Ligand-directed bias of G protein signaling at the dopamine D₂ receptor

SUMMARY

G protein Coupled Receptors (GPCRs) represent the largest family of drug targets. Upon activation GPCRs signal primarily via a diverse set of heterotrimeric G proteins. Most GPCRs can couple to several different G protein subtypes. However, how drugs act at GPCRs contributing to selectivity of G protein recognition is poorly understood. Here, we examined the G protein selectivity profile of the dopamine D₂ receptor (D₂), a GPCR targeted by antipsychotic drugs. We show that D₂ discriminates between 6 individual members of the G_i/o family and its profile of functional selectivity is remarkably different across its ligands which all engaged D₂ with a distinct G protein coupling pattern. Using structural modeling, receptor mutagenesis and pharmacological evaluation, we identified residues in the D₂ binding pocket that shape these ligand-directed biases. We further provide pharmacogenomic evidence that natural variants in D₂ differentially impact its G protein biases in response to different ligands.

eTOC Blurb

Moo et al. report that the dopamine D2 receptor engages different G proteins when activated by dopamine and antipsychotics. They elucidate the structural mechanism underlying G protein bias and show how it can be altered by natural variants and rationally designed mutations.

Graphical Abstract

INTRODUCTION

G protein-coupled receptors (GPCRs) form the largest class of receptors that recognize diverse types of ligands, including ions, small molecules and peptides, linking them to downstream signals (Alexander et al., 2019). They are prominent drug targets, responsible for the therapeutic effect of ~30–40% of approved drugs (Hauser et al., 2017; Hauser et al., 2018; Insel et al., 2019; Santos et al., 2017). All canonical GPCRs transduce their signals by activating heterotrimeric G proteins that entails the release of the GTP-bound Gα subunit from the inseparable Gβγ complex (Flock et al., 2015; Olsen et al., 2020; Wettschureck and Offermanns, 2005). Mammalian genomes encode a diverse set of 16 Gα proteins, divided into four major families based on sequence homology and function: G_i/o, G_s, G_q/11 and G_12/13 which uniquely regulate downstream effectors (Hermans, 2003; Milligan and Kostenis, 2006). It is now well-established that individual GPCRs can couple to multiple different Gα proteins and that specificity of this recognition is the key parameter defining GPCR properties (Inoue et al., 2019; Masuho et al., 2015b; Michel and Charlton, 2018). In addition, GPCRs can signal by engaging β-arrestin adaptors that upon activation scaffold a variety of signaling molecules also generating cellular responses (Shenoy and Lefkowitz, 2011; Shukla et al., 2011; Wootten et al., 2018).

One of the most exciting developments in understanding GPCRs has been the realization that they can differentially route the signals depending on the nature of the ligand engaged, a phenomenon termed biased signaling or functional selectivity (Kenakin, 2011, 2019; Spongier et al., 1993; Urban et al., 2007). Biased agonists are theorized to stabilize different receptor conformations, which exhibit different affinity for signal transducers to produce distinct cellular responses (Kenakin and Miller, 2010; Michel and Charlton, 2018; Smith et al., 2018; Wootten et al., 2018). The concept of biased signaling offers an attractive possibility for achieving selective pharmacological modulation of GPCR signaling outcomes (Hoffmann et al., 2008). This may allow to dissociate desirable therapeutic effects associated with favorable signaling pathways from the side-effects driven by unfavorable signals, a concept fueling an immense drug discovery effort for identifying and characterizing GPCR ligands with biased signaling properties (Kenakin, 2019; Kenakin and Christopoulos, 2013; Violin et al., 2014).

The best-known example of biased signaling involves differential recruitment of β-arrestin vs. activation of G proteins, a discrimination that many GPCRs have been shown to be capable of (Rajagopal et al., 2010; Reiter et al., 2012; Wacker et al., 2017). Exploiting this phenomenon, a number of synthetic biased ligands has been developed for several receptors. They are capable of preferential engagement of either β-arrestin or G protein and show substantial promise for enhancing the therapeutic effect window (Becker et al., 2016; Bonifazi et al., 2019; Ryba et al., 2017; Singla et al., 2017; Viscusi et al., 2016). Ultimately, the concept of biased singling can be generalized to differential engagement of various downstream pathways initiated by GPCRs which occurs upon alteration of the balance in activation of different proximal signaling mediators (Urban et al., 2007). Considering that G proteins are universal transducers of signals engaged by most GPCRs in a physiological setting, their differential activation has been suggested to greatly contribute to biased signaling (Kenakin, 2019). Indeed, several examples attest to a possibility that GPCRs can differentially engage G proteins in a biased fashion (Bonifazi et al., 2019; Masuho et al., 2015b; Randáková et al., 2020). However, biased GPCR signaling across G proteins remains poorly explored, especially from the angle of understanding the actions of drugs in clinical use. Furthermore, the impact of genetic variations that prominently affect many GPCRs on their G protein selectivity and signaling bias is largely an uncharted territory (Hauser et al., 2018).

Receptors for dopamine are among the most canonical GPCRs with important roles in physiology and disease (Beaulieu et al., 2015). Dysfunction in dopamine signaling has been linked to numerous neuropsychiatric disorders, including schizophrenia, depression and Parkinson’s disease (Beaulieu and Gainetdinov, 2011). In mammals, there are two subfamilies of dopamine receptors. The stimulatory D₁ class includes dopamine D₁ and D₅, which activate primarily the G_s/olf subfamily of G proteins and signal to increase production of the second messenger, cAMP. The D₂ class comprises inhibitory dopamine D₂, D₃, and D₄ receptors, which couple to the G_i/o proteins and engage several signaling mediators, (Neve et al., 2004). Most members of the G_i/o family, except for sensory subunits Gα_t1, Gα_t2 and Gα_gust, are ubiquitously expressed across the nervous system (Jiang and Bajpayee, 2009; Syrovatkina et al., 2016; Uhlen et al., 2015). However, activation of individual members can produce varying outcomes. For example, Gα_i1, Gα_i2 and Gα_i3 proteins inhibit most adenylyl cyclase (AC) isoforms, reducing cAMP production (Sunahara et al., 1996). In contrast, Gα_o does not directly interact with AC and instead acts as signal modifier via the release of Gβγ subunits which differentially modulate activities of various AC isoforms (Dessauer et al., 2017). Gα_z has a number of unique properties, including extremely long lifetime in the active state, insensitivity to pertussin toxin and ability to stimulate potassium channels (Fields and Casey, 1997; Ho and Wong, 2001).

Among the dopamine receptors, D₂ is a major drug target for the treatment of schizophrenia and many last generation atypical antipsychotic drugs in clinical use act as D₂ partial agonists (Ginovart and Kapur, 2012; Mauri et al., 2014; Miyamoto et al., 2012). Interestingly, these drugs produce a range of the distinct efficacies and side effects. One particularly interesting hypothesis with respect to explaining varying effects relates to D₂ signaling bias (Mailman and Murthy, 2010; Urs et al., 2017). Biased signaling of D₂ has been investigated intensely, in respect to its activation of G protein vs. β-arrestin pathways. Studies suggest that β-arrestin2 recruitment is essential for the antipsychotic activity of D₂ acting compounds, yet there is conflicting evidence in respect to β-arrestin bias and antagonism in effectiveness of psychosis treatment (Allen et al., 2011; Madras, 2013; Masri et al., 2008; Urs et al., 2017). Hence, the link between antipsychotic efficiency and signaling bias of D₂ remains unclear.

In this study, we report that dopamine and several antipsychotics with partial agonism at D₂ are able to induce biases amongst different Gα subtypes at the D₂. We identified the structural determinants in D₂ involved in determining G protein signaling biases induced by various ligands. Finally, we showed that genetic variants of D₂ display biased G protein activation. Taken together, our study highlights the importance of studying the initial step of receptor activation to produce a comprehensive profile of ligand signaling and provide information that can aid in the design of better D₂-targeted therapeutics. These findings also suggest that ligand directed bias across G protein subtypes may be a universal phenomenon applicable to many other GPCRs.

RESULTS

Dopamine activates D₂ with distinct G protein selectivity profile favoring Gα_o.

Previous studies established that D₂ couples exclusively to the Gi/o family of Gα proteins with seemingly varying efficiency across individual subtypes (Beaulieu et al., 2015; Masuho et al., 2015b). To follow up on previously observed qualitative differences in D₂ coupling to G proteins (Masuho et al., 2015b), we quantitatively measured the ability of D₂ to activate different non-sensory Gα_i/o family members: Gα_i1, Gα_i2, Gα_i3, Gα_oA, Gα_oB and Gα_z using a NanoBRET-based biosensor approach (Masuho et al., 2015a)(Fig. 1A). Stimulation of D₂ with a saturating concentration of the endogenous agonist, dopamine, revealed that it differentially activated various G proteins (Fig. 1B). We have previously established equivalence of stoichiometric ratios of all G protein heterotrimers used in this assay as well as specificity of the signal not observed when G protein subunits or receptors are skipped from the transfection (Masuho et al., 2015a). With these assay conditions, we observed the largest BRET signal amplitudes for Gα_oA and Gα_oB followed by Gα_i1–3 and low but significant activation of Gα_z (Fig. 1C). Because the extent of G protein activation measured by the signal amplitude may be influenced by intrinsic G protein properties, we focused our subsequent analysis on evaluating the kinetics of the BRET signal generation (Fig. 1D). The onset kinetics reflects the direct measure of the catalytic reaction of G protein activation and thus provides a more accurate measure of GPCR activity. Using exponential fitting of the rising phase of the BRET response we found that application of saturating dopamine concentration activated Gα_oA and Gα_oB the fastest (2.43 ± 0.27 s⁻¹ and 2.28 ± 0.15 s⁻¹, respectively), followed by the markedly slower activation of Gα_i1–3 proteins (Gα_i1, 1.12 ± 0.01 s⁻¹; Gα_i3, 0.74 ± 0.04 s⁻¹; Gα_i2, 0.58 ± 0.06 s⁻¹) and very slow Gα_z activation (0.03 ± 0.01 s⁻¹).

Figure 1. Activation profile of Gai/o proteins by D₂ stimulated by dopamine.

(A) Schematic representation of the NanoBRET assay used to measure D2R G protein activation. Gβγ is tagged with Venus and masGRK3ct is tagged with Nano luciferase (Nluc). Agonist stimulation of D2 leads to dissociation of Gαβγ trimer. Gβγ-Venus then associates with Gβγ effector mimetic masGRK3ct-Nluc to product BRET response. Different Gα responses are recorded by transient transfection with Gα of choice. (B) Representative traces of the effect of 100 μM dopamine on G_i1, G_i2, G_i3, G_oA, G_oB, G_z activation and no G protein transfected by D₂ measured over 40 seconds. Four independent experiments were performed in triplicates. (C) Maximum BRET response of dopamine for D2R individual G protein response. (D) Activation kinetics of dopamine for D2R individual G protein response. (E) Dopamine concentration-response curves for D2R Gi/o protein activation. Three independent experiments were performed in duplicates. (F) Potency (pEC₅₀) of dopamine at D2R individual G protein response. (G) Correlation plot showing the relationship between D activation rate and F potency of dopamine G_i/o activation. (H) The differences in activation rate of G_i/o proteins are illustrated in web plot constructed from data in D. * denotes significantly different when compared to value for G_i1, * denotes significantly different when compared to value for G_i2, * denotes significantly different when compared to value for G_i3, * denotes significantly different when compared to value for G_oA, * denotes significantly different when compared to value for G_oB, * denotes significantly different when compared to value for G_z, p<0.05, one-way ANOVA with Tukey’s post hoc test. The values in B-G are expressed as mean ± SEM. See also Figure S1.

We have previously noted that G protein activation kinetics can be used to approximate the potency of GPCR responses (Masuho et al., 2015b), suggesting that dopamine-stimulated D₂ activates various Gα_i/o subtypes with different potencies. To directly test this possibility, we conducted concentration-response studies measuring BRET response amplitudes mediated by individual Gα_i/o proteins at different dopamine concentrations (Fig. 1E). The analysis of the data showed that potencies of the dopamine responses mediated by different Gα_i/o subtypes also varied and had the same rank order as kinetic differences. Namely, the Gα_o signaling was the most sensitive, followed by Gα_i1–3 and least response sensitivity was observed for the Gα_z (Fig. 1F). Furthermore, we found a nearly perfect correlation (R² = 0.93) between activation rates and pEC₅₀ values for dopamine mediated by different Gα_i/o subtypes (Fig. 1G). Importantly, the Gα_oA-B > Gα_i1–3 > Gα_z rank order in activation rates was maintained across dopamine concentrations (Fig. S1). As the concentration of dopamine declined to the lowest range, D₂ lost coupling to Gα_z first, followed by abolished coupling to Gα_i1–3 while maintaining detectable activation of Gα_oA-B. These observations indicate that potency of D₂ responses to dopamine varies depending on the Gα subtype transducing the signal and stems from differences in efficiencies with which D₂ can catalyze the activation of its individual G protein substrates (Fig. 1H). Therefore, dopamine induces a distinct signaling profile at D₂ with differential selectivity amongst the members of the Gα_i/o family.

Differential ligand-directed G protein biases at the D₂.

Given the multitude of small-molecule agonists developed for D₂ and their documented differences in signaling properties such as G protein vs. β-arrestin bias (Allen et al., 2011; Bonifazi et al., 2019; Free et al., 2014), we next investigated if they have different Gα selectivity bias of D₂. To address this experimentally, we evaluated a panel of D₂-targeting drugs with divergent chemical scaffolds (Fig. 2A; Table S1). When used at the saturating concentrations, all ligands produced smaller responses as compared to dopamine, with the maximum BRET amplitudes ranging from 14% to 90% of dopamine response consistent with their partial agonism (Fig. S2). As expected, we observed that the rank order of efficacy defined by maximal BRET amplitudes of Gα_oA-B > Gα_i1–3 > Gα_z remained equivalent to dopamine responses consistent with the notion that response efficacies are mostly driven by intrinsic Gα subtype properties rather than their activation by D₂.

Figure 2. Antipsychotics show distinct Gi/o protein activation profile.

(A) The ability of five antipsychotics with divergent chemical scaffolds (MLS1547, bifeprunox, cariprazine, aripiprazole and brexpiprazole) to activate G_i1, G_i2, G_i3, G_oA, G_oB and G_z at D₂ were measured over 60 seconds. (B) 100 μM of each agonist was used. The data was normalized to the maximal response induced by the agonist at that pathway. Four independent experiments were performed in triplicates. (C) Activation kinetics of agonists for D2R individual G protein response. * denotes significantly different when compared to value for G_i1, * denotes significantly different when compared to value for G_i2, * denotes significantly different when compared to value for G_i3, * denotes significantly different when compared to value for G_oA, * denotes significantly different when compared to value for G_oB, * denotes significantly different when compared to value for G_z, p<0.05, one-way ANOVA with Tukey’s post hoc test. + denotes significantly different when compared to corresponding values for dopamine, p<0.05, two-way ANOVA with Dunnett’s post hoc test. The values are presented as mean ± SEM. (D) The differences in activation rate of G_i/o proteins are illustrated in web plot constructed from data in C. The values are represented as mean only. (E) The 1/τ activation rates from C were normalized to reference agonist dopamine, and then normalized to reference pathway GoA activation to obtain 1/τ activation rate ratio. The values are showed in web plots. Open circles indicate significant differences when compared to 1/τ activation rate of dopamine at the same pathway, p<0.05, determined by one-way ANOVA with Tukey’s post hoc test. See also Figure S2, Table S1.

In contrast, analysis of the activation kinetics (1/τ), as the main parameter defining catalytic D₂ action and its response potencies revealed substantial heterogeneity in the behavior of the ligands, which displayed distinct activation profile of Gα subtypes (Fig. 2B). The response profile triggered by MLS1547 largely resembled that of dopamine with all Gα_i/o subtypes being engaged (Fig. 2C, D). It still activated Gα_o better than Gα_i, however the extent of this preference was diminished relative to dopamine. Bifeprunox was also able to activate all Gα_i/o subtypes, however it no longer favored Gα_oA-B over Gα_i1–3 (Fig. 2C, D). Similar lack of Gα_i vs. Gα_o preferences was exhibited by cariprazine and aripiprazole, which additionally lost their ability to activate Gα_z. Finally, brexpiprazole showed an extreme preference for both Gα_o isoforms and was unable to activate any other G proteins (Fig. 2C, D). None of these agonists showed G protein activation in the absence of transfected D₂ (Fig. S2) showing that the effects are driven by changes in the D₂ properties.

To probe whether G protein biases seen with different ligands are related to the induction of various agonist states we evaluated the effects of antipsychotics in the antagonist mode studying their ability to inhibit signaling via Gα_i1 or Gα_oA initiated by dopamine. As a reference, we used a canonical D₂ antagonist, haloperidol, which showed equivalent potency in inhibiting both Gα_i1 and Gα_oA signaling (Fig. S3). All representative antipsychotics used in these experiments behaved similar to haloperidol and were equipotent in inhibiting D₂ signaling via both G proteins, suggesting that their biased signaling properties may indeed be related to the induction of specific agonist states.

In order to determine whether the observed changes in G protein selectivity truly reflect biases in G protein activation, we compared the response profiles induced by antipsychotics in reference to dopamine selectivity profile (Fig 2E), an approach commonly used to estimate biased agonism (Kenakin et al., 2012). In this analysis, the coupling profile elicited by dopamine is set to an equal value for each G protein and deviations produced by other ligands become apparent as increases or decreases in values relative to the dopamine. Analysis of the data showed that different agonists indeed generated distinct fingerprints of G protein selectivity different from the profile triggered by dopamine. The unifying theme of these changes was a modulation of either Gα_i or Gα_o selectivity of D₂. Overall, these results show that D₂ exhibits a clear ligand-directed bias in the activation of G proteins.

Dopamine-induced G protein bias of D₂ is determined by ligand-binding residues.

Differences in G protein bias of D₂ observed from ligands with varied molecular structures suggest that interactions with different residues in the binding pocket can contribute to selective recognition of different Gα_i/o subtypes. To test this hypothesis, we sought to identify residues determining Gα selectivity bias by modeling of ligand-receptor structure complexes constructed from induced-fit docking in an active state homology model of D₂ generated with the GPCRdb pipeline (Pandy-Szekeres et al., 2018) (Fig. 3). All five antipsychotics form the canonical salt-bridge to D114^3×32 (GPCRdb generic residue numbering scheme (Isberg et al., 2015) in superscript) and aromatic edge-to-face π-π interactions to F389^6×51 and F390^6×52 (Fig. S4A); all of which are conserved residues and interactions of aminergic ligand-receptor complexes (Peng et al., 2018). All ligands except MLS1547 share binding mode in the dopamine-binding (orthosteric) pocket (Vass et al., 2019; Wang et al., 2017) lined by side chains of eight interacting residues: V115^3×33, C118^3×36, I184^45×52, S193^5×43, S197^5×461, H393^6×55, Y408^7×34, T412^7×38 and Y416^7×42. In contrast, all ligands but cariprazine display two spatially different binding modes in sites extending towards Y408^7×34 in TM7 or F110^3×28 in TM3 and capped by W100^23×50 in the first extracellular loop (ECL1) (Fig. 3 and Fig. S4B–C). Notably, MLS1547 flips not only horizontally, but also vertically, in the two best scoring binding poses. Our selection is reinforced by the experimental evidence that residues S193^5×43 and S197^5×461, which are conserved in all dopamine receptor subtypes, influence agonist binding and efficacy (Chemel et al., 2012; Cummings et al., 2010; Fowler et al., 2012; Sartania and Strange, 1999) and that residue H393^6×55 influences D₂ signaling bias and efficacy (Tschammer et al., 2011).

Figure 3. Putative binding modes of antipsychotics in D₂.

One (cariprazine) or two proposed binding modes for the five studied antipsychotics from induced-fit docking in a D₂ active state homology model. Antipsychotics are displayed as thin sticks with carbon atoms colored as in Figure 2 with a slight nuance difference between the two suggested poses for the same compounds. D₂ is depicted as beige cartoon and thick sticks for the backbone and the carbon atoms of residues selected for mutagenesis. Other atoms have the following colors: oxygen – red, nitrogen – blue, sulphur – yellow and chlorine – green. The receptor is viewed from TM1+2 and, for clarity reasons, these two helices together with hydrogen atoms are not displayed. Additionally, only one receptor conformation is depicted, though, induced-fit docking produces varying amino acid rotamers corresponding to individual binding poses. GPCRdb generic numbers accompany the D₂ sequence numbers in superscript (Isberg et al., 2015). See also Figure S4, Table S4. Figure prepared with the PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC.

To evaluate their roles as determinants for ligand-dependent biased signaling, we designed 17 single-point mutants of the 11 antipsychotics interacting residues and evaluated the effect of the mutations on dopamine-induced activation of representative members along the axis of bias: Gα_i1 and Gα_oA (Fig. 4). We found that all D₂ mutants were present on the surface in levels comparable to the wild-type receptor (Fig. S5), ensuring that any alterations in their function would not be due to differences in expression or localization. Most of the 17 mutations significantly slowed the G protein activation kinetics of D₂ when activated by dopamine (Fig. 4A–C; Table S2). Nine mutations affected Gα_i1 activation while 13 slowed activation of Gα_oA. Three mutations (I184^45×52F, Y408^7×34F and T412^7×38V) had no significant impact on the activation of either Gα, while one mutation (S197^5×461A) completely abrogated the functional activity of D₂ on both. In order to evaluate whether the mutations affected Gα_o vs. Gα_i signaling balance thereby changing receptor preferences we calculated the G protein bias (Fig. 4E). With this metric we found that ten of the mutations significantly affected selectivity of D₂ diminishing its preference for Gα_oA over Gα_i1. The degree of this effect varied amongst the affected mutants and in two cases (S193^5×43A and Y416^7×42W) completely eliminated the Gα_oA preference seen in the wild-type D₂ making respective receptors equipotent in activation of both the Gα_i and Gαo branches. Dopamine binding affinity was only affected by the mutations, S197^5×461A and Y416^7×42W (Table S2). In summary, we uncovered a range of spatially and physicochemically different determinant residues for the Gα_o versus Gα_i protein selectivity of dopamine-activated D₂.

Figure 4. D₂ mutants identified using structural modeling display dopamine-induced G_i/o proteins bias.

(A) Snake plot of D2R with mutations studied highlighted in their respective colors. (B) The ability of 100 μM dopamine to activate G_i1 and G_oA at D2R mutations were measured with BRET G protein activation assay over 60 seconds. The data were normalised to the maximal response induced by dopamine at G_oA activation. Activation kinetics of dopamine for D₂ WT and mutants (C) G_i1 and (D) G_oA protein response. * denotes significantly different values when compared to D₂ WT, p<0.05, determined by one-way ANOVA with Tukey’s post hoc test. (E) G_oA - G_i1 bias was determined by normalising 1/τ of G_oA to G_i1 activation pathway. Open circles denote significantly different values when compared to D₂ WT, p<0.05, determined by one-way ANOVA with Tukey’s post hoc test. Data are presented as mean ± SEM of three experiments performed in triplicates. ND: not determinable. See also Figure S5, S6, Table S2.

Determinants underlying ligand-induced alteration of G protein biases.

We next evaluated if mutation of the four ligand interacting residues in the two extended binding sites would influence, or even switch, the Gα_i1 vs. Gα_oA bias of the most structurally distinct ligands we evaluated: MLS1547, bifeprunox and aripiprazole on D₂ (Fig. 5; Table S3). Remarkably, the W100A^23×50, F110A^3×28, Y408A^7×34 and S193A^5×43 mutations exhibited a diverse spectrum of effects on the different ligands and G proteins. MLS1547-D₂ signaling was affected significantly by all four mutations, but in different ways, as W100A^23×5, S193A^5×43 and Y408A^7×34 selectively reduced Gα_oA activation, whereas F110A^3×28 abolished signaling through both Gα_i1 and Gα_oA. Aripiprazole-D₂ signaling changes were more selective, as only the S193A^5×43 and F110A^3×28 mutations had effects which again differed – augmenting and diminishing the responses of the two Gα subtypes, respectively. Finally, bifeprunox-D₂ signaling was only influenced by the Y408A^7×34 mutation, which results in a slight reduction specific for Gα_oA.

Figure 5. Antipsychotics display G_i/o protein bias at D2R mutants.

(A) The ability of 100 μM of MLS1547, bifeprunox and aripiprazole to activate G_i1 and G_oA at D₂ mutations were measured over 60 seconds. The data were normalized to the maximal response induced by dopamine at G_oA activation. Activation kinetics of antipsychotics for D2R WT and mutants (B) G_i1 and (C) G_oA protein response. * denotes significantly different values when compared to D₂ WT, p<0.05, determined by one-way ANOVA with Tukey’s post hoc test. (D) G_oA - G_i1 bias was determined by normalizing 1/τ of G_oA to G_i1 activation pathway. Open circles denote significantly different values when compared to D2 WT, p<0.05, determined by one-way ANOVA with Tukey’s post hoc test. Data are presented as mean ± SEM of three experiments performed in triplicates. ND not determinable. # denotes that the value was constrained to G_oA bias because G_i1 pathway was not activated. See also Table S3.

To better understand the impact of mutations on the agonist bias of Gα preferences we calculated G protein bias (Fig. 5D). This analysis revealed that mutations produced selective bidirectional effects on Gα selection by D₂ in an agonist selective manner. For instance, both W100A^23×50 and Y408A^7×34 flipped the selectivity of D₂ to a strong preference for Gα_i1 instead of Gα_oA when stimulated with MLS1547. In contrast the Y408A^7×34 mutation made aripiprazole selective for Gα_oA at the D₂ by abolishing Gα_i1 coupling. Similar results were observed using an orthogonal assay platform, the TRUPATH technology (Fig. S6) (Olsen et al., 2020). Taken together, these observations identified structural determinants in the dopamine-binding site of D₂ that uniquely modulate the G protein bias of antipsychotics.

Genetic variants influence agonist-directed bias in G proteins selectivity of D₂.

Since natural GPCR mutations have been linked to different drug therapy efficacy, toxicity and diseases (Hauser et al., 2018; Insel et al., 2007), we next tested if reported naturally occurring genetic variations in D₂ alter the G protein selectivity of dopamine and antipsychotics causing or perturbing the medication of disease, respectively. V154^4×44I (rs10489422; Allele Frequency: 0.00004602), located at TM4 close to the intracellular domain of the receptor, has been associated with myoclonus dystonia without noting any functional alterations of traditional measures for agonist or antagonist affinities and functional responses (Klein et al., 1999; Klein et al., 2000). S311^ICL3C (rs1801028; Allele Frequency: 0.02572, the most frequently observed missense mutation in D₂) (Karczewski et al., 2020), in the middle of intracellular loop 3 (ICL3), has been associated with an increased risk for schizophrenia (Arinami et al., 1996; Glatt and Jonsson, 2006; Yao et al., 2015) and described to lower the affinity of dopamine and diminish its efficacy in inhibiting cAMP synthesis (Cravchik et al., 1996). We have then examined the signaling properties of both mutants in our functional assays determining the impact on the Gα biases. The V154^4×44I variant selectively increases dopamine- and aripiprazole-D₂ activation of Gα_i1 and augments MLS1547-D₂ Gα_oA signaling (Fig. 6A, B). Thus, this variant produces two types of D₂ signaling changes that switch from Gα_o to Gα_i selectivity across all three agonists. In contrast, the S311^ICL3C variant only significantly changes the response of aripiprazole for which it augments both Gα_i1 and Gα_oA signaling. Evaluation of the signaling bias indicated significant change with Gα_i over Gα_o selectivity induced by both mutations in response to aripiprazole, and S311C showed preference for Gα_i signaling when activated by dopamine (Fig. 6C). We conclude that naturally occurring variants in D₂ can change the balance of its G protein preferences in an agonist selective manner. These changes may lead to altered drug responses with physiological implications.

Figure 6. Natural genetic variants of D2R show different G protein activation.

Activation kinetics of 100 μM dopamine, MLS1547 and aripiprazole for D₂ WT and the natural genetic variants V154I and S311C at (A) G_i1 and (B) G_oA protein response. * denotes significantly different values when compared to D₂ WT, p<0.05, determined by one-way ANOVA with Tukey’s post hoc test. (C) GoA - Gi1 bias factor was determined by normalizing 1/τ of GoA to G_i1 activation pathway. Open circles denote significantly different values when compared to D2 WT, p<0.05, determined by one-way ANOVA with Tukey’s post hoc test. Data are presented as mean ± SEM of three experiments performed in triplicates.

DISCUSSION

Biased G protein signalling profile of the D₂ dopamine receptor

In this study we show, that the prototypical GPCR, the D₂ dopamine receptor shows prominent ligand-dependent functional selectivity at the level of G protein activation based on its ability to discriminate between individual members of the Gα_i/o subfamily. The endogenous agonist dopamine is capable of triggering D₂ activity on all non-sensory members of the Gα_i/o subfamily: Gα_oA, Gα_oB, Gα_i1, Gα_i2, Gα_i3 and Gα_z but has an intrinsic signaling profile of preferring Gα_o over Gα_i and Gα_z activation. It should be noted that D₂ is exquisitely selective for the Gα_i/o members and does not show appreciable activation of other G protein subtypes (Masuho et al., 2015b) making this investigation an exhaustive profiling of changes in D2-Gα preferences. Our results indicate that dopamine-bound D₂ is functionally selective at the first step of signaling following receptor activation, prior to any consequent downstream second messenger system. This measure of functional outcome allows for direct comparison between diverse G protein pathways, uncovering new bias activities, different from previous bias studies that focused on downstream second messenger readouts (Allen et al., 2011; Chen et al., 2016; Tschammer et al., 2011). Even though the individual Gα_i/o members are classified in the same Gα family, their activation have significantly diverse functional outcomes pertaining to effector engagement selectivity and kinetic profiles (Ho and Wong, 2001; Jiang and Bajpayee, 2009), suggesting that bias within the Gα_i/o subfamily may have physiological and/or therapeutic implications (Anderson et al., 2020).

Our results are consistent with several prior studies that noted differences in G protein activation by the D₂ and its preference for Gα_o (Bonifazi et al., 2019; Klein Herenbrink et al., 2016; Masuho et al., 2015a; Sanchez-Soto et al., 2016). We substantially extended these observations by providing a systematic comparative analysis of all Gα subunits coupled to D₂ including the often neglected Gα_z protein that binds to D2 in vivo (Leck et al., 2006). Uniquely, we also systematically compare the G protein profiles engaged by dopamine to those in response to a structurally diverse class of other D2 ligands to reveal their differential biases. We report that partial agonists with varying degrees of clinical efficacy also have distinct G protein activation profiles compared to the full agonist dopamine. In contrast to the clear preference for Gα_o signaling of dopamine, partial agonists bifeprunox and cariprazine have a more balanced Gα_i/o activation profile. Importantly, we observe that partial agonists do not simply subdue the signaling via the weaker coupled Gα subtypes, which would be expected to proportionally scale down the response to result in a perceived bias due to loss of D₂ coupling to less preferred substrates, a phenomenon observed for μ-opioid (Gillis et al., 2020) and M₁ muscarinic (Masuho et al., 2015b) receptors. Instead, we observe a rank-order change in the D₂ preferences for individual Gα subtypes induced by antipsychotics as compared to dopamine reflecting a true functional bias. Together with previous findings, our results form a comprehensive picture of the G protein biased signaling profile of the D₂ and uncovers its potential for modulation by drugs.

Determinants of G protein discrimination by the D₂ receptor

From the structural standpoint, biased signaling is thought to be underpinned by the ability of ligands to stabilize specific receptor conformation that lead to differential engagement of signaling partners and cellular outcomes (Kenakin, 2011). The partial agonists included in this study are bivalent molecules that are able to simultaneously occupy the orthosteric binding site and one of two spatially distinct secondary extended binding pockets (Free et al., 2014; Klein Herenbrink et al., 2019; Szabo et al., 2014). The nature of the head group can allosterically influence ligand binding mode in both spaces and thus affect the global receptor conformation (Weichert et al., 2015). Besides aripiprazole and cariprazine, all partial agonists have a distinct chemical scaffold with diverse primary and secondary structures. Indeed, molecular dynamics has shown that dopamine and aripiprazole are able to induce different conformations of important GPCR structural motifs, including the extracellular loops, ligand binding sites and intracellular G protein-binding domains, contributing to their biased signaling profiles (Kling et al., 2014).

Our study presents new insights into the structural basis of the identified D₂ biased signaling. Dopamine at the wild type receptor has a clear Gα_o bias over Gα_i subtypes, and based on docking to a D₂ homology model we were able to pinpoint amino acid residues involved in setting this selectivity and overriding it. As only an antagonist bound D₂ structure (Wang et al., 2018) was available when the current project was started, we modelled an active state D₂ receptor to take structural changes in the binding cavity into account. Two additional inactive (Fan et al., 2020; Im et al., 2020) and two active state (Yin et al., 2020; Zhuang et al., 2021) D₂ structures are now available. Comparison of all five structures to our model shows a similar overall location of the binding site residues. Furthermore, the induced-fit docking protocol employed herein compensates for different side chain rotamers making it likely to obtain similar binding poses independent of D₂ structure employed. Nevertheless, mutational experiments based on the model showed that the S193^5×43A mutation in the orthosteric site was able to completely abolish the intrinsic bias of D₂ towards Gα_o indicating its importance in dictating G protein preferences of the receptor. We also find that mutating residues in the extended binding pockets that indirectly coordinate dopamine reduced D₂ bias for Gα_o but did not erase its preference over the Gα_i upon activation by dopamine. This structural information can be utilized to design ligands that differentially engage the residues with the hope of generating small molecules mimicking the signaling profile on the wild-type receptor. Of note, only one of the two proposed binding modes for MLS1547 and aripiprazole provides a structural explanation for the marked effect of the F110^3×28A mutation, which should be taken into account using our results for structure-based design of bias ligands. We envision that the mutants with altered G protein selectivity that we describe could be used for in vivo studies to explore the behavioral consequence of changing the balance of Gα_i/o signaling at D₂.

Significant insights are generated from the analysis of how mutations alter the D₂ responses to antipsychotics introducing another dimension of biased signaling. Previous data showed that aripiprazole displayed bias towards cAMP over pERK1/2 and mutating ligand binding site residues did not cause a significant change in bias (Klein Herenbrink et al., 2019; Szabo et al., 2014). However, we find that aripiprazole, MLS1547 and bifeprunox have drastically different profiles of G protein activation at the D₂ mutants making the ligands functionally selective for Gα_i signaling. Most noteworthy, the F110^3×28A mutation displayed no G protein activation by MLS1547. The residues identified are also able to affect antipsychotic activity, creating different antipsychotic signaling profiles at the mutant receptors. Overall, we were able to identify receptor microdomains that participate in mediating ligand-specific conformations associated with biased signaling, and also detected residues that are not involved in this process.

Pharmacogenomic implications of analyzing D₂ signaling biases

It is interesting to consider our observations with rationally designed mutations in D₂ that altered its G protein selectivity profile from the perspective of naturally occurring variants. It is possible that drug-induced biases can be broadly extended when one considers non-synonymous polymorphisms in D₂. We have evaluated two such representative naturally occurring alterations in D₂, V154^4.44I and S311^ICL3C, and found that they also have vastly different G protein signaling profiles relative to wild type receptor. Generally, dopamine was less biased for the Gα_o pathway, while MLS1547 had enhanced bias towards Gα_o with these variants. Meanwhile, the D₂ genetic variants preferred the Gα_i pathway when the receptor was activated by aripiprazole. Interestingly, S311^ICL3C was found in patients with psychotic affective disorders, making it plausible that people with these mutations could be prescribed antipsychotics as part of their treatment (Arinami et al., 1996). Therefore, with information on more genetic variants, it may be required to examine putative antipsychotics in genotyped-matched clinical trials. Additionally, it would be ideal to study patient’s genotype before initializing treatment to prevent unexpected drug responses and adverse reactions. The outcome highlights the importance of a personalized approach in prescribing medications based on genetic make-up, a concept also applicable to other GPCR variants (Hauser et al., 2018).

Conclusion

In summary, we propose that it is important to consider biased signaling across various Gα subtypes when profiling ligands and genetic variants, and when developing drugs with better therapeutic efficacy and reduced side effects. We provide a framework for the analysis that includes direct evaluation of GPCR selectivity for the spectrum of G proteins using proximal assays, predictive structural modeling and the functional surveillance of ligand panels across receptors carrying mutations and natural variants. With the rising importance of personalized medicine, pharmacogenomic discrimination among related G proteins should provide an additional dimension to the development and prescription of antipsychotics.

STAR METHODS

Resource Availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Kirill Martemyanov (kirill@scripps.edu).

Materials availability

Plasmids generated in this study will be distributed upon request without restriction.

Data and code availability

The source data for raw traces of GPCR responses used for quantitative analysis reported in this study have not been deposited in a publicly available repository because of their trivial nature and custom format they are generated in. They have been archived locally on the institutional cloud service. To request access, please contact the Lead Contact. The paper does not report any original code. Any additional information to reanalyse the data reported in this paper is available from the lead contact upon request.

Experimental Model and Subject Details

Cell lines

HEK293T/17 cells were obtained from ATTC (Manassas, VA) and grown in DMEM supplemented with 10% FBS, minimum Eagle’s medium non-essential amino acids, 1mM sodium pyruvate, and antibiotics (100 units/ml penicillin and 100 μg/ml streptomycin) at 37°C in a humidified incubator containing 5% CO₂.

Method Details

cDNA constructs

Flag-tagged dopamine D2 receptors (NM_000795) containing the hemagglutinin signal sequence (KTIIALSYIFCLVFA) at the N terminus was a gift from Dr. Abraham Kovoor. The pCMV5 plasmids encoding rat Gα_oA, rat Gα_i1, rat Gα_i2 and rat Gα_i3, were gifts from Dr. Hiroshi Itoh. pcDNA3.1+ plasmids encoding Gα_oB (AH002708) and Gα_z (J03260) were purchased from the cDNA Resource Center (www.cDNA.org). Venus 156–239-Gβ₁ (amino acids 156–239 of Venus fused to a GGSGGG linker at the N terminus of Gβ₁ without the first methionine (NM_002074)) and Venus 1–155-Gγ₂ (amino acids 1–155 of Venus fused to a GGSGGG linker at the N terminus of Gγ₂ (NM_053064)) were gifts from Dr. Nevin A. Lambert (Hollins et al., 2009). The masGRK3ct-Nluc-HA and PTX-S1 were reported previously (Gulati et al., 2018; Masuho et al., 2015a; Raveh et al., 2010). HiBiT tagged D₂ and D₂ mutants were obtained from Genscript. TRUPATH assay related cDNA was a gift from Bryan Roth (Addgene kit #1000000163) (Olsen et al., 2020). GenBank accession number for each sequence is given in parentheses.

Cell transfection

On the day of transfection, HEK293T/17 cells were split into 2×10⁶ cells onto 35mm plates in culture medium supplemented with 10μg/ml Matrigel. Four hours later, the cells were transfected using desired constructs (total of 5μg DNA) with Lipofectamine LTX (5μL) and Plus (5μL) reagent. For BRET G protein activation assay, the cells were transfected with the Venus 156–239-Gβ (1), Venus 1–155-Gγ₂ (1), and masGRK3ct-Nluc (1) constructs in addition to D2 WT or mutants (1) and the Gα of interest: GαoA (2), GαoB (1), Gαi1 (1), Gαi2 (2), Gα_i3 (1.5), Gα_z (1.5). PTX-S1 (1) was added to cells transfected with Gα_z to avoid coupling of endogenous Ga_i/o to the receptor (the number in parentheses indicates the ratio of transfected DNA, ratio 1 = 0.21μg). The empty vector pcDNA3.1(+) was used to normalize the amount of DNA to 5μg in each transfection.

BRET G protein activation assay

Agonist-dependent BRET measurements between Venus-Gβ₁γ₂ and masGRK3ct-Nluc were performed as described previously (Masuho et al., 2015a; Masuho et al., 2015b). 16 to 24 hours after transfection, HEK 293T/17 were washed once with BRET buffer (PBS containing 0.1% glucose and 0.5mM MgCl₂) and then were gently detached using BRET buffer. The cells suspension was centrifuged at 500g for five minutes and were resuspended in BRET buffer. Approximately 50,000 to 100,000 cells were added into 96-well flat-bottomed white microplates (Greiner Bio-One). The Nano-Glo Luciferase assay substrate was prepared according to manufacturer’s instructions. The substrate was dissolved in BRET buffer immediately before use and added to the cells at a final dilution factor of 1:250, then a 5 second baseline measurement was recorded. 100μM of agonist was injected into each well and activation of D₂ was measured at 20ms intervals using a LUMIstar plate reader (BMG LabTech). All measurements were performed at room temperature. The BRET ratio was calculated as ratio of light emitted at 535nm by light emitted at 475nm. The average baseline value was subtracted from the agonist-stimulated BRET value to obtain the ΔBRET ratio.

Receptor cell surface expression HiBiT assay

Cell surface expression of D₂ WT and mutants were determined using Promega’s NanoGlo® HiBiT Extracellular Detection System. HEK 293T/17 cells were transfected with 0.21 μg HiBiT-tagged D₂ and incubated overnight. The next day, cells were added at 100,000 cells per well in culture medium without phenol-red into 96-well flat-bottomed white microplates. Nanoglo HiBiT extracellular reagent containing substrate and buffer was prepared per manufacturer’s protocol. 25 μL of reagent was added into each well to a final volume of 50 μL. Luminenscene was measured every 2.5 minutes for 20 minutes using the LUMIstar plate reader to ensure maximum, steady signal was achieved. Data was normalized to mock-transfected cells as background control.

TRUPATH assay

5×10⁶ cells were plated into poly-D-lysine coated 96-well flat-bottomed white microplates (Greiner Bio-One) and transfected using a 6:1:1:1 ratio of receptor:Gα-RLuc8:Gβ:Gγ-GFP2 (0.4μg per 1 ratio). A combination of GαoA-RLuc8, Gβ3 and Gγ8-GFP2, or Gαi1-RLuc8, Gβ3 and Gγ9-GFP2 was used. The next day, culture media was removed and replaced immediately with 80μL of assay buffer (Hank’s balanced salt solution (HBSS) + 20mM HEPES + 1mM CaCl2 + 1mM MgCl2, pH 7.4), and 10μL of freshly prepared 50μM coelenterazine 400a (Nanolight Technologies). After a 5 minutes incubation period, cells were treated with 10μL of ligand addition. Plates were read using a LUMIstar plate reader (BMG Labtech), at integration times of 1 second per well. Plates were read at 1-minute interval, over 10 minutes. Measurements over the read was averaged and used in all analyses. BRET2 rations were calculated as the ratio of the GFP2 emission (510nm) to Rluc8-coelenterazine 400a (395nm) emission.

Radioligand binding assay

7×10⁶ cells were plated onto 150mm dish and transfected using a 1:6 ratio of receptor and polyethyleneimine as transfection reagent (10μg per 1 ratio). 48 hours later, the cells were harvested with PBS containing 2mM EDTA and centrifuged at 500g for 5 minutes. The resulting pellet was resuspended in ice-cold binding assay buffer (20mM HEPES, 100mM NaCl₂, 6mM MgCl₂, 1mM EGTA and 1mM EDTA, pH 7.4) and homogenised using UltraTurrax homogenizer before being centrifuged at 1000g for 10 minutes. After centrifugation, the supernatant was recentrifuged at 30000g for 60 minutes at 4°C. The resulting pellet was resuspended in assay buffer and stored at −80°C. Membrane protein concentration was determined using the BCA protein assay kit (Thermo Scientific).

The radioligand binding assay was performed using [³H]spiperone in a 96-well format. 4nM [³H]spiperone was incubated with 10μg/well protein in a total volume of 200μL binding buffer containing 100μM GppNHp and 0.1% ascorbic acid for 3 hours at −80°C. Non-specific binding was determined using 10μM haloperidol. After the incubation period, the binding reaction was terminated by rapid filtration through GF/C unifilters (PerkinElmer) using a 96-well Packard FilterMate cell harvester (PerkinElmer), followed by 3 washes with 250μL ice-cold 0.9% NaCl. After drying at 50°C, microscintillation fluid was added to the dried filters and the boud radioactivity was measured in a Packard TopCounter microplate scintillation counter.

Ligand-receptor docking and mutation design

The active state D₂ homology model was created with the automated protocol from the G protein-coupled database, GPCRdb.org (Pandy-Szekeres et al., 2018), using the β₁ receptor structure with isoprenaline, (PDB ID 6H7J)(Warne et al., 2019) as the main template. Available active state structures were ranked based on sequence similarity to the 7TM helices plus ECL2 and the mentioned template ranked second was selected due to the high similarity of the co-crystalized ligand to dopamine.

Incorporating isoprenaline from the template into the model and deleting the OH- and iPr-substituents gave the dopamine molecule in the D₂ binding site. All molecular modelling was performed within using applications in the Schrödinger Drug Discovery Software Suite (Schrödinger Release 2019–2, Schrödinger, LLC, New York, NY, 2019). The D₂ structure model was prepared for docking with the Protein Preparation Wizard using default settings. The chemical structures of the five antipsychotics were downloaded from the PubChem database (http://pubchem.ncbi.nlm.nih.gov/ – CIDs: 1093278, 208951, 11154555, 60795 and 11978813)(Kim et al., 2019) and prepared for docking with default settings in LigPrep and docked in the D₂ model using the standard Induced-fit Docking Protocol. The binding site center was defined by the ligand from the main template and the ligand length was set to ≦ 18 Å and XP precision was used in the re-docking step, while all other settings were default. Among the top three ranked output poses with similar IFD scores, four of the antipsychotics produced two markedly different binding modes, which were selected as possible and for cariprazine only the highest-ranking pose was selected.

17 mutations in 11 sequence positions were designed based on the binding modes obtained from the homology model and the induced fit docking. Residue positions predicted to, directly or indirectly, influence ligand binding were selected and mutant amino acids were chosen to remove/introduce interactions or change the shape, size or property of the binding site as described in Table S4.

Quantification and Statistical Analysis

Data Analysis

The maximum BRET amplitude and the rate constants (1/τ) of activation phase were obtained by fitting a single exponential curve to the traces with Clampfit Ver. 10.3 software (Molecular Devices).

Dose response curves were analyzed using Prism 8.4 (GraphPad Software Inc.). The results were fitted using the following three parameter equation:

$$\mathrm{Response}=\mathrm{Bottom}+\frac{\mathrm{Top}-\mathrm{Bottom}}{1+{10}^{(\log {\mathrm{EC}}_{50}-\log \text{[A]})}}$$

Where Top and Bottom represent the maximal and minimal asymptote of the concentration response curve, [A] represents the concentration of agonist and EC₅₀ is the concentration of agonist required to elicit a response halfway between top and bottom. To exclude the impact of system and observational bias effects on the observed agonism of each agonist, the 1/τ activation rate constant of reference agonist, dopamine was normalized to the 1/τ activation rate constant of each agonist. Then the ratio was normalized to 1/τ of reference pathway, Gα_oA activation from 1/τ of the pathways to yield 1/τ activation rate constant ratio. To calculate bias factor of D2 mutants, the 1/τ of GαoA pathway was subtracted by 1/τ of Gαi1 pathway. IC50 values obtained from the radioligand inhibition curves were converted to Ki values using the Cheng and Prusoff equation.

Statistical analysis

One-way ANOVA followed by Tukey’s post hoc test was used for comparing results and determining statistical significance. The statistical analysis performed using GraphPad Prism Ver 8.4.2. Statistical details of the experiments can be found in the figure legend.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
Dulbecco’s modified Eagle’s medium	Thermo Fisher Scientific	Cat #11965–092
Fetal bovine serum	Genesse Scientific	Cat #25–550
Sodium pyruvate	Thermo Fisher Scientific	Cat #11360–070
MEM non-essential amino acids	Thermo Fisher Scientific	Cat #11140–050
Penicillin-streptomycin	Thermo Fisher Scientific	Cat# 15140–122
Matrigel	Corning	Cat# 356230
Lipofectamine LTX and Plus reagent	Thermo Fisher Scientific	Cat# 15338–100
Dulbecco’s phosphate-buffered saline	MilliporeSigma	Cat# D5652
Hank’s balanced salt solution	Thermo Fisher Scientific	Cat# 14175053
Guanosine 5′-[β,γ-imido]triphosphate trisodium salt hydrate	Sigma-Aldrich	Cat# 3655
Dopamine hydrochloride	MilliporeSigma	Cat# H8502
Aripiprazole	Cayman Chemicals	Cat# 19989
Cariprazine	Cayman Chemicals	Cat# 31446
Brexpiprazole	Cayman Chemicals	Cat# 22906
MLS1547	Sigma-Aldrich	Cat# SML1331
Bifeprunox mesylate	Sigma-Aldrich	Cat# SML1670
Haloperidol	Sigma-Aldrich	Cat# H1512
Critical commercial assays
Nano-Glo Luciferase Assay substrate (furimazine)	Promega	Cat# N1120
Nano-Glo HiBiT Extracellular Dectection System	Promega	Cat# N2420
[3H]Spiperone	Perkin Elmer	Cat# NET1187250UC
Coelentrazine 400a	Cayman Chemicals	Cat# 16157
Deposited data
β1 structure	(Warne et al., 2019)	PDB ID 6H7J
Aripiprazole chemical structure	(Kim et al., 2019)	CID: 60795
Cariprazine chemical structure	(Kim et al., 2019)	CID: 11154555
Brexpiprazole chemical structure	(Kim et al., 2019)	CID: 11978813
MLS1547 chemical structure	(Kim et al., 2019)	CID: 1093278
Bifeprunox mesylate chemical structure	(Kim et al., 2019)	CID: 208951
Experimental models: Cell lines
HEK293T/17	ATCC	ATCC: CRL-11268
Recombinant DNA
Plasmid: Flag-D2R	Dr Abraham Kovoor, Univesity of Rhode Island	N/A
Plasmid: GαoA	Dr Hiroshi Itoh, Nara Institute of Science and Technology	N/A
Plasmid: GαoB	cDNA resource center	Cat# GNA0OB0000
Plasmid: Gαi1	Dr Hiroshi Itoh	N/A
Plasmid: Gαi2	Dr Hiroshi Itoh	N/A
Plasmid: Gαi3	Dr Hiroshi Itoh	N/A
Plasmid: Gαz	cDNA resource center	Cat# GNA0Z00000
Plasmid: Venus-156-239-Gβ1	(Hollins et al., 2009)	N/A
Plasmid: Venus-1-155-Gγ2	(Hollins et al., 2009)	N/A
Plasmid: masGRK3ct-Nluc-HA	(Gulati et al., 2018)	N/A
PTX-S1	(Raveh et al., 2010)	N/A
Plasmid: HiBiT-D2R	This paper	N/A
Plasmid: HiBiT-D2R W100A	This paper	N/A
Plasmid: HiBiT-D2R W100L	This paper	N/A
Plasmid: HiBiT-D2R F110A	This paper	N/A
Plasmid: HiBiT-D2R F110L	This paper	N/A
Plasmid: HiBiT-D2R V115I	This paper	N/A
Plasmid: HiBiT-D2R C118S	This paper	N/A
Plasmid: HiBiT-D2R C118V	This paper	N/A
Plasmid: HiBiT-D2R I184F	This paper	N/A
Plasmid: HiBiT-D2R I184S	This paper	N/A
Plasmid: HiBiT-D2R S193A	This paper	N/A
Plasmid: HiBiT-D2R S197A	This paper	N/A
Plasmid: HiBiT-D2R H393N	This paper	N/A
Plasmid: HiBiT-D2R Y408A	This paper	N/A
Plasmid: HiBiT-D2R Y408F	This paper	N/A
Plasmid: HiBiT-D2R Y408V	This paper	N/A
Plasmid: HiBiT-D2R T412V	This paper	N/A
Plasmid: HiBiT-D2R Y416W	This paper	N/A
Plasmid: HiBiT-D2R V154I	This paper	N/A
Plasmid: HiBiT-D2R S311C	This paper	N/A
Plasmid: Gαi1-RLuc8	(Olsen et al., 2020)	Cat# 1000000163
Plasmid: GαoA-RLuc8	(Olsen et al., 2020)	Cat# 1000000163
Plasmid: Gβ3	(Olsen et al., 2020)	Cat# 1000000163
Plasmid: Gγ8-GFP2	(Olsen et al., 2020)	Cat# 1000000163
Plasmid: Gγ9-GFP2	(Olsen et al., 2020)	Cat# 1000000163
Software and algorithms
GraphPad Prism8.4.2	GraphPad software	https://www.graphpad.com/
Clampfit 10.3	Molecular Devices	http://www.moleculardevices.com/products/software/pclamp.html
Microsoft Excel	Microsoft	https://www.microsoft.com/en-ww/microsoft-365/excel
Schrödinger Drug Discovery Software Suite	Schrödinger, LLC	Version 2019–2
PyMOL	Schrödinger, LLC	Version 2.4
Other
GPCRdb	https://gpcrdb.org/	N/A
PubChem	http://pubchem.ncbi.nlm.nih.gov	N/A

SIGNIFICANCE.

The dopamine D2 receptor (D₂) is an extensively studied prototype of G protein-coupled receptors (GPCRs) and is a prominent target for the development of antipsychotic drugs. Although widely used in the clinic, antipsychotics exhibit a range of the poorly explained pharmacological effects while sharing a common property: partial agonism at D₂. The variability of D₂ properties require their empirical prescription and limit the development of precise interventions. Explaining how different drugs with common pharmacological properties elicit different effects also presents a conundrum in basic receptor biology. The discovery of drug-induced functional discrimination among G proteins activated by the D₂ variants reported here should prompt the design of precise pharmacological tools and personalized prescriptions for the treatment of schizophrenia.

Highlights.

• The D₂ dopamine receptor differentially activates various Gαi/o proteins

• Dopamine and antipsychotic drugs produce distinct Gα activation profiles

• Rationally designed mutations in D₂ can override G protein biases

• Genetic variants in D₂ alter its G protein preferences