Abstract
Biased agonism is having a major impact on modern drug discovery, and describes the ability of distinct G protein–coupled receptor (GPCR) ligands to activate different cell signaling pathways, and to result in different physiologic outcomes. To date, most studies of biased agonism have focused on synthetic molecules targeting various GPCRs; however, many of these receptors have multiple endogenous ligands, suggesting that “natural” bias may be an unappreciated feature of these GPCRs. The μ-opioid receptor (MOP) is activated by numerous endogenous opioid peptides, remains an attractive therapeutic target for the treatment of pain, and exhibits biased agonism in response to synthetic opiates. The aim of this study was to rigorously assess the potential for biased agonism in the actions of endogenous opioids at the MOP in a common cellular background, and compare these to the effects of the agonist d-Ala2-N-MePhe4-Gly-ol enkephalin (DAMGO). We investigated activation of G proteins, inhibition of cAMP production, extracellular signal-regulated kinase 1 and 2 phosphorylation, β-arrestin 1/2 recruitment, and MOP trafficking, and applied a novel analytical method to quantify biased agonism. Although many endogenous opioids displayed signaling profiles similar to that of DAMGO, α-neoendorphin, Met-enkephalin-Arg-Phe, and the putatively endogenous peptide endomorphin-1 displayed particularly distinct bias profiles. These may represent examples of natural bias if it can be shown that they have different signaling properties and physiologic effects in vivo compared with other endogenous opioids. Understanding how endogenous opioids control physiologic processes through biased agonism can reveal vital information required to enable the design of biased opioids with improved pharmacological profiles and treat diseases involving dysfunction of the endogenous opioid system.
Introduction
Opioids are the most widely used and most effective analgesics available. However, their use is associated with a large array of side effects (Benyamin et al., 2008). Opioids that are currently used as therapeutics, such as morphine, predominantly exert both their analgesic effects and undesirable effects through activation of the µ-opioid receptor (MOP) subtype (Matthes et al., 1996; Cox et al., 2015). It is now accepted that chemically distinct ligands binding to the same G protein–coupled receptor (GPCR) can stabilize the receptor in multiple active conformations, which results in differential activation of cell signaling pathways and, eventually, in different physiologic outcomes, a phenomenon known as biased agonism (Kenakin et al., 2012). Biased agonism can be exploited to design drugs that selectively activate signaling pathways, leading to the desired physiologic effects while minimizing on-target side effects elicited by activation of other signaling pathways via the same receptor subtype (Gesty-Palmer et al., 2009; Walters et al., 2009; Valant et al., 2014).
Biased agonism at the MOP has been extensively studied (McPherson et al., 2010; Al-Hasani and Bruchas, 2011; Raehal et al., 2011; Pradhan et al., 2012; Rivero et al., 2012; Kelly, 2013; Williams et al., 2013; Thompson et al., 2015). However, evidence of biased agonism at the MOP is mostly limited to qualitative analyses of two signaling events with a limited number of ligands. Evidence of the involvement of β-arrestin 2 (β-arr2) in the mediation of the adverse effects of morphine has prompted the search for opioids that do not induce β-arr2 recruitment. Indeed, enhanced morphine analgesia with reduced respiratory and gastrointestinal side effects was reported in β-arr2 knockout mice (Bohn et al., 1999; Raehal et al., 2005; Maguma et al., 2012). Recently, ligands with impaired β-arr2 recruitment have been reported to provide potent analgesia with less severe side effects (Groer et al., 2007; DeWire et al., 2013; Soergel et al., 2014). However, the improved pharmacological profiles of these ligands may still be due to their partial agonism, with lower overall efficacy than morphine, rather than true “bias” away from β-arr2 recruitment (Thompson et al., 2015).
To guide discovery efforts to generate drugs with therapeutically relevant biased agonism profiles, it is necessary to quantify this phenomenon. In any system, the observed biased agonism can be confounded by “system bias,” which reflects the differing coupling efficiencies of the receptor for each signaling pathway, and by “observational bias,” which results from differing assay sensitivities and conditions (Kenakin and Christopoulos, 2013). Bias imposed by the ligand on the conformation of the receptor is the only source of bias that can be chemically optimized to improve its therapeutic profile. Therefore, it is important to quantify biased agonism in such a way that excludes system and observational bias to reveal the unique signaling profile that is induced by the ligand. Several analytical methods to quantify biased agonism have been developed. The method recently described by Kenakin et al. (2012), based on the Black and Leff (1983) operational model, can be applied to concentration-response curves to obtain a single parameter that describes bias between signaling pathways in a system-independent manner.
Several studies have quantified biased agonism at the MOP, but these have been limited to comparison of efficacies for G protein activation versus β-arr2 recruitment (McPherson et al., 2010; Molinari et al., 2010; Frolich et al., 2011; Rivero et al., 2012). However, differential activation of other signaling events including receptor internalization, mitogen-activated protein kinase, and protein kinase C (PKC) activation has been reported (Williams et al., 2013), but these signaling events are rarely considered when describing bias of the MOP, or when attempting to link biased agonism with physiologic effects. Moreover, bias between G protein activation and β-arrestin recruitment may not necessarily predict differential activation of such downstream signaling events. Therefore, it is important to study biased agonism at multiple signaling pathways to encompass multiple aspects of the signaling properties of a ligand.
To date, descriptions of biased agonism have mainly focused upon the actions of synthetic ligands. However, the existence of multiple endogenous ligands targeting the same GPCR suggests the potential for “endogenous” biased agonism. Indeed, biased agonism by endogenous peptide ligands has been observed at several GPCRs (Zidar, 2011; Rajagopal et al., 2013; Zhao et al., 2013). Such observations may explain, in part, the apparent redundancy of such systems. However, the potential for bias within the endogenous opioid system has not been explored extensively. The endogenous ligands for opioid receptors are small peptides generated by cleavage of precursor proteins: pro-enkephalin, pro-opiomelanocortin, and pro-dynorphin. Pro-enkephalin–derived and pro-opiomelanocortin–derived peptides, enkephalins, and endorphins, respectively, are generally localized to regions with MOP or δ-opioid receptor expression and are involved in the numerous physiologic processes mediated by these receptors. Similarly, the pro-dynorphin–derived peptides, the dynorphins and neoendorphins, are generally localized to similar regions as the κ-opioid receptor; however, it is likely that, in some regions, dynorphins may also activate the MOP. All of the opioid peptides have varying affinities for all three opioid receptors, and none are significantly selective for one receptor subtype (Mansour et al., 1995; Janecka et al., 2004). Finally, the precursor genes for endomorphin-1 (endo-1) and endomorphin-2 (endo-2) are still unknown, and therefore, these two ligands remain “putatively endogenous”; however, they are the most selective and potent opioid peptides for the MOP (Zadina et al., 1997). Such diversity in the endogenous opioid system suggests that biased agonism may play a role in control of normal MOP-mediated physiologic processes. Understanding the fundamental basis for ligand bias at the MOP and determining whether differences in the expression and release of endogenous opioids can underlie the development and maintenance of disease may offer promising avenues to address and provide mechanistic insight for the development of safer opioid-based therapies.
In the current study, we investigated the existence of biased agonism at the MOP by endogenous and putatively endogenous opioids. The ability to activate several signal transduction pathways for a range of opioid peptides and three reference ligands was assessed. Bias between each signaling pathway was quantified to obtain unique biased agonism profiles for these ligands. The results show that, although most endogenous opioids possess similar biased agonism profiles to one another, α-neoendorphin (α-neo) and the putative endogenous opioid endo-1 display unique biased agonism profiles. Our studies also provide a foundation for future studies aimed at linking these profiles to physiology of the opioid system.
Materials and Methods
Chinese hamster ovary (CHO) FlpIn cells and Dulbecco's modified Eagle's medium (DMEM) were purchased from Invitrogen (Mulgrave, VIC, Australia). Fetal bovine serum was purchased from ThermoTrace (Melbourne, VIC, Australia). Hygromycin-B was purchased from Roche Applied Science (Dee Why, NSW, Australia). Sure-Fire cellular extracellular signal-regulated kinase 1/2 (ERK1/2) assay kits were a generous gift from TGR BioSciences (Adelaide, SA, Australia). AlphaScreen reagents for ERK1/2 assays, [35S]guanosine 5′-O-[γ-thio]triphosphate ([35S]GTPγS; >1000 Ci/mmol), [3H]diprenorphine, and Ultima Gold scintillation cocktail were purchased from PerkinElmer Life Sciences (Melbourne, VIC, Australia). All endogenous opioid peptides were purchased from Mimotopes (Melbourne, VIC, Australia). All other chemicals were purchased from Sigma-Aldrich (St. Louis, MO). Morphine HCl was from GlaxoSmithKline (Boronia, VIC, Australia). Renilla Luciferase (Rluc)–tagged MOPr (MOP-Rluc) was a gift from Laura Bohn (Scripps Research Institute, Jupiter, Florida).
Cell Culture and Generation of Stable Cell Line.
Cells were maintained and cultured in high-glucose DMEM containing 10% fetal bovine serum and 600 μg/ml hygromycin B at 37°C under a humidified atmosphere containing 5% CO2. cDNA encoding the wild-type human MOP was obtained from the Missouri University of Science and Technology (http://www.cdna.org) and was provided in pcDNA3.1+. Sequence of the human MOP was amplified by polymerase chain reaction and cloned into the Gateway entry vector pENTR/D-TOPO, using the pENTR directional TOPO cloning kit, according to the manufacturer’s instructions (Life Technologies, Mulgrave, VIC, Australia). The construct was subsequently transferred into the Gateway destination vector pEF5/frt/V5/dest using the LR Clonase enzyme mix (Invitrogen), and the constructs were used to transfect FlpIn CHO cells (Invitrogen), which ensure constant expression of the MOP across a cell population. Cells were selected using 600 μg/ml hygromycin B to generate cell lines stably expressing MOP.
Membrane Preparation.
Cells from 10 175-cm2 flasks were grown until approximately 90% confluence and harvested using 2 mM EDTA in phosphate-buffered saline (137 mM NaCl, 2.7 mM KCl, 4.3 mM Na2HPO4, and 1.5 mM KH2PO4). Cells were pelleted by centrifugation for 10 minutes at 1200g, and the pellets were resuspended in 20 ml of phosphate-buffered saline containing 20 mM HEPES and 10 mM EDTA at pH 7.4. All subsequent steps were performed at 4°C. The cell suspension was homogenized using a Polytron homogenizer (PT 1200 CL; Kinematica, Basel, Switzerland), with three 10-second bursts separated by cooling on ice. The cell homogenate was centrifuged for 5 minutes at 1700g, and the supernatant was transferred to new tubes and further centrifuged (90 minutes, 38,000g). The pellet was resuspended in 5 ml of buffer (20 mM HEPES and 0.1 mM EDTA, pH 7.4) and briefly homogenized to ensure uniform consistency. Membranes were aliquoted and stored at −80°C. The protein concentration was determined using a Bradford assay and bovine serum albumin as a standard.
Saturation Radioligand Binding Assay.
Cell membranes (20 μg) were incubated in buffer (50 mM Tris, 100 mM NaCl, 3 mM MgCl2, pH 7.4) containing increasing concentrations of [3H]diprenorphine (0.01–10 nM) at 25°C for 60 minutes. The reaction was terminated by rapid filtration through glass fiber filters (GF/B) with a Brandel cell harvester (Brandel Inc, Gaithersburg, MD) and washing with cold saline. Nonspecific binding was determined using 1 mM naloxone, and radioactivity was determined by liquid scintillation counting using Ultima Gold scintillation cocktail and a Tri-Carb 2900TR liquid scintillation counter (PerkinElmer).
[35S]GTPγS Binding Assay.
Cell membranes (10 μg) were incubated in buffer (20 mM HEPES, 100 mM NaCl, 4 mM MgCl2, pH 7.4) containing 1 μm GDP, 0.01% bovine serum albumin (BSA), protease inhibitors (1 μM captopril, 1 μM phosphoramidon, and 1 μM amastatin), and increasing concentrations of agonist for 30 minutes at 30°C. A 100-μl volume of [35S]GTPγS (0.1 nM final concentration) was then added, and the incubation was continued for a further 30 minutes. The reaction was terminated by rapid filtration through glass fiber filters (GF/B) with a Brandel cell harvester and washing with cold saline. Radioactivity was determined by liquid scintillation counting using Ultima Gold scintillation cocktail and a Tri-Carb 2900TR liquid scintillation counter (PerkinElmer). All experiments were performed in triplicate.
Inhibition of Forskolin-Induced cAMP Levels.
The ability of ligands to inhibit forskolin-induced cAMP production was assessed in FlpIn CHO-MOP cells transiently transfected to express the CAMYEL cAMP bioluminescence resonance energy transfer (BRET) biosensor (Jiang et al., 2007). FlpIn CHO-MOP cells were grown overnight in 10-cm dishes. Transient transfection was performed using polyethylenimine at a 6:1 ratio of DNA. Twenty-four hours after transfection, FlpIn CHO-MOP cells were seeded in white 96-well plates (Culturplates; PerkinElmer) and incubated overnight. The following day, cells were rinsed and preincubated in Hanks’ balanced saline solution with 0.01% BSA and protease inhibitors (1 μM captopril, 1 μM phosphoramidon, and 1 μM amastatin) for 30 minutes at 37°C. Cells were then incubated with the Rluc substrate coelenterazine-h, final concentration 5 µM, for 5 minutes, followed by a further 5-minute incubation with increasing concentrations of agonists. Forskolin was added to a final concentration of 10 µM. After 5 minutes, the yellow fluorescent protein (YFP) and the Rluc emissions were measured using a LumiSTAR Omega instrument (BMG Labtech, Ortenberg, Germany), which allows for sequential integration of the signals detected at 475 ± 30 and 535 ± 30 nm, using filters with the appropriate band pass. Data are presented as a BRET ratio, calculated as the ratio of YFP to Rluc signals, and expressed as the percentage of the forskolin-induced signal.
BRET Assays.
Agonist-induced recruitment of β-arrestins to the MOP and MOP proximity to the plasma membrane maker KRas were examined using a BRET-based method. Parental FlpIn CHO cells were transfected to coexpress MOP C-terminally tagged with Rluc and β-arr1–Venus, β-arr2–Venus, or KRas-Venus at a 1:4 DNA ratio. The assay was performed as described earlier for the cAMP BRET-based assay. For β-arrestin recruitment assays, agonists were added after 5 minutes of preincubation with coelenterazine-h and then incubated for an additional 5 minutes, and the BRET ratio was determined. For receptor trafficking (KRas BRET) assays, cells were incubated with agonists for 60 minutes in the presence of protease inhibitors, and coelenterazine-h added 10 minutes prior to detection of BRET. Data are expressed as d-Ala2-N-MePhe4-Gly-ol enkephalin (DAMGO) response for β-arrestin assays and percent vehicle response for KRas assays.
ERK1/2 Phosphorylation Assay.
Cells were seeded at 4 × 104 cells/well in clear 96-well plates and grown for 5 hours. Cells were washed twice with 200 μl of phosphate-buffered saline and incubated in serum-free DMEM overnight at 37°C in 5% CO2. ERK1/2 phosphorylation (pERK) was detected using the AlphaScreen ERK1/2 SureFire protocol (TGR Biosciences). The assay was performed at 37°C, and cells were preincubated with 0.01% BSA and protease inhibitors (1 μM captopril, 1 μM phosphoramidon, and 1 μM amastatin) followed by addition of ligands to a final volume of 200 μl. Time course experiments were initially performed to determine the time at which maximal pERK was detected following agonist stimulation (5 minutes). Ligand-stimulated pERK was terminated by removal of media and drugs, followed by the addition of 100 μl of SureFire lysis buffer per well and agitation of lysates for 5 minutes at room temperature. Five microliters of lysate was added in a 384-well white ProxiPlate (PerkinElmer). A mixture of SureFire activation buffer, SureFire reaction buffer, and AlphaScreen beads was prepared in a ratio of 100:600:3 (v/v/v) and added to the lysate for a lysate:mixture ratio of 5:8 (v/v). Plates were incubated for 1–1.5 hours at 37°C before the fluorescence signal was measured on a Fusion-α plate reader (PerkinElmer) using standard AlphaScreen settings. For all experiments, 10 µM DAMGO (100%) and vehicle (0%) were used to normalize pERK curves.
Data Analysis.
Agonist concentration-response curves were fitted empirically to a three-parameter logistic equation using Prism 6.0 (GraphPad Software Inc., La Jolla, CA) where bottom and top are the lower and upper plateaus, respectively, of the concentration-response curve, [A] is the molar concentration of agonist, and EC50 is the molar concentration of agonist required to generate a response halfway between the top and the bottom:
(1)To compare agonist profiles and to quantify stimulus bias, agonist concentration-response data were fitted to the following form of the operational model of agonism (Black and Leff, 1983):
(2)where Em is the maximal possible response of the system, basal is the basal level of response, KA denotes the functional equilibrium dissociation constant of the agonist (A) for the receptor, τ is an index of the signaling efficacy of the agonist and is defined as RT/KE (where RT is the total number of receptors and KE is the coupling efficiency of each agonist-occupied receptor), and n is the slope of the transducer function that links occupancy to response. The analysis assumes that the maximal system responsiveness (Em) and the transduction machinery used for a given cellular pathway are the same for all agonists, such that the Em and transducer slope (n) are shared between agonists. Data for all compounds for each pathway were fit globally to determine values of KA and τ.
The ratio τ/KA was determined as a logarithm [i.e., log(τ/KA)] and is referred to herein as the transduction coefficient which represents a single fitted parameter sufficient to describe agonism and bias for a given pathway [i.e., biased agonism can result from either a selective affinity (KA) of an agonist for a given receptor state(s) and/or a differential coupling efficacy (τ) toward certain pathways].
To cancel the impact of cell-dependent effects on the observed agonism at each pathway, the log(τ/KA) values were then normalized to that determined for DAMGO at each pathway to yield a normalized transduction coefficient, Δlog(τ/KA), which was calculated as follows:
(3)Finally, to determine the actual bias of each agonist between different signaling pathways, the Δlog(τ/KA) values were evaluated statistically between the pathways. The ligand bias of an agonist for one pathway, j1, over another, j2, is given as:
(4)A lack of biased agonism as compared with the reference agonist DAMGO will result in values of ΔΔlog(τ/KA) not significantly different from 0 between pathways. To account for the propagation of error associated with the determination of composite parameters using eqs. 3 and 4, the following equation was used:
(5)All potency (pEC50) and transduction ratio [ΔΔlog(τ/KA)] parameters were estimated as logarithms. Fold changes in bias were calculated by first converting values of ΔΔlog(τ/KA) to the corresponding antilog value:
(6)The distribution of antilog values does not conform to a normal (Gaussian) distribution, whereas the logarithm of the measure is approximately Gaussian (Motulsky and Ransnas, 1987; Leff et al., 1990; Christopoulos, 1998). Thus, as the application of t tests and analyses of variance assume Gaussian distribution, estimating the parameters as logarithms allows valid statistical comparison. All results are expressed as the mean ± S.E.M. Statistical analysis was performed using a two-way unpaired Student’s t test to make pairwise comparisons between bias factors for a given ligand and DAMGO, where P < 0.05 was considered to be statistically significant.
To visualize bias between multiple pathways at once, webs of bias were constructed. ΔΔτ/KA values between a reference pathway, cAMP, and all other pathways were obtained using DAMGO as the reference ligand as follows:
(7)
(8)Principal Component Analysis.
Principal component analysis (PCA) is a dimensionality reduction method that uses transformations to project a high-dimensional set of data into a lower dimensional set of variables called principal components (PCs). The PCs extract the important information from the data, revealing the data’s internal structure in a way that best explains its variance (Wold et al., 1987). PCs are ranked according to the percentage of total variance in the data they explain. The first PC explains a maximal amount of total variance in the data. Each succeeding PC explains a maximal amount of the remaining variation, without being correlated with the preceding components. PCA was applied using singular value decomposition as implemented in the package scikit-learn (Pedregosa et al., 2011); the script used for the analysis and plotting can be found at https://github.com/thomas-coudrat/pca_analysis.
Results
The ability of the MOP to activate several signal transduction pathways in response to 10 endogenous peptides, endo-1, endo-2, and three synthetic ligands was assessed in FlpIn CHO cells stably expressing the human MOP. Receptor expression levels were determined by saturation binding assays with [3H]diprenorphine (Bmax 0.72 ±0.04 pmol/mg protein; Supplemental Fig. 1). The selected opioid peptides included members of all three of the main classes of endogenous opioids (enkephalins, dynorphins, and endorphins) as well as the putative endogenous ligands, the endomorphins (Table 1). Quantification of bias between each pathway was performed using DAMGO as the reference ligand. Morphine was included as an additional control ligand, as the literature suggests that it may be a biased agonist at the MOP compared with DAMGO (Raehal et al., 2011; Pradhan et al., 2012; Williams et al., 2013), and finally, the signaling profile of the peripherally restricted MOP agonist loperamide was also investigated. All incubations in the different signaling assays were performed in the presence of endopeptidase inhibitors (see Materials and Methods) to prevent peptide degradation.
Endogenous and selective opioid peptides used in this study
Endogenous Opioids Differentially Inhibit Adenylyl Cyclase and Recruit β-Arrestin 2.
MOPs are primarily coupled to Gαi G proteins that mediate inhibition of adenylyl cyclase (AC), resulting in a decrease in the levels of intracellular cAMP. The ability of ligands to inhibit forskolin-induced cAMP production was assessed in FlpIn CHO-MOP cells transiently transfected to express the CAMYEL cAMP BRET-based biosensor (Fig. 1, A and B; Supplemental Table 1). All ligands inhibited forskolin-induced cAMP stimulation. Such inhibition was abolished by incubation of the cells with pertussis toxin (unpublished data) demonstrating the Gαi dependence of this effect. However, concentration-response curves could not be obtained for dynorphin A (dynA) and dynA 1–13. Although these two peptides stimulated the production of cAMP at high concentrations (Fig. 1B), this effect was not mediated by the MOP, as it was not blocked by the MOP antagonist naloxone, and was still observed in untransfected FlpIn CHO cells (Supplemental Fig. 2).
Biased agonism of endogenous opioids between inhibition of forskolin-induced cAMP and β-arr2 recruitment. (A and B) Inhibition of forskolin (Fsk)-induced cAMP in CHO-MOP cells. (C and D) Recruitment of β-arr2 in FlpIn CHO cells. Data normalized to the 10 μM DAMGO response. Data are expressed as the mean ± S.E.M. of at least three separate experiments. (E) Bias factors for all agonists between cAMP and β-arr2 recruitment (Supplemental Table 2). *P ≤ 0.05; **P ≤ 0.005, different from DAMGO, as determined by two-tailed t test. Bias factors for DynA and DynA 1–13 not shown as log[τ/KA] values could not be calculated for the cAMP assay. NC, not calculable.
We then examined recruitment of β-arr2 to the MOP using a BRET assay. CHO FlpIn cells were cotransfected with Rluc-tagged MOP and YFP-tagged β-arr2. All endogenous opioids as well as DAMGO and loperamide stimulated recruitment of β-arr2. In contrast, morphine displayed partial agonism by only stimulating 28% of the DAMGO-mediated response (Fig. 1, C and D; Supplemental Table 1).
To quantify biased agonism at these two signaling pathways, bias factors were calculated as described in Materials and Methods (Fig. 1E; Supplemental Table 2). Both loperamide and the endogenous opioid, α-neo, showed significant bias toward inhibition of AC over recruitment of β-arr2 compared with the reference ligand DAMGO. Morphine and all the other opioid peptides tested were not significantly biased compared with DAMGO.
Differential Recruitment of β-Arrestin Isoforms by Opioid Ligands.
Morphine has previously been shown to preferentially stimulate recruitment of β-arr2 over β-arr1 (Bohn et al., 2004). Differential recruitment of arrestins may have significant effects on downstream signaling, as different β-arrestin isoforms have different functions (DeWire et al., 2007; Nobles et al., 2011). We examined the potential for endogenous opioids to stimulate differential recruitment of β-arrestin isoforms to the MOP. For this, β-arrestin recruitment assays were repeated using YFP-tagged β-arr1. All the ligands tested stimulated recruitment of β-arr1, with the exception of morphine, which did not stimulate any detectable recruitment (Fig. 2, A and B; Supplemental Table 1). Of note, the BRET signal was much lower for the β-arr1 assays compared with the BRET signal obtained for β-arr2. Although this could be interpreted as a compromised recruitment of β-arr1 for all the ligands, we cannot exclude the possibility that this is due to lower BRET efficiency between the different arrestin isoforms and the Rluc-tagged receptor. Bias factors between recruitment of β-arr1 and β-arr2 showed that, in addition to morphine, endo-1 differentially recruits β-arrestins in comparison with the reference ligand DAMGO (Fig. 2C; Supplemental Table 2). Endo-1 showed significant bias toward recruitment of β-arr2 over β-arr1. Since morphine did not stimulate any detectable β-arr1 recruitment, a bias factor could not be calculated. However, this does not necessarily indicate that morphine is biased toward β-arr2 recruitment. Since all the ligands showed a lower response in the β-arr1 assay than for β-arr2, a partial agonist such as morphine is expected to stimulate little to no response in the β-arr1 assay. Hence, our results suggest that bias between β-arrestin isoforms can be due to the lower sensitivity of β-arr1 assays and/or lower coupling efficiency of β-arr1 recruitment to the MOP.
Bias between β-arr1 and β-arr2 recruitment. (A and B) Recruitment of β-arr1 in FlpIn CHO cells. Data are normalized to the 10 μM DAMGO response and expressed as means ± S.E.M. of three separate experiments. (C) Bias factors for all agonists between β-arr1 and β-arr2 (Supplemental Table 2). *P ≤ 0.05, different from DAMGO as determined by two-tailed t test. NC, not calculable.
Biased ERK1/2 Phosphorylation Correlates with Bias between cAMP and β-Arrestin Recruitment.
Next, we examined activation of ERK1/2. Agonist-induced stimulation of pERK was measured in FlpIn CHO-MOP cells using the AlphaScreen phospho ERK1/2 assay. All agonists strongly stimulated pERK in CHO-MOP cells (Fig. 3, A and B; Supplemental Table 1). This effect was blocked by naloxone and was absent in untransfected CHO cells (unpublished data). Bias factors between pERK and all the previous signaling pathways (inhibition of cAMP production, β-arr1 and β-arr2 recruitment) were calculated (Fig. 3, C and D; Supplemental Table 2). There were several ligands that displayed biased agonism toward cAMP inhibition over pERK compared with DAMGO, including morphine, loperamide, endo-1, α-neo, and dynorphin B. In line with this result, there was also a strong correlation between bias toward cAMP inhibition over pERK and the bias toward cAMP inhibition over β-arr2 recruitment (Pearson r = 0.893, P < 0.0001; Fig. 3E). However, β-arrestin–dependent pERK may not be specifically mediated via β-arr2 only, as a similar correlation was observed for bias between pERK and β-arr1 recruitment (Supplemental Fig. 3). The lack of bias between pERK and β-arr1 or 2 recruitment for nearly all of the ligands suggests that a component of pERK in CHO cells is β-arrestin–dependent. The only exception to this was morphine, which was biased toward β-arr2 and cAMP and away from pERK.
Bias toward β-arrestin recruitment determines pERK bias. (A and B) pERK in FlpIn CHO-MOP cells. Data are normalized to the 10% fetal bovine serum (FBS) response and expressed as means ± S.E.M. of at least three separate experiments. Bias factors between pERK and cAMP (C) and β-arr2 (D). *P ≤ 0.05; **P ≤ 0.005, as determined by two-tailed t test compared with DAMGO. (E) Two-tailed Pearson correlation was calculated between bias factors for cAMP–β-arr2 in Fig. 1E and cAMP-pERK excluding morphine. NC, not calculable.
Differential Activation of G Protein–Mediated Signaling.
Although we used the inhibition of AC as a measure of G protein–mediated signaling, AC activity is regulated by numerous other signaling effectors in addition to Gαi subunits. These include Ca2+, Gβγ subunits, and A-kinase anchoring protein. In addition to this, AC is differentially regulated by various Gα subunits (Willoughby and Cooper, 2007), which in turn are differentially modified by regulators of G protein signaling (Traynor, 2012). We measured direct G protein activation in FlpIn CHO-MOP membrane preparations using [35S]GTPγS binding assays and quantified bias between [35S]GTPγS binding and inhibition of forskolin-induced cAMP (Fig. 4). Several ligands showed bias toward cAMP over [35S]GTPγS binding in comparison with DAMGO, including morphine, Met-enkephalin (Met-enk), endo-1, endo-2, and α-neo. Interestingly, some of these, endo-1 and α-neo, consistently showed bias toward cAMP over all other signaling pathways, whereas Met-enk and endo-2 did not. This suggests that bias toward cAMP by these two sets of ligands may be driven by different effectors. Modulation of AC activity by βγ subunits and A-kinase anchoring protein did not contribute to bias between the cAMP and [35S]GTPγS binding, as inhibitors for these proteins had no effect on endo-1 or DAMGO inhibition of cAMP stimulation (Supplemental Fig. 4).
Bias between G protein activation and AC inhibition. (A and B) Agonist-induced [35S]GTPγS binding in CHO-MOP membrane preparations. Data are normalized to the 10 μM DAMGO response and expressed as means ± S.E.M. of three separate experiments. (C) Bias factors between cAMP and GTPγS assays. *P ≤ 0.05; **P ≤ 0.005, as determined by two-tailed t test compared with DAMGO. NC, not calculable; resp., response.
Receptor Localization/Trafficking.
The inability of morphine to induce MOP internalization was the first indication of differential actions of MOP agonists (Arden et al., 1995; Keith et al., 1996). It is now apparent that this compromised internalization is associated with different modes of desensitization for internalizing versus noninternalizing agonists. MOP trafficking was examined using BRET to measure the proximity between the Rluc-tagged MOP and Venus-tagged KRas construct, used as a plasma membrane marker. Before stimulation with agonists, MOP and KRas are in close proximity, allowing BRET to occur. Upon MOP activation, receptor redistribution, clustering, and internalization, the distance between these two proteins increases and results in a reduction in the BRET signal. A time-course experiment was initially performed to determine the time at which a maximal reduction in BRET was observed following agonist stimulation, which occurred at 60 minutes (unpublished data). All ligands except morphine induced a reduction in BRET between the MOP and KRas at 60 minutes (Fig. 5; Supplemental Table 1). Bias between receptor trafficking and the other signaling pathways for Met-enkephalin-Arg-Phe (Met-enk-RF) could not be quantified since a full concentration-response curve could not be obtained for this ligand in the KRas BRET assay. As Met-enk-RF gave similar responses to Met-enk in all the signaling pathways interrogated previously, this result suggests that Met-enk-RF is biased away from this pathway. Since receptor trafficking is dependent on recruitment of β-arrestins, bias factors were calculated between KRas BRET and β-arr1 and 2 recruitment. No ligands showed bias between KRas BRET and recruitment of β-arr1. Only endo-1 showed significant bias toward recruitment of β-arr2 and away from receptor trafficking. This result is in line with the fact that this ligand had previously shown significant bias toward β-arr2 over β-arr1 when these two pathways were compared (Fig. 2; Supplemental Table 2).
Endogenous opioids decrease the association of the MOP with KRas. (A and B) BRET between MOR-Rluc and KRas-Venus. Data are expressed as the percentage of vehicle, and as means ± S.E.M. of at least three separate experiments. Bias factors between internalization measured using KRas BRET assay and β-arr1 (C), β-arr2 (D), cAMP (E), and GTPγS (F). *P ≤ 0.05; **P ≤ 0.005, as determined by two-tailed t test compared with DAMGO. NC, not calculable.
We next compared the bias between KRas BRET and other signaling pathways. Endo-1 and α-neo, which previously showed bias toward cAMP inhibition over β-arr1 and 2 recruitment, now displayed significant bias toward cAMP over receptor trafficking. Surprisingly, endo-2, which previously showed no bias between cAMP and recruitment of β-arrestins, was now biased toward cAMP over KRas BRET. This shows that endo-2 is slightly biased toward β-arrestin recruitment over receptor trafficking, and this has significant consequences on downstream signaling. Bias between recruitment of β-arrestins and KRas BRET indicates that the β-arrestins recruited to endo-2–activated MOP receptors may be mediating activation of an alternative signaling pathway that has not been captured by the signaling pathways interrogated in this study. Interestingly, the differences in bias between cAMP and KRas BRET or [35S]GTPγS and KRas BRET track with the bias between cAMP and [35S]GTPγS binding. Therefore, these results highlight the fact that the bias between G protein signaling and receptor trafficking depends on the endpoint measured for G protein signaling.
Comprehensive Biased Agonism Profiles.
In the context of biased agonism, given the complex signaling pathways that lie downstream of the MOP, it is evident that the comparison of ligand action between two pathways gives only a limited picture of drug action. Thus, to allow visualization of the action of different ligands across all the pathways tested in this study, “webs of bias” were generated. These webs of bias allow the clustering of ligands into different activity profiles across the entire data set. For this, bias factors between cAMP and all other pathways were calculated using DAMGO as the reference ligand (ΔΔτ/KA) and represented in a single multiaxial graph. Morphine and loperamide are already known to possess different signaling properties compared with DAMGO; hence, they were included in this study as positive controls to validate our analytical tools. As expected, morphine and loperamide each showed characteristic fingerprints, different from that of DAMGO (Fig. 6B). Importantly, two additional opioids, endo-1 and α-neo, also displayed unique signaling profiles, and showed bias across multiple different signaling pathways when compared with DAMGO. Met-enk and endo-2 also showed bias compared with DAMGO, but were similar to one another. Met-enk-RF also displayed a unique signaling profile. Despite its bias profile being very similar to DAMGO at most signaling pathways, Met-enk-RF only stimulated a very small response in the KRas BRET assay, which indicates that Met-enk-RF is biased away from receptor trafficking; however, the degree of bias could not be quantified. All other endogenous ligands displayed activity profiles that were very similar to DAMGO (Fig. 6A).
Webs of bias of endogenous opioid peptides and reference ligands at the MOP. (A) Ligands with profiles similar to DAMGO. (B) Ligands with profiles that differ from that of DAMGO. τ/KA values were normalized to the reference ligand DAMGO and to the cAMP assay. Statistically significant differences (P ≤ 0.05) are denoted by black circles as determined by two-tailed t test. For the purposes of visualization only, a τ/KA for Met-enk-RF in the internalization assay was estimated using the incomplete KRas concentration-response curve. (C) PCA of all bias factors excluding morphine, DynA, DynA 1–13, and Met-enk-RF.
Another method to visualize and evaluate the overall signaling profiles is to perform PCA (see Materials and Methods). PCA identifies values in the data set that contribute the most variability, the principal components, which in this case are the bias factors that reveal the largest differences between ligands. Ligands that show similar biased agonism to one another will cluster together. PCA of all the bias factors (Fig. 6C; Supplemental Table 2) showed that most of the endogenous ligands cluster closely with DAMGO. In contrast, loperamide, α-neo and endo-1, and to a smaller extent Met-enk and endo-2 are separated, indicating that these ligands display an overall unique pattern of bias, consistent with their signaling profiles in the webs of bias. Of note, the first principal component (PC1) of the analysis, which accounts for the greatest source of variability between ligands, only accounts for 56% of the variability (Fig. 6C; Supplemental Table 3). PC2 contributes 26% to the variability between ligands. Indeed, PC1 and PC2 together only account for 82% of the variability. Interestingly, both of the first two principal components are composed mainly of bias factors between G protein– and β-arrestin–mediated signaling endpoints (Supplemental Table 4). This shows that bias between G protein activation and β-arrestin recruitment is a major determinant of the biased agonism characteristics of a ligand. However, as these bias factors are separated into two uncorrelated principal components, this also means that there is a diverse spectrum of bias between G protein– and β-arrestin–mediated signaling pathways. Altogether, this highlights the multidimensional nature of biased agonism and the fact that quantification of bias should be expanded beyond examination of only two pathways, such as G proteins versus β-arrestins, to cover the whole spectrum of possible signaling characteristics.
Discussion
The data presented here show that endogenous opioids acting on the MOP can exhibit diverse signaling profiles. Examining bias across multiple pathways has highlighted the complex nature of biased agonism at the MOP, and has revealed another level of complexity of bias that extends beyond differential activation of G proteins and β-arrestin recruitment. Several endogenous opioids, including α-neo, Met-enk, Met-enk-RF, and the putative endogenous opioids endo-1 and endo-2, displayed biased agonism compared with DAMGO across multiple signaling pathways, whereas the rest of the endogenous ligands displayed profiles that were similar to that of DAMGO. In particular, endo-1 and α-neo displayed markedly different signaling profiles. α-Neo is considered a κ-opioid receptor agonist, and as such, no studies have examined its actions at the MOP. The physiologic actions and signaling endo-1 at the MOP, conversely, have been extensively studied (Fichna et al., 2007). Endo-1 produces physiologic effects similar to most opioids, such as analgesia, inhibition of gut motility, respiratory depression, and the development of tolerance (Goldberg et al., 1998; Tonini et al., 1998; Lonergan et al., 2003; Yu et al., 2007). Differences between the physiologic effects of endo-1 and endo-2 have been described. However, these differences are usually small, and it cannot be ruled out that they are due to different degradation rates of the peptides rather than biased agonism (Sakurada et al., 2002; Terashvili et al., 2007). Interestingly, endo-1 has been shown to produce antinociceptive cross-tolerance to morphine whereas endo-2 does not, suggesting potential differences in the mechanisms of tolerance produced by these two ligands. As both endo-1 and endo-2 stimulate receptor internalization (McConalogue et al., 1999), they would both be expected to produce tolerance via a similar mechanism to that of DAMGO. This suggests that tolerance produced by endo-1 and endo-2 may involve differential activation of signaling pathways unrelated to receptor internalization, or alternatively, that the mechanisms of development of tolerance to these ligands are cell type–dependent.
Few other studies have examined biased agonism of endogenous opioids, and the results from these different studies are not consistent. Morse et al. (2011) examined a range of endogenous and exogenous opioids in HEK293 cells using dynamic mass redistribution to measure MOP global responses. In that study, all endogenous opioids, except dynA 1–13, exhibited similar profiles. Our analysis suggests that, although dynA 1–13 displayed some bias compared with the other endogenous ligands, this is limited evidence to suggest that dynA 1–13 is particularly unique compared with the other endogenous opioids. Rivero et al. (2012) quantified bias between [35S]GTPγS binding and β-arr2 recruitment in HEK293 cells for some endogenous ligands, and found that endo-1 and endo-2 were biased toward β-arr2 recruitment compared with leu-enkephalin (McPherson et al., 2010; Rivero et al., 2012; Thompson et al., 2015). In our study, only endo-1 was biased toward β-arr2 over [35S]GTPγS binding (Supplemental Fig. 5). Various reasons can explain the different results. First, the different methodologies and time points used to measure β-arr2 recruitment may affect bias, since at later time points differential rates and modes of desensitization may affect the level of β-arr2 recruitment. In addition, the different cell backgrounds used in each study may also change the bias of a ligand, as the expression levels of the various proteins involved in the activation of the signaling pathways examined may differ across cell lines and have an impact in the measures of bias (Thompson et al., 2015).
Morphine is often described as a ligand with compromised ability to recruit β-arrestins (Whistler and von Zastrow, 1998; Bohn et al., 2004; Molinari et al., 2010; Chen et al., 2013). As a partial agonist, morphine is expected to give a lower response in signaling pathways with low coupling efficiency and/or assays with low sensitivity. However, understanding whether these observations refer to a biased action requires systematic quantification of biased agonism. In our study, morphine did not show significant bias between β-arr2 and cAMP, although it was significantly biased toward β-arr2 when compared with [35S]GTPγS binding. This is similar to results obtained by Rivero et al. (2012), where morphine was slightly biased toward β-arr2, although not significantly (Thompson et al., 2015).
We observed a correlation between pERK and β-arrestin recruitment, suggesting that β-arrestin recruitment is involved in pERK in CHO cells. This correlation was seen with both β-arr1 and β-arr2, indicating that both β-arrestins are capable of mediating a component of ERK activation. β-Arrestin–dependent pERK has also been shown previously in transgenic cell lines and primary cells (Macey et al., 2006; Miyatake et al., 2009). The one exception to this correlation was morphine. This indicates that morphine stimulates pERK via a different pathway that is independent of β-arrestins. A likely candidate is PKC, as morphine-stimulated pERK in HEK293 cells has been suggested to be PKC-dependent (Zheng et al., 2008). However, to date this has not been demonstrated in CHO cells.
In our study, endo-1 preferentially recruited β-arr2 over β-arr1. Differential recruitment of β-arrestins to the MOP by endo-1 has not been shown previously, but has been demonstrated for morphine, which promotes recruitment of β-arr2 and little or no recruitment of β-arr1 (Bohn et al., 2004; Groer et al., 2007). However, as mentioned previously, as all ligands gave a lower response in the β-arr1 assay, a partial agonist is expected to have a very low to undetectable signal. Biased activation of β-arrestins may result in differential activation of downstream signaling, as different β-arrestins can act as scaffolds for different signaling complexes (DeWire et al., 2007; Nobles et al., 2011). β-Arr1, but not β-arr2, is required for MOP ubiquitination, and has also been shown to promote MOP dephosphorylation more rapidly than β-arr2 (Groer et al., 2011). In addition, the downstream functions of β-arrestins may vary with different ligands. Differential engagement of G protein receptor–coupled kinases and different patterns of ligand-induced MOP phosphorylation have been demonstrated previously (Schulz et al., 2004; Doll et al., 2011; Lau et al., 2011; Chen et al., 2013). These receptor phosphorylation patterns play a vital role in determining receptor interactions with β-arrestins and may direct β-arrestin functions (Zidar et al., 2009). The differential role of β-arr2 recruitment in response to morphine compared with other opioids has been shown in the β-arr2 knockout mice, where the improved pharmacological profile is only observed with morphine (Bohn et al., 1999; Raehal et al., 2005). Although morphine-induced β-arrestin recruitment results in activation of different signaling pathways, whether the differential recruitment of β-arrestins induced by endo-1 mediates biased activation of signaling pathways downstream of β-arrestin is still unknown.
Agonists were assessed for bias between G protein activation and inhibition of AC, a classic downstream G protein–mediated effect. Several ligands showed bias toward inhibition of AC over [35S]GTPγS binding in comparison with DAMGO, indicating differential activation of G protein–mediated signaling pathways. There are not many examples where [35S]GTPγS assays have been performed in conjunction with cAMP assays in the same cellular background to assess MOP function (Zaki et al., 2000; Vigano et al., 2003). Zaki et al. (2000) reported differing relative potencies of some agonists between the two assays, but concluded that this was due to the different coupling efficiencies. There is also some evidence of ligand-dependent engagement of different Gα subunits with MOP (Sanchez-Blazquez et al., 2001; Massotte et al., 2002; Clark et al., 2006; Saidak et al., 2006). However, these differences are small, and since most methods used in these studies require overexpression or addition of purified G proteins, the real preferences for different subunits may be masked by differences in stoichiometry between receptor and G proteins. It is also possible that the bias observed between these signaling effectors is due to differences in signaling kinetics. As measurements for [35S]GTPγS binding assays are usually taken after longer incubations with the ligand than for cAMP assays, ligands with slower association kinetics or that stimulate different rates of desensitization can display bias between two signaling endpoints. Further experiments are required to elucidate whether bias between [35S]GTPγS binding and AC inhibition or other signaling pathways is due to differential regulation of G protein–mediated signaling pathways, different signaling kinetics, or different ligand association/dissociation rates.
The bias between cAMP inhibition and [35S]GTPγS also correlates with the bias observed between MOP trafficking and these signaling pathways. Several ligands showed bias between trafficking and cAMP, but not between trafficking and [35S]GTPγS. Morphine and possibly Met-enk-RF may also be biased toward cAMP inhibition and [35S]GTPγS binding over trafficking. However, as bias factors could not be quantified for morphine and Met-enk-RF between trafficking and other signaling pathways, we cannot exclude the possibility that these ligands simply have lower efficacy, in which case the lower response in a low-sensitivity assay, such as trafficking, is to be expected. There were no ligands showing significant bias between trafficking and recruitment of β-arr1, and only endo-1 was biased toward β-arr2 over receptor trafficking. This result is in line with the fact that receptor trafficking is highly dependent on recruitment of β-arrestins.
In summary, we have performed a systematic quantitative analysis of biased agonism at the MOP by endogenous and putatively endogenous opioid peptides across multiple different signaling pathways. This work has revealed that opioid peptides display a variety of different biased agonism profiles, some of which are unique. α-Neo and endo-1 display particularly distinct biased agonism profiles, and may have different signaling properties and physiologic effects in vivo compared with other endogenous opioids. Biased agonism profiles, combined with different degradation rates, expression patterns in the body, and differing selectivities for opioid receptor subtypes, may engender tremendous diversity in endogenous opioid activity and lead to finely tuned physiologic processes. Although the impact of MOP-biased agonism in the control of normal physiologic processes still remains to be explored, our findings provide a pharmacological framework to progress our understanding of ligand redundancy of the opioid system. A greater understanding of how endogenous opioids control physiologic processes through biased agonism will reveal vital information required to enable the design of biased opioids with improved pharmacological profiles.
Authorship Contributions
Participated in research design: Thompson, Lane, Christopoulos, Canals.
Conducted experiments: Thompson.
Performed data analysis: Thompson, Coudrat.
Wrote or contributed to the writing of the manuscript: Thompson, Lane, Sexton, Christopoulos, Canals.
Footnotes
- Received March 4, 2015.
- Accepted May 26, 2015.
M.C. is a Monash Fellow. A.C. and P.M.S. are Principal Research Fellows of the National Health and Medical Research Council (NHMRC) of Australia. J.R.L. is a Monash University Larkins Fellow and an RD Wright Biomedical Career Development Fellow (NHMRC). This work was funded in part by the NHMRC [Program Grant APP1055134 to A.C. and P.M.S.] and the Monash Institute of Pharmaceutical Sciences Large Grant Support Scheme (M.C.). G.L.T. scholarship was funded by the Australian Commonwealth Government Department of Defence.
↵
This article has supplemental material available at molpharm.aspetjournals.org.
Abbreviations
- AC
- adenylyl cyclase
- BRET
- bioluminescence resonance energy transfer
- BSA
- bovine serum albumin
- CHO
- Chinese hamster ovary
- DAMGO
- d-Ala2-N-MePhe4-Gly-ol enkephalin
- DMEM
- Dulbecco's modified Eagle's medium
- dynA
- dynorphin A
- endo-1
- endomorphin-1
- endo-2
- endomorphin-2
- GPCR
- G protein–coupled receptor
- GTPγS
- guanosine 5′O-(γ-thio)triphosphate
- Met-enk
- Met-enkephalin
- MOP
- µ-opioid receptor
- α-neo
- α-neoendorphin
- PC
- principal component
- PCA
- principal component analysis
- pERK
- ERK1/2 phosphorylation
- PKC
- protein kinase C
- Rluc
- Renilla Luciferase
- YFP
- yellow fluorescent protein
- Copyright © 2015 by The American Society for Pharmacology and Experimental Therapeutics
References
In this issue
Endogenous Biased Agonism at the μ-Opioid Receptor
