Gα_s signalling of the CB₁ receptor and the influence of receptor number

Abstract

Background and Purpose

CB₁ receptor signalling is canonically mediated through inhibitory Gα_i proteins, but occurs through other G proteins under some circumstances, Gα_s being the most characterized secondary pathway. Determinants of this signalling switch identified to date include Gα_i blockade, CB₁/D₂ receptor co‐stimulation, CB₁ agonist class and cell background. Hence, we examined the effects of receptor number and different ligands on CB₁ receptor signalling.

Experimental Approach

CB₁ receptors were expressed in HEK cells at different levels, and signalling characterized for cAMP by real‐time BRET biosensor –CAMYEL – and for phospho‐ERK by AlphaScreen. Homogenate and whole cell radioligand binding assays were performed to characterize AM6544, a novel irreversible CB₁ receptor antagonist.

Key Results

In HEK cells expressing high levels of CB₁ receptors, agonist treatment stimulated cAMP, a response not known to be mediated by receptor number. Δ⁹‐THC and BAY59‐3074 increased cAMP only in high‐expressing cells pretreated with pertussis toxin, and agonists demonstrated more diverse signalling profiles in the stimulatory pathway than the canonical inhibitory pathway. Pharmacological CB₁ receptor knockdown and Gα_i1 supplementation restored canonical Gα_i signalling to high‐expressing cells. Constitutive signalling in both low‐ and high‐expressing cells was Gα_i‐mediated.

Conclusion and Implications

CB₁ receptor coupling to opposing G proteins is determined by both receptor and G protein expression levels, which underpins a mechanism for non‐canonical signalling in a fashion consistent with Gα_s signalling. CB₁ receptors mediate opposite consequences in endpoints such as tumour viability depending on expression levels; our results may help to explain such effects at the level of G protein coupling.

Abbreviations

2‐AG

2‐arachidonoyl glycerol

AC

adenylate cyclase

AEA

anandamide/N‐arachidonoylethanolamine

AM6544

1‐(2,4‐dichlorophenyl)‐5‐(4‐(4‐isothiocyanatobut‐1‐yn‐1‐yl)phenyl)‐4‐methyl‐N‐(piperidin‐1‐yl)‐1H‐pyrazole‐3‐carboxamide

BAY

BAY59,3074

CAMYEL

cAMP sensor YFP‐Epac‐Rluc

HA

haemagglutinin

PEI

polyethylenimine

PTX

pertussis toxin

THC

Δ⁹‐tetrahydrocannabinol

Introduction

Despite longstanding consensus as to its canonical Gα_i‐mediated signalling pathway, the type one cannabinoid receptor (CB₁) is among several GPCRs now understood to transduce signals through more than one G protein effector – a capacity sometimes referred to as ‘promiscuous G protein coupling’ (reviewed in Maudsley et al., 2005). Environmental and pharmacological conditions that have been found to affect CB₁ receptor signalling in this way include blockade of the canonical Gα_i pathway with pertussis toxin (PTX) (Glass and Felder, 1997; Bonhaus et al., 1998; Scotter et al., 2010), the class of CB₁ agonist (Bonhaus et al., 1998; Lauckner et al., 2005), receptor oligomers (Kearn et al., 2005; Bagher et al., 2016) and cell background (McIntosh et al., 2007).

The CB₁ receptor was first seen to demonstrate Gα_s‐like signalling behaviour under conditions of co‐stimulation with the type two dopamine receptor (D₂) in primary rat striatal neurons: on stimulation with quinpirole (a specific D₂ receptor agonist) or HU210 (a CB₁ receptor agonist) alone, these neurons demonstrated the expected inhibition of forskolin‐induced cAMP, but when HU210 was applied in addition to quinpirole, this inhibitory signal was reversed in a concentration‐dependent manner (Glass and Felder, 1997). In both primary cells and transfected cell lines stably expressing CB₁ receptors, PTX pretreatment has also been found to unmask this Gα_s‐like signalling phenotype, in the absence of D₂ receptor co‐expression/co‐stimulation (Glass and Felder, 1997; Bonhaus et al., 1998; Scotter et al., 2010). These observations gave rise to the hypothesis that Gα_s‐like CB₁ receptor signalling arose in circumstances when the availability of Gα_i was limited (‘Gα_i exhaustion’), either because Gα_i was directly inactivated by PTX or because it was sequestered by other Gα_i‐coupled receptors (of which the D₂ receptor is an example). However, Jarrahian et al. (2004) have reported that the mere co‐expression of CB₁ and D₂ receptors is sufficient for a CB₁ receptor‐mediated Gα_s‐like cAMP phenotype to be unmasked, a finding which appears to require either G protein pre‐coupling (Ferre, 2015) or high levels of D₂ receptor‐mediated constitutive signalling for the Gα_i exhaustion hypothesis to hold. Subsequently, co‐immunoprecipitation data led Kearn et al. (2005) to propose that the CB₁ receptor signalling switch is derived pleiotropically, from altered protein conformations caused by a physical interaction between D₂ and CB₁ receptors – a receptor heterodimer. Characterization of the CB₁/D₂ receptor heterodimer has continued using other approaches (Marcellino et al., 2008; Przybyla and Watts, 2010; Bagher et al., 2016). It remains noteworthy that the two hypotheses of the signalling switch mechanism need not be mutually exclusive, and both promiscuous (low‐efficacy activation of a non‐preferred G protein species due to lack of availability of the preferred species; i.e. Gα_i exhaustion) and pleiotropic (stoichiometric/conformational) mechanisms have precedent in the literature (Zhu et al., 1994; Laugwitz et al., 1996; Cordeaux et al., 2000; Jin et al., 2001).

The exact identity of the G protein subtype that mediates the novel effects was not originally shown directly (Glass and Felder, 1997), although the stimulatory cAMP effect and absence of PTX sensitivity suggested CB₁ receptor coupling to Gα_s. Abadji et al. (1999) demonstrated that the CB₁ receptor contains a conserved two amino acid motif that mediates Gα_s coupling of the β‐adrenoceptor, but Bash et al. (2003) provided the first direct evidence of some Gα_s protein involvement by demonstrating that anti‐Gα_s antibodies could knock down CB₁ receptor‐mediated Ca²⁺ flux. Very recently, Eldeeb et al. (2016) showed specific, direct Gα_s protein involvement in CB₁ receptor signalling using a scintillation proximity variant of a [³⁵S]‐GTPγS assay, in a study that underscores the relevance of the CB₁ receptor‐mediated Gα_s signalling pathway by demonstrating its presence under experimental conditions where agonism does not induce a nett stimulatory cAMP signal.

Recently, we have observed Gα_s‐like signalling of CB₁ receptor in assays where the receptor is expressed transiently. This expression method frequently results in very high transgene expression primarily due to cellular uptake of multiple copies of the transfected construct. In stably expressing cell lines, high levels of transgene expression can be produced by molecular engineering of an N‐terminal signal sequence tag such as the preprolactin signal sequence (pplss). This is a short (30 amino acid) peptide, which substantially increases the efficiency of nascent protein secretion by acting as a recognition sequence for the signal recognition particle (SRP) (Kurzchalia et al., 1986), but is cleaved during maturation and therefore does not alter protein function (Belin et al., 1996). Based on our observations, we hypothesized that CB₁ receptor number is a novel determinant of the Gα_i‐Gα_s signalling switch and have therefore systematically characterized the relationship between receptor expression and CB₁ receptor signalling. A panel of six chemically distinct CB₁ agonists was compiled to allow delineation of ligand‐dependence of the two signalling pathways.

Methods

Molecular biology

The pEF4A (Thermo Fisher Scientific, Waltham, MA, USA) plasmid encoding human CB₁ chimerized with three N‐terminal haemagglutinin (HA) tags (3HA‐hCB₁) has been described previously (Cawston et al., 2013). The coding region for bovine pplss was amplified by PCR from a plasmid kindly gifted by Professor Ken Mackie (Indiana University, Bloomington, IN, USA) with primers designed to incorporate KpnI restriction sites at either end of the amplicon. The pplss sequence was then incorporated into the 3HA‐hCB₁ pEF4A plasmid by non‐directional KpnI restriction cloning, and sequence verified.

Cell culture and transfection

Cell culture media and reagents were purchased from Thermo Fisher Scientific, and plasticware was purchased from Corning (Corning, NY, USA). HEK293 cell lines were cultured in DMEM supplemented with 10% FBS and an appropriate selection of antibiotics and were cultured in 5% CO₂, at 37°C in a humidified incubator.

The HEK cell line stably expressing 3HA‐hCB₁ receptors has been described previously (Cawston et al., 2013). This cell line was used as the background to generate a HEK cell line that co‐expresses CB₁ and D₂ receptors, and is first described in Hunter et al. (2016). The HEK cell line stably expressing the human 3HA‐hCB₂ receptor has also been described previously (Grimsey et al., 2011). The pplss‐3HA‐hCB₁ pEF4A construct (described above) was linearized and stably transfected into the same wild‐type HEK cell background using Lipofectamine 2000. A clonal population of cells expressing receptors at qualitatively high levels was then isolated and propagated for subsequent experiments. Wild‐type HEK Flp‐In™‐293 cells (Thermo Fisher Scientific) were also stably transfected with a modified pcDNA3L‐His‐cAMP sensor Venus‐Epac‐Rluc8 (V8‐CAMYEL) plasmid (Hunter et al., 2017) (as distinct from the original CAMYEL biosensor which was used in assays described below, using the same random incorporation method of stable transfection (i.e. the Flp Recombination Target site (FRT) was left unoccupied). This cell line was used for assays involving transient receptor expression and FACS.

Assays for cAMP where CB₁ receptors were expressed stably

Cellular cAMP was measured using a commercially available kinetic BRET assay (CAMYEL), as previously described (Jiang et al., 2007; Cawston et al., 2013). In brief, sufficient cells were seeded in 10 cm dishes such that they would be approximately 60% confluent the next day. The day after seeding, culture medium was replaced and 5 μg of an expression construct encoding the CAMYEL biosensor (pcDNA3L‐His‐CAMYEL) was transfected using linear polyethylenimine (PEI, MW 25 kDa; Polysciences, Warrington, PA, USA), where the transfection ratio (DNA : PEI) was 1:6. Approximately 24 h after transfection, transfected cells were lifted from their culture dishes by trypsinization and plated in poly‐D‐lysine (Sigma Aldrich) coated, white 96‐well CulturPlates™ (Perkin Elmer, Waltham, MA, USA) at a density of 8 × 10⁴ cells per well. For experiments involving PTX pretreatment to irreversibly inactivate Gα_i proteins, cells were plated in half the usual volume of culture medium, and then 6 h later, 2× concentrated PTX (Sigma Aldrich) or PTX vehicle (50% glycerol, 50 mM Tris–HCl pH 7.5, 10 mM glycine, 500 mM NaCl) was prepared in culture medium and added on top of the plating medium to give a final concentration of 100 ng·mL⁻¹ PTX. The cells were then cultured overnight (>16 h if treated with PTX) prior to beginning detection. Cells were washed once with HBSS and then equilibrated for 30 min in HBSS supplemented with 1 mg·mL⁻¹ BSA (ICPBio, Auckland, NZ, USA). Coelenterazine H (NanoLight Technologies, Pinetop, AZ, USA) was added to a final concentration of 5 μM for 5 min prior to dispensing drugs (2.5 μM forskolin and agonist, etc.) and initiating luminescence reading at 460 nm (Rluc) and 535 nm (YFP) in either a VICTOR™ X Light luminometer (Perkin Elmer), or a LUMIstar® Omega luminometer (BMG Labtech, Ortenberg, Germany). Each series of stimulations was detected in real‐time for approximately 20 min. All drugs (including coelenterazine H) were prepared in HBSS +1 mg·mL⁻¹ BSA, and all incubations and stimulations were performed at 37°C.

CAMYEL assays involving pharmacological CB₁ receptor knockdown with AM6544 were performed as noted above, except that an AM6544 serial dilution series was prepared in serum‐free medium supplemented with 1 mg·mL⁻¹ BSA and applied to cells for 1 h. The AM6544 dilutions were then removed and wells washed twice with HBSS, before the commencement of the 30 min equilibration in HBSS +1 mg·mL⁻¹ BSA as described above.

For CAMYEL assays involving transient transfection of GNAI1 pcDNA3.1+ (or empty pcDNA3.1+) DNA, the cell lines of interest were seeded into poly‐D‐lysine coated, white CulturPlates™. Approximately 24 h later, an in‐well Lipofectamine 2000 protocol was used to transiently transfect either GNAI1 in pcDNA3.1+ (cDNA Resource Center, Bloomsburg, PA, USA) or empty pcDNA3.1+ (Thermo Fisher Scientific), followed by a second Lipofectamine 2000 in‐well transfection of CAMYEL biosensor DNA approximately 6–8 h later. Lipofectamine reagent was selected for these assays because it results in higher transfection efficiency than linear PEI. Cells were cultured thus for 48 h post‐transfection, prior to stimulation and detection in accordance with the protocol described above.

Cell staining and FACS for cAMP assays where receptor was expressed transiently

The aforementioned HEK Flp‐In™ cell line stably expressing the V8‐CAMYEL biosensor was plated in a 10 cm dish at a density such that the cells would be approximately 60% confluent the next day. The day after plating, transient transfection of the pplss‐3HA‐hCB₁ pEF4A construct was performed using Lipofectamine 2000, in order to produce a population of V8‐CAMYEL biosensor‐expressing cells, which also expressed CB₁ receptors over a wide range of expression levels. Following transfection, cells were cultured for 2 days, then detached with Versene and resuspended in culture medium. The cells were sedimented and washed in culture medium, then labelled with mouse anti‐HA clone 16B12 monoclonal antibody (1:500; Covance, Princeton, NJ, USA) and secondary Alexa Fluor® 647 goat anti‐mouse antibody (1:300; Thermo Fisher Scientific). Following the secondary antibody incubation, the cells were washed and resuspended to a concentration of approximately 2.5 × 10⁶ mL⁻¹ in HBSS supplemented with 1 mg·mL⁻¹ BSA. FACS was performed using a FACSVantage Cell Sorter (Becton Dickinson, Franklin Lakes, NJ, USA). A ‘no primary antibody’ control was used to define the 647 background (low signal detection limit), and gating was used such that ‘above background’ 647 staining intensity was assessed during sorting. The transiently expressed population of cells was sorted into six sub‐populations on the basis of receptor expression (647 staining intensity). The sorted cells were then sedimented and resuspended in HBSS +1 mg·mL⁻¹ BSA, and counted. A white 96‐well CulturPlate™ was prepared with diluted coelenterazine H (see CAMYEL assay details above) and drugs (forskolin and CP55,940) such that 1 × 10⁴ cells could be subsequently dispensed into each well to give the desired drug concentrations, and detection in a VICTOR™ X Light luminometer was initiated immediately, as described above for other CAMYEL assays. The FACS and subsequent cAMP assay was repeated in three independent experiments.

Cell membrane preparation and quantification

HEK cells were harvested when semi‐confluent in 175cm² flasks, with ice‐cold 5 mM EDTA in PBS. Cells were then sedimented, and cell pellets were snap frozen at −80°C. Pellets were thawed on ice and resuspended in sucrose buffer (200 mM sucrose, 50 mM Tris–HCl pH 7.4, 5 mM MgCl₂, 2.5 mM EDTA), and then thoroughly homogenized by plunging (40 strokes) with a tight‐fitting pestle in a glass dounce homogenizer. The homogenate suspension was sedimented by centrifugation at 1000 × g for 10 min (‘P1 pellet’), and the supernatant subsequently sedimented by centrifugation at 26 916 × g for 30 min (‘P2 pellet’). Membrane pellets were resuspended in sucrose buffer, aliquoted and stored at −80°C. Multiple membrane preparations were produced for each cell line, in order that biological (independent) replicates of binding assays reflected cell line variability relevant to functional assays, rather than just experimental error.

Membrane protein concentrations were quantified by Amido Black 10B protein assay. In brief, 1 μL of membrane preparation was spiked into 1 M Tris–HCl pH 7.5 (contained 2% w^.v^‐1 SDS), and protein was precipitated by addition of concentrated trichloroacetic acid. The precipitated protein samples, alongside standards of 0–20 μg BSA, were then spotted onto 0.45 μm HA filters of a 96‐well MultiScreen‐HA Filter Plate (Merck Millipore, Billerica, MA, USA) and stained with 0.1% w^.v^‐1 amido black 10B dye (Sigma Aldrich) in a 45:10:4 solution of methanol : glacial acetic acid : milliQ water. The stain was washed off with milliQ water, then destained twice with a 90:2:8 solution of methanol : glacial acetic acid : milliQ water and washed again with milliQ. The stain was eluted with a solution of 25 mM NaOH, 0.05 mM EDTA and 50% v^.v^‐1 absolute ethanol, and absorbance measurements taken for BSA standards and unknowns at 630 nm on an EnSpire® plate reader (Perkin Elmer). To increase confidence in interpolated sample protein concentrations, the protein assay was performed at least twice on each membrane preparation.

Radioligand binding assays

Competition binding assays

Homogenate competition/displacement binding assays were performed on ‘P2’ HEK cell membrane preparations (see above). In brief, concentration series of non‐tritiated drugs, dilutions of [³H]‐CP55,940 and [³H]‐ SR141716A (rimonabant) (Perkin Elmer) and P2 membrane dilutions were prepared separately in binding buffer (50 mM HEPES pH 7.4, 1 mM MgCl₂, 1 mM CaCl₂, 2 mg·mL⁻¹ BSA) and dispensed into 96‐well, polypropylene V‐shaped plates (Hangzhou Gene Era Biotech Co Ltd, Zhejiang, China), with the membrane dilution dispensed last. Final reaction volumes were 200 μL per well. Plates were sealed and incubated for 1 h at 30°C. During the incubation, the 1.2 μm pore fibreglass filters of a 96‐well Harvest Plate (Perkin Elmer) were blocked with a solution of 0.1% w^.v^‐1 branched PEI (Sigma Aldrich) in water. At the conclusion of the incubation, the PEI solution was washed through the filters using a filter plate vacuum manifold (Pall Corporation, Port Washington, NY, USA), and all filters were washed with 200 μL ice cold wash buffer (50 mM HEPES pH 7.4, 500 mM NaCl, 1 mg·mL⁻¹ BSA). The 200 μL binding assay mixture was then transferred onto the filter plate (under vacuum); V‐well plate wells were immediately washed once with wash buffer, and then unbound ³H‐ligand was washed through the filter by washing each well three more times with wash buffer. Harvest plate filters were then dried overnight, and the underside of the plate was sealed. Irgasafe Plus scintillation fluid (Perkin Elmer) was then dispensed to each well, and after a 30 min delay, scintillation was read for 2 min per well in a Wallac MicroBeta® TriLux Liquid Scintillation Counter (Perkin Elmer), with counts corrected for detector efficiency.

Saturation binding assays to characterize AM6544

Saturation binding assays were performed in the Makriyannis laboratory to characterize the novel irreversible antagonist of CB₁ receptors, AM6544 (Figure 4A) (Patent US8084451, 2011). Irreversible labelling assays were performed as described previously (Janero et al., 2015; Hua et al., 2016). HEK293F cell P2 membranes containing the hCB₁ receptor were purified as described previously (Xu et al., 2005) and were resuspended at a concentration of 1 mg·mL⁻¹ in TME buffer (25 mM Tris Base, 5 mM MgCl₂, 1 mM EDTA, pH 7.45) containing 1 mg·mL⁻¹ w^.v^‐1 BSA and incubated with 10× K_i (~100 nM) of AM6544 or SR141716A (or DMSO vehicle) at 30°C for 1 h with gentle agitation. The reaction was terminated by centrifugation at 27 000 × g for 10 min, followed by removal of excess unbound ligand by washing. Pellets were resuspended in TME buffer containing 1 mg·mL⁻¹ BSA and incubated at 30°C for 30 min with gentle agitation. This centrifugation and resuspension was repeated. After a final centrifugation, membrane pellets were resuspended in TME containing 1 mg·mL⁻¹ BSA, and 25 μg of protein were added to each assay well of a 96‐well plate. [³H]‐CP55,940 was diluted in TME/BSA to yield ligand concentrations ranging from 1.58 to 50 nM, and [³H]‐SR141716A was diluted in TME/BSA to yield ligand concentrations ranging from 3.77 to 119 nM. Nonspecific binding was assayed in the presence of 5 μM unlabelled CP55,940 or SR141716A. Membranes were incubated at 30°C for 1 h with gentle agitation. Samples were then transferred to Unifilter GF/B filter plates and filtered using a Packard Filtermate‐96 Cell Harvester (Perkin Elmer, Shelton, CT, USA). Filter plates were washed four times with ice‐cold wash buffer (50 mM Tris base, 5 mM MgCl₂ containing 5 mg·mL⁻¹ BSA, pH 7.4). Bound radioactivity was quantified in a Packard Top Count Scintillation Counter. Nonspecific binding was subtracted from the total bound radioactivity to calculate specific binding of [³H]‐CP55,940 and [³H]‐SR141716A.

Figure 4.

Chemical structure of AM6544 (A), and non‐specific binding‐corrected [³H]‐SR141716A (B) and [³H]‐CP55,940 (C) saturation binding curves on hCB₁ HEK293F membranes pretreated for 1 h with vehicle (‘V’, DMSO), SR141716A (100 nM) or AM6544 (100 nM) and then washed three times before the incubation with the radioligand. Representative data are presented, demonstrating mean ± SEM of technical duplicates.

Whole cell binding assays

For whole cell binding assays, pplss‐3HA‐hCB₁ HEK cells were plated in poly‐D‐lysine coated 24‐well plates and were cultured overnight. The next day, cell confluency was approximately 80%. For live cell assays to determine AM6544 washout (‘live cell affinity/irreversibility’), a concentration series of AM6544 was prepared in serum‐free medium supplemented with 1 mg·mL⁻¹ BSA and incubated with the cells under conditions described above for the CAMYEL assays (volume adjusted for same well surface area : volume ratio, and 37°C for 1 h). At the end of the AM6544 pretreatment, wells were washed twice with warm HBSS, and the binding assay was initiated. [³H]‐SR141716A was diluted in serum‐free medium supplemented with 1 mg·mL⁻¹ BSA; dilution series of non‐tritiated drugs were prepared in this mixture where applicable and dispensed into the 24‐well plate. Cells were incubated with radioligand for 1 h at 30°C. At the end of the incubation, the plate was moved to ice, well contents were removed with a vacuum and wells were gently washed twice with ice‐cold PBS. Cells were lysed with 250 μL per well 0.1 M NaOH in milliQ water. Lysates (200 μL per well) were moved to 4 mL scintillation tubes and diluted in excess volumes of scintillation fluid. Scintillation was read in the TriLux Counter (as above) following a 6 h delay, to allow time for the aqueous lysate to fully disperse in the scintillation fluid.

Phospho‐ERK assays

Characterization of the activation by phosphorylation of ERK (pERK) was performed using the commercially available AlphaScreen SureFire kit (Perkin Elmer). In brief, cells were plated in 96‐well culture plates and then after 24 h of culture were serum starved for >16 h (in serum free medium supplemented with 1 mg·mL⁻¹ BSA) prior to stimulation, in the presence or absence of PTX (100 ng·mL⁻¹). Drugs were prepared in serum‐free medium supplemented with 1 mg·mL⁻¹ BSA, and then pERK time courses were performed with plates resting on a barely submerged stage in a 37°C waterbath. At the conclusion of the time course, plates were placed on ice, well contents were removed by aspiration, and cells were immediately lysed in 30 μL of lysis buffer. Subsequently, AlphaScreen detection was performed according to manufacturer instructions, and plates were read in a CLARIOstar® plate reader (BMG Labtech, Ortenberg, Germany).

Data and statistical analysis

BRET cAMP time course data were analysed using AUC analysis, as described in Finlay et al. (2016). Data were normalized to basal (0%) and forskolin alone (100%) in order to enable a combination of independent experiments (accounting for small differences in forskolin‐induced cAMP levels between replicates). Concentration–response parameters were obtained by fitting three‐parameter (Hill coefficient constrained to 1) nonlinear regression curves. Where concentration–response curves could not be fitted, connecting lines were drawn between data points to help visualize trends, including for U‐shaped responses. Plots show representative data (mean ± SEM) of technical replicates, unless otherwise specified. All statistical analyses (see below) were performed using data from biological (independent) replicates. Operational analysis for ligand bias (van der Westhuizen et al., 2014) and all plots and curve‐fits were obtained using GraphPad Prism v6 (GraphPad Software Inc., La Jolla, CA, USA).

Homologous binding data were modelled utilizing the predefined equations in GraphPad Prism, producing log K_D and B_MAX values. The resulting B_MAX values (in disintegrations per min) were converted to milli‐Curies per sample point. The specific activity of [³H]‐CP55,940 (175 Ci·mmol⁻¹) was then used to convert bound radioligand into mmol radioligand bound, and this was normalized for protein amount.

The data and statistical analysis comply with the recommendations on experimental design and analysis in pharmacology (Curtis et al., 2015). Power analysis for cAMP assays was performed using data from pilot experiments, for both EC₅₀ (minimum effect difference of 0.7 log M) and E_MAX (minimum effect difference of 20% of the matched forskolin response). The values for α and power were 0.05 and 0.80 respectively. All final datasets were tested for compliance with parametric test assumptions of normality (Shapiro–Wilk) and equality of variance (Levene Median). If the tested population failed to meet either assumption, data were transformed and re‐tested. Where statistical tests were utilized (one‐ or two‐way ANOVA, as appropriate, unless otherwise specified), data were obtained from at least five biological replicates of each experiment. Where statistically significant differences (P < 0.05) were detected across the entire group in the test, the Holm–Šídák post hoc test was used. All statistical analysis was performed using SigmaPlot™ v12.5 (Systat Software Inc., San Jose, CA, USA). Data presented in figures are either ‘representative’ (illustrative data from a single experiment, consistent with data from independent replicates) or ‘combined’ (summary data, where all replicates' results were averaged) of five independent experiments unless otherwise indicated. This approach was necessary in order to preserve the statistical meaningfulness of parameters from concentration–response curves, as explained in detail by Hall and Langmead (2010).

Drugs

Forskolin, CP55,940, WIN55,212‐2 and BAY59‐3074 (BAY) were purchased from Tocris Bioscience (Bristol, UK); anandamide (AEA) and 2‐arachidonoyl glycerol (2‐AG) were purchased from Cayman Chemical Company (Ann Arbour, MI, USA); and (‐)‐trans‐Δ⁹‐tetrahydrocannabinol (THC) was purchased from THC Pharm GmbH (Frankfurt, Germany). U0126 was purchased from Cell Signalling Technology (Danvers, MA, USA), and PMA and quinpirole hydrochloride were purchased from Sigma Aldrich (St Louis, MO, USA). SR141716A (rimonabant) was a gift from Roche (Basel, Switzerland). Drug stocks were prepared in absolute ethanol (CP55,940, WIN55,212‐2, AEA, 2‐AG, THC, BAY) or DMSO (forskolin, AM6544, SR141716A) and were stored in aliquots at −80°C prior to use. Drug aliquots used for experiments involving serial dilutions were always single‐use, to minimize assay drift due to solvent evaporation or chemical decay from freeze–thaw cycles. Vehicle controls for serial dilutions were maintained constant within experiments, and cell exposure to solvents was minimized (0.1–0.3%). In order to assuage concerns about endogenous cannabinoids in serum, all drug stimulations were performed in the absence of FBS and following a period of serum starvation.

Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY (Southan et al., 2016), and are permanently archived in the Concise Guide to PHARMACOLOGY 2015/16 (Alexander et al., 2015).

Results

Gα_s‐mediated signalling unmasks agonist bias and is specific for CB₁ receptors

Using the CAMYEL assay, cAMP concentration–response curves were obtained for six CB₁ agonists in (not pplss‐tagged) 3HA‐hCB₁ HEK cell line, following pretreatment both in the absence and presence of PTX. These data revealed that all six agonists act with substantial efficacy in the Gα_i pathway (Table 1; Figure 1A), with THC and BAY, reported partial agonists (De Vry et al., 2004; Pertwee, 2008), inhibiting cAMP with significantly lesser efficacies than CP55,940 (P < 0.05). Following PTX pretreatment, however, inhibition of cAMP synthesis was no longer evident. Instead, the majority of agonists stimulated cAMP synthesis above that produced by forskolin alone. Agonist potencies were significantly reduced in comparison with their potencies at inhibiting cAMP production (interaction effects, P < 0.05; Table 1). Wider variations in E_MAX values were also observed (Figure 1B), with significant differences in efficacies proceeding 2AG (most efficacious) >WIN55,212‐2 > AEA ≈ CP55,940 in the stimulatory pathway (P < 0.05; Table 1). Interestingly, THC and BAY did not measurably stimulate Gα_s‐like cAMP signalling (reliable concentration–response curves could not be fitted to enable statistical comparisons of E_MAX values). Operational analysis (Table 2) was performed to compare agonists' signalling in the Gα_i pathway (Figure 1A) with signalling in the Gα_s pathway (Figure 1B), but bias was not detected for any ligand (Table 2). To determine whether the signalling switch was specific for the CB₁ receptor, concentration‐responses with traditional full agonists were performed for hCB₂ receptors (Figure 1C) and hD₂ receptors (Figure 1D) following pretreatment in the absence and presence of PTX. No evidence was seen for non‐Gα_i signalling, as PTX pretreatment resulted in a loss of all receptor‐mediated cAMP signalling.

Table 1.

CB₁ receptor agonist potencies/efficacies under various stimulation conditions (summary data)

	3HA‐hCB1 HEK
	‐PTX			+PTX
	pEC50 (M) (± SEM)	EMAX (% of FSK) (± SEM)	n	pEC50 (M) (± SEM)	EMAX (% of FSK) (± SEM)	n
CP55,940	9.19 (0.18)	43.79 (1.28)	5	8.04 (0.03)	134.86 (5.82)	5
AEA	7.47 (0.14)	48.18 (3.61)	5	5.36 (0.19)	142.06 (4.69)	5
WIN55,212‐2	7.82 (0.11)	51.29 (2.83)	5	5.93 (0.11)	173.03 (3.28)	5
2‐AG	6.99 (0.05)	55.39 (2.48)	5	5.20 (0.16)	192.31 (6.14)	5
THC	7.74 (0.17)	59.97 (3.50)	5	Not Measurable	Not Measurable	5
BAY	7.25 (0.03)	62.93 (3.09)	5	Not Measureable	Not Measureable	5

	pplss‐3HA‐hCB1 HEK
	‐PTX			+PTX
	pEC50 (M) (± SEM)	EMAX (% of FSK) (± SEM)	n	pEC50 (M) (± SEM)	EMAX (% of FSK) (± SEM)	n
CP55,940	7.42 (0.07)	230.00 (12.23)	5	7.93 (0.11)	288.66 (8.29)	5
AEA	5.79 (0.14)	183.33 (7.03)	5	6.24 (0.10)	287.96 (11.25)	5
WIN55,212‐2	6.169 (0.06)	341.98 (12.69)	5	6.82 (0.07)	331.22 (7.22)	5
2‐AG	5.41 (0.08)	326.82 (13.46)	5	5.99 (0.06)	327.64 (17.94)	5
THC	Not Measurable	Not Measurable	5	6.78 (0.06)	184.59 (5.66)	5
BAY	Not Measureable	Not Measureable	5	6.20 (0.17)	186.17 (12.08)	5

Figure 1.

Concentration–response curves for cAMP formation showing 3HA‐hCB₁ HEK signalling on stimulation with 2.5 μM forskolin (‘F’; FSK) and a panel of six agonists, following >16 h pretreatment in the absence (A) or presence (B) of PTX. Panel (C) shows the signalling of 3HA‐hCB₂ HEK cells stimulated with CP or WIN, following pretreatment with or without PTX; panel (D) shows the signalling of 3HA‐hCB₁/FLAG‐D₂ HEK cells stimulated with quinpirole, following pretreatment with or without PTX. Representative data are presented, demonstrating mean ± SEM of technical duplicates. Curves were generated by AUC analysis of kinetic CAMYEL biosensor data, which were normalized to basal (0%) and forskolin (100%).

Table 2.

Operational analysis with bias factors (ΔΔLogR ± error) comparing Gα_i (no PTX) with Gα_s (with PTX) signalling of moderate expression 3HA‐hCB₁ HEK cells

	CP55,940	WIN55,212‐2	AEA	2‐AGa	THC	BAY
ΔΔLogR	0.29 ± 0.37	0.19 ± 0.29	0.53 ± 0.27	0.00 ± 0.29	NDb	NDb

^a2‐AG was used as the reference ligand.

^bAs no Gα_s signal was observed for THC or BAY, a bias factor could not be determined.

High CB₁ receptor expression is sufficient to induce nett Gα_s signalling

To determine whether receptor number was sufficient to result in a switch in cAMP signalling from mostly Gα_i‐mediated (inhibitory) to mostly Gα_s‐mediated (stimulatory), cAMP concentration–response stimulations were performed on the HEK cell line stably expressing pplss‐3HA‐hCB₁ (see below for quantification of receptor expression differences). In these cells (Figure 2A), the effect of CB₁ receptor activation on cAMP signalling was stimulatory, with most agonists demonstrating greater Gα_s pathway efficacies than those seen following PTX‐treatment of the non‐pplss CB₁‐expressing HEK cells described above (all P < 0.05). THC and BAY again produced negligible stimulation of cAMP (Figure 2A, Table 1). PTX pretreatment of the pplss‐3HA‐hCB₁ HEK cells did not increase maximum efficacy in the Gα_s pathway for WIN55,212‐2 or 2AG (P > 0.05; Figure 2B, Table 1), despite potencies appearing to increase for all agonists for which curves could be fitted in both pathways (possibly indicating some degree of signalling pathway competition at sub‐maximal signalling activities). However, PTX pretreatment did result in increased efficacy for some agonists: most notably THC and BAY (which shifted from no nett change in cAMP, to low‐efficacy stimulation of cAMP following PTX treatment), and also (to a lesser extent) AEA and CP55,940 (all P < 0.05; Figure 2A, B; Table 1).

Figure 2.

Concentration–response curves for cAMP formation showing pplss‐3HA‐hCB₁ HEK signalling, on stimulation with 2.5 μM FSK (‘F’; FSK) and a panel of six agonists, following >16 h pretreatment in the absence (A) or presence (B) of PTX. Representative data are presented, demonstrating mean ± SEM of technical duplicates. Curves were created by AUC analysis of kinetic CAMYEL biosensor data, which were normalized to basal (0%) and FSK (100%).

To quantify the difference in receptor expression between the 3HA‐hCB₁ and pplss‐3HA‐hCB₁ HEK cell lines, and to determine whether changes in receptor number influenced ligand affinity, homologous competition binding assays were performed with [³H]‐CP55,940 on membrane homogenates from each of the two cell lines of interest, and pK_d and B_MAX were calculated. The pK_d of CP55,940 was not significantly different between the two cell lines, at 8.82 (SEM ± 0.11) and 8.76 (SEM ± 0.12) respectively (n = 6, t‐test P > 0.05). B_MAX was therefore calculated using a shared average pK_d of 8.79 (SEM ± 0.08). Membranes from the clonal 3HA‐hCB₁ HEK cell line contained 1005.96 fmol·mg⁻¹ (SEM ± 85.87, n = 6) of CB₁, while membranes from the clonal pplss‐3HA‐hCB₁ cell line were found to contain 2713.57 fmol·mg⁻¹ (SEM ± 319.20, n = 6) of CB₁ receptors; a 2.7‐fold difference in receptor number. Homologous competition assays for [³H]‐CP55,940 were also performed in membranes from the 3HA‐hCB₂ cell line (Figure 1C) to determine whether the lack of stimulatory signalling following PTX treatment could be explained by a substantially lower receptor number. However, membranes from this cell line were found to contain substantially greater CB₂ receptor protein than the 3HA‐hCB₁ HEK cells, at 2541.08 fmol·mg⁻¹ (SEM ± 91.77, n = 6; pK_d 9.33, SEM ± 0.12).

The pplss motif is not an inherent determinant of the switch to Gα_s signalling

To further investigate the dependence of cAMP signalling on CB₁ receptor expression level and confirm that the observed stimulatory cAMP signalling was not the result of the N‐terminal pplss tag, HEK Flp‐In™ cells stably expressing V8‐CAMYEL biosensor were transiently transfected with the pplss‐3HA‐hCB₁ construct to generate a population of cells over which receptor expression ranged widely. Cells were then sorted by FACS into six subpopulations on the basis of surface CB₁ receptor expression level determined by antibody binding to N‐terminal HA tags (647 staining intensity, Figure 3A). The sorted cells were stimulated (CP55,940 concentration‐response in the presence of 5 μM forskolin) and cAMP levels assayed. In cells with ‘very low’ to ‘moderate’ expression, a progressive increase in the magnitude of Gα_i‐associated cAMP signalling was observed as receptor staining increased (Figure 3B). Further increases in receptor expression led to alterations in the overall signalling profile; less efficacious Gα_i signalling at ‘intermediate’ receptor expression, a U‐shaped signalling response at ‘high’ receptor expression (where lower CP55,940 concentrations exerted nett Gα_i cAMP effects and higher concentrations exerted nett Gα_s cAMP effects) and finally a nett switch in cAMP phenotype to Gα_s signalling at ‘very high’ receptor expression (Figure 3B).

Figure 3.

FACS frequency distribution (A), showing live‐gated HEK Flp‐In™ cells stably expressing V8‐CAMYEL, sorted into six subpopulations on the basis of transient pplss‐3HA‐hCB₁ surface expression level (647 fluorophore staining intensity). Panel (B) shows cAMP concentration–response curves for CP55,940 in the presence of 5 μM forskolin (FSK; ‘F’), performed on cells from each sorted cell population. Representative data are presented, demonstrating mean ± SEM of technical duplicates.

Characterization of AM6544: an irreversible antagonist of CB₁ receptors

Pharmacological knockdown involves irreversible binding of an antagonist to the orthosteric binding site, effectively eliminating receptors from the pool available to orthosterically bind ligand. As there are currently no well‐described irreversible CB₁ antagonists available commercially, the binding of a novel CB₁ irreversible antagonist, AM6544 (Figure 4A), was characterized under both homogenate and whole cell binding conditions, using [³H]‐SR141716A as the probe so that conditions could be kept as constant as possible between homogenate and whole cell assays. Homogenate binding assays were performed on membranes from the same cell line used for whole cell binding assays, pplss‐3HA‐hCB₁ HEK, and also in membranes from hCB₁ receptor‐expressing HEK293F cells (Figure 4B, C). These hCB₁ receptor‐expressing HEK293F cells were not used in any functional studies reported in this study.

In homogenate binding assays, the pK_d of [³H]‐SR141716A was determined empirically by homologous competition and was found to be 9.00 (SEM ± 0.24, n = 5). This value was then used for fitting one‐site, heterologous binding curves for AM6544 displacement. The resulting pK_i for AM6544 was 8.93 (SEM ± 0.08, n = 5). This result was confirmed by replication in the Makriyannis laboratory on HEK293F cells expressing hCB₁ receptors (pK_i 8.38 ± 0.22). Irreversibility of AM6544 in membrane homogenates was then determined by saturation binding assay, with a comparison of SR141716A versus AM6544 washout at ~100× K_i (100 nM). After extensive washing to remove unbound and reversibly bound ligand, AM6544 caused a 79% reduction in [³H]‐SR141716A B_MAX, relative to B_MAX in vehicle pretreated hCB₁ HEK293F membranes (Figure 4B). Similarly, AM6544 caused an 83% reduction in [³H]‐CP55,940 B_MAX, relative to that of vehicle pretreated membranes (Figure 4C). This demonstrates that AM6544 bound irreversibly to hCB₁ receptors at a concentration of 100 nM, such that [³H]‐SR141716A or [³H]‐CP55,940 were unable to engage AM6544‐occupied hCB₁ receptors.

Under whole cell binding conditions, however, the pK_d of [³H]‐SR141716A was right‐shifted, to 7.27 (SEM ± 0.06, n = 5), and similarly, the affinity of AM6544 was also reduced: pK_i 6.25 (SEM ± 0.06, n = 5). AM6544 irreversible binding was also estimated using the whole cell binding method and was interpreted as being the concentration of ligand that was able to prevent [³H]‐SR141716A binding when applied as a 1 h pretreatment and subjected to washing prior to the [³H]‐SR141716A incubation. Under these conditions, the pEC₅₀ of AM6544 knockdown was estimated to be 5.45 (SEM ± 0.11, n = 5).

Functional characterization of AM6544 under pretreatment conditions (as used in receptor knockdown studies, see below) in a cAMP signalling assay revealed inverse agonist behaviour, comparable with the effects of SR141716A. As shown in Supporting Information Figure S1, assay potency under these washout conditions was consistent with the aforementioned high pEC₅₀ for receptor knockdown.

Systematic pharmacological receptor knockdown reverses pplss‐3HA‐hCB₁ Gα_s signalling

As further evidence to demonstrate the receptor number‐dependence of the CB₁ receptor cAMP signalling phenotype, the irreversible CB₁ antagonist, AM6544, was used to systematically reduce receptor number in both the 3HA‐hCB₁ and pplss‐3HA‐hCB₁ HEK stable cell lines. Following pharmacological knockdown, standard concentration–response curves were performed for CP55,940, WIN55,212‐2 and THC in the presence of 2.5 μM forskolin. The cAMP signalling of 3HA‐hCB₁ HEKs showed decreases in agonist potency and efficacy following pretreatment with μM concentrations of AM6544 (Figure 5A, C, E). However, in pplss‐3HA‐hCB₁ HEK cells, AM6544 pretreatment produced a concentration‐dependent loss of nett Gα_s‐like signalling and progressive increase in Gα_i‐like cAMP signalling. Pretreatment with intermediate concentrations of AM6544 (e.g. 1 μM) resulted in U‐shaped agonist concentration responses, as lower concentrations of agonist caused inhibition (e.g. 10 nM CP55,940, Figure 5B; 1 μM WIN55,212‐2, Figure 5D) but higher concentrations in the same agonist‐response curve resulted in cAMP increases. In cells stimulated with THC (Figure 5F), an agonist which does not cause Gα_s‐like signalling at any concentration in the absence of PTX, inhibitory signalling was also restored following pretreatment with high‐concentrations of AM6544. Notably, pretreatment with concentrations of AM6544 that were sufficient to restore Gα_i signalling resulted in lower potency agonist responses than the curves arising from signalling in the absence of antagonist, perhaps indicating some degree of competitive binding despite attempted washout of non‐irreversibly bound AM6544.

Figure 5.

Concentration–response curves for cAMP formation showing 3HA‐hCB₁ HEK cells (A, C, E) and pplss‐3HA‐hCB₁ HEK cells (B, D, F) signalling on stimulation with 2.5 μM forskolin (FSK; ‘F’) and CP (A, B), WIN (C, D) or THC (E, F), following pretreatment in the presence of increasing concentrations of the irreversible CB₁ receptor antagonist AM6544. Representative data are presented, demonstrating mean ± SEM of technical duplicates. Curves were created by AUC analysis of kinetic CAMYEL biosensor data, which were normalized to basal (0%) and forskolin (100%).

Supplementation of Gα_i protein restores inhibitory CB₁ receptor signalling

If the switch to Gα_s signalling at high receptor levels is due to Gα_i exhaustion, then it would follow that expression levels of either the receptor or the Gα_i protein may drive the change in CB₁ receptor signalling. CB₁ receptors are known to interact with and signal through all three Gα_i subtypes (Mukhopadhyay and Howlett, 2005); therefore, Gα_i1 expression was supplemented by transfecting untagged GNAI1 (or matched empty vector) into the 3HA‐hCB₁ and pplss‐3HA‐hCB₁ HEK cell lines. cAMP assays revealed no obvious changes to agonist concentration‐responses in the 3HA‐hCB₁ HEK cell line (whose nett cAMP signal is inhibitory under normal conditions, Figure 6A, C, E). However, in the pplss‐3HA‐hCB₁ cells, increased Gα_i1 expression resulted in substantially reduced CP55,940 and WIN55,212‐2‐dependent Gα_s signalling efficacy, with inhibitory efficacy evident at some agonist concentrations (Figure 6B, D), and restoration of nett inhibitory signalling on stimulation with THC.

Figure 6.

Concentration–response curves for cAMP formation showing 3HA‐hCB₁ HEK cells (A, C, E) and pplss‐3HA‐hCB₁ HEK cells (B, D, F) signalling on stimulation with 2.5 μM forskolin (FSK; ‘F’) and CP (A, B), WIN (C, D) or THC (E, F), following transfection either with supplementary GNAI1 (encoding Gα_i1) or empty vector. Representative data are presented, demonstrating mean ± SEM of technical duplicates. Curves were created by AUC analysis of kinetic CAMYEL biosensor data, which were normalized to basal (0%) and forskolin (100%).

Constitutive cAMP signalling in stable 3HA‐hCB₁ and pplss‐3HA‐hCB₁ HEK cells is Gα_i‐linked

The divergence of agonist‐induced signalling for CB₁ receptors at different levels of expression led us to investigate the nature of CB₁ receptor constitutive signalling in each cell line. As expected, in 3HA‐hCB₁ cells SR141716A concentration‐dependently inhibited constitutive Gα_i‐mediated inhibition of cAMP, resulting in an increase in cAMP. Following PTX pretreatment, this effect was lost (Figure 7A), indicating that the effect SR141716A inhibited was indeed Gα_i‐linked. Interestingly, SR141716A treatment of pplss‐3HA‐hCB₁ HEK cells resulted in a much more pronounced inhibition of Gα_i activity, as demonstrated by the much larger increase in cAMP. This effect was wholly PTX‐sensitive (Figure 7B).

Figure 7.

Concentration–response curves showing inhibition by the inverse agonist SR141716A of cAMP signals derived from CB₁ receptor constitutive activity, in both 3HA‐hCB₁ (A) and pplss‐3HA‐hCB₁ (B) HEK cells. SR141716A concentration–response curves were performed each in the presence and absence of forskolin (‘F’) and PTX. Representative data are presented, demonstrating mean ± SEM of technical duplicates. Curves were created by AUC analysis of kinetic CAMYEL biosensor data, which were normalized to basal (0%) and forskolin (100%).

pERK, a pathway mediated through multiple G protein subtypes, provides further evidence of Gα_i‐independent signalling

Time course experiments were performed for activation (phosphorylation) of ERK in both cell lines. In the lower‐expressing 3HA‐hCB₁ cells, high concentrations of the agonists CP55,940 and WIN55,212‐2 elicited similar pERK maxima of approximately 40% of a 5 min 100 nM PMA stimulation (Figure 8A, C), while 10 μM THC elicited a partially efficacious response of approximately 20% of the PMA response (Figure 8E). In each instance, this effect was essentially wholly PTX‐sensitive, indicating that the pERK responses observed were Gα_i‐mediated. Interestingly, in the pplss‐3HA‐hCB₁ cells, the same stimulation conditions elicited larger efficacy responses that were predominantly not PTX‐sensitive. Specifically, stimulation with either CP55,940 or WIN55,212‐2 (Figure 8B, D) resulted in pERK signals of approximately 60% of the PMA response in the absence of PTX, and this effect was attenuated by just 10–20% when Gα_i was inhibited with PTX. Under these conditions, THC stimulation (Figure 8F) demonstrated low efficacy in the 3HA‐hCB₁ cells, and a relatively greater proportion of the pERK signal (approximately half) was sensitive to PTX. Time course differences are also apparent in these data: 3HA‐hCB₁ HEK cells showed pERK maxima at approximately 3–4 min stimulation times, and this signal decayed rapidly to basal within 10 min, while the pplss‐3HA‐hCB₁ cells showed an earlier increase in signal (pERK levels at 1 min stimulation already tended to be above basal), a peak at 3–4 min stimulation, but then a delayed decay, as pERK levels tended to plateau at approximately 6–7 min stimulation and gradually returning toward basal pERK state over a period longer than the 30 min of the assay.

Figure 8.

Time course of pERK activation in 3HA‐hCB₁ HEK cells (A, C, E) and pplss‐3HA‐hCB₁ HEK cells (B, D, F) on stimulation with CP (A, B), WIN (C, D) or THC (E, F), following >16 h pretreatment in the absence or presence of PTX. Combined data (n = 5) were normalized to the level of pERK induced by treatment with PMA for 5 min (100%) and U0126 for 30 min (0%).

Discussion and conclusions

In this study, CB₁ receptor expression level is shown to be a novel determinant of signalling outcome, and this effect was found to be dependent on the cannabinoid agonist used and the expression level of Gα_i protein. Assays for cAMP signalling revealed that some ‘full’ agonists of the CB₁ receptor, particularly CP55,940, elicit maximal signalling in just one of the two pathways characterized, an observation consistent with the findings of Bonhaus et al. (1998). Additionally, the two CB₁ partial agonists included in the cAMP screen in this study, THC and BAY, showed limited ability to stimulate cAMP, requiring both PTX pretreatment and high CB₁ receptor expression levels for partial Gα_s‐like efficacy to be unmasked. The receptor number‐driven switch from Gα_i to Gα_s signalling was shown to be specific to CB₁ receptors and could be reversed both by systematic pharmacological knockdown with the novel irreversible antagonist AM6544 (Patent US8084451, 2011) and by increasing the expression level of Gα_i1 protein. Finally, characterization of pERK activation was undertaken to reveal the relative contributions of Gα_i activation and alternative pathways to a downstream pathway. These data showed that most (but not all) of the signalling in the lower expressing 3HA‐hCB₁ HEK cells is Gα_i‐mediated and that most of the signalling in the higher expressing pplss‐3HA‐hCB₁ HEK cells is not Gα_i‐mediated.

To date, the signalling phenotype of the CB₁ receptor has been viewed as almost entirely Gα_i‐mediated under normal circumstances. At best, this is an oversimplification that has arisen because the easiest, and by far most common, means for assaying Gα_i and Gα_s protein activity is changes in forskolin‐stimulated cAMP. An obvious but inherent limitation of this endpoint, however, is that it is not equipped to distinguish contributions of the different G protein subtypes to the total signal, a problem that originates from the mutually antagonistic effects that Gα_i and Gα_s activity exert on the cAMP pathway. Indeed, recent work by Eldeeb et al. (2016) would suggest that nett cAMP signalling does not necessarily reflect CB₁ receptor‐mediated G protein activity, as Gα_s protein cycling was elicited concentration‐dependently following agonist stimulation under conditions where the nett cAMP effect was inhibitory. The use of PTX to irreversibly and specifically inactivate Gα_i proteins is a widely used approach that aids in cAMP phenotype characterization. However, this approach is of limited utility in the context of CB₁ receptor characterization because both the circumstances under which Gα_s activity occurs, and the extent to which it exists within canonical signalling, remain uncharacterized. Without a reliable means to eliminate Gα_s activity, it is also imponderable as to whether PTX unmasks ‘secondary’ signalling pathways with magnitude equal to that in the absence of PTX, or whether it effectively drives greater efficacy in the non‐preferred pathway simply because of reduced competition of non‐preferred effectors for active receptors. The levels of CB₁ receptors present in these cell lines are not dissimilar to that observed in native tissue. Where the pplss‐3HA‐hCB₁ cells used in this study contained approximately 2700 fmol·mg⁻¹ hCB₁ receptor protein (vs. approximately 1000 fmol·mg⁻¹ for the lower‐expressing cell line), whole rat brain ‘sausage’ homogenate sections have been shown to contain approximately 1158 fmol·mg⁻¹ of rCB₁ receptor protein (Herkenham et al., 1991), and membranes prepared from mouse brain contained 1810 fmol·mg⁻¹ of mCB₁ receptor protein (Abood et al., 1997), where neither study attempted to account for differences in density across different brain regions. Experiments to quantify cannabinoid binding in guinea pig forebrain homogenates revealed the presence of as much as 5658 fmol·mg⁻¹ of CB₁ receptor protein, as determined by binding of the CB₁‐selective radioligand [³H]‐SR141716A, although estimates using nonselective radioligands ([³H]‐CP55,940 and [³H]‐WIN55,212‐2) reported 4281 and 2032 fmol·mg⁻¹ protein respectively (Ross et al., 1998).

The concept of functional selectivity adds further complexity to the characterization of CB₁ receptor signalling (see Urban et al., 2007 for a review). Qualitative means of estimating bias are frequently confounded by variables such as cell background (e.g. different Gα_i and Gα_s expression levels) or, in cAMP signalling, concentrations of forskolin (higher concentrations will overstate Gα_i activity but understate Gα_s activity, and vice versa). An empirical method of analysis that purports to objectively describe functional selectivity (not be influenced by the aforementioned variables) is the Operational model. This approach produces parameters for affinity, K_A, and efficacy, τ, based on the classical principle that the two levels of interaction that they represent (agonist‐receptor interactions and receptor‐signalling proteins interactions, respectively) are contingent for profiling a given signalling response (Black and Leff, 1983). A value obtained from these variables, termed the transduction coefficient, Δlog(τ/K_A), may therefore be considered an agglomeration of both interactions to allow approximation of an agonist's intrinsic efficacy (‘stimulus per receptor’) (Kenakin et al., 2012). Comparisons of the transduction coefficients of different pathways (for example, Gα_i and Gα_s signalling) are subsequently represented as single values in the form of a bias factor, ΔΔlog(τ/K_A). Operational analysis was performed on the cAMP data included in this study, for 3HA‐hCB₁ HEK cells in the absence and presence of PTX (Table 2). These data did not reveal significant bias for any agonist. However, it must be noted that neither THC nor BAY elicited any cAMP effects when cells had been pretreated with PTX, and therefore, concentration–response curves could not be fitted, nor operational parameters obtained. A simple EC₅₀ comparative analysis suggests that if THC and BAY underwent the same potency shift as the other ligands (approximately 100‐fold), then responses would still be detectable within this assay. It would follow that the apparent inability of these ligands to signal in the Gα_s‐like pathway must indicate either a much larger shift in potency or an actual lack of Gα_s signalling efficacy. Either way, these explanations indicate that THC and BAY are both biased toward the inhibitory pathway and against the stimulatory pathway.

In this study, experiments of two different designs were performed to demonstrate the role of receptor number in the reported signalling switch for CB₁ receptors. The first, sorting of HEK cells transiently transfected with a construct for pplss‐3HA‐hCB₁ by FACS (Figure 3), revealed in ‘very low’ to ‘medium’‐expressing cells a progressive increase in inhibitory cAMP signalling, then (with continued increases in receptor expression) a gradual loss of efficacy in the inhibitory pathway, and ultimately nett stimulatory cAMP signalling. The reduction in Gα_i signalling at ‘intermediate’ receptor expression probably indicates the stimulatory signalling component acting competitively with Gα_i for receptors – as more receptors are activated, Gα_i protein saturation may be occurring as Gα_s recruitment increases. Significantly, this suggests that the switch in CB₁ receptor signalling from Gα_i to Gα_s is dependent on the expression of all three proteins, and potentially other accessory proteins. We hypothesize that the CB₁ receptor possesses high affinity for Gα_i, but that this is a relatively low‐efficacy signalling pathway. Conversely, the CB₁ receptor has lower affinity for Gα_s, but this pathway has higher intrinsic activity and is therefore capable of overwhelming the effects of Gα_i. Certainly, the availability and abundance of G protein effectors has already been shown to contribute to alterations in signalling for other GPCRs (Nasman et al., 2001). It is also possible that cAMP increases results from increased activation of Gα_i‐insensitive adenylate cyclase (AC) isoforms. HEK cells express the transcripts of multiple isoforms of AC (Atwood et al., 2011). Different AC isoforms are modulated by different Gα subtypes: all nine AC isoforms are sensitive to stimulation by Gα_s/Olf family proteins, but some isoforms (e.g. AC2, 3, 4 and 7) are not directly regulated by Gα_i (Dessauer, 2009).

The second assay type that demonstrated the CB₁ receptor specificity of the switch involved systematic pharmacological knockdown with the novel antagonist AM6544. In order to determine compound affinity, homogenate binding competition‐displacement assays were performed and revealed low nanomolar‐range affinity. However, in live cell functional assays, micromolar concentrations of AM6544 were required in order for receptor knockdown to be observed (Figure 5) – a difference in concentration that is incongruent with simple occupancy. Further affinity characterization involved whole cell binding assays, which were performed under matched conditions to functional assays (Figure 5). As reported, this assay type revealed a right‐shifted micromolar‐range pK_i and a similar right‐shift in SR141716A affinity, emphasizing the dependence of ligand binding on assay conditions. AM6544 was found to concentration‐dependently cause a switch of nett stimulatory cAMP signalling back to canonical inhibitory signalling in the pplss‐3HA‐hCB₁ cells (Figure 5B, D, F), although pretreatment with even very high concentrations (100 μM) failed to completely abolish signalling, probably indicating both that only a portion of receptors were AM6544‐occupied and therefore irreversibly blocked under these assay conditions, and that the cAMP pathway has a substantial degree of receptor reserve.

Experiments to increase Gα_i expression by transfection were performed to investigate the influence of G protein expression level on the signalling switch (Figure 6). No substantial differences were observed in the signalling of the 3HA‐hCB₁ cells, which signal through Gα_i under usual circumstances (Figure 6A, C, E). In the pplss‐3HA‐hCB₁ cells, supplementary Gα_i1 expression dramatically altered the concentration–response curves, producing U‐shaped concentration‐response curves for CP55,940 and WIN55,212‐2 and resulting in an inhibitory curve for THC (Figure 6B, D, F). The incomplete reversal of signalling with CP55,940 and WIN55,212‐2 probably arises due to our assay design, wherein partial transfection efficiency resulted in only a proportion of cells receiving the GNAI1 construct by transfection, and therefore, a range of Gα_i1 expression levels were assayed, including a proportion of cells with no additional Gα_i1 expression. Interestingly, however, in these assays, agonist efficacies align with Gα pathway bias: CP55,940 exhibits partial efficacy in the Gα_s‐like pathway in the absence of PTX and demonstrates almost no nett signalling at 1 μM CP55,940 following GNAI1 transfection (Figure 6B); WIN55,212‐2 exhibits full efficacy in the Gα_s‐like pathway and demonstrates less pronounced Gα_i rescue in cells overexpressing Gα_i (Figure 6D); and THC exhibits no efficacy in the Gα_s‐like pathway in the absence of PTX, while also being most amenable to demonstrating nett Gα_i rescue following GNAI1 transfection, being the only agonist for which an inhibitory concentration–response curve could be fitted (Figure 6F). Similarly, it is worth noting that at the ratios at which CB₁ receptors and its signalling effectors are expressed following GNAI1 transfection in this assay, functional bias is genuinely druggable: for example, 10 μM WIN55,212‐2 (Figure 6D) results in a nett stimulatory cAMP signalling response, while 1 μM THC (Figure 6F) results in a nett inhibitory signal.

Characterization of the differential constitutive signalling of the 3HA‐hCB₁ and pplss‐3HA‐hCB₁ HEK cells was based on the Gα_i exhaustion hypothesis that under unstimulated conditions, irrespective of receptor expression level, receptors are less likely to be limited for their preferred G protein effector because a minority of receptors are in an active conformation at any moment. The inverse agonist SR141716A induced concentration‐dependent stimulation of cAMP in both cell lines (Figure 7), and these effects were PTX‐sensitive, indicating that these were completely mediated by Gα_i. Intriguingly, this results in a situation in pplss‐3HA‐hCB₁ cells where stimulation with both agonists and inverse agonists has the same effect on signalling, to increase cAMP, albeit by different mechanisms.

Although cAMP assays facilitated substantial characterization of the interplay between receptor number and signalling outcome, the mutually antagonistic consequences of Gα_i and Gα_s activation on cAMP levels limit ability to discern activation states for each distinct pathway. Characterization of pERK activation in 3HA‐hCB₁ and pplss‐3HA‐hCB₁ HEK cells was undertaken in order to better observe an effect of concurrent activity in the Gα_i pathway and alternative Gα_s‐like pathway and also to examine a downstream consequence of the CB₁ receptor signalling switch. In a result which further extends understanding of the mechanism of the signalling switch, the pERK signal in 3HA‐hCB₁ cells was PTX sensitive in every condition (Figure 8A, C, E), but in pplss‐3HA‐hCB₁ cells, it was predominantly PTX‐independent, suggesting that the alternative signalling pathway may not simply exist additively with Gα_i signalling (as required by the Gα_i exhaustion hypothesis) but may replace it, in agreement with a pleiotropic switch mechanism (Maudsley et al., 2005). Significantly, however, it is possible that the pathway mediating the non‐PTX‐sensitive pERK signal in the pplss‐3HA‐hCB₁ cells is not the same as the Gα_s‐like signal that gives rise to stimulatory cAMP signalling: CB₁ receptor coupling to Gα_q has also been reported previously (Lauckner et al., 2005), and this pathway has also been shown to elicit ERK phosphorylation responses (Asimaki and Mangoura, 2011).

Taken together, the data we present support more than one mechanism for the switch from the canonical Gα_i pathway to the alternative Gα_s‐like pathway: the central idea of receptor number driving the switch and the fact that it can be reversed competitively by altering G protein expression strongly support receptor coupling promiscuity and affinity/efficacy interactions that underpin the Gα_i hypothesis; however, the apparent change in signalling balance between Gα_i and other effectors in pERK characterization (as opposed to the alternative effectors simply adding to the total signal) better supports a pleiotropic switch mechanism. The potential for more than one signalling switch mechanism is significant from a biomedical perspective. This is because only the pleiotropic mechanism is directly targetable pharmacologically by design of a biased agonist (it requires that stabilization of a particular receptor conformation is sufficient to drive differential signalling), whereas the Gα_i exhaustion mechanism requires both receptor agonism and a secondary means to either promote coupling of the favoured G protein or diminish the activity of the disfavoured one. In development of CB₁ receptor‐targeted interventions, the ability to target one Gα pathway over another will be of interest therapeutically. For example, in an in vitro model of Huntington's disease, CB₁ receptor Gα_s‐mediated increases in cAMP have been found to exacerbate cell death by promoting aggregation of mutant huntingtin (Scotter et al., 2010). Similarly, in a different in vitro Huntington's disease model, CB₁ receptor‐mediated Gα_i activity has recently been suggested to be therapeutically beneficial, as Gα_i‐mediated pERK is substantially reduced in diseased cells compared with matched controls, and agonists that demonstrated Gα_i bias in those assays were shown to improve cell viability (Laprairie et al., 2016).

One of the greatest remaining obstacles in biomedical sciences generally is the difficulty in relating easily characterized near‐signalling phenotypes with whole cell‐level endpoints, as mid‐stream signalling pathways are relatively poorly characterized and diverge greatly between different cell types and disease states – indeed, a recent review by Velasco et al. (2016) observed that oncogenically transformed cells may exhibit different signalling phenotypes to non‐transformed cells and that the molecular reasons for this are unknown. An early study into the effects of cannabinoids on cancer growth revealed that THC treatment inhibited growth of C6 rat glioma cells and induced cell death (Sanchez et al., 1998), a finding now supported by other studies (Gomez del Pulgar et al., 2002; Salazar et al., 2009). Confusingly, however, more recent data suggests that blockade of CB₁ receptors can also lead to apoptosis in glioma cells and primary human cells (Ciaglia et al., 2015), meaning that agonism and antagonism of CB₁ receptors may not confer opposite effects on cell fate, even in related cell types. Additionally, high expression of CB₁ receptors has been shown to correlate with poor prognosis in prostate cancer (Chung et al., 2009; Fowler et al., 2010) as well as stage two microsatellite stable colorectal cancer (Gustafsson et al., 2011). These endpoints have not been associated with near‐signalling phenotypes, but one study to date has revealed that astrocytoma subclones signal through the pro‐survival Akt pathway in a manner dependent on high expression of CB₁ receptors (Cudaback et al., 2010), a finding that was replicated in prostate cancer tissue (Cipriano et al., 2013). Relatedly, an up‐regulation of CB₁ receptors has been reported in a variety of other cancers, including Hodgkin lymphoma (Benz et al., 2013), human epithelial ovarian tumours (Messalli et al., 2014) and stage four colorectal cancer (Jung et al., 2013) and has been shown to correlate with disease severity in the latter two cases. Thus, substantial conflicts remain in the literature and emphasize the complex interconnections between CB₁ receptor near signalling phenotypes and downstream endpoints such as cell fate. Our study may help to provide a basis for deconvoluting the conflicting data regarding CB₁ receptor near‐signalling and cell fate, though much more research is required in order to understand the roles that the signalling switch plays in normal physiology and disease and the ways in which downstream endpoints are affected.

Author contributions

D.B.F. designed and performed experiments, analysed the data and wrote the paper. E.E.C., M.R.H. and A.K. designed and performed experiments. N.L.G. analysed the data and wrote the paper. V.K.V. designed and performed experiments, wrote the paper and generated the AM6544 compound. A.M. generated the AM6544 compound. M.G. designed experiments, analysed the data and wrote the paper.

Conflict of interest

The authors declare no conflicts of interest.

Declaration of transparency and scientific rigour

This Declaration acknowledges that this paper adheres to the principles for transparent reporting and scientific rigour of preclinical research recommended by funding agencies, publishers and other organisations engaged with supporting research.

Supporting information

Figure S1 Cyclic AMP concentration–response curves showing 3HA‐hCB1 HEK cell signalling on stimulation with vehicle or 2.5 μM FSK, following pretreatment in the presence of increasing concentrations of the CB1 irreversible antagonist AM6544. Representative data. Curves were created by area‐under‐the‐curve analysis of kinetic CAMYEL biosensor data which was normalized to basal (0%) and FSK (100%).

Finlay, D. B. , Cawston, E. E. , Grimsey, N. L. , Hunter, M. R. , Korde, A. , Vemuri, V. K. , Makriyannis, A. , and Glass, M. (2017) Gα_s signalling of the CB₁ receptor and the influence of receptor number. British Journal of Pharmacology, 174: 2545–2562. doi: 10.1111/bph.13866.