Assessment of select synthetic cannabinoid receptor agonist bias and selectivity between the type 1 and type 2 cannabinoid receptor

The first synthetic cannabinoid receptor agonists (SCRAs) were designed as tool compounds to study the endocannabinoid system’s two predominant cannabinoid receptors, CB1R and CB2R. Unfortunately, novel SCRAs now represent the most rapidly proliferating novel psychoactive substances (NPS) of abuse globally. Unlike ∆9-tetrahydrocannabinol, the CB1R and CB2R partial agonist and the intoxicating constituent of Cannabis, many SCRAs characterized to date are full agonists of CB1R. Gaining additional insight into the pharmacological activity of these SCRAs is critical to assess and regulate NPSs as they enter the marketplace. The purpose of this study was to assess select SCRAs recently identified by Canadian police, border service agency, private companies and the illicit market as potential CB1R and CB2R agonists. To this end, fifteen SCRAs were screened for in vitro activity and in silico interactions at CB1R and CB2R. Several SCRAs were identified as being highly biased for cAMP inhibition or βarrestin2 recruitment and receptor subtype selectivity between CB1R and CB2R. The indazole ring and halogen-substituted butyl or pentyl moieties were identified as two structural features that may direct βarrestin2 bias. Two highly-biased SCRAs—JWH-018 2′-napthyl-N-(3-methylbutyl) isomer (biased toward cAMP inhibition) and 4-fluoro MDMB-BINACA (biased toward βarrestin2 recruitment) displayed unique and differential in vivo activity in mice. These data provide initial insight into the correlations between structure, signalling bias, and in vivo activity of the SCRAs.

In 2018 Canada became the first G7 country to legalize Cannabis sativa for medical and recreational purposes. It is critically important that we gain a more comprehensive understanding of cannabinoid pharmacology in order to reduce harm and make full use of this plant's medical potential. Biologically active compounds in Cannabis are referred to as 'phytocannabinoids' . The two best-known phytocannabinoids are ∆ 9 -tetrahydrocannabinol (THC) and cannabidiol (CBD). Our bodies also naturally produce endogenous cannabinoids anandamide (AEA) and 2-arachidonoylglycerol (2-AG). These cannabinoids act to modulate the body's cannabinoid receptors: CB1R and CB2R, limiting neurotransmitter release throughout the brain and inflammatory processes, respectively 1 . Beyond these naturally occurring compounds, a wide and structurally diverse array of synthetic cannabinoid receptor agonists (SCRAs) have been produced (Fig. 1) 1 . These were originally intended to aid in drug development and as tool compounds to help better understand the body's endogenous cannabinoid system 1,2 . Unfortunately, many of these compounds are now available through illegal markets where they are sold as psychoactive and intoxicating drugs of misuse known colloquially as 'spice' or 'K-2' 1,2 .
SCRAs pose serious health risks to Canadians as they can lead to severe vomiting, chest pain, increased heart rate, vision blackouts, headaches, kidney damage, agitation, high blood pressure, and psychosis 3,4 . Acute toxicities of SCRAs such as JWH 018 and JWH 073 are thought to be mediated by CB1R activation [5][6][7] ; however, novel SCRAs continue to emerge on the market without previous characterization in the scientific literature 4 . In Canada, the Controlled Drugs and Substances Act (CDSA) provides a legislative framework for the control of substances that can alter mental processes and that may cause harm to the health of an individual or society when misused or diverted to the illegal market. Many SCRAs are controlled under Item 2 of Schedule II to the Inhibition of FSK-stimulated cAMP. All fifteen SCRAs displayed agonist activity for the inhibition of FSKstimulated cAMP at CB1R. The majority of these were full agonists of CB1R-dependent cAMP inhibition relative to the reference agonist CP55940, and displayed similar potency compared to that of CP55940 as well (Fig. 3, Table 1). ∆ 9 -THC was assayed as a control partial agonist and displayed significantly lower efficacy and potency relative to CP55940 (Fig. 3, Table 1). One SCRA-AB-PINACA N-(2-fluoropentyl) isomer-was a partial agonist of CB1R-dependent cAMP inhibition relative to CP55940, but retained high potency (Fig. 3, Table 1). Two compounds-AM 2232 and EG 018-displayed significantly lower potency relative to CP55940, but retained full (i.e., 100%) efficacy (Fig. 3, Table 1). Importantly, cAMP accumulation was not completely inhibited by the maximum effects observed with SCRAs, indicating that supramaximal responses could have been detected had they occurred ( Supplementary Fig. 2a).
Bias analyses. Given the high potency and efficacy observed for most SCRAs in the cAMP inhibition assay and comparatively low potency and efficacy observed for the majority of these compounds in the βarrestin2 recruit-   c. d.
hCB1R cAMP inhibition  Table 1; or fit to the operational model to calculate bias (Fig. 5). Data are mean ± S.E.M. n ≥ 6 independent experiments performed in triplicate. c. d.
hCB1R βarrestin2 recruitment  Table 1; or fit to the operational model to calculate bias (Fig. 5). Data are mean ± S.E.M. n ≥ 6 independent experiments performed in triplicate. Note the difference in y-axis scale between (a) and all other panels.  Table 1). With one exception, MAM2201 N-(5-chloropentyl) analog, the potency of these compounds at CB2R for cAMP inhibition was not different compared to that of CP55,940 ( Fig. 7, Table 1). ∆ 9 -THC was assayed as a control partial agonist and displayed significantly lower efficacy relative to CP55,940 ( Fig. 7, Table 1). As was observed for CB1R, cAMP accumulation was not completely inhibited by the maximum effects observed with SCRAs, indicating that supramaximal responses could have been detected had they occurred ( Supplementary Fig. 2b).
CB1R versus CB2R selectivity. With regard to receptor affinity, six SCRAs displayed greater affinity at  Table 1). Interestingly, AB PINACA N-(2-fluoropentyl) isomer was unique in displaying a higher affinity for CB1R but only partially  Table 1; or fit to the operational model to calculate bias (Fig. 9). Data are mean ± S.E.M. n ≥ 6 independent experiments performed in triplicate.  Table 1).
In silico docking to CB1R and CB2R. Following our cell culture-based work, we wanted to determine whether the structures of these SCRAs might explain their ligand bias and receptor subtype selectivity. Binding of each SCRA to CB1R and CB2R was modelled via induced fit docking in Schrödinger using G protein-bound structures for each receptor (CB1R PDB 6N4B 20 ; CB2R PBD 6PT0 21 ). All fifteen cannabinoids studied bound to the orthosteric sites of CB1R and CB2R using these models (JWH 018 2′-naphthyl-N-(3-methylbutyl) isomer and 4-fluoro MDMB-BINACA shown in Fig. 10). Two interactions were conserved between both CB1R and CB2R in terms of the binding of SCRAs: the hydrogen bond with Ser383 7.39 (CB1R) or Ser285 7.39 (CB2R) and the T-shaped π-π interaction His178 2.65 (CB1R) and His95 2.65 (CB2R). Furthermore, many common interactions were observed between homologs amino acid residues at CB1R and CB2R, which suggests that the binding sites are well conserved between both receptors (Tables 2 and 3). www.nature.com/scientificreports/ Because the amino acid sequence and binding sites are well conserved between CB1R and CB2R, the interaction energy calculated for each cannabinoid represents a predictive metric that can be used to estimate and compare the relative affinity of each cannabinoid at each receptor (Tables 2 and 3 However, three notable exceptions were found in our data: (1) 4-fluoro MDMB-BINACA displayed a higher affinity for CB2R in [ 3 H]CP55,940 binding assays, but greater affinity was predicted to CB1R in this model, and greater potency and efficacy was observed for this compound at CB1R for βarrestin2 recruitment; (2) AM 2232 displayed a higher affinity for CB1R in [ 3 H]CP55,940 binding assays, but cAMP inhibition and modeling data suggest this compound is CB2R-selective; and (3) FDU-NNEI displayed a higher affinity for CB1R in [ 3 H]CP55,940 binding assays, but greater affinity was predicted for CB2R in these models.
Using these models and focusing on CB1R, if we compare the highest affinity CB1R agonist (Cl 2201) to the weakest affinity CB1R agonist (JWH 145) we can understand the structural features that make a high affinity CB1R agonist. Firstly, the main structural difference between Cl 2201 and JWH 145 is that in Cl 2201 the benzene ring is fused to the pyridine vs in JWH 145 it is a branched benzene ring. Because of this, the two cannabinoids only share one similar interaction: Phe268 EL2 . In fact, JWH 145 only has two electrostatic interactions with CB1R. However, if we compare Cl 2201 with MAM2201 N-(5-chloropentyl) analog, which are very similar in structure, we can likely see what interactions are important for Cl 2201's high affinity for CB1R. The major difference between these two structures is that MAM 2201 N-(5-chloropentyl) analog has a methyl substitution from the naphthalene group, whereas Cl 2201 has a chlorine substitution at this position. MAM 2201 also bears a choro-pentyl whereas Cl 2201 has a fluorine. The binding poses of these cannabinoids are identical except for the Arg182 EC1 halogen bond exhibited by Cl 2201 but not by MAM2201. Thus, this halogen bond with extracellular loop one may give rise to the high affinity of Cl 2201 at CB1R relative to other SCRAs. We also attempted to model ligand affinity to CB1R in either G protein-permissive or G protein-absent states using these in silico models and two selective highly biased CB1R SCRAs: JWH 018 2′-naphthyl-N-(3-methylbutyl) isomer (lacking βarrestin2 activity) (Fig. 10a,e) and 4-fluoro MDMB-BINACA (βarrestin2 biased) (Fig. 10c,g). Modelling was carried out with CB1R 6N4B in the presence and absence of Gα i in order to simulate a G protein-accommodating state (i.e. a model of G protein bias) and a non-G protein-accommodating state (i.e. a model of non-G protein bias). JWH 018 2′-naphthyl-N-(3-methylbutyl) isomer bound with greater estimated affinity to CB1R with Gα i present (greater negative interaction energy); this is congruent with the ligand's G protein bias (Table 4). 4-fluoro MDMB-BINACA, on the other hand, bound with greater estimated affinity to CB1R without Gα i present; this is congruent with the ligand's observed βarrestin2 bias (Table 4). These in silico assessments of SCRA affinity with and without G protein serve as validation of our cell culture experiments, but do not themselves represent a determination of bias. Overall, in silico modeling of SCRAs to cannabinoid receptors was largely congruent with cell signaling experiments but cannot fully model the dynamic and multi-faceted aspects of ligand-receptor interaction necessary to predict receptor subtype selectivity.
In vivo activity. Finally, we wanted to determine whether CB1R SCRAs with observable βarrestin2 bias in cell culture and in silico experiments produced an in vivo effect profile that was distinct from G protein-biased Table 3. Summary of the interactions between the residues of the CB2R (PDB: 6PT0) and each SCRA. The right column lists the interaction energy of each SCRA.  www.nature.com/scientificreports/ and non-biased ligands. To do so, two compounds were assessed in a triad of cannabinoid-based outcomes in adult, male, C57Bl/6 mice: the highly βarrestin2-biased 4-fluoro MDMB-BINACA and JWH 018 2′-napthyl-N-(3-methylbutyl) isomer, which was inactive in the CB1R βarrestin2 assay. CP55,940 and THC were also assessed as positive controls for full and partial agonism at CB1R in vivo, respectively (Fig. 11). Time points for in vivo activity of i.p. administered cannabinoids were chosen to be 5 min, 15 min, and 20 min for catalepsy, body temperature, and nociception, respectively, based on the previously-published findings of ourselves and others 19,22,23 and time course experiments conducted with CP55,940 ( Supplementary Fig. 3). 4-fluoro MDMB-BINACA was a full agonist of all three measured responses: catalepsy (Fig. 11a), hypothermia (Fig. 11b), and anti-nociceptive response (Fig. 11c). Producing substantial increases in catalepsy and reductions in body temperature at even the lowest doses tested (0.1 mg/kg i.p.) and producing effects in all three responses that were greater than THC. The responses to 4-fluoro MDMB-BINACA were greater than those observed for CP55,940 in catalepsy and body temperature measurements, but not different in the anti-nociception assay (Fig. 11). In contrast to 4-fluoro MDMB-BINACA, JWH 018 2′-napthyl-N-(3-methylbutyl) isomer was a partial agonist in the catalepsy assay (Fig. 11a), did not change body temperature (Fig. 11b), and was a full agonist in the anti-nociception assay (Fig. 11c). The SCRAs 4-fluoro MDMB-BINACA and JWH 018 2′-napthyl-N-(3-methylbutyl) isomer were further tested for their in vivo activity using the elevated plus maze (EPM), which measures locomotive effects (total arm entries) and anxiolytic effects (open arm entries and % time in the open arms). One mg/kg of CP55,940 or 4-fluoro MDMB-BINACA reduced total arm entries compared to vehicle or THC treatment (Fig. 12a-c); whereas THC or JWH 018 2′-napthyl-N- (3-methylbutyl) isomer treatment increased open arm entry and did not change total arm entries (Fig. 12a-c). Mice treated with 4-fluoro MDMB-BINACA spent approximately the same amount of time in the open arm as vehicle-treated mice; but CP55,940-and 4-fluoro MDMB-BINACAtreated mice spent the majority of their time in the central quadrant immobile (Fig. 12d-f). In contrast, THC or JWH 018 2′-napthyl-N-(3-methylbutyl) isomer treatment was associated with a greater amount of time spent in the open arms, indicative of an anxiolytic effect (Fig. 12d-f). Therefore, these two SCRAs which had opposing biases for and against βarrestin2 relative to G protein signaling also differed in their in vivo activity such that βarrestin2 bias was associated with greater cataleptic and hypothermic activity, and G protein bias was associated with greater anxiolytic activity.

Discussion
In general, we observed that most SCRAs tested behaved as CB1R and CB2R full agonists in the cAMP inhibition assay. However, many did not fully displace [ 3 H]CP55,940 from CB1R or CB2R in the competition binding assay. A notable exception to this was AB-PINACA N-(2-fluoropentyl) isomer, a partial agonist for the CB1R cAMP inhibition response and a full agonist of βarrestin2 recruitment. SCRAs generally displayed little to no consistent selectivity between CB1R and CB2R. This is likely due to a high degree of shared receptor homology and topology between the CB1R and CB2R orthosteric binding sites 20,21,[24][25][26] .
In this study, we did not explore the pharmacology of SCRAs at receptors other than CB1R and CB2R. Naturally-occurring phytocannabinoids such as CBD and THC have affinity for the transient receptor potential vanilloid 1 (TRPV1) and transient receptor potential ankyrin 1 (TRPA1) Ca 2+ channels, orphan GPCRs, and serotonin receptors, among others 2 . Similarly, the SCRA WIN55,212-2 has been shown to desensitize TRPV1 and TRPA1 to produce analgesia 27 . Although it is not known whether the SCRAs studied here are ligands for these other 'non-canonical' cannabinoid receptors, modulation of these receptors' activity by SCRAs could affect nociception, inflammation, behaviour, mood, and appetite 27 and therefore explain some of their in vivo activity beyond CB1R and CB2R.
A select few of the SCRAs assayed here produced potent and supramaximal responses in the βarrestin2 recruitment assay at both CB1R and CB2R. Specifically, SCRAs with indazole cores rather than indole cores and halogen-substituted butyl or pentyl groups, such as 4-fluoro MDMD-BINACA, were the most-potent and most-efficacious agonists of βarrestin2 recruitment as CB1R and CB2R. Inclusion of an indole core rather than an  15,[28][29][30] . Further, the indazole core has been shown to have a stabilizing effect on the "twin toggle switch" that affords CB1R-Gα i interactions 17 . It may be that the stabilization of the "twin toggle" in combination with the increased affinity provided by halogen-substituted butyl or pentyl groups promotes sustained receptor activation and therefore, βarrestin2 recruitment 17 . Previous work has found that many SCRAs structurally and functionally related to the subset studied here display ligand bias and functional selectivity for βarrestin2 recruitment relative to the inhibition of cAMP [14][15][16][17][18][19][31][32][33][34] . Our data extend previous findings to suggest that these structural elements specifically confer βarrestin2 bias because these structural differences did not dramatically change potency and efficacy in the cAMP assay. Building from our cell culture data, we observe that the majority of in silico interactions made between SCRAs and either CB1R or CB2R are highly similar, with key homologous residues such as Ser383 7.39 (CB1R)/Ser285 7.39 (CB2R) and His178 2.65 (CB1R)/His95 2.65 (CB2R) present in our models and crystal structures of these receptors bound to SCRAs 20,21,24,26 . The binding pocket's shape and position are retained and consistent for SCRAs at both CB1R and CB2R, with slight differences in amino acid interactions observed between cAMP-biased and βarrestin2-biased compounds (Tables 2 and 3). Perhaps most-intriguing is the observation that interaction energies differed depending on the G protein's inclusion or exclusion in our model. The highly βarrestin2-biased ligand 4-fluoro MDMB-BINACA incurred a 10.56 kcal/mol binding energy penalty in the presence of the G protein (Table 4). In contrast, the G protein-biased JWH 018 2′-naphthyl-N-(3-methylbutyl) isomer gained 3.78 kcal/mol in interaction energy in the presence of the G protein (Table 4). Although these static in silico models cannot recapitulate the dynamic interactions occurring between ligand and receptor in vivo, they do add support to the growing body of evidence for unique binding interactions for βarrestin2-biased ligands at the cannabinoid receptors 29,30 .
How do the kinetics of binding differ between the SCRAs studied here and the more benign partial CB1R and CB2R agonists such as THC, AEA, and 2-AG? Data presented here suggest that despite many SCRAs being full agonists, they incompletely displace CP55,940 in vitro. In silico modelling of CB1R and CB2R indicates that while SCRAs occupy a similar site in the cannabinoid receptors, specific amino acid residue interactions differ from ligand-to-ligand. SCRAs such as the indole-and indazole-core ligands largely focused on here appear to interact with the cannabinoid receptors via unique receptor confirmations that differentially affect receptormediated signaling. Specific SCRAs with isothiocyanate substitutions have been designed to covalently bind CB1R (e.g. AM6538 and AM841) 25,26,35 , but no evidence to date indicates that other SCRAs covalently bind the www.nature.com/scientificreports/ cannabinoid receptors [14][15][16][17][18][19][31][32][33][34] . The unique binding modes of SCRAs may dramatically increase on-rate or slow off-rate of ligands from the receptors and consequently alter the pharmacodynamics of SCRAs 34,36 , such as the βarrestin2 bias and catalepsy observed with low concentrations of 4-fluoro MDMB-BINACA. This work and previous studies point to specific amino acids within CB1R and CB2R that may dictate ligand bias and kinetics, such as Ser383 7.39 and His178 2.65 (CB1R) 20,21,[24][25][26] . Planned future in silico and in vitro studies will explore the on-and off-rate binding kinetics of SCRAs with mutagenized and wild-type receptors. The translation of observed in vitro bias to in vivo effects remains a key limitation in the study of biased agonism 37,38 . Here, we observed that the highly βarrestin2-biased ligand 4-fluoro MDMB-BINACA was the most potent and efficacious agonist of catalepsy and hypothermia among the cannabinoids tested. In contrast, the G protein-biased JWH 018 2′-naphthyl-N-(3-methylbutyl) isomer was less efficacious in the catalepsy assay, did not produce hypothermia, and reduced anxiety-like behaviours in the EPM. Two related compounds, MDMB-FUBINACA and 5-fluoro AMB, have similarly been shown to produce profound hypothermic responses in Long-Evans rats at these doses 15 . Structural analogues of these ligands, such as XLR-11 39 , 5-fluoro CUMYL-P7AICA 16 , CUMYL-4CN-BINACA 40 , and AB-CHMINACA, and AB-PINACA 41 have shown similar cannabimimetic effects in rodent models-including profound hypothermia and pro-convulsant effects. Here, in vitro and in silico experiments utilized human CB1R and CB2R, whereas in vivo experiments were conducted in mice, which represents a limitation of our study. Subtle differences in CB1R and CB2R-dependent βarrestin recruitment have been documented between species 42 ; however, ligand affinity and activity, and sequence similarity have been shown to be highly similar between human and mouse CB1R and CB2R 11,18,19,22 . Another possible reason for varying activity in vivo could be differences in non-specific plasma protein binding among SCRAs. However, pharmacokinetic analyses of AM 2201 and JWH-018 N-(5-hydroxypentyl) in rats show that these compounds have similar plasma protein binding, rates and extents of metabolism, and their efficacy correlates with plasma concentrations 43 . Similarly, a recent meta-analysis in humans found no significant difference in free plasma serum concentrations of 65 SCRAs across 117 reports published between 2009 and 2020 [reviewed in 44 ]. As a limitation of our data, equivalent mouse pharmacokinetic data do not exist for 4-fluoro MDMB-BINACA and JWH 018 2′-naphthyl-N-(3-methylbutyl) and such data are beyond the scope of the present study. However, existing data presented here and elsewhere indicate that differences in SCRAs activity are pharmacodynamic rather than pharmacokinetic [45][46][47] . Although previous studies have not correlated the βarrestin2 bias of such ligands with specific in vivo effects, it is possible that βarrestin2-biased signaling may selectively enhance hypothermic and cataleptic responses relative to other cannabimimetic effects such as anti-nociception. Additional research is required to test this hypothesis directly.
Is βarrestin2 bias responsible for the on-target pharmacodynamic toxicities associated with SCRAs 39,41,48 ? Specifically, are hypothermia and catalepsy the functional repercussions of supramaximal βarrestin2 recruitment via the cannabinoid receptors? Increasing data from other GPCR systems, such as the µ-opioid receptor and serotonin 5HT2A receptor, indicate that undesirable effects of receptor activation may be the result of functional selectivity toward βarrestin2 and away from G protein-mediated signaling cascades 37,38,49 . In the case of the µ-opioid receptor, respiratory depression, constipation, and possibly addictive potential, appear to be βarrestin2-dependent 38 . For 5HT2A, hallucinogenic effects of lysergic acid diethylamide (LSD)-as modelled in rodents-may be βarrestin2-dependent 49 . As with the second question posed above, understanding the molecular mediators of the SCRAs' toxicities may aid in limiting their harm. Our work has previously shown that βarrestin-biased ligands produce CB1R internalization, overall CB1R downregulation, and reduced cell viability in cultured neurons 11,18 . Therefore, it is critical that we determine whether supra-physiological βarrestin2 recruitment by some SCRAs 15,16,40 augments CB1R internalization and downregulation, and reduces cell viability 11,18 . Whether βarrestin2 bias dictates the differential signaling of SCRAs in vivo in both acute and chronic treatment paradigms must also be determined. In human case studies, illicit SCRA use confirmed by blood analysis is linked to delirium, agitation, lethargy, immobility, and tachycardia and associated cardiovascular events [reviewed in 44,50 ]. These clinical data include compounds we observed to be G protein-biased (e.g. AM 2232) and βarrestin2-biased (e.g. 5-fluoro NPB-22), but the relationship between SCRA potency and toxicity in humans is uncharacterized at this time. We observed profound immobility with the βarrestin2-biased SCRA 4-fluoro MDMB-BINACA, but did not assess cardiovascular physiology in mice. Understanding such adverse effects, and differentiating them from the partial agonism of plant-derived phytocannabinoids such as THC, may aid in our understanding of the unique toxicities associated with SCRAs 39,41 . Future mouse studies will be conducted utilizing wild-type and βarrestin2 knockout mice to determine the consequence of SCRA βarrestin2 bias on neuronal viability, desensitization, cognition, movement, and cardiac function; as has been done for both the study of novel cannabinoids and for understanding the functional consequences of ligand bias at other receptor systems in the past 37 .
The cannabinoid receptors are capable of accommodating a wide array of structurally diverse agonists. This study describes some of the SAR of SCRAs in vitro, in silico, and in vivo at CB1R and CB2R. Future extensions of this research will need to explore a greater variety of cannabinoid-targeted receptors, such as those mentioned above. As data on the SCRAs continue to accumulate, we hope to improve our ability to predict the pharmacodynamic profiles of SCRAs based on their chemical structures.

Methods
Compounds. All compounds were purchased from Cayman Chemicals (Ann Arbor, MI) with the exception of ∆ 9 -tetrahydrocannabinol, which was purchased from Toronto Research Chemicals (Toronto, ON). [ 3 H] CP55,940 (174.6 Ci/mmol) was obtained from PerkinElmer (Guelph, ON). All other reagents were obtained from Sigma-Aldrich (Oakville, ON) unless specifically noted. Compounds were dissolved in DMSO (final con- www.nature.com/scientificreports/ Council on Animal Care 56 and approved by the Animal Research Ethics Board and the Scientific Merit Review Committee for Animal Behaviour at the University of Saskatchewan. The following steps were taken to ensure our study was in keeping with the ARRIVE guidelines: (1) power analyses were conducted to determine the minimum number of animals required for the study; (2) animals were purchased-rather than bred-to limit animal waste; and (3) all assessments of animal behaviour were made by individuals blinded to the treatment group 56,57 . Modified tetrad testing (catalepsy, body temperature, tail flick assay) was conducted according to previously described methods, which are summarized briefly here 19 . Time points for testing were chosen based on previously published findings 19,22,23 and a time course experiment conducted with 10 mg/kg CP55,940 ( Supplementary  Fig. 3). Catalepsy was assessed in the ring holding assay 5 min following injection with mouse forepaws clasped to a 5 mm ring positioned 5 cm above the surface of the testing space. The length of time the ring was held was recorded (s) to a maximum time of 60 s (i.e. MPE = 60 s). Internal body temperature was measured 15 min after injection. Anti-nociception was determined by assessing tail flick latency 20 min after injection. Mouse tails were placed ~ 1 cm into 52 ± 2 °C water and the time until the animal removed its tail was recorded as tail flick latency (s) up to 20 s (i.e. MPE = 20 s).
EPM was conducted according to methods adapted from Marks et al. 58 . The EPM was built from a plywood base painted blue with blue corrugated plastic floor and walls. The floor of the maze was composed of perpendicular interlocking arms (110 cm long × 10 cm wide) with two arms having 45 cm walls (closed arms) and two arms having no walls and a 0.

Statistical analyses. [ 3 H]CP55
,940 radioligand saturation binding data are provided as raw counts per minute (cpm) bound and fit to a one site total binding non-linear gression (GraphPad) (Supplementary Fig. 1).
[ 3 H]CP55,940 radioligand competition binding data are provided as % change from maximal 3 H bound (i.e. 100%). Data for HitHunter cAMP and PathHunter βarrestin2 data are shown as % of maximal CP55,940 response (i.e. 100%). Estimates of K i , EC 50 , E min , and E max were determined using non-linear regression with variable slope (four parameters) (GraphPad, Prism, v. 8.0). The operational model of Black and Leff 59 was used to estimate bias (∆∆LogR) with CP55,940 as the reference agonist 11,18,19 . One-way analysis of variance (ANOVA), followed by Tukey's post-hoc test, was used for statistical analyses (p < 0.05 determined to be significant); and Bartlett's test was used to confirm homogeneity of variance (GraphPad). Values are presented as the mean ± the standard error of the mean (SEM) or 95% confidence interval (CI), as indicated in tables and figure legends.