Compared to eticlopride bound in the D₃R structure, risperidone in the D₂R structure penetrates deeper into the binding site, with its benzisoxazole moiety occupying a sub-pocket that eticlopride does not reach. By examining the D₂R/risperidone structure, we found that the benzisoxazole moiety is enclosed by eight residues in D₂R, which are identical among all D₂-like receptors (i.e. D₂R, D₃R, and D₄R): Cys118^3.36 (superscripts denote Ballesteros-Weinstein numbering Ballesteros and Weinstein, 1995), Thr119^3.37, Ile122^3.40, Ser197^5.46, Phe198^5.47, Phe382^6.44, Trp386^6.48, and Phe390^6.52. Notably, three of these residues (Ile122^3.40, Phe198^5.47, and Phe382^6.44) on the intracellular side of the OBS that we previously defined (Michino et al., 2015a), accommodate the F-substitution at the tip of the benzisoxazole ring in a small cavity (termed herein as the Ile^3.40 sub-pocket) (Figure 2a). Both Ile122^3.40 and Phe382^6.44 of this Ile^3.40 sub-pocket are part of the conserved Pro^5.50-Ile^3.40-Phe^6.44 motif that undergoes rearrangement upon receptor activation (Rasmussen et al., 2011), and we have found that the I122^3.40A mutation renders D₂R non-functional (Klein Herenbrink et al., 2019; Wang et al., 2018). Interestingly, this Ile^3.40 sub-pocket is collapsed in both the D₃R and D₄R structures (Sibley and Shi, 2018; Figure 2b,c). We noted that this collapse is associated with rotation of the sidechain of Cys^3.36: In the D₂R/risperidone structure, the sidechain of Cys^3.36 faces the OBS, whereas in the D₃R/eticlopride and D₄R/nemonapride structures, it rotates downwards to partially fill the Ile^3.40 sub-pocket (Figure 2a–c).

To test our hypothesis that these observed differences in the crystal structures are due to the binding of antagonists bearing different scaffolds but not intrinsic divergence of D₂-like receptors, we compared the binding modes of three non-selective antagonists in D₂R. We reverted three thermostabilizing mutations introduced for crystallography (I122^3.40A, L375^6.37A, and L379^6.41A) back to their WT residues, established WT D₂R models in complex with risperidone, spiperone, or eticlopride, and carried out extensive MD simulations (see Materials and methods, Figure 1—figure supplement 1 and Table 1).

In our prolonged MD simulations of the WT D₂R/risperidone complex (>65 µs, Table 1), we observed that risperidone stably maintains the binding pose captured in the crystal structure, even without the thermostabilizing mutations (Figure 2d). Thus, the I122^3.40A mutation has minimal impact on the binding pose of risperidone. Interestingly, in the simulations of the WT D₂R model in complex with spiperone, a butyrophenone derivative, the F-substitution on the butyrophenone ring similarly occupies the Ile^3.40 sub-pocket as risperidone (Figure 2e). Note that the F-substitutions in risperidone and spiperone are located at similar distances to the protonated N atoms that interact with Asp^3.32 (measured by the number of carbon atoms between them, Figure 1—figure supplement 1) and these two ligands appear to be optimized to occupy the Ile^3.40 sub-pocket.

In contrast, in our simulations of the D₂R/eticlopride complex, the eticlopride pose revealed in the D₃R structure (PDB: 3PBL) is stable throughout the simulations and does not protrude into the Ile^3.40 sub-pocket (Figure 2f). Consistent with the difference in the crystal structures noted above (Figure 2a,b), when risperidone and spiperone occupy the Ile^3.40 sub-pocket, the sidechain of Cys118^3.36 rotates away with its χ1 rotamer in gauche-, while in the presence of the bound eticlopride, this rotamer is stable in trans (Figure 2—figure supplement 1).

To validate these computational findings regarding the occupation of the Ile^3.40 sub-pocket, we mutated Ile122^3.40 of WT D₂R to both Trp and Ala and characterized how these mutations affect the binding affinities for spiperone, risperidone, and eticlopride (Table 2). We hypothesized that the bulkier sidechain of Trp at position 3.40 would hamper the binding of spiperone and risperidone since they occupy the Ile^3.40 sub-pocket but have no effect on eticlopride binding, while the smaller Ala should not affect the binding of spiperone or risperidone. Consistent with this hypothesis, the I122W mutation decreased the binding affinities of risperidone (13-fold) and spiperone (6-fold) compared to WT but had no effect on that of eticlopride. In contrast, the I122A mutation did not affect the affinities of spiperone or risperidone, which is consistent with our simulation results that show the I122A mutation has minimal impact on risperidone binding. In contrast, I122A caused a threefold increase in the affinity of eticlopride, suggesting that the I122A mutation may promote an inactive conformation of D₂R that favors eticlopride binding. Together these results support our proposal that different antagonist scaffolds may favor distinct inactive conformations of D₂R.

Ligand binding in D₂-like receptors can be modulated by Na⁺ bound in a conserved allosteric binding pocket coordinated by Asp^2.50 and Ser^3.39 (Michino et al., 2015b; Neve, 1991; Wang et al., 2017). Note that the aforementioned Cys^3.36 and Ile^3.40 are adjacent to the Na⁺ coordinating Ser^3.39; thus, we further hypothesized that the occupation of the Ile^3.40 sub-pocket by spiperone or risperidone makes them insensitive to Na⁺. To test this hypothesis, we simulated D₂R/risperidone, D₂R/spiperone, D₂R/eticlopride, and D₂R/(-)-sulpiride complexes in the presence versus absence of bound Na⁺ (Table 1). Interestingly, the occupancy of the Ile^3.40 sub-pocket by either spiperone or risperidone was unaffected by the presence or absence of bound Na⁺ (Figure 2—figure supplement 1). In contrast, while the poses of eticlopride and (-)-sulpiride are highly stable in the presence of bound Na⁺, they oscillated between different poses in the absence of Na⁺. These oscillations are associated with the sidechain of Cys^3.36 swinging back and forth between the two rotamers, suggesting an important role of Na⁺ binding in stabilizing the poses of eticlopride and (-)-sulpiride and the configuration of the Ile^3.40 sub-pocket (Figure 2—figure supplement 1). Interestingly, the previous MD simulations described by Wang et al. indicated that nemonapride’s binding pose in D₄R is more stable in the presence of bound Na⁺ as well (Wang et al., 2017).

Consistent with these computational results, we have previously shown that spiperone binding is insensitive to the presence of Na⁺, while the affinities of eticlopride and sulpiride are increased in the presence of Na⁺ (Michino et al., 2015b). In this study, we performed binding experiments in the absence or presence of Na⁺ and found the affinity of risperidone to be unaffected, in accordance with this hypothesis (Table 2).

Together these findings support our hypothesis that the ability of a ligand to bind the Ile^3.40 sub-pocket relates with its sensitivity to Na⁺ in binding, due to allosteric connections between the sub-pocket and the Na⁺ binding site.

To further investigate the functional impact of these conformational differences surrounding the OBS, we used a bioluminescence resonance energy transfer (BRET) assay, which measures conformational changes of the Go protein heterotrimer following activation by D₂R (Michino et al., 2017), to evaluate the inverse agonism activities of several representative D₂R ligands. These ligands can be categorized into two groups according to their sensitivities to Na⁺ in binding at D₂R, which have been characterized either in our current study or in previous studies (Michino et al., 2015b; Neve, 1991; Newton et al., 2016). While risperidone, spiperone, and (+)-butaclamol have been found to be insensitive to Na⁺ in binding, (-)-sulpiride, eticlopride, and raclopride show enhanced binding affinities in the presence of Na⁺. Using quinpirole as a reference full agonist, we found that the Na⁺ insensitive ligands display significantly greater inverse agonism (< −30% that of the maximal response of quinpirole) relative to the Na⁺-sensitive ligands (> −15% that of the maximal response of quinpirole, Figure 3). These observations are consistent with findings from earlier [³⁵S]GTPγS binding experiments of Roberts and Strange in which (+)-butaclamol, risperidone, and spiperone were found to inhibit significantly more [³⁵S]GTPγS binding than raclopride and (-)-sulpiride (Roberts and Strange, 2005). Of note, these [³⁵S]GTPγS-binding experiments were performed in the absence of Na⁺.

Based on these functional data together with the different binding modes revealed by our computational simulations, we propose that ligands that occupy the Ile^3.40 sub-pocket exhibit a greater level of inverse agonism as compared to those that do not. Therefore, across the tested inverse agonists there is a negative relation between ligand sensitivity to Na⁺ and the extent of inverse agonism at D₂R. The differential occupation of the Ile^3.40 sub-pocket is the structural basis for the Na⁺ sensitivity, which contributes significantly to the extent of inverse agonism of the tested ligands.

By occupying the Ile^3.40 sub-pocket, the benzisoxazole moiety of risperidone pushes the conserved Phe^6.52 away from the binding site in the D₂R/risperidone structure compared to its position in the D₃R/eticlopride structure. This interaction is responsible for positioning the aromatic cluster of TM6 and TM7 (Trp^6.48, Phe^6.51, Phe^6.52, His^6.55, and Tyr^7.35) in D₂R differently from its configurations in the D₃R and D₄R structures, resulting in an overall outward positioning of the extracellular portion of TM6 in D₂R (Figure 4—figure supplement 1). On the extracellular side of the OBS, the space near Ser^5.42 and Ser^5.43 that accommodates the bulky substitutions of the benzamide rings of the bound eticlopride and nemonapride in the D₃R and D₄R structures is not occupied by risperidone in D₂R, which is likely associated with the inward movement of the extracellular portion of TM5 in D₂R relative to those in the D₃R and D₄R structures (Figure 1).

To evaluate whether these conformational rearrangements are due to the minor divergence in these regions of the receptors or to the ligand-binding site plasticity that accommodates ligands bearing different scaffolds, we compared the resulting conformations of D₂R bound with risperidone or eticlopride. We observed the same trend of rearrangements of the transmembrane segments surrounding the OBS in the resulting receptor conformations from our D₂R/risperidone and D₂R/eticlopride simulations (Figure 4a), that is, an inward movement of TM6 and outward movement of TM5 in the presence of the bound eticlopride (Figure 4b,c). Without such movements in D₂R/eticlopride, Ser193^5.42 and Ser194^5.43 would clash with the bound eticlopride (Figure 4a). These findings further support our inference that differences between the D₂R and D₃R inactive structures are largely due to the different scaffolds of the bound non-selective ligands.

In addition to differences in the transmembrane segments surrounding the OBS, there are also substantial differences in the configuration of EL2 in the D₂R and D₃R structures. EL2 between TM4 and TM5 is connected to TM3 via a disulfide bond formed between Cys^EL2.50 (see Materials and methods and Figure 5—figure supplement 1 for the indices of EL1 and EL2 residues) and Cys^3.25. The conformation of EL2, the sequence of which is not conserved among aminergic GPCRs, is expected to be dynamic. Indeed, in the D₂R/risperidone structure, the sidechains of residues 176^EL2.40, 178^EL2.46, 179^EL2.47, and 180^EL2.48, which are distal to the OBS were not solved, likely due to their dynamic nature. Curiously, the portion of EL2 C-terminal to Cys182^EL2.50 (residues 182^EL2.50-186^EL2.54), which forms the upper portion of the OBS that is in contact with ligand, is in a helical conformation in the D₂R/risperidone structure.

Strikingly, in our MD simulations of D₂R complexes, we found that this helical region showed a tendency to unwind (Video 1). The unwinding of EL2 involves a drastic rearrangement of the sidechain of Ile183^EL2.51, which dissociates from a hydrophobic pocket formed by the sidechains of Val111^3.29, Leu170^4.60, Leu174^EL2.38, and Phe189^5.38. Specifically, the unwinding process is initiated by the loss of a hydrogen-bond (H-bond) interaction between the sidechain of Asp108^3.26 and the backbone amine group of Ile183^EL2.51 formed in the D₂R/risperidone structure (Figure 5—figure supplement 2b, step (i). When this interaction is broken, the orientation of residues 182^EL2.50-186^EL2.54 deviates markedly from that of the crystal structure, losing its helical conformation (see below). Subsequently, the sidechain of Ile183^EL2.51 rotates outwards and passes a small steric barrier of Gly173^EL2.37 (Figure 5—figure supplement 2b, step (ii), and in some trajectories makes a favorable hydrophobic interaction with the sidechain of Ala177^EL2.45. In a few long trajectories, Ile183^EL2.51 rotates further toward the extracellular vestibule where it can make favorable interactions with hydrophobic or aromatic residues from the N terminus, or the bound risperidone (Video 1). Consequently, residues 182^EL2.50-186^EL2.54 are in a fully extended loop conformation while Ile184^EL2.52 tilts under EL2 (Figure 5—figure supplement 2b, step (iii).

In the D₃R structure, the aligned residue for Asp108^3.26 of D₂R is conserved as Asp104^3.26; its sidechain forms an interaction not with Ile182^EL2.51 but rather with the sidechain of Asn173^EL2.39, which is also conserved in D₂R as Asn175^EL2.39. In the D₄R, the aligned two residues (Asp109^3.26 and Asn175^EL2.39) are conserved as well, their sidechains are only 4.3 Å away in the D₄R structure, a distance slightly larger than the 3.2 Å in the D₃R structure. Even though these residues are conserved in D₂R, the interaction in D₃R (and potentially in D₄R), between Asp^3.26-Asn^EL2.39, is not present in the D₂R structure in which the aligned Asn175^EL2.39 faces lipid (Figure 5—figure supplement 2a). However, in a few of our long D₂R simulations, Asn175^EL2.39 gradually moves inwards and approaches Asp108^3.26 (Figure 5—figure supplement 2b, step (iv). At this point, the EL2 conformation of D₂R is highly similar to that of D₃R (Figure 5—figure supplement 2c), suggesting that EL2 is dynamic and can exist in both conformations.

We evaluated the tendency of the EL2 helix to unwind in each of the simulated D₂R complexes by measuring the stability of the backbone H-bond between Ile183^EL2.51 and Asn186^EL2.54, a key stabilizing force of the helix (Figure 5a). When we plotted the Ile183^EL2.51-Asn186^EL2.54 distance against the Asp108^3.26-Ile183^EL2.51 distance for each D₂R complex (Figure 5b), we found that the loss of the Asp108^3.26-Ile183^EL2.51 interaction increases the probability of breaking the Ile183^EL2.51-Asn186^EL2.54 H-bond, that is the unwinding of EL2. Interestingly, in all our simulated D₂R complexes, EL2 has a clear tendency to unwind, regardless of the scaffold of the bound ligand (Figure 5c,d, Videos 1–3). Note that in the D₃R/eticlopride simulations, the aligned residues Ser182^EL2.51 and Asn185^EL2.54 do not form such a H-bond, and EL2 is always in an extended conformation (Figure 5b–d). This tendency of EL2 to transition toward the extended conformation is also present in our simulations of D₂R in complex with a partial agonist, aripiprazole, whereas EL2 in the D₃R complexes with partial agonists (R22 and S22) remains in the extended conformation (Table 1 and Figure 5—figure supplement 3). Interestingly, Asp104^3.26 and Ser182^EL2.51 can move into interacting range in the D₃R/eticlopride simulations, and the Ser182^EL2.51-Asn185^EL2.54 interaction can sporadically form in the D₃R/R22 simulations – both raise the possibility that the extended conformation of D₃R EL2 may transition to a helical conformation.

Interestingly, in one of our long MD trajectories of the D₂R/risperidone complex, EL2 evolved into a conformation that has a helical N-terminal portion and an extended C-terminal portion (Video 4 and Figure 5—figure supplement 4). This conformation is not observed in either of the D₂R/risperidone and D₃R/eticlopride structures but is similar to that of the 5-HT_2AR/risperidone structure, further demonstrating the dynamics of this loop region (Figure 5—figure supplement 4).

In marked contrast to the obvious trend toward unwinding of EL2 in all our simulated D₂R complexes, in our recent simulations of MhsT, a transporter protein with a region found by crystallography to alternate between helical and unwound conformations (Malinauskaite et al., 2014), we failed to observe any spontaneous unwinding over a similar simulation timescale (with the longest simulations being ~5–6 µs) when the region was started from the helical conformation (Abramyan et al., 2018; Stolzenberg et al., 2017). This shows how difficult it can be to capture known dynamics in simulations and suggests that the C-terminal helical conformation of EL2 in D₂R represents a higher energy state than the extended conformation, which allows for observation of the transitions in a simulation timescale not usually adequate to sample folding/unfolding events (Piana et al., 2011).

We have previously shown that the divergence in both the length and number of charged residues in EL1 among D₂R, D₃R, and D₄R is responsible for the selectivity of more extended ligands (Michino et al., 2013; Newman et al., 2012). Another striking difference in the D₂R, D₃R, and D₄R structures is the position of the conserved Trp^EL1.50 in EL1. Trp100^EL1.50 is in a much more inward position in the D₂R structure, making a direct contact with the bound risperidone (Figure 6a), Trp101^EL1.50 in D₄R interacts with the bound nemonapride that has an extended structure, whereas Trp96^EL1.50 in D₃R is not in contact with eticlopride (Figure 6b). Thus, we asked whether these distinct positions of Trp^EL1.50 are due to the divergence in EL1 among these receptors (Michino et al., 2013) or due to the multiple inactive conformations that differentially accommodate the binding of non-selective ligands of divergent scaffolds.

When residues 182^EL2.50-186^EL2.54 of EL2 are in a helical conformation, in the D₂R/risperidone simulations, we found that there is more room in the extracellular vestibule and the position of Trp100^EL1.50 is flexible and can adopt several positions and orientations (Figure 6c,e,f). In the D₂R/eticlopride simulations, Trp100^EL1.50, which cannot interact with eticlopride, shows more flexibility than that observed in the presence of risperidone and can move to a similar position like that of Trp96^EL1.50 in the D₃R structure (Figure 6—figure supplement 1 and Video 2). Interestingly, in this position, the conformation of Trp^EL1.50 can be stabilized by the disulfide bond of EL2 (Ioerger et al., 1999) (as shown in Video 2) or by interaction with the N terminus, which was truncated in the receptor construct used in the determination of the crystal structure. In the D₂R/spiperone simulations, the phenyl substitution on the triazaspiro[4.5]decane moiety protrudes toward the interface between TM2 and TM3, and contacts Trp100^EL1.50, which is flexible as well and can adopt a position that is even further away from the OBS than that of Trp96^EL1.50 in the D₃R structure (Figure 6—figure supplement 1).

In contrast, when EL2 is in an extended conformation like that in D₃R, it restricts the flexibility of Trp100^EL1.50 (Video 3). This trend is consistent with the D₃R/eticlopride simulations in which we do not observe any significant rearrangement of Trp96^EL1.50 (Figure 6d,e,f).

Thus, we infer that the distinct conformation of Trp100^EL1.50 in the D₂R structure is a combined effect of the helical EL2 conformation and the favored interaction that Trp100^EL1.50 can form with the bound risperidone in the crystal structure, the latter of which however, has a limited influence on the binding affinity of risperidone (Wang et al., 2018), consistent with the unstable interaction between risperidone and Trp100^EL1.50 in our simulations (Figure 6, Video 2). Indeed, in the fully extended EL2 conformation in which Ile183^EL2.51 rotates to face the extracellular vestibule, Ile183^EL2.51 makes a direct contact with the bound risperidone, whereas Trp100^EL1.50 loses its interaction with the ligand entirely (Video 1). Nevertheless, risperidone retains all other contacts in the OBS. In the recently reported 5-HT_2AR/risperidone structure (PDB: 6A93) Kimura et al. (2019), risperidone has a very similar pose in the OBS as that in the D₂R structure, occupying the Ile^3.40 sub-pocket as well. However, on the extracellular side of the OBS, EL2 in the 5-HT_2AR/risperidone complex is in an extended conformation and the EL2 residue Leu228^EL2.51 contacting risperidone aligns to Ile183^EL2.51 of D₂R, whereas the conserved Trp141^EL1.50 does not interact with risperidone in the 5-HT_2AR. It is tempting to speculate that the EL2 and EL1 dynamics we observe in the D₂R/risperidone simulations represents a more comprehensive picture, as the divergent interactions shown in the extracellular loops of the 5-HT_2AR/risperidone and D₂R/risperidone structures may not result from differences in the protein sequences of this dynamic region between these two receptors but rather two different static snapshots due to differences in the crystallographic conditions (Note risperidone has similarly high affinities for both D₂R and 5HT_2AR; Kimura et al., 2019; Wang et al., 2018).

Thus, the plasticity of the OBS and the dynamics of the extracellular loops appear to be two relatively separated modules in ligand recognition. To the extent of our simulations, we did not detect strong ligand-dependent bias in the EL2 dynamics as we did for the OBS. However, when EL2 is helical, the EL1 dynamics are sensitive to the bound ligand (compare Figure 6 and Figure 6—figure supplement 1); when EL2 is extended, it restricts EL1 dynamics (Figure 6).

To further investigate the dynamics and coordination of EL2 and EL1 loops, we mutated Leu94^2.64, Trp100^EL1.50, and Ile184^EL2.52, and evaluated the effects of the L94A, W100A, and I184A, mutations on the binding affinities of eticlopride, risperidone, and spiperone. As shown in Figure 6—figure supplement 2, Leu94^2.64 and Trp100^EL1.50 are closely associated in both the D₂R and D₃R structures, while Ile184^EL2.52 interacts with Trp100^EL1.50 only in the D₂R structure. In our time-resolved energy transfer (Tr-FRET) binding experiments, using a fluorescently labeled spiperone derivate (spiperone-d2) as a tracer ligand, we found that both L94A and W100A significantly reduced the affinities of all tested antagonists, whereas I184A only reduced the affinity of eticlopride while it improved that of risperidone (Table 3). Thus, the effects of the L94A and W100A mutations have similar trends, which appear independent of the effect of I184A. Indeed, for Trp100 to switch between the positions in the D₂R and D₃R structures, it must pass the steric hinderance of the sidechain of Leu94; thus, some effects of the L94A mutation may reflect its perturbation of the positioning of Trp100, and vice versa.

These findings support our conclusions that the close interaction between Ile184^EL2.52 and Trp100^EL1.50 revealed by the D₂R/risperidone crystal structure is not necessary for the stabilization of the risperidone pose. Indeed, in our simulations, EL2 has significant intrinsic dynamics and transitions from the helical to unwound conformation independent of the bound ligands (see above). When it is in an extended conformation, Ile184 is dissociated from Trp100.

Virtual screening has been widely used as an initial step in drug discovery for novel ligand scaffolds. To this end, we found that D₂R can significantly change its binding site shape to accommodate antagonists bearing different scaffolds, while EL2 is intrinsically dynamic. Thus, it is necessary to comprehensively consider the binding site conformations in virtual screening campaigns against D₂R, because limiting the screening to only a single conformation will miss relevant ligands. Indeed, the strategy of ensemble docking, in which each ligand is docked to a set of receptor conformers, has been adapted in recent virtual screening efforts (Amaro et al., 2018).

To characterize the OBS conformational ensemble sampled by D₂R in complex with ligands bearing different scaffolds in the context of EL2 dynamics, we clustered the OBS conformations in our representative D₂R/eticlopride and D₂R/risperidone MD trajectories in which EL2 transitioned from helical to unwound conformations (see Materials and methods). As expected, the OBS conformations in these two complexes are significantly different and can be easily separated into distinct clusters. For the clustering results shown in Table 4, the average pairwise RMSDs of the OBS residues (apRMSDs, see Materials and methods) between the D₂R/eticlopride and D₂R/risperidone clusters are >1.1 Å, which are similar to that between the D₂R and D₃R structures (1.2 Å), while the apRMSDs within each cluster is smaller than those between any two clusters (Figure 7). Interestingly, at this level of clustering, when the two clusters for each complex are ~0.8–0.9 Å apRMSD away from each other, the extended and helical conformations of EL2 are always mixed in a cluster (Table 4). This observation suggests that the helical versus extended EL2 conformations are not closely associated with the OBS conformations.

Thus, while the centroid frames from each cluster can form an ensemble for future virtual screening for the primary scaffold occupying the OBS, in order to discover novel extended ligands that protrude out of the OBS to interact with EL2 and EL1 residues (Michino et al., 2015a), additional frames that cover both helical and extended EL2 conformations from each cluster will have to be used to screen for the optimal extensions of the primary scaffold.

Based on a systematic analysis of aminergic receptors, we found a Trp in the middle of EL1 and the disulfide-bonded Cys in the middle of EL2 are the most conserved residues in each segment, and defined their residue indices as EL1.50 and EL2.50, respectively (Michino et al., 2015a), In this study, for the convenience of comparisons among D₂R, D₃R, and D₄R, and 5-HT_2AR, based on the alignments of EL1 And EL2 shown in Figure 5—figure supplement 1, we index the EL1 and EL2 residues of each receptor in the same way as the Ballesteros-Weinstein numbering, for example the residues before and after the EL2.50 are EL2.49 and EL2.51, respectively. Note the indices for the shorter sequences are not necessarily be consecutive, given the gaps in the alignment.

The D₂R models in this study are based on the corrected crystal structure of D₂R bound to risperidone (PDB: 6CM4) (Wang et al., 2018). We omitted T4 Lysozyme fused into intracellular loop 3. Three thermostabilizing mutations (Ile122^3.40A, L375^6.37A, and L379^6.41A) were reverted to their WT residues. The missing N terminus in the crystal structure was built de novo using Rosetta (Bradley et al., 2005), and then integrated with the rest of the D₂R model using Modeller (John and Sali, 2003). Using Modeller, we also extended two helical turns at the TM5 C terminus and three residues at the TM6 N terminus of the structure and connected these two ends with a 9 Gly loop, similar to our experimentally validated treatment of D3R models (Michino et al., 2017). The position of the Na⁺ bound in the canonical Na⁺-binding site near the negatively charged Asp^2.50 was acquired by superimposing the Na⁺-bound structure of adenosine A_2A receptor (Liu et al., 2012) to our D₂R models.

The binding poses of risperidone and eticlopride were taken according to their poses in the D₂R (Wang et al., 2018) and D₃R (Chien et al., 2010) structures, respectively. Docking of spiperone in our D2R model was performed using the induced-fit docking (IFD) protocol (Sherman et al., 2006) in the Schrodinger software (release 2017–2; Schrodinger, LLC: New York NY). Based on our hypothesis regarding the role of the Ile^3.40 sub-pocket in the Na⁺ sensitivity (see text), from the resulting poses of IFD, we choose the spiperone pose with the F-substitution on the butyrophenone ring occupying the Ile^3.40 sub-pocket. Note that in risperidone and spiperone the F-substitutions have similar distances to the protonated N atoms that interact with Asp^3.32 (measured by the number of carbon atoms between them, Figure 1—figure supplement 1).

MD simulations of the D₂R and D₃R complexes were performed in the explicit water and 1-palmitoyl-2-oleoylphosphatidylcholine (POPC) lipid bilayer environment using Desmond MD System (version 4.5; D. E. Shaw Research, New York, NY) with either the OPLS3e force field (Roos et al., 2019) or the CHARMM36 force field (Best et al., 2012; Klauda et al., 2010; MacKerell et al., 1998; MacKerell et al., 2004) and TIP3P water model. For CHARMM36 runs, the eticlopride parameters were obtained through the GAAMP server (Huang and Roux, 2013), with the initial force field based on CGenFF assigned by ParamChem (Vanommeslaeghe et al., 2010). The system charges were neutralized, and 150 mM NaCl was added. Each system was first minimized and then equilibrated with restraints on the ligand heavy atoms and protein backbone atoms, followed by production runs in an isothermal–isobaric (NPT) ensemble at 310 K and one atm with all atoms unrestrained, as described previously (Michino et al., 2017; Michino et al., 2015b). We used Langevin constant pressure and temperature dynamical system (Feller et al., 1995) to maintain the pressure and the temperature, on an anisotropic flexible periodic cell with a constant-ratio constraint applied on the lipid bilayer in the X-Y plane. For each condition, we collected multiple trajectories, the aggregated simulation length is ~392 μs (Table 1).

While the majority of our D₂R simulations in this study used the OPLS3e force field, to compare with the D₃R simulations using CHARMM36 that have been continued from the previously reported shorter trajectories (Michino et al., 2017; Michino et al., 2015b), we carried out the D₂R/eticlopride simulations using both the OPLS3e and CHARMM36 force fields (see Table 1). We did not observe significant differences and pooled their results together for the analysis.

Distances and dihedral angles of MD simulation results were calculated with MDTraj (version 1.8.2) (McGibbon et al., 2015) in combination with in-house Python scripts.

To characterize the structural changes in the receptor upon ligand binding, we quantified differences of structural elements between the D₂R/eticlopride and D₂R/risperidone conditions (using last 600 ns from a representative trajectory for each condition), by applying the previously described pairwise interaction analyzer for GPCR (PIA-GPCR) (Michino et al., 2017). The subsegments on the extracellular side of D₂R were defined as following: TM1e (the extracellular subsegment (e) of TM1, residues 31–38), TM2e (residues 92–96), TM3e (residues 104–113), TM4e (residues 166–172), TM5e (residues 187–195), TM6e (residues 364–369), and TM7e (residues 376–382).

For the PIA-GPCR analysis in Figure 4 and the distance analysis in Figure 6, we used the set of ligand-binding residues previously identified by our systematic analysis of GPCR structures. Specifically, for D₂R, they are residues 91, 94, 95, 100, 110, 111, 114, 115, 118, 119, 122, 167, 184, 189, 190, 193, 194, 197, 198, 353, 357, 360, 361, 364, 365, 367, 368, 376, 379, 380, 383, 384, 386, and 387; for D₃R, they are residues 86, 89, 90, 96, 106, 107, 110, 111, 114, 115, 118, 165, 183, 188, 189, 192, 193, 196, 197, 338, 342, 345, 346, 349, 350, 352, 353, 362, 365, 366, 369, 370, 372, and 373.

For the clustering of the OBS conformations, we used representative D₂R/eticlopride and D₂R/risperidone MD trajectories in which EL2 transitioned from the helical to unwound conformations. For each complex, using the Ile183-Asn186 distance as a criterion to differentiate the EL2 conformation (Figure 5), 1000 MD frames with helical EL2 conformations and another 1000 frames with extended EL2 conformations were randomly selected. For these 4000 frames, the pairwise RMSD of the backbone heavy atoms of the OBS residues defined in Michino et al. (2015a), except for Ile184^EL2.52, were calculated. The resulting 4000 × 4000 matrix was used to cluster these frames using the k-mean algorithm implemented in R. We chose nstart to be 20 to assure the convergence of cluster centroids and boundaries. We chose the clustering level to be 4, so that the average pairwise RMSDs (apRMSDs) between the D₂R/eticlopride and D₂R/risperidone clusters are similar to that between D₂R and D₃R structures (1.2 Å), while all the apRMSDs within a cluster are smaller than those between any given two clusters. The same frame selection and clustering procedure was repeated to 20 times. The averages of these 20 runs for the compositions of each cluster were reported in Table 4.

The MSM analysis was performed using the pyEMMA program (version 2.5.5) (Scherer et al., 2015). To characterize the dynamics of EL2 of D₂R, specifically the transitions between helical and extended conformations of its C-terminal portion, we focused on a key hydrogen bond formed in the helical conformation between the backbone carbonyl group of Ile183 and the backbone amine group of Asn186. Thus, for each of the simulated conditions, the distance of Ile183-Asn186 (Ser182-Asn185 in D₃R) was used as an input feature for the MSM analysis. We discretized this feature into two clusters – distances below and above 4 Å (i.e. EL2 forming a helical conformation and unwinding). Implied relaxation timescale (ITS) (Swope et al., 2004) for the transition between these clusters was obtained as a function of various lag times. Convergences of ITS for the MSMs for all conditions was achieved at a lag time of 300 ns (Figure 5—figure supplement 6), which we further used to estimate Bayesian Markov models with 500 transition matrix samples (Trendelkamp-Schroer and Noé, 2013). The maximum likelihood transition matrix was used to calculate the transition and equilibrium probabilities (π) shown in Figure 5 and Figure 5—figure supplement 3.

Site-directed mutagenesis was performed using the Quickchange method using pEF5/DEST/FRT plasmid encoding FLAG-SNAP-D_2SR as the DNA template. The mutagenesis was confirmed, and the full coding region was checked using Sanger sequencing at the DNA Sequencing Laboratory (University of Nottingham). Stable cell lines were generated using the Flp-In recombination system (Invitrogen).

FlpIn CHO cells (Invitrogen) stably expressing WT or mutant SNAP-D2s cells were cultured before the preparation of cell membrane as described before (Klein Herenbrink et al., 2019). All stable cell lines were confirmed to be mycoplasma free. For saturating binding assays cell membranes (Mutant or WT SNAP-D_2s-FlpIn CHO, 2.5 µg) were incubated with varying concentrations of [³H]spiperone and 10 µM haloperidol as a non-specific control, in binding buffer (20 mM HEPES, 100 mM NaCl, 6 mM MgCl₂, 1 mM EGTA, and 1 mM EDTA, pH 7.4) to a final volume of 200 μL and were incubated at 37°C for 3 hr. For competition binding assays, cell membranes (SNAP-D_2s-FlpIn CHO, 2.5 µg) were incubated with varying concentrations of test compound in binding buffer containing 0.2 nM of [³H]spiperone to a final volume of 200 µL and were incubated at 37°C for 3 hr. Binding was terminated by fast-flow filtration using a Uniplate 96-well harvester (PerkinElmer) followed by five washes with ice-cold 0.9% NaCl. Bound radioactivity was measured in a MicroBeta2 LumiJET MicroBeta counter (PerkinElmer). Data were collected from at least three separate experiments performed in triplicate and analysed using non-linear regression (Prism 7, Graphpad software). For radioligand saturation binding data, the following equation was globally fitted to nonspecific and total binding data:(1)Y=BmaxAA+KA+NSAwhere Y is radioligand binding, B_max is the total receptor density, [A] is the free radioligand concentration, K_A is the equilibrium dissociation constant of the radioligand, and NS is the fraction of nonspecific radioligand binding. The B_max of the SNAP-tagged D2SRs we as follows; WT = 7.95 ± 1.63 pmol.mg⁻¹, 6.39 ± 1.04 pmol.mg⁻¹, 4.37 ± 0.92 pmol.mg⁻¹, 2.61 ± 0.50 pmol.mg⁻¹.

For competition binding assays, the concentration of ligand that inhibited half of the [³H]spiperone binding (IC₅₀) was determined by fitting the data to the following equation:(2)Y=Bottom+(Top-Bottom)1+10X-LogIC50nHwhere Y denotes the percentage specific binding, Top and Bottom denote the maximal and minimal asymptotes, respectively, IC₅₀ denotes the X-value when the response is midway between Bottom and Top, and nH denotes the Hill slope factor. IC₅₀ values obtained from the inhibition curves were converted to K_i values using the Cheng and Prusoff equation. No statistical methods were used to predetermine sample size.

The Go-protein activation assay uses a set of BRET-based constructs previously described (Michino et al., 2017). Briefly, HEK293T cells were transiently co-transfected with pcDNA3.1 vectors encoding (i) D₂R, (ii) Gα_oA fused to Renilla luciferase 8 (Rluc8; provided by Dr. S. Gambhir, Stanford University, Stanford, CA) at residue 91, (iii) untagged Gβ1, and (iv) Gγ2 fused to mVenus. Transfections were performed using polyethyleneimine (PEI) at a ratio of 2:1 (PEI:total DNA; weight:weight), and cell culture was maintained as described previously (Bonifazi et al., 2019). After ~48 hr of transfection, cells were washed with PBS and resuspended in PBS + 0.1% glucose + 200 µM Na Bisulfite buffer. Approximately 200,000 cells were then distributed in each well of the 96-well plates (White Lumitrac 200, Greiner bio-one). 5 μM Coelenterazine H, a luciferase substrate for BRET, was then added followed by addition of vehicle and test compounds using an automated stamp transfer protocol (Nimbus, Hamilton Robotics) from an aliquoted 96-well compound plate. Following ligands were used – quinpirole, eticlopride, raclopride, and (-)-sulpiride (Tocris Bioscience), (+)-butaclamol, dopamine, and risperidone (Sigma Aldrich), and Spiperone (Cayman chemicals). mVenus emission (530 nm) over RLuc 8 emission (485 nm) were then measured after 30 min of ligand incubation at 37°C using a PHERAstar FSX plate reader (BMG Labtech). BRET ratio was then determined by calculating the ratio of mVenus emission over RLuc eight emission.

Data were collected from at least nine independent experiments and analyzed using Prism 7 (GraphPad Software). Drug-induced BRET, defined by BRET ratio difference in the presence and absence of compounds, was calculated. Concentration response curves (CRCs) were generated using a non-linear sigmoidal dose-response analyses using Prism 7 (GraphPad Software). CRCs are presented as mean drug-induced BRET ± SEM. E_max bar graphs are plotted as the percentage of maximal drug-induced BRET by quinpirole ± SEM.

Materials: Spiperone-d2, SNAP-Lumi4-Tb and 5x SNAP/CLIP labeling medium were purchased from Cisbio Bioassays. Eticlopride hydrochloride was purchased from Tocris Bioscience. Saponin was purchased from Fluka/Sigma-Aldrich. Bromocriptine, haloperidol, risperidone, spiperone, pluronic-F127, Gpp(NH)p, DNA primers, Hanks Balanced Salt Solution H8264 (HBSS) and phosphate buffered saline (PBS) was purchased from Sigma-Aldrich.