Development of novel biosensors to study receptor-mediated activation of the G-protein α subunits Gs and Golf

Gαs (Gs) and Gαolf (Golf) are highly homologous G-protein α subunits that activate adenylate cyclase, thereby serving as crucial mediators of intracellular signaling. Because of their dramatically different brain expression patterns, we studied similarities and differences between their activation processes with the aim of comparing their receptor coupling mechanisms. We engineered novel luciferase- and Venus-fused Gα constructs that can be used in bioluminescence resonance energy transfer assays. In conjunction with molecular simulations, these novel biosensors were used to determine receptor activation–induced changes in conformation. Relative movements in Gs were consistent with the crystal structure of β2 adrenergic receptor in complex with Gs. Conformational changes in Golf activation are shown to be similar to those in Gs. Overall the current study reveals general similarities between Gs and Golf activation at the molecular level and provides a novel set of tools to search for Gs- and Golf-specific receptor pharmacology. In view of the wide functional and pharmacological roles of Gs- and Golf-coupled dopamine D1 receptor and adenosine A2A receptor in the brain and other organs, elucidating their differential structure–function relationships with Gs and Golf might provide new approaches for the treatment of a variety of neuropsychiatric disorders. In particular, these novel biosensors can be used to reveal potentially therapeutic dopamine D1 receptor and adenosine A2A receptor ligands with functionally selective properties between Gs and Golf signaling.

The G s family of G␣ proteins, comprised of two highly homologous G s and G olf subtypes, positively couple to adenylate cyclase (thus, "s" for stimulatory). Upon activation, both G s and G olf promote cAMP production and subsequent signaling events such as activation of the PKA cascade. G s is ubiquitously expressed in most organs, whereas G olf is mainly restricted to the brain. Moreover, within the brain, G s and G olf exhibit distinct expression patterns. G s is uniformly expressed throughout the brain, except in the striatum where its expression is very low. In contrast, G olf is highly expressed in the striatum and olfactory tubercle, as well as in the hippocampus and cerebellum to a lesser extent (41). The contrast in tissue expression for G s and G olf is quite dramatic and unique among other G␣ homologs (e.g. G i versus G o , G q versus G 11 , and G 12 versus G 13 ) (1), making G s and G olf fascinating molecular targets with regard to their corresponding functions, particularly in terms of D 1 receptor (D1R)-mediated 3 and A 2A receptor (A2AR)-mediated signaling in the striatum compared with other brain regions.
Conformational changes associated with GPCR activation have been revealed in remarkable detail by the crystal structure of agonist-bound ␤ 2 adrenergic receptor (␤2AR) in complex with G s , by complementary spectroscopy studies, as well as by related molecular dynamics studies (2)(3)(4)(5). However, the extent to which conformational changes in G protein are conserved in living cells, as well as across different receptors and different G-protein isoforms, remains unclear. In particular, little is known about the G s homolog G olf in terms of its functional similarities and differences. Despite their different expression patterns, the high degree of homology in amino acid sequences (89% identity) has led to the assumption that G s and G olf function essentially identically at both the molecular and cellular levels. For this reason, as well as the fact that G olf expression is typically poor in heterologous cells, G s functional assays have been used as surrogates for G olf activation, begging the question of just how similar these processes are. The answer may provide specific ways to target selectively physiological functions mediated by either G s or G olf signaling.
In the current study, we first focused on ␤2AR-G s activation in intact cells to investigate conformational changes of different domains of G s . Using bioluminescence resonance energy transfer (BRET)-based assays, we assessed movements both within the G protein, as well as between the receptor and G␣ subunit.  cro ARTICLE Using a library of novel G s biosensors with either luciferase or GFP variants inserted at various positions throughout the structure, we studied conformational changes in living cells and compared the results to the crystal structures of the closed and open conformations of G s (2,6). We then studied conformational changes in G s induced by activation of the D1R, A2AR, and ␤ 1 adrenergic receptor (␤1AR). Next, taking advantage of the significant homology, we created a series of G olf biosensor constructs with insertions at the same nine positions used for G s . Agonist-induced conformational changes in G olf were compared with those in G s . Finally, G olf assay optimization was carried out for D1R. Our analysis using these G s biosensors suggests that conformational changes within the G s heterotrimer are similar when induced by different G s -coupled receptors. Comparison between the G s and G olf sensor readouts also indicates a very similar regulation of activation by endogenous agonists. Using this set of G s and G olf biosensors, the efficacy and potency of agonists, as well as the activation preference between G s and G olf , can be studied in relation to structural changes and subsequent effector activation. Our results with D1R establish that these biosensors represent a novel pharmacological tool to study structure-function relationships comparing G s and G olf .

Sensor insertion positions to assess the open and closed conformations of G s
A dramatic structural change is apparent between the closed (PDB code 1AZT) and open (PDB code 3SN6) states of the G s heterotrimer, particularly in the ␣-helical domain (Fig. 1A). To detect such conformational changes upon G-protein activation, biosensors were constructed in which Rluc or mVenus were inserted at nine different insertion positions in the loop motifs of different domains in G s (Fig. 1, A and B). The insertion positions (i.e. loop regions) were selected to avoid structural perturbations. Position 7 is located between the N terminus and ␣N. Positions 67 and 71 are situated in the linker-loop motif, which was not resolved in the crystal structures; they are in the hinge domain that connects the Ras-like catalytic domain and ␣-helical domain. Insertion of GFP at position 71 of G s has been functionally validated previously (7). Positions 99, 154, and 175 are located in the ␣-helical domain (position 99, proximal; positions 154 and 175, distal), whereas positions 305, 338, and 349 are located in the Ras-like domain, avoiding the catalytic core. Insertion of mVenus (YFP variant) or Rluc8 at these positions led to similar levels of expression based on levels of fluorescence for the mVenus constructs and luminescence for the RLuc8 constructs (data not shown).

G s biosensors detect distinct conformational changes upon activation
Relative movements between G s and G ␥2 upon receptor activation were studied, similarly to previous analysis of G i activation using G i biosensors with insertions at positions 60, 91, and 122 (aligned with positions 67, 99, and 131 in G s , respectively) (8,9). Of the nine insertion constructs we created, when co-expressed with ␤2AR ( Fig. 2A), G s with insertions at positions 305, 338, and 349 failed to show significant isoproterenol-induced BRET changes, although the fluorescence and luminescence levels were not significantly different from other biosensors. Insertion positions, 67, 71, 99, and 154, on the other hand, produced substantial agonist-induced BRET changes. When coexpressed with ␤2AR ( Fig. 2A), isoproterenol increased BRET between Gs67-Rluc and ␥2-GFP10 or Gs71-Rluc and ␥2-GFP10, consistent with greater proximity of the sensors in the two subunits. In contrast, Gs99-Rluc and ␥2-GFP10 or Gs154-Rluc and ␥2-GFP10 decreased BRET, indicating an increase in distance. When the donor-acceptor pair was reversed, the directions of BRET change in G s -Venus-␥2-Rluc remained the same (supplemental Fig. S1). Furthermore, when co-expressed with D1R (Fig. 2B), the directions of change for all the positions were consistent with the ␤2AR results. Activation of adrenergic ␤1AR and adenosine A2AR also showed the same directionality as D1R and ␤2AR (supplemental Fig. S2).
Venus-fused G s constructs were also tested for their use in measuring receptor-G␣ engagement (supplemental Fig. S3). With both ␤2AR and D1R, sensor position 154 showed the largest dynamic range for agonist-induced effects, ϳ3-fold greater than that observed for the previously studied position 71 insertion (7-9) (supplemental Fig. S3).

Simulated conformational trajectories reveal a movement in the hinge loop
To provide a structural context to our BRET results, we used the closed crystal structure of G s (6) and the open conformation of the G s crystal structure in complex with ␤2AR (2) as beginning and ending structures, respectively, to simulate domain movement between the closed and open crystal structures of G s (Fig. 2, C and D). A missing loop (residues 66 -72) of the closed G s crystal structure was built using the Rosetta loop prediction algorithm. The best scoring conformation was extracted and equilibrated in the context of the protein by a 20-ns all-atom MD simulation. Adiabatic biased MD was then employed to generate a continuous, low-energy transition path starting from the closed G s crystal structure and reducing the root mean square deviation from the open G s crystallographic conformation. Distances between the C␣ atoms of experimental insertion points for the different configurations are reported in supplemental Table S1,

Novel G␣ s and G␣ olf biosensors G olf movement extrapolated from novel G olf biosensors corresponds to that of G s
As mentioned above, the G olf subunit is widely expressed in the striatum, where it is critical to the function of D1R. Taking advantage of its 89% sequence identity to G s , luciferase or mVenus was inserted at the same nine positions explored above (Fig. 1B). Given their enriched expression in striatum (12, 14 -17), ␤2 and ␥7 constructs were used to study G olf activation. Similar to the G s results, both G olf -Rluc-␥7-GFP10 and G olf -Venus-␥7-Rluc configurations revealed an increase in BRET values at the hinge region (position 69) and a decrease in BRET values or a lack of response in the ␣-helical domain (posi-tions 100 and 155) for ␤2AR receptor activation ( Fig. 3A and supplemental Table S2). The same directionalities were observed for D1R, further supporting the conservation of domain movements of these homologous G proteins when activated by different receptors (Fig. 3A and supplemental Table  S2). The results are also consistent with a large displacement of the ␣-helical domain in G olf , similar to that observed in G s (Fig.  2, C and D) and to the crystal structure of the active complex (2).

Development of a D1R-G olf assay
To establish a reliable assay for drug screening at the D1R with regard to G olf coupling, different configurations of BRET . The ␣-helical and Ras-like domains of G␣ are in blue and red, respectively, whereas G␤ and G␥ are in light gray and dark gray. B, amino acid sequence alignment between G olf and G s short. Identical and homologous residues are highlighted in yellow and green, respectively. Insertion positions for G s as well as G olf are enclosed by rectangles, and the residue numbers for G s are shown above.

Novel G␣ s and G␣ olf biosensors
were tested (supplemental Table S2 and Fig. 4). Fig. 4 shows direct comparisons between G s and G olf biosensors in the activation and engagement modes. For G␣-␥ activation assays, the relative potency and efficacy differences between dopamine, a full agonist, and norepinephrine, a less potent agonist, were tested. The potency differences between the two agonists were similar for activation of G s and G olf (Fig. 4, A and B, and supplemental Table S3). The engagement assays also demon-

Novel G␣ s and G␣ olf biosensors
strated a tight agreement of relative potency and efficacy between dopamine and norepinephrine in G s and G olf (Fig. 4, C and D, and supplemental Table S3).
The engagement (i.e. D1R-Rluc-G olf -Venus) BRET configuration was pursued for optimization because of its larger dynamic window. Different ␤-␥ subunit combinations were tested (supplemental Fig. S5). Although the ␤1-␥7 and ␤2-␥7 combinations showed a similar dynamic range, ␤2-␥7 was chosen for the rest of the studies because of the established expression overlap in striatum (10 -13). One of the crucial factors for successful G olf BRET assay regardless of configuration was coexpression of the G-protein chaperone Ric8 (18 -20), which robustly enhanced the dynamic range of agonist-induced BRET (supplemental Fig. S4). Because luciferase expression, detected by luminescence, does not differ significantly with and without Ric8 co-expression, the dramatic change in dynamic range of BRET may have to do with chaperone activities of Ric8, possibly rescuing misfolding or aiding proper localization of the G olf biosensor to the receptor complex rather than simply enhanc-ing its expression (supplemental Fig. S4). Overall, cross-comparison of the D1R-G␣ engagement and G␣-␥ activation assays validates the potency and efficacy range of the four assays tested and thus their utility in pharmacological characterization of D1R activation.

Creation of novel homology-based G i1 and G q biosensors
Because the Ras-like domain, hinge region, and ␣-helical domain are well-conserved in other classes of G␣ subunits (21), the relative movements upon activation were compared in other G␣ subunits. The same three sensor insertions (i.e. G s equivalent of positions 67, 99, and 154) were made in G i1 and G q at the aligned amino acid residues (Fig. 5A). Upon transfection with dopamine D 2 receptor (D2R) and using dopamine as ligand, for G i1 , the hinge region (position 60) moves closer to the ␥ subunit, whereas the ␣-helical domain (positions 91 and 145) moves away from the ␥ subunit (Fig. 5B), similar to our results in G s (Fig. 2). The conformational changes in the hinge and ␣-helical domains of G i1 are consistent with previous

Novel G␣ s and G␣ olf biosensors
reports with insertions at positions 60 and 91 (8,9). Upon transfection with muscarinic M 1 receptor (M1R) and using carbachol as ligand, G q sensors also revealed the same directionalities of BRET change for the ␣-helical domain (positions 97 and 150) (Fig. 5C) but with a very robust dynamic range for drug response with the position 97 sensor. However, the insertion at position 66 (equivalent to 67 in G s ) yielded very little agonistinduced BRET (Fig. 5C, orange curve), although the luminescence was similar (data not shown), suggesting a possible structural difference in the hinge loop of G q .
Finally, using the Venus-fused G i1 or G q sensors, receptor-G␣ engagement BRET was assessed for D2R or M1R (supplemental Fig. S6). Consistent with the G s results, insertion at position 150 of G q (aligned to 154 in G s ) gave the largest efficacy window, as well as higher potency when compared with the previously characterized position 97 (22) (supplemental Fig.  S6B). This trend was not maintained with G i1 , where insertion at position 91 (aligned with position 99 in G s ) produced the most robust sensor (supplemental Fig. S6A). Taken together, these results with G-protein activation BRET have established generally conserved movements of the ␣-helical domain in three different classes of G␣ subunits, albeit with subtle differences.

Discussion
The seminal work leading to the crystal structure of the active ␤2AR-G s heterotrimer complex has enabled comparison between open and closed structures of the G protein, as well as interactions of the G protein with the receptor, providing molecular details of key conformational changes associated Figure 5. A, amino acid sequence alignment among G i1 , Gi2, Gi3, GoA, G olf , G s short, and G q . Identical, highly homologous, and homologous residues are highlighted in yellow, blue, and green, respectively. Insertion positions equivalent to G s short positions 67, 99, and 154 are indicated by arrows. Novel constructs made for this study are indicated by check marks. B and C, dose-response curves of G i1 protein activation BRET for D2R with dopamine (B) and G q protein activation for muscarinic M1R with carbachol (C

Novel G␣ s and G␣ olf biosensors
with the activation process (2,6,23). A series of relevant structure-function studies have pointed to the large displacement of the ␣-helical domain as a central mechanism, albeit not sufficient, for the promotion of GDP-GTP exchange (3)(4)(5). Although the ␣-helical domain may undergo spontaneous fluctuation between the open and closed states, insertion of the ␣5 helix of G s into the intracellular vestibule of the ␤2AR promotes opening of the ␣-helical domain. The pronounced decrease in BRET values in living cells indicates a distancing event between the ␣-helical domain and the ␥ subunit, consistent with an opening movement from three different amino acid positions of the ␣-helical domain (positions 99, 154, and 175). The movement in the loop structures, which serve as a hinge between the ␣-helical and Ras-like domains, is therefore an important feature that links the displacement of the ␣-helical domain with G s activation. Presumably because of the highly flexible nature of the linker loop, this region was not resolved in the crystal structure (2,6). Based on our MD simulations, we hypothesize that the transition between the closed and open states of the Gprotein subunit may involve an outward protruding movement of the linker loop (positions 67, 71), along with the overall structural changes that enable ␣-helical domain opening. The negative BRET change between the myristoylated ␣N loop (position 7) and ␥ subunit is also consistent with the displacement of ␣N between the opened and closed G s crystal structures.
Similarly, ␣-helical domain displacement has been proposed for the G i and G q proteins as an activation mechanism (21). In addition to previously studied positions (G s71 , G i160 , G i191 , and G q97 ) (7-9, 22, 24), we have created novel fusion constructs at G s positions 67, 99, and 154 and equivalent positions in G i and G q . Because of the conserved structural domains (i.e. ␣-helical, linker loop, Ras-like domains), not surprisingly, our results mostly coincide with previous studies. In the activation configuration, only G q66 failed to display positive BRET changes compared with G i16 0 or G s67 , possibly because of a difference in the linker-loop structure that does not generate a protruding movement in G q . Overall, the G␣-␥ BRET assay demonstrates the conserved nature of ␣-helical domain movement across three different G␣ protein subtypes and strengthens the case for these assays as robust sensors of agonist-induced activation in living cells.
In line with its specific brain distribution, G olf is involved in olfaction and basal ganglia function (12,25,41). Mutations in the GNAL gene encoding G␣olf have been implicated in movement disorders in humans (26 -29). Because of their high homology, G olf is generally considered to function similarly to G s in terms of its ability to stimulate adenylate cyclase. Although some kinetic difference in GTP hydrolysis has been suggested between G s and G olf in ␤2AR (30), to our knowledge, there has not been a thorough molecular study of its activation. The current study is the first to focus on direct comparison of D1R-G s and D1R-G olf coupling and activation. Our new findings indicate that: 1) conformational changes upon activation are similar for G s and G olf ; 2) Ric8B is required for heterologous expression of G olf biosensors, as reported previously (18 -20); and 3) the ␤2-␥7 pair confers the largest dynamic range for G olf engagement BRET in agreement with previous studies showing a dependence on co-expression of G olf , ␤2, and ␥7 subunits for striatal D1R and A2AR signaling (10,13). The directionalities of the G olf activation BRET at different insertion positions are for the most part consistent with the G s results, but overall the dynamic range of agonist response is not as robust as for G s . It is worth considering that there may be subtle differences between ␤2AR and D1R in G olf activation because their EC 50 values for the 69, 72, and 155 position sensors show a trend of difference, although these did not reach statistical significance.
Notably, although the Rluc or Venus expression level (measured by luminescence or fluorescence) is similar between the G s and G olf biosensors, the efficiency of folding or localization of the G olf sensors may be impaired because the basal BRET is lower for D1R-Rluc-G olf -Venus than for D1R-Rluc-G s -Venus. This may explain the lack of agonist response for the position 100 insertion in G olf -Rluc. The expression of G olf and G olf fusion constructs is likely challenging because the accessory molecules that are present in neurons may be missing in heterologous cells. Studies have indicated that co-expression of Ric8B and HSP70, both chaperone proteins, enhance the expression of both olfactory receptor and its G olf signaling (31). In our hands, Ric8B increased the BRET dynamic range of BRET of G olf constructs, although HSP70 did not. Perhaps expression of other accessory proteins might help to increase further the dynamic range of the G olf assays.
In summary, our novel G olf assay represents a useful screening method for G olf signaling in heterologous cells. The G s and G olf assays presented herein can be used in parallel for pharmacological investigation of receptors relevant in neuropsychiatric disorders, including both D1R and A2AR.

BRET assay
Three modes of BRET assays were performed to detect receptor ligand-induced events for 1) G␣-␥ protein activation, 2) G␥-␣ protein activation, and 3) receptor-G␣ engagement. 1) The G␣-␥ protein activation assay uses a RLuc-fused G␣-protein subunit and GFP10-fused G␥ protein for a resonance energy transfer (RET) pair. FLAG-tagged receptor and untagged G␤ constructs were co-transfected. 2) Similarly the G␥-␣ protein activation assay uses a RLuc-fused G␥ protein subunit and GFP10-fused G␣ protein for a RET pair. FLAGtagged receptor and untagged G␤ constructs were co-transfected.
3) The receptor-G␣ engagement assay uses RLuc-fused receptor and mVenus-fused G␣ protein for the RET pair. Untagged G␤ and G␥ constructs were co-transfected. As reported previously (35), cells were harvested, washed, and resuspended in PBS. Approximately 200,000 cells/well were distributed in 96-well plates, and 5 M coelenterazine H (substrate for BRET1) or 5 M coelenterazine 400a (substrate for BRET2) was added to each well. Three minutes after addition of coelenterazine, ligands (dopamine (Sigma), L-(Ϫ)-norepinephrine (Sigma), 5Ј-N-ethylcarboxamidoadenosine (Tocris), isoproterenol (Tocris), or carbachol (Tocris)) were added to each well. The fluorescence of the acceptor was quantified (for Venus excitation at 500 nm and emission at 530 nm for 1-s recording or for GFP10 excitation at 405 nm and emission at 515 nm for 1-s recording for GFP10) in a Mithras LB940 (Berthold Technologies, Bad Wildbad, Germany) to confirm constant expression levels across experiments. In parallel, luminescence and BRET1 signal from the same batch of cells was determined as the ratio of the light emitted by Venus (530 nm) over that emitted by coelenterazine H (485 nm) or luminescence and BRET2 signal from the same batch of cells was determined as the ratio of the light emitted by GFP10 (515 nm) over that emitted by coelenterazine 400a (400 nm). The results are calculated for the BRET change (BRET ratio for the corresponding ligand minus BRET ratio in the absence of the ligand). E max values are expressed as the basal subtracted BRET change in the dose-response graphs. The fluorescence and luminescence counts (arbitrary units) were similar in different experiments using the same construct. The data and statistical analyses were performed with Prism 5 (GraphPad software).

Sequence homology alignment
Amino acid sequence homology analysis was performed using Vector NTI Advance (Invitrogen). Identical residues are highlighted yellow, and homologous residues are highlighted green.

Molecular modeling and computer simulations
The closed (PDB code 1AZT), and open (PDB code 3SN6, chains A, B, and G) crystal structures of G s were used for MD simulations. Missing loop residues (at positions 66 -72) of the closed G s crystal structure were built using the Rosetta loop prediction algorithm (37). The best scoring conformation was extracted and equilibrated in TIP3P waters by a 20-ns MD simulation using all-atom description and the Charmm27 force field (38). The open and closed structures were used as templates to model G olf by homology (39). To investigate the changes in conformation between the inactive and active conformations, an adiabatic biased MD simulation (40) was performed starting from the protein inactive state, using the RMSD from the active state model as a collective variable. Briefly, a steep repulsive bias was applied when the RMSD from the target state increased above the minimum value reached during the simulation. Specifically, the applied potential was where the collective variable is the RMSD to the active state, and CV Min ͑R͑t͒͒ ϭ min 0ϽsϽt CV͑R͑s͒͒ ϩ ͑t͒ (Eq. 2) Similar to a ratchet and pawl system, propelled by thermal motion, the biasing potential does not exert work on the system and ensures that the obtained trajectory is a low-free energy path connecting the initial and final states. The simulation was stopped after 20 ns. Simulations were performed with Gromacs 4.6 with Plumed 2.0. The simulation was carried out in the NPT ensemble, using v-rescale thermostat and Parrinello-Rahman barostat to maintain temperature and pressure constant. Electrostatics was calculated with the particle-mesh Ewald algorithm, and non-bonded interactions were cut-off at 1.2 nm. A time step of 2 fs was used. Distances between the C atoms of insertion points of the experimental probes were monitored during simulation.
Author contributions-H. Y. designed, conducted, and analyzed the molecular biology and BRET experiments. D. P. and M. F. conducted and analyzed the molecular dynamics simulation. N. S. C. performed the molecular biology work. H. Y. and J. A. J. wrote the manuscript. All authors contributed to reviewing the results and writing the manuscript.