In order to crystallize the D3R, it was modified by introducing a point mutation in the transmembrane domain [Leu119^3.41Trp (15)] to enhance thermal stability (16), and replacing most of the third cytoplasmic loop (ICL3) (Arg222 to Arg318) with T4-lysozyme (D3R-T4L) (17). Further stabilization of the receptor was achieved by purifying with the antagonist eticlopride, which conferred the highest thermostability compared with five other ligands (18) (table S2). The engineered receptor retained near native ligand binding properties (table S3) and crystallized from a lipidic mesophase in an orthorhombic space group. Diffraction data were anisotropic, extending to 2.9 Å in the c* direction and 3.6 Å in the a* direction. Overall, the structure was determined at 3.15 Å and included all data up to 2.9 Å where an improvement in map quality was observed (see fig. S8 and table S1). The structure was determined with two receptors arranged in an antiparallel orientation in the asymmetric unit of the crystal (fig. S2). Both copies of the receptor are very similar (RMSD of 0.6 Å for the seven-transmembrane (TM) bundle) and will be treated identically in the discussion except where noted otherwise. The N-terminal 31 residues are not included in the deposited structure as they do not have interpretable density. The main fold of the D3R consists of the canonical seven-TM bundle of α-helices (Fig. 1A), which resembles previously solved GPCR structures (19–22). Subtleties in the orientations of these helices, as well as differences in the intracellular and extracellular portions of the receptor, confer the pharmacological and biochemical properties unique to the D3R.

The extracellular region in general is characterized by high sequence diversity among the GPCR family, which translates into high structural diversity in terms of the presence of varied secondary structure elements and the presentation of individual amino acids in the binding pocket (23–24). In the D2R and D3R, for instance, the second extracellular loop (ECL2) is much shorter than in the β-adrenergic receptors (βARs) and lacks the helical secondary structure. Interestingly, the portion of ECL2 in D3R (182–185) that contributes to the ligand binding pocket is quite similar to that in the βARs in both spatial positioning relative to bound ligand, and in the presentation of side chains in the ligand binding pocket. In the D3R, a disulfide bond is formed between Cys355 and Cys358 in ECL3 in addition to the canonical disulfide bond bridging ECL2 (Cys181) and helix III (Cys103^3.25) (25). Comparison of the D3R structure to the β₂AR structure reveals small shifts in the helical bundle; for example, helices VI and VII are shifted by ~3.5 Å (inward) and ~1.8 Å (outward), respectively, (Fig. 1B), whereas extracellular tips of helices III and V are about 3.5 Å closer to each other in the D3R as compared with the β 2AR structure. The latter shift can be explained by the fact that a segment of ECL2 connecting the tips of helixes V and III through a C181-Cys103^3.25 disulfide bond in D3R and other D2-like receptors is one amino acid shorter than in β₂AR and D1-like dopamine receptors (see fig. S4B, D).

A common feature thought to be important in many class A GPCRs is the ionic lock - a salt bridge between the charged Arg^3.50 in the conserved “D[E]RY” motif and Asp/Glu^6.30 at the cytoplasmic side of helices III and VI. This interaction is observed in all of the inactive rhodopsin crystal structures (Fig. 2A) (26–27) and has been implicated through mutagenesis as a major factor in stabilizing the receptors in the inactive conformation (28–29). Despite the presence of residues capable of forming this interaction, the ionic lock has not been found in any of the other GPCR crystal structures published to date (19–22) (Fig. 2C–E). The absence of this interaction is puzzling given its presumed importance and has been thought to be partly attributable to the inclusion of the T4L fusion protein within ICL3, which may induce a non-native helical conformation within this region. However, the presence of an intact ionic lock in both molecules in the D3R structure establishes the possibility of forming this interaction in the presence of T4L (Fig. 2B). The propensity for formation of the ionic lock, therefore, may indicate different distributions of conformational states in different receptors that may have direct implications on basal signaling activities. Differences between two molecules observed in the crystallographic asymmetric unit may highlight particular areas of receptor structure conformational flexibility. In chain A, ICL2 forms a 2.5 turn α-helix which runs parallel to the membrane (Fig. 1A). The observation of this α-helix in only one copy of the receptor may be because of the conformational dynamics of ICL2 and the associated regions (30), as in chain B, ICL2 is unstructured and the intracellular ends of helices IV and V are shifted ~2.9 Å closer to each other relative to their positions in chain A (fig. S3C). The two different conformational states of ICL2 observed in the D3R structure suggests that this helix is transient, raising the possibility that interactions between ICL2 and the receptor ionic lock may modulate the signaling properties of the D3R and perhaps contribute to the tolerance property in D3R signaling that persists after agonist is removed (31).

Strong electron density was observed for eticlopride in the binding cavity (fig. S3A, B), which is similar to the β₂AR pocket (Fig. 3C, D) as expected for receptors that bind closely related catecholamine ligands (32). The similarity includes a number of conserved side chains in the core binding site deep in the seven-TM bundle (10 of 18 eticlopride contact residues are conserved in the β₂AR), and open access to this site through a crevice from the extracellular side. Compared with the β₂AR, however, a part of the D3R access crevice is blocked by the inward shift of helices V and VI, and access to the ligand binding pocket is controlled by side chains of helices I, II, III, VII and ECL2.

Eticlopride occupies the part of the binding pocket defined by side chains from helices III, V, VI and VII (Fig. 1, ​,3A,3A, and table S4) that largely overlaps with the carazolol binding site in the β₂AR (Fig. 1B). The tertiary amine in the ethyl-pyrrolidine ring of eticlopride, is likely charged at physiological pH and forms a salt bridge (2.8 Å) to the carboxyl of Asp110^3.32, which is highly conserved in all aminergic receptors (Fig. 3A, B). This salt bridge is structurally and pharmacologically critical for high-affinity ligand binding to the aminergic subfamily of GPCRs (4, 33). Another key component of the eticlopride pharmacophore is a substituted aromatic ring connected to the pyrrolidine by an amide bond that fits tightly within a hydrophobic cavity formed by Phe345^6.51 and Phe346^6.52 in helix VI, Val189^5.39, Ser192^5.42, and Ser193^5.43 in helix V, and Val111^3.33 in helix III, as well as Ile183 in ECL2. Polar substituents (e.g. OH, OCH₃) in the phenyl ring form intramolecular hydrogen bonds with both the N and O of the amide, thereby maintaining the compound in an almost planar conformation (Fig. 3A, B), consistent with the small molecule crystal structure determination (13).

Of the 18 eticlopride contact residues in the D3R structure, 17 are identical in the D2R (Val350^6.56 is an isoleucine in D2R), whereas 5 differ in the D4R (see fig. S4). Qualitatively, this agrees with the finding that eticlopride, and some of its analogues, share similar affinities for the D2R and D3R with lower binding affinities for D4R. Mutation of 4 divergent residues in D2R to the aligned D4R residues led to a 3-order of magnitude enhancement of binding to a D4R-selective antagonist (34). Most of the differences in ligand binding specificity between D4R and D2R/D3R can therefore be explained by the differences in physicochemical properties of the contact side chains, as the mutated residues included three of the five nonconserved, eticlopride-contact residues - Val91^2.61Phe, Phe110^3.28Leu, and Tyr408^7.35Val.

The structural determinants of pharmacological specificity in the D3R and D2R are more subtle considering that the residues lining the binding pocket are essentially identical. In accordance with high conservation of the eticlopride binding site between D3R and D2R, the available SAR data suggests that, in order to achieve targeted selectivity (>100 fold), the ligand must extend towards the extracellular opening of the binding pocket (see (12) for a review). The D3R-selective pharmacophore consists of an extended aryl amide connected to an amine-containing scaffold by a relatively flexible 4-carbon linker (fig. S1) (35). Previous efforts to rationalize the structural basis of D3R selectivity have naturally focused on regions that are not conserved, with primary attention being given to ECL2, which has previously been implicated in ligand binding to the D2R (4, 36). Indeed, in chimeric studies, ECL2 has been found to play a role in both enantioselectivity and D3R-selectivity of a number of compounds in which the butylamide linker is functionalized (37). In addition, roles for both ECL2 and ECL1 have been demonstrated for the D3R-selective tetrahydroisoquinoline, SB 269,652 (fig. S1) (38).

To explore the structural basis of selectivity, we created a homology model of D2R based on the D3R structure (see Supplementary methods). Eticlopride could be reproducibly docked to the D3R structure and D2R model in highly similar orientations to that in the crystal structure. However, alignment of residues of D3R and D2R indicates substantial differences in their extracellular electrostatic surfaces that could impact binding of other longer and bulkier ligands (figs. S5, S6). Docking studies with the D3R-selective antagonist R-22 (37) revealed that the core amine-containing substituent (2,3-diCl-phenylpiperazine) binds in the essentially same binding pocket as eticlopride, whereas the indole-2-carboxamide terminus is oriented toward the extracellular part of the binding pocket comprised of ECL2/ECL1 and the junction of helices I, II and VII, defining a second extracellular binding pocket (orange ellipse in Fig. 4A) that includes conserved Tyr373^7.43 and Glu95^2.65 (Fig. 4B). However, the residue at 1.39, which is spatially positioned between Tyr373^7.43 and Glu95^2.65 is divergent (Tyr36^1.39 in D3R and Leu41^1.39 in D2R) (Fig. 4C, D). Moreover, Tyr36^1.39 is located in a stretch of five non-conserved residues at the extracellular end of helix I. Indeed, 44% of the extracellular half of helix I from 1.35 to 1.50 is not conserved between D2R and D3R (fig. S6), which should lead to functionally relevant changes in packing in D2R at the junction of helices I, II and VII (Fig. 4C, D, and fig. S7), consistent with previous structure-function investigations (39–41). The lack of conservation of Thr368^7.38 (Phe in D2R), which forms a hydrogen bond with the backbone of the conserved Tyr365^7.35 in the D3R, may also contribute to a shift in the relative position of helices I and VII (Fig. 4C, D) (28).

Such differences in packing and backbone configuration between the D2R and D3R, even when relatively subtle, are expected to lead to changes in selectivity even without changes in ligand contact side chains in the binding pocket. Indeed, molecular dynamics simulations of the D2R in an explicit lipid bilayer (see Supplementary methods) suggest a reorganization of ECL3 and helices I/II/VII that alters the configuration of the second binding pocket (Fig. 4D and fig. S7). Accordingly, the distance between the conserved residues Glu95^2.65 (in the second binding pocket) and Tyr373^7.43 (between the orthosteric binding site and the second binding pocket) is ~1 Å greater in the D3R than in the D2R because of distinct 2.65-1.39-7.43 interactions (Fig. 4C–D, and fig. S7), representing subtle but critical differences in the relative disposition between the orthosteric binding site and the second binding pocket in the D2R and D3R (Fig. 4B).

The crystal structure of the human D3R provides an opportunity to identify subtle structural differences, at the molecular level, between closely related GPCRs that can be exploited for novel drug design. In particular, the structural observation of an extracellular binding pocket, which may interact with bitopic or allosteric ligands, highlights the importance of the extracellular loops that were once thought to only provide superficial definition to ligand binding. Highly D2R and D3R subtype selective molecules will provide the tools necessary to parse behavioral actions associated with individual subtypes and identify mechanisms underlying side effects, resulting in improved medications for the treatment of neuropsychiatric disorders, including drug abuse.

Supplemental Materials