.m., n = 6; Ki is the inhibition constant) determined from competition binding assays with 3H-naloxone.

Source Data

Full size image

Several of the agonists that show a state preference have previously been identified as biased agonists that prefer G-protein signalling over β-arrestin2 recruitment (for example, oliceridine18, PZM2119, herkinorin20, buprenorphine21,22,23 and SR-1701824). When tested in the cellular assays expressing the human MOR that were used to evaluate the biased agonism of SR-1701824, both muzepan1 and muzepan2 show no preference between GTPγS binding and β-arrestin2 recruitment (Extended Data Fig. 3a and Extended Data Table 2). Moreover, the exchange selectivity for GTP release over binding is maintained at the human receptor (Extended Data Fig. 3b, Extended Data Table 3). Therefore, whereas many of the compounds that showed selectivity for release over binding also show preference for G-protein binding over β-arrestin2 recruitment, the correlation is not absolute.

To demonstrate the physiological significance of agonist-induced GTP release, the experiment was repeated in mouse spinal cord membranes. In the binding experiment, DAMGO promotes only a 40% stimulation in the native tissue ((1.4 ± 0.01)-fold; P < 0.001, paired t-test versus baseline; Extended Data Fig. 4a). We determined that the sodium-free conditions lead to very high levels of GTPγS binding, making it difficult to see an effect of DAMGO on release. This is not unexpected, as there are relatively low levels of MOR in the system, as reflected by the approximately 40% stimulation in the binding studies (Extended Data Fig. 4a). Therefore, to isolate the MOR-accessible G-protein pool, we used DAMGO in the presence of sodium and 35S-GTPγS in the pulse phase and diluted 100-fold as part of the chase (Methods and ref. 14). We demonstrate that this is feasible in the CHO-mMOR cell line, where the potencies of DAMGO are similar to those in the sodium-free loading conditions, although the potency for the release function is slightly decreased (19 nM binding versus 43 nM release; P < 0.05, t-test; Extended Data Fig. 4b). In mouse spinal cord membranes, when 100 nM DAMGO is included in the pretreatment period, there is an increase of about 10% in 35S-GTPγS loading (P < 0.001; paired t-test), but no change is evident in the spinal cord of MOR-knockout mice (Extended Data Fig. 4c). We therefore took this modest stimulation as representative of MOR-mediated GTPγS loading in the mouse spinal cord.

In mouse spinal cord membranes, DAMGO is more potent in stimulating 35S-GTPγS binding than muzepan1 and muzepan2, whereas all agonists are full agonists (Fig. 3a). In the release paradigm, DAMGO loses potency, whereas muzepan1 and muzepan2 gain potency (Fig. 3b). Notably, the efficacy obtained by muzepan1 and muzepan2 reach the 10% maximal effect anticipated in the pulse loading (Extended Data Fig. 4c). This is in contrast to DAMGO, which does not reach this plateau, suggesting that in spinal cord, the enkephalin-like agonist may be selective against release. No significant effects were observed in spinal cord membranes from MOR-knockout mice (Extended Data Fig. 4d); therefore, the effects are likely to be due to MOR activation.

Fig. 3: Bidirectional GTP exchange in mouse spinal cord membranes.
figure 3

a, 35S-GTPγS binding and release induced by indicated agonists. Data points show mean ± s.e.m.; potencies are presented as pEC50 with 95% confidence interval. Binding: DAMGO: n = 8; muzepan1: n = 12; muzepan2: n = 4. Release: DAMGO: n = 8; muzepan1: n = 9; muzepan2: n = 4. b, Comparison of potencies between the binding and release exchange function for each agonist shown in a. Data are mean with 95% confidence interval. Unpaired, two-tailed, t-test of the mean of individual experiments.

Source Data

Full size image

In mice, muzepan1 and muzepan2 are brain penetrant as they can be detected in brain 1 h following intraperitoneal injection of 3 mg kg−1 of compound (muzepan1: 463 ± 83 nM, muzepan2: 493 ± 187 nM, n = 3; Extended Data Fig. 5a). We therefore tested the compounds for antinociceptive efficacy in the hot plate and warm water tail immersion assays and compared them with morphine (Fig. 4a). The potencies determined by the effect at 1 hour correlate with their rank order potency in GTPγS binding (as well as release), although we note that at higher doses, muzepan1 remains nearly maximally efficacious for the 4 h duration of the test (Fig. 4b). Notably, none of the agonists have effects in MOR-knockout mice (Fig. 4b,c; for morphine see ref. 24).

Fig. 4: Potency in mouse hot plate and warm water tail immersion nociceptive assays.
figure 4

a–c, Morphine (a), Muzepan1 (b) and Muzepan2 (c) were tested at the indicated doses (in mg kg−1, intraperitoneal injection (IP)) in the hot plate (left) and tail flick (middle) assays in wild-type and MOR-knockout (MOR-KO) mice. Right, potency was determined by comparing the response at 1 h based on the percentage of MPE calculated from the baseline (BL) and a cut-off time of 20 s for the hot plate and 30 s for the tail flick. Potency curves are extended to convey 0% (determined by baseline) and 100% (determined by maximum cut-off time used to estimate the ED50); potencies are presented with 95% confidence interval in the legends. a, Morphine: n = 6 (3 mg kg−1 and 24 mg kg−1), n = 10 (6 mg kg−1 and 12 mg kg−1). b, Muzepan1: n = 8 (3 mg kg−1), n = 5 (6 mg kg−1 and 12 mg kg−1), n = 7 (24 mg kg−1); MOR-KO: n = 5. c, Muzepan2: n = 4 (6 mg kg−1), n = 10 (12 mg kg−1), n = 6 (24 mg kg−1); MOR-KO: n = 3. HP, hot plate assay; TF, tail flick assay.

Source Data

Full size image

Since muzepan1 and muzepan2 promote GTP release more potently than GTP binding, we considered whether a sub-efficacious dose would alter responsiveness to conventional opioid analgesics. When co-administered with 12 mg kg−1 morphine, a low dose of muzepan1 (3 mg kg−1) enhances and prolongs morphine-induced antinociception, and exceeds the calculated additive individual effects of each drug in both nociceptive tests (Fig. 5a). Furthermore, the inclusion of muzepan1 at varying doses of morphine produces the same enhancement, significantly improving the potency (median effective dose (ED50)) of morphine by about two-fold (P < 0.0001; Fig. 5b and Extended Data Fig. 6) in both assays. Muzepan2 (12 mg kg−1), which alone produces very little antinociception, also enhances and prolongs the response of morphine in a manner that is greater than additive (Fig. 5c).

Fig. 5: GTP-release-selective agonists enhance and prolong opioid-induced antinociception in mice.
figure 5

a, Hot plate (top) and tail flick (bottom) assays with muzepan1 (Muze1; 3 mg kg−1, intraperitoneal) and morphine (Mor; 12 mg kg−1, intraperitoneal) alone and combined. The calculated sum effect of both drugs (Σ) is shown for comparison. Right, mean (± s.e.m.) area under the curve (AUC). One-way ANOVA. b, AUC for 4 h hot plate (top) and tail flick (bottom) assays following treatment with different doses of morphine (Extended Data Fig. 6). Legend indicates the ED50 (with 95% confidence interval). c, Muzepan2 (Muze2; 12 mg kg−1, intraperitoneal) also enhances morphine (12 mg kg−1, intraperitoneal)-induced antinociception (analysis as in a). d, Muzepan1 (3 mg kg−1, intraperitoneal) prolongs fentanyl (Fent; 0.3 mg kg−1)-induced antinociception in male and female mice (analysis as in a). See Extended Data Table 4 for two-way repeated measures ANOVA analyses of time course data for a–d and the number and sex of mice in each assay (n = 6–14; individual mouse data are shown as symbols in the bar charts) as well as the results of post hoc ANOVA analyses for drug effect over time.

Source Data

Full size image

To assure that the compounds are not indirectly enhancing morphine effects by competing with morphine metabolism, which occurs via glucuronidation25, we also used fentanyl, which is metabolized by CYP3A426. Notably, neither compound competes with CYP3A4 as determined by in vitro competition studies (muzepan1: 11%, muzepan2: <10% inhibition at 10 µM; Extended Data Fig. 5b). Whereas only male mice were tested with morphine owing to sex-dependent differences in morphine metabolism and sensitivity27, both male and female mice were tested with fentanyl, since the responses that we measured are comparable between the sexes24,28. When combined, muzepan1 dramatically prolongs the efficacy of fentanyl over time in a manner that is greater than the predicted additive effects of both compounds (Fig. 5d; see Extended Data Table 4 for statistical analyses of all time course data in Fig. 5).

Notably, the effects of muzepans on opioid-mediated antinociception resemble those of adding a positive allosteric modulator to an MOR agonist. Therefore, we tested whether muzepan1 could act allosterically by measuring its concentration-dependent effect on fentanyl-induced GTPγS binding in spinal cord membranes. Since we do not observe a leftward shift in fentanyl potency, we cannot conclude that muzepan1 behaves as a positive allosteric modulator at MOR (Extended Data Fig. 7).

Whereas it is desirable to enhance the antinociceptive effects of analgesic opioids, it is not desirable to enhance the respiratory suppression and bradycardia associated with these agonists. Therefore, we tested the effect of muzepan1 alone and in combination with fentanyl in mouse pulse oximetry and heart rate monitoring studies (Fig. 6). Alone, muzepan1 produces respiratory suppression and bradycardia at 24 mg kg−1 and 48 mg kg−1 but not at 3 mg kg−1; by contrast, fentanyl at lower doses (0.3 mg kg−1 and 2 mg kg−1) produces marked decreases in arterial oxygen saturation and heart rate (two-way repeated measures ANOVA, see Extended Data Table 5 for all comparisons). The response to the combination of 3 mg kg−1 muzepan1 with 0.3 mg kg−1 or 2 mg kg−1 fentanyl does not differ from the effect produced by fentanyl alone in either assay. Furthermore, increasing the dose of muzepan1 to 24 mg kg−1, a dose that produced mild but significant effects alone (two-way repeated measures ANOVA; Fig. 5 and Extended Data Table 5), is also not additive with fentanyl. Therefore, inclusion of muzepan1 with fentanyl prolongs fentanyl-induced antinociception, whereas respiratory and cardiac effects are not enhanced.

Fig. 6: Muzepan1 does not enhance fentanyl-induced respiratory suppression or bradycardia in mice.
figure 6

a,b, Muzepan1 produces respiratory suppression (a) and bradycardia (b) at intraperitoneal doses of 24 mg kg−1 and 48 mg kg−1 but not at 3 mg kg−1 in male and female mice. Neither dose alters fentanyl (0.3 mg kg−1 or 2 mg kg−1, intraperitoneal) effects when combined. c,d, The AUC for the 1.5 h after drug treatments (doses in mg kg−1, intraperitoneal, indicated in brackets) for oxygen saturation (c) and heart rate (d). One-way ANOVA with Tukey’s post hoc test. See Extended Data Table 5 for two-way repeated measures ANOVA for time course data in a,b and number and sex of mice in each assay (n = 4–8; individual mouse data are shown as symbols in the bar charts).

Source Data

Full size image

Discussion

The state-dependent interaction between GPCRs and G proteins that we reveal here implies an ongoing cycle that would affect the availability of the G proteins and their ability to engage with different effectors. Moreover, steric hindrance produced by a constitutive G-protein cycle may prevent the receptor from interacting with other effectors, particularly with direct binders, such as β-arrestins. It is attractive to speculate that persistence of the cycle could prevent the recruitment of β-arrestins. This is consistent with the observation that many of the agonists that produce state-dependent increases in potency and/or efficacy of GTP release, have been characterized as G-protein-signalling-biased agonists (that is, those that are biased against recruiting β-arrestins). However, here we present compounds that diverge from this correlation. Specifically, we focus on two full agonists that do not show bias against β-arrestin recruitment, and demonstrate that the release state selectivity cannot be the only contributor to G-protein signalling bias. Although the exceptions can be identified, it remains possible that changes in the G-protein cycling kinetics will affect GPCR signalling bias.

Our study demonstrates that all opioid agonists tested in the transfected cell system promote GTP release and that the potencies for both release and binding are similar for many agonists. In mouse spinal cord membranes, agonists are less potent in the binding assays compared with in the transfected cell system; this is probably due to the low representation of MOR-expressing neurons in the gross dissection of the spinal cord. In the same preparations, DAMGO-induced release is significantly diminished, and this decrease may be amplified by the low number of receptors in the preparation. Regardless of the concentration of receptor, the release induced by muzepans remains efficacious, and their potency is significantly improved, suggesting that the robust effect on release is preserved even in this heterogenous endogenous system.

In this system, we have utilized a nonhydrolysable form of GTP (GTPγS) to demonstrate that the GEF function of the GPCR is reversible and that agonists that show a preference for the reversal can have effects at sub-efficacious doses in vivo. It is important to acknowledge that, in the endogenous setting, we must consider that the GTP bound to the G protein is not stabilized and is still subject to hydrolysis by the intrinsic GTPase activity of the G protein. Further, this can be acted upon by other endogenous GAPs such as regulator of G-protein signalling (RGS) family proteins. However, if the receptor can promote GTP binding and release independent of the energy expenditure of hydrolysis, it may present a means to affect G-protein coupling kinetics which would change the effect of such GAPs in the system.

Our findings show that the receptor can facilitate GTP release; however, the context of the environment and the nature of the ligand will determine the ultimate consolidation of signalling. Therefore, the bidirectionality of the exchange function of a GPCR will affect the overall kinetics and energy landscape of GPCR activation. Thus, we propose a modification of the two-state model of receptor activation (as we present in Fig. 1a). We believe the net effect of an agonist on the fate of the G protein is a sum of the hydrolysis and the exchange, and not simply the forward reaction limited to energy-consuming hydrolysis.

The nature of the GTPγS binding assay favours the detection of coupling to inhibitory G proteins owing to their abundance in cells; moreover, the actions at the MOR can be blocked by pertussis toxin, further implicating the Gαi/o class of proteins14. It remains to be determined whether this occurs for other G proteins (such as, Gs or Gq proteins). In addition to having a role in determining preference for G-protein signalling, the release cycle of one G protein could shift the preference of the receptor away from another G protein. Effectively, the release cycle of a previously coupled G protein would also compete with the coupling of another class of G protein (that is, the inability to change coupling from Gi to Gq). One benefit of this approach is that it allows for the simultaneous assessment of two active states of the receptor in a single receptor population and that it can be amenable to testing in tissue. In the spinal cord, we note that DAMGO, an MOR-selective enkephalin analogue, is more potent in promoting GTP binding than release, whereas in the CHO overexpression cell lines, the DAMGO effects are nearly equivalent in both assays. Other GPCRs examined in the CHO cell lines may also prove to have differential affinities for one active state over the other when tested in the endogenous setting.

Our studies have shown that compounds that prefer the release state of the MOR are able to enhance and prolong the effects of opioid analgesics in mouse thermal nociception assays. We favour a model in which the receptor engages in multiple exchange events, keeping the G protein nearby to rapidly exchange nucleotide, and allowing another cycle to begin limiting the expenditure of energy through GTPase activity. This would enable the receptor-mediated G-protein activation to be reset without waiting for the completion of the GTPase cycle and, further, it keeps the proteins in close proximity. Of note, SR-17018 is also release-preferring, as well as a partial agonist, and a G-protein-signalling-biased agonist (over β-arrestin2 recruitment). In mice, it produces antinociception without respiratory suppression; moreover, it does not lead to tolerance in several pain assays24,29,30. In addition to being a G-protein-signalling-biased partial agonist, SR-17018 has also been shown to be a noncompetitive agonist that stabilizes G-protein signalling16. Moreover, treatment of morphine-tolerant mice with SR-17018 leads to a reversal of morphine tolerance while suppressing signs of withdrawal and restoring morphine sensitivity in mice29. It is not yet clear what exact properties lead to SR-17018 restoration of sensitivity but the identification of this property and the further generation of probe compounds that preserve or eliminate each property independently will aid our studies to disentangle desirable opioid effects from unwanted side effects.

Notably, muzepan1 produces respiratory suppression and bradycardia at 24 and 48 mg kg−1, which may not be unexpected, as it is a full agonist that does not show a preference for G-protein signalling over β-arrestin2 recruitment. However, whereas it is not surprising that the inactive dose of 3 mg kg−1 muzepan1 does not produce an additive effect with 0.3 or 2 mg kg−1 fentanyl, it is surprising that 24 mg kg−1 dose of muzepan1, which alone does suppress activity, does not enhance or prolong the effects of 2 mg kg−1 fentanyl on respiratory suppression or bradycardia. There remains the potential that we have reached a ceiling effect with the mice when combining the higher doses of each drug. However, we do note that the median lethal dose (LD50) for fentanyl has been recorded at 113 mg kg−1 (intraperitoneal) in C57BL/6 mice, suggesting that there should be potential to detect an additive decrease in oxygen saturation and heart rate31. Notably, the muzepans have been explored here for their robust ability to produce GTP release, and they may not be considered safer opioids at this time, as we clearly show that significant respiratory suppression can be observed with muzepan1 at high doses. Moreover, these probe compounds have not been evaluated for other opioid side effects, such as addiction liability, tolerance or dependence, nor have they been evaluated for general safety or toxicity. However, developing agonists that preserve the release preference, while also implementing other favourable properties, such as slow onset pharmacokinetics, partial agonism and fine-tuning preference for different signalling partners (β-arrestin versus G protein) is underway to improve the safety profile of such ligands.

With the examples provided here, we can conclude that at nearly inactive doses, muzepan1 enhances fentanyl-induced antinociception but does not enhance fentanyl-induced respiratory suppression in mice. We speculate that by favouring a release state of the receptor, either the receptor or the G proteins may have less opportunity to engage with secondary effectors that promote the cardiovascular side effects while perpetuating signalling that leads to antinociception. For example, a G protein that is cycling between GTP-bound and GDP-bound states may have more or fewer opportunities to interact with the Gβγ subunits, thereby preventing their further interaction with ion channels or other effectors, such as regulatory kinases or scaffolding proteins. Alternatively, the perpetual engagement between the receptor and G protein could sterically prevent engagement with other proteins, such as β-arrestins or other G-protein types. In this manner, one downstream signalling pathway may become more efficient while another is disengaged. In the case of SR-17018, a preference for promoting G-protein binding over β-arrestin2 recruitment as well as a preference over GIRK activation have been demonstrated24,32,33. It remains to be determined what preferences will be desirable; however it is evident that the environment of the receptor in the tissue in which it mediates its response will dictate any preference for a secondary cascade.

By demonstrating that there are changes in both the rank order potency and efficacy of agonists, we demonstrate that there is a change in activity of the receptor for the GTP binding versus GTP release functions. Further, we conclude that the GTP binding and release events are independently regulated by receptor activity and that an agonist has the opportunity to not only promote G-protein activation but also to modulate G-protein activity in a state-selective manner. We demonstrate that a GTP-releasing agonist can enhance and prolong the effects of conventional opioid agonists in antinociception without enhancing respiratory suppression or bradycardia in mice. Therefore, the selective modulation of potency and efficacy in the release function can change the dynamics of the drug effects in vivo. More broadly, since a ligand can induce a preference for exchange selectivity at a particular receptor, the agonist could thereby effectively regulate the availability of the Gα subunit for subsequent protein interactions and downstream signalling events. Overall, these findings demonstrate that for the opioid receptors, and possibly for all GPCRs, drug action will be a composite of its ability to promote both GTP binding and GTP release from the G protein.

Methods

Chemicals

Compounds were obtained from the following vendors: Sigma-Aldrich: oxymorphone (O-004-1ML), loperamide (1448005), methadone hydrochloride (M0267), morphine sulfate pentahydrate (M8777), sufentanil citrate (SML0535), herkinorin (5.08018.0001), buprenorphine (B9275), serotonin hydrochloride (H9523), carbachol (C4382), naloxone hydrochloride (N7758) and GDP (G7127). Cayman Chemical: PZM21 (20576-10), fentanyl citrate (22659) and GTPγS (35098). DAMGO (11711) was from Tocris Bioscience. Oliceridine (TRV-130; 510256) was from MedKoo Biosciences. Dynorphin A 1-17 (3195) was from Fisher Scientific. Somatostatin-14 was custom synthesized by CPC Scientific. Met-enkephalin (30-0-10) was from American Peptide Company. SR-17018, muzepan1 and muzepan2 were made in house. 35S-GTPγS (NEG030H001MC) was from Revvity. 3H-(-)naloxone was provided by the National Institute on Drug Abuse (NIDA) Drug Supply Program.

Animals

Male and female C57BL6/J (JAX:000664) and male MOR-KO (JAX:007559) mice were purchased from The Jackson Laboratory. MOR-KO mice were maintained by homozygous breeding. Mice were housed in groups of 2–5 and maintained on a 12 h:12 h light:dark cycle with food and water ad libitum. All adult mice were naive and at least ten weeks old prior to injection. For thermal antinociception tests, investigators were blinded to drugs and doses being administered. Mice were administered drugs intraperitoneally at a volume of 10 µl per g; for combinations of drugs, a single solution was prepared. All mice were used in accordance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals with approval by The Herbert Wertheim UF Scripps Institute of Biomedical Technology and Innovation Animal Care and Use Committee.

Cell lines

Chinese Hamster Ovary (CHO-K1) cells were purchased from ATCC. Human MOR (hMOR), mouse MOR (mMOR) and human KOR (hKOR) cells have been described previously24. For the other cell lines, receptor constructs were purchased from cDNA Resource Center including SST2R (SSTR20TN00), 5-HT1AR (5TR01ATN00) and M2R (MAR020TN00). The SSTR2R, 5-HT1AR and M2R cell lines were produced by electroporation of a pcDNA3.1 vector containing the N-terminally HA-tagged receptor into the parental cell line. A BD FACSAria3 flow cytometer was utilized to select for individual cells expressing receptor utilizing an anti-HA AlexaFluor 488 conjugate antibody (1:200). hMOR, hKOR, SST2R, 5-HT1AR and M2R cell lines were maintained under geneticin selection (500 µg µl−1). The mMOR cell line was maintained under puromycin selection (500 µg µl−1). Cells were maintained in 1:1 DMEM:F12 media supplemented with 10% heat-inactivated fetal bovine serum (HI-FBS) at 37 °C with 5% CO2. Prior to use in assays, cells were serum-starved for 30 min (hMOR, hKOR, 5-HT1AR and M2R) or 2 h (mMOR and SST2R) then removed from the plate with 5 mM EDTA in PBS with a scraper. Pellets were rinsed with PBS and frozen in 1.5 ml tubes at −80 °C until use. All cell lines were verified to be mycoplasma-free by monthly testing.

35S-GTPγS binding and release

35S-GTPγS binding

35S-GTPγS binding in cell lines was performed similarly as previously described16,24. In brief, for CHO-hMOR cells, pellets were homogenized with a Potter–Elvehjem Teflon-on-glass Dounce homogenizer in homogenization buffer (10 mM Tris (pH 7.4), 100 mM NaCl, 1 mM EDTA) then pelleted (20,000g, 4 °C, 30 min). All other cell lines (CHO-mMOR, CHO-hKOR, CHO-hSST2R, CHO-h5-HT1AR and CHO-hM2R) were homogenized in 10 mM Tris (pH 7.4), 1 mM EDTA. All reactions were performed with 0.1% DMSO and 0.1 nM 35S-GTPγS (specific acitivity = 1,250 Ci mmol−1; Revvity) in 50 mM Tris (pH 7.4), 100 mM NaCl, 5 mM MgCl2, 1 mM EDTA with differing quantities of protein and GDP. For CHO-hMOR, CHO-mMOR and CHO-5-HT1AR cells, 10 µg protein and 10 µM GDP was used. CHO-M2R cell reactions were performed with 3 µg protein and 3 µM GDP. CHO-SST2R cell reactions were performed with 10 µg protein and 20 µM GDP. CHO-hKOR cell reactions were performed with 15 µg protein and 3 µM GDP. Reactions were terminated by rapid filtration through GF/B filters with cold water after a 1 h incubation. Filters were punched into 96-well Opti-plates (Revvity) and dried overnight. Radioactivity was measured with 100 µl per well MicroScint-20 on a MicroBeta 2 (Revvity). For determination of bias between GTPγS and β-arrestin2 recruitment, 35S-GTPγS binding was performed in CHO-hMOR cells exactly as described24 and is shown in Extended Data Fig. 3a.

For 35S-GTPγS binding in C57BL/6J and MOR-KO spinal cord, tissue was homogenized via a Polytronic Tissue Tearor (BioSpec Products, 985370) and then a glass-on-glass Dounce homogenizer in homogenization buffer (10 mM Tris (pH 7.4), 1 mM EDTA). Homogenate was pulled through a 28G insulin needle before pelleting at 20,000g, 4 °C, 30 min. Reactions were performed with 10 µg, 10 µM GDP, 0.1% DMSO and 0.1 nM 35S-GTPγS and incubated for 1 h at 25 °C. Reactions were terminated as described above.

35S-GTPγS release in sodium-free conditions

Membranes were prepared in 10 mM Tris (pH 7.4), 1 mM EDTA as described for 35S-GTPγS binding in (10 mM Tris (pH 7.4), 1 mM EDTA). For 35S-GTPγS loading (the ‘pulse’) of CHO-hMOR, CHO-mMOR and CHO-5-HT1AR cells, 1 mg of protein was incubated with 1 nM 35S-GTPγS and 10 µM GDP in 20 ml of 50 mM Tris (pH 7.4), 5 mM MgCl2, 1 mM EDTA for 1 h at 25 °C. Identical conditions were used for CHO-SST2R cells except GDP was increased to 20 µM. For CHO-M2R cells, 0.3 mg of protein was incubated in the same conditions with 3 µM GDP. For CHO-hKOR cells, 1.5 mg of protein was incubated with 3 µM GDP. Release was performed by diluting tenfold into 50 mM Tris (pH 7.4), 100 mM NaCl, 5 mM MgCl2, 1 mM EDTA supplemented with 1 µM cold GTPγS and GDP corresponding to the receptor utilized. Release was performed for 1 h at 25 °C then terminated as described for 35S-GTPγS binding. See Supplementary Fig. 1 for a schematic.

35S-GTPγS release following 100 nM DAMGO-stimulated loading

For CHO-mMOR cells and C57BL/6J spinal cord studies relying on 100 nM DAMGO for loading, membranes were prepared as described for the sodium-free loading conditions. Then, 1 mg of protein was incubated with 1 nM 35S-GTPγS, 10 µM GDP, and 100 nM DAMGO in 2 ml of 50 mM Tris (pH 7.4), 100 mM NaCl, 5 mM MgCl2, 1 mM EDTA for 1 h at 25 °C. Release was performed in large-volume 96-well plates with 10 µg protein, 10 µM GDP, 0.1% DMSO, at 2 ml final volume and incubated for 1 h at 25 °C. Reactions were terminated as described above for 35S-GTPγS binding. See Supplementary Fig. 2 for a schematic.

Radioligand binding

3H-(-)naloxone binding studies were performed as previously described16. Membranes were prepared via homogenization with a Polytronic Tissue Tearor then glass-on-glass Dounce homogenization in homogenization buffer (50 mM Tris (pH 7.4), 1 mM EDTA). Homogenate was pulled through a 28G insulin needle before pelleting at 20,000g, 4 °C, 30 min. Binding was performed with 10 µg membrane in 10 mM Tris (pH 7.4) containing 1% DMSO and approximately 2 nM 3H-naloxone (1.50–2.04 nM; specific acitivity = 48.19 Ci mmol−1) at a final volume of 200 µl. Reactions were incubated for 1 h at 25 °C then filtered through GF/B fiberglass filters with cold 10 mM Tris (pH 7.4) by rapid filtration over GF/B filters and washed with cold 10 mM Tris buffer. Filters were punched into white, 96-well OptiPlate and dried overnight. Radioactivity was quantified using 100 µl per well MicroScint-20 on a MicroBeta2. The Kd of 3H-(-)naloxone determined by homologous competition in these studies is 0.83 (0.33–1.1) nM, n = 6.

β-arrestin2 recruitment

β-arrestin2 recruitment was performed as previously described24. In brief, U2OS-β-arrestin2-hMOR PathHunter cells were plated at a density of 5,000 cells per well in a 384-well, white-walled assay plate in OptiMEM supplemented with 1% HI-FBS and incubated at 37 °C with 5% CO2 for 16–20 h. Drug was prepared in PBS and cells were treated for 90 min at 37 °C. β-arrestin2 recruitment was determined using the PathHunter Detection Kit and luminescence was measured using a BioTek Synergy Neo2 multimode plate reader (BioTek).

Cytochrome P450 inhibition

Inhibition studies were carried out with 10 µM compound incubated with human liver microsomes and selective marker substrates (1A2, phenacetin demethylation to acetaminophen; 2C9, tolbutamide hydroxylation to hydroxytolbutamide; 2D6, bufuralol hydroxylation to 4′-hydroxybufuralol; 3A4, midazolam hydroxylation to 1′-hydroxymidazolam). After a 10 min incubation, the reaction was terminated and the percent inhibition was determined as previously described34.

Antinociception

Thermal antinociception was performed as previously described24. Prior to testing, mice were habituated to the testing room for 1 h. The tail flick test was determined as the amount of time until a mouse rapidly flicked its tail when placed 2–3 cm into a 49 °C water bath with a cut-off applied at 30 s. The hot plate test was measured using a 52 °C hot plate analgesia meter (Columbus Instruments) and forepaw or hindpaw licking or flicking were observed with a maximum latency of 20 s to prevent tissue damage.

Respiration and heart rate measures

Respiratory and heart rate parameters were simultaneously measured using the MouseOx Plus pulse oximeter (Starr Life Sciences) as previously described16,24. Two days prior to testing, mice were shaved around the neck and habituated for 30 min to the collars and 50 ml conical tubes which were modified to restrain the mice. The following day, the mice were habituated to the collars and restraint for 30 min. On testing day, the basal vital signs of the mice were determined for 30 min then animals were injected and monitored for 90 min.

Pharmacokinetics

Male C57BL6/J mice were injected intraperitoneally with muzepan1 or muzepan2 at the doses indicated and blood was collected at indicated time points. Brains were collected following cervical dislocation, and snap frozen in liquid nitrogen. Samples were subjected to liquid chromatography (Shimadzu)–tandem mass spectrometry from AB Sciex. Pharmacokinetic parameters were calculated using a noncompartmental model24 (Phoenix WinNonlin, Pharsight).

Data analysis

Concentration response studies were analysed by nonlinear regression analysis following normalization (baseline = 0 and maximum response = 100%). For all of the studies, we used the mean of the individual experiments to generate the potency (logEC50) and efficacy (Emax) values as presented as pEC50 with 95% confidence interval in the figures and graphs and as pEC50 with s.e.m. in the table. Both potency and efficacy parameters were produced using three-parameter nonlinear regression with adaptation of the equation:

$$\mathrm{Response}=\mathrm{basal}+\frac{{E}_{\max }-\mathrm{basal}}{{10}^{({\mathrm{logEC}}_{50}-X)}+1}$$

where X is the agonist concentration in log molar units and logEC50 is agonist potency in log molar units. Statistical analyses comparing binding and release parameters, for each compound, was performed by unpaired t-test. For the MOR studies, DAMGO was tested in parallel for all compounds and was used for normalization (baseline = 0, DAMGO at 10 µM = 100%). Statistical comparisons between binding and release parameters were performed by unpaired t-test comparing the individual parameters determined in each experiment; the number of replicates are indicated in the table.

In addition, a form of the operational model frequently applied to bias analysis was employed as the binding and release assays were considered independent measures of agonist activity17,35. The equation takes the form:

$$\mathrm{Response}=\mathrm{basal}+\frac{{E}_{\max }-\mathrm{basal}}{1+{\left(\frac{1+{10}^{(X+\log K)}}{{10}^{(X+\log {R}_{\text{reference}}+\Delta \log R)}}\right)}^{n}}$$

where basal and Emax describe the system limits, and n defines the transducer slope. For the reference agonist DAMGO, logK and ΔlogR are held constant at zero. In this case, the logRreference for the reference agonist reduces to the plogEC50. For full test agonists the logRreference is held constant, from the fit of the reference agonist, and the ΔlogR is permitted to float. The logK is held constant at zero for all full agonists. For partial test agonists, the logRreference is again held constant and the ΔlogR and logK are permitted to float.

For the determination of the transduction efficiency, the ΔlogR was determined for each individual assay with DAMGO serving as the reference agonist. The ΔΔlogR was determined by unpaired t-test between the ΔlogR from the G-protein release assay and the ΔlogR of the G-protein binding assay. The same approach was used to determine the bias factor comparing the ΔlogR in G-protein binding versus the ΔlogR in β-arrestin2 recruitment in the CHO-hMOR cells.

In radioligand binding studies, naloxone competition was fit to the homologous (naloxone) or heterologous (muzepan1, muzepan2) competition equation:

$$\mathrm{Binding}=\mathrm{bottom}+\frac{{B}_{\max }\times [{}^{3}\text{H-naloxone}]}{[{}^{3}\text{H-naloxone}]+{10}^{\log (X)}+{K}_{\mathrm{naloxone}}}$$

where, for 3H-naloxone binding, ‘bottom’ and Bmax are the non-specific and maximum binding, [3H-naloxone] is the radioligand concentration, Knaloxone is the naloxone equilibrium dissociation constant, and X is the cold naloxone concentration in molar units. For muzepan1 and muzepan2, competition data were fit to the heterologous competition equation:

$$\mathrm{Binding}=\mathrm{bottom}+\frac{\mathrm{Top}\,-\,\mathrm{bottom}}{1+\frac{{10}^{\log (X)}}{{10}^{{K}_{i}\times \left(1+\frac{[{}^{3}{\rm{H}}-\mathrm{naloxone}]}{{K}_{\mathrm{naloxone}}}\right)}}}$$

where parameter definitions are shared between the two equations. In the heterologous competition equation, Top is the maximum observed binding, Ki is the molar affinity constant of the competitive ligand, and X is the concentration of the competitive ligand. Experiments were run together and both [3H-naloxone] and Knaloxone are held constant for the analysis.

Antinociception

A maximum possible effect (%MPE) was calculated as 100% × [(baseline response − test response)/(cut-off time − baseline response)]. For the determination of potency, the %MPE was compared at the 1 h time point by nonlinear regression analysis in GraphPad Prism (v.10.4), sharing the Hill slope and constraining the bottom to 0 and the top to 100%. For the comparison of morphine potency with and without muzepan1, the AUC was determined from the %MPE over the 4 h testing period; this was normalized to the maximum possible effect (the AUC if all points reached 100%) and fit the nonlinear regression analysis. Statistical comparisons of the logED50 were made between two curves in Prism using an extra-sum-of-squares F test.

Respiration and heart rate

A two-way repeated measures ANOVA was used to compare drug effects (35–120 min) as a function of time and the results are presented in Extended Data Table 5. In addition, the AUC was determined by normalizing to the first 30 min of habituation for the following drug effect over 1 h. These values were then compared by one-way ANOVA comparing to vehicle with a Dunnett’s post hoc test, or between drug treatments (fentanyl versus fentanyl plus 3 mg kg−1 or 24 mg kg−1 muzepan1; 3 groups, Tukey’s post hoc test).

Reporting summary

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