A bitopic agonist bound to the dopamine 3 receptor reveals a selectivity site

Abstract

Although aminergic GPCRs are the target for ~25% of approved drugs, developing subtype selective drugs is a major challenge due to the high sequence conservation at their orthosteric binding site. Bitopic ligands are covalently joined orthosteric and allosteric pharmacophores with the potential to boost receptor selectivity and improve current medications by reducing off-target side effects. However, the lack of structural information on their binding mode impedes rational design. Here we determine the cryo-EM structure of the hD₃R:Gα_Oβγ complex bound to the D₃R selective bitopic agonist FOB02-04A. Structural, functional and computational analyses provide insights into its binding mode and point to a new TM2-ECL1-TM1 region, which requires the N-terminal ordering of TM1, as a major determinant of subtype selectivity in aminergic GPCRs. This region is underexploited in drug development, expands the established secondary binding pocket in aminergic GPCRs and could potentially be used to design novel and subtype selective drugs.

Developing subtype selective drugs for GPCRs is a major focus of research. Here, Arroyo-Urea et al. point to an unexploited selectivity site in aminergic receptors, as seen in the dopamine 3 receptor bound to a bitopic agonist.

Introduction

While G protein-coupled receptors (GPCRs) form the largest family of drug targets, accounting for more than a third of FDA-approved drugs¹, developing subtype-selective drugs is a major challenge. This is especially true for aminergic GPCRs, which include 42 receptors (dopamine, serotonin, adrenaline, histamine, acetylcholine, and trace amine receptors) with high sequence similarity. In the most closely related aminergic receptor subtypes, sequence identity often exceeds 80% of the orthosteric binding site (OBS) residues. Such conservation supports neurotransmitter promiscuity² between subtypes, but results in undesired off-target side effects of drugs that only bind in the OBS³. Controlling drug selectivity for aminergic receptors has the potential to improve current therapies and it could be achieved by the design of bitopic molecules^4–6. These are ligands generated by covalently joining two pharmacophores, a primary pharmacophore (PP), that usually targets the OBS, and a secondary pharmacophore (SP), that targets an allosteric or secondary binding pocket (SBP) generally divergent in sequence and/or structure within the target receptor^4–7. Hence, bitopic molecules have been proposed to have a separate “message-address” system wherein an agonist/antagonist, the message, is linked to a pharmacophore binding to the SBP, which contains the address^5,8. Indeed, several bitopic compounds with enhanced receptor selectivity have been developed for GPCRs^9–12. Overall, bitopic ligands present a rational approach to develop molecules with enhanced functionality and selectivity, however there is scarce structural information on their binding modes and development relies heavily on structure-activity relationships together with computational and synthetic strategies^13–16. Here we aim to understand the molecular basis of a selective bitopic molecule that distinguishes between two closely related aminergic GPCRs, the human dopamine D₂ receptor (D₂R) and dopamine D₃ receptor (D₃R). These receptors share 78% sequence identity at the transmembrane segment and 100% identity at the OBS, making their pharmacological distinction a notoriously hard challenge^17,18. D₂R and D₃R differ in brain distribution and signaling properties and are both targeted by current antipsychotics and drugs for the treatment of neurological diseases (such as Parkinson’s disease^19,20). Although agonists with some selectivity exist^20,21, new subtype selective molecules are likely to help understand their physiological role as well as provide leads for improved therapeutics. Indeed, selective activation of D₃R may yield neuroprotective effects in the treatment of Parkinson’s disease, hence harboring potential in the treatment of neurodegeneration^21,22.

In this work, we determine the cryo-electron microscopy (cryo-EM) structure of the human D₃R bound to a bitopic and full agonist (FOB02-04A) and coupled to a Gα_Oβγ heterotrimer. Together with functional assays, mutagenesis, docking studies, and molecular dynamics (MD) simulations, we determine the binding mode and basis for the D₃R selectivity of this compound. The bitopic molecule occupies the OBS and protrudes towards the outside of the ligand binding pocket to contact an allosteric site at the extracellular vestibule of D₃R formed by TM2-ECL1 and TM1. This region is of high sequence and structural variability and expands the established aminergic SBP, opening new avenues to develop subtype-selective bitopic drugs, potentially across other aminergic GPCRs.

Results

Overall cryo-EM structure of the hD₃R:Gα_Oβγ:scFv16 bound to a bitopic ligand

The D₃R:G_O heterotrimer:FOB02-04A complex was produced by co-expressing the hD₃R (L^3.41W mutation following Ballesteros–Weinstein numbering²³), the dominant negative Gα_O subunit²⁴, Gβ₁ and Gγ₂ in insect cells (see “Methods”) (Supplementary Fig. 1A). The D₃R^L3.41W was previously used in structural studies²⁵ and was validated in this work using cellular BRET assays²⁶, where it displayed a virtually identical ligand-induced activation as the wild-type D₃R (Supplementary Fig. 1B). The bitopic FOB02-04A was synthesized as previously described⁹ and was added before complex solubilization from insect cell membranes. The scFv16²⁷ (which binds to the Gα_O:Gβ interface) was incorporated prior to size exclusion chromatography. The structure of the complex was then solved by single-particle cryo-EM (Fig. 1 and Supplementary Fig. 2). Positioning the ligand binding pocket at the center of the cryo-EM box improved the resolution at the D₃R extracellular region (Methods and Supplementary Fig. 3), and allowed to classify two cryo-EM models containing two FOB02-04A conformations – Conformation A (to a global resolution of 3.05 Å) and B (global resolution of 3.09 Å), which mainly differed in the position of the bitopic SP and residues around the SBP (Fig. 1 and Supplementary Fig. 2). We will initially focus on Conformation A unless otherwise stated since Conformation B was concluded to be a non-productive antagonistic conformation (see below). Both final cryo-EM maps were of sufficient quality to build confidently the D₃R, the Gαβγ proteins, the scFv16, and the bitopic FOB02-04A ligand (Supplementary Fig. 4 and Supplementary Table 1). Both D₃R conformations were built from residues H29^1.32 (Conformation A)/Y32^1.35(Conformation B) to C400 with missing residues for intracellular loop 3 (ICL3) (missing residues including I223^5.73 to R323^6.29). No cholesterol (or cholesterol hemisuccinate) or lipid molecules were found around the transmembrane part of the receptor, consistent with previously reported structures of the D₂R^28–30 and D₃R^25,30,31 and in contrast to D₁R, D₄R, and D₅R where cholesterol was bound to the transmembrane segment^30,32,33.

Fig. 1. Overall cryo-EM reconstruction of the D₃R-G_O:FOB02-04A complex.

A, B Cryo-EM maps for the D₃R-G_OA:FOB02-04A complex in Conformation A (A) and B (B) are shown with an inset into the ligand binding site from the top view. Cryo-EM density is colored according to subunit with the bitopic ligand colored in red (Conformation A) and green (Conformation B). C, D Coordinates for Conformation A (C) and B (D) for both complexes are shown as cartoons and colored by subunit with the bitopic ligand colored in red (Conformation A) and green (Conformation B).

Activation mechanism and G_O coupling of the D₃R bound to FOB02-04A

The D₃R:Gα_Oβγ:FOB02-04A displays the characteristic structure of a GPCR:G protein complex, with resemblance to the previously determined structures of D₃R coupled to a G_i heterotrimer^30,31 (e.g., RMSD of 1.036 Å for 1022 Cα for the pramipexole bound structure (PDB 7CMU). No major conformational changes were found at the D₃R when comparing its structure when bound to pramipexole (PDB 7CMU), PD128907 (PDB 7CMV), rotigotine (PDB 8IRT) or FOB02-04A (0.535 Å RMSD over 253 Cα in the pramipexole bound as example) aside from the ordering of the extracellular region of TM1 (see below). The D₃R activation induced by FOB02-04A follows the canonical conformational changes³⁴, i.e., a downward shift of the toggle switch W342^6.48, a conformational change of the PIF (I118^3.40, F338^6.44), DRY (D127^3.49, R128^3.50, Y129^3.51) and NPxxY (N379^7.49, P380^7.50, P Y383^7.53) motifs, which end up with an ~ 9 Å outward swing of the cytoplasmic end of TM6 and inward movement of TM7 toward the core of the receptor as compared to the inactive state²⁵ (Supplementary Fig. 5). The coupling of the G_O heterotrimer to the D₃R occurs through two interfaces: a first major interface located between the Gα_O C-terminal α5, that engages mainly with the intracellular part of TM3, TM5 and TM6 of the D₃R (I344^G.H5.16, L348^G.H5.20, C351^G.H5.23, L353^G.H5.25, and Y354^G.H5.26 in Gα_O packing against R128^3.50, A131^3.53, V132^3.54, I211^5.61, L215^5.65, R218^5.68, R222^5.72, R323^6.29, K326^6.32, A327^6.33 and M330^6.36 in D₃R) with contributions from TM7 and TM2 (Fig. 2A). Of note, from MD simulations spanning five independent 0.6 µs runs of the D₃R bound to FOB02-04A and coupled to Gα_Oβγ within a membrane bilayer, alternating salt bridge interactions occurred between D341^G.H5.13 (superscript denotes CGN numbering system³⁵) of the Gα_O C-terminal α5 and the guanidinium groups of R218^5.68 and R222^5.72 in D₃R (Supplementary Fig. 6). A second interface is located at the intracellular loop 2 (ICL2), which makes interactions in a pocket formed by the Gα_O N-terminal helix, the C-terminal α5 and the loops connecting the β-strands. The interaction is also held together by unspecific electrostatic charges between the receptor and the Gα protein conserved among G_i/O coupled receptors^30,31.

Fig. 2. Coupling of D₃R to G_O heterotrimer.

A Interaction details of the D₃R:G_O interface when bound to FOB02-04A. Cryo-EM density of the C-terminal α5 is shown as mesh. B Interaction details of α5 interaction of G_O (turquoise, cryo-EM; orange, MD simulations) and G_i (violet, PDB 7JVR) with ICL2 of D₃R (yellow, cryo-EM structure; orange, MD simulation) and D₂R (blue, PDB 7JVR). Mutation at residue G350^G.H5.22 to D in G_O-MD is highlighted in dark orange and shown with an asterisk. C Comparison of the C-terminal α5 interaction of G_i (PDB 7CMU) and G_O (D₃R-G_O:FOB02-04A). D Interaction details of N-terminal G_O (turquoise) and G_i (violet) with ICL2 of D₃R. E Concentration-response curve of D₃R upon G_OA and G_i1 activation by FOB02-04A using the TRUPATH assay. pEC₅₀ values are means ± SEM of five independent experiments performed in technical triplicate. F pEC₅₀ values for D₃R in response to G_OA mutants activation by FOB02-04A using the TRUPATH assay. Data are presented as means ± SEM of three (G_OA I28^G.HN.52E, N194^G.s2s3.02D, Y354^G.H5.26F)^, four (G_OA-V334^G.H5.06F, G350^G.H5.22D,) and five (G_OA WT) independent experiments performed in technical triplicate. *p < 0.05 (one-way ANOVA with Dunnett post hoc analysis) for G350^G.H5.22D (p = 0.048). All source data within this figure is provided as a Source Data file.

The D₃R has been shown to couple preferentially to G_O compared to G_i³⁶. We have validated the D₃R G_O/G_i preference by using cellular BRET assays in HEK293T cells²⁶, which displays a ~ 135-fold difference in potency between G_O and G_i (Fig. 2E). The current D₃R:G_O structure allows us to compare it with the previously determined D₃R:G_i complex to search for potential differences that could explain such D₃R coupling preference (Fig. 2C, D). Overall, both structures exhibit a similar interface area, with D₃R-G_O having only a slightly lower buried surface area than D₃R:G_i (959.4 Ų and 1051.8 Ų for G_O and G_i coupled D₃R, respectively). However, a smaller interface area is usually seen in G_O vs G_i couplings irrespective of selectivity^37,38. In addition, both structures present a similar outward swing in TM6 irrespective of G_O or G_i coupling (Fig. 2C), in line with previous observations of the same receptor coupled to different Gα proteins keeping the magnitude of TM6 outward swing^39,40. However, differences occur when looking at the C-terminal α5 interactions of G_O vs G_i. In the case of Gα_O, the terminal Y354^G.H5.26 points toward TM5, in contrast to its equivalent F354^G.H5.26 in Gα_i, which is sandwiched between R323^6.29 and K345^G.H5.17 (this residue is specific for Gα_i, A345^G.H5.17 in Gα_O) (Fig. 2C). Furthermore, previous studies suggested that native ICL contacts are essential to achieve G_O selectivity in D₃R^36,41. Structural differences were also found at the interaction made by ICL2 where, in G_O, Q139^34.54 moves away from the α5 of G_O to interact with K32^G.hns1.03 in the αN (Fig. 2D). Such interaction was further confirmed in MD simulations, whose interacting distance remained constant along the five trajectories spanning 0.6 µs each (Supplementary Fig. 6). In addition, the interaction between Q144^ICL2 and E28^G.HN.52 in the G_i αN is lost when coupled to G_O due to a replacement of E28^G.HN.52 by isoleucine as well as the slight difference in the positioning of G_O with respect to the receptor. In order to understand the residues responsible for the G_O selectivity at the D₃R we identified all residues involved at the D₃R:Gα protein interface which differed between G_O and G_i and we reverted them one at a time to G_i over the Gα_O background, which involved I28E^G.HN.52, N194L^G.s2s3.02, V334F^G.H5.06, G350D^G.H5.22, and Y354F^G.H5.26 (Fig. 2F and Supplementary Fig. 5). Overall, only mutation of residue G350^G.H5.22, located at the C-terminal α5, had a significant impact on its own and hence this residue is the most determinant of the D₃R:G_O selectivity (Fig. 2F). G350^G.H5.22 from Gα_O packs against ICL2, which in Gα_i the bulkier G350D^G.H5.22 substitution might present steric strains, hindering G_i coupling. In fact, MD simulations suggest that a second rotamer of H140^34.55 is sampled that occupies the gap left at G350^G.H5.22, and such rotamer would clash with the G350D^G.H5.22 substitution in Gα_i, all together promoting the coupling to G_O vs G_i (Fig. 2B and Supplementary Fig. 5). Furthermore, in D₂R (a receptor without Gα_i/O selectivity) ICL2 is displaced outward from the receptor core, yielding a wider cavity and posing no steric restrictions to either Gα_O or Gα_I coupling at this position (Fig. 2B). Overall, ICL2 seems to play a relevant role in determining the Gα_i/O selectivity at the D₃R.

The binding mode of the bitopic agonist FOB02-04A at D₃R

Bitopic molecules are composed of a PP (binding at the OBS), an SP (binding at the allosteric site), and a linker. FOB02-04A is a full agonist bitopic molecule composed of a non-catechol PP (based on PF592,379, an aminopyridinyl-based scaffold), an SP with an indole-amide group, and a (1 R,2S)-cyclopropyl linker moiety whose chirality has been optimized for ligand binding and selectivity^9,42,43 (Fig. 3A). The cryo-EM density allowed modeling of the three components of the bitopic ligand. Unlike other agonists, which target the bottom of the pocket exclusively, FOB02-04A binds to the OBS and runs along a narrow channel towards the allosteric site in the extracellular vestibule, interacting with residues from TM1-3 and TM5-7 (Fig. 3B–D). The SP of FOB02-04A is found protruding out of the tight channel to bind in the extracellular vestibule of D₃R, occupying most of the ligand binding pocket, in contrast to pramipexole which only occupies 23% of the pocket volume (Fig. 3B, C). Each component of the bitopic molecule (PP, linker, and SP) occupies a different region within the D₃R pocket, overall defined by a combination of hydrophobic and polar interactions, as described in Fig. 3E.

Fig. 3. Binding of the bitopic FOB02-04A to the D₃R receptor.

A Schematic of dopamine, pramipexole, rotigotine and the bitopic FOB02-04A ligand shown as sticks and colored by component. B Binding of the secondary pharmacophore (SP) (sticks, dark red) to a groove-shaped pocket at the D₃R (yellow, surface representation) formed by ECL1 and TM1. C Two views of a comparison of FOB02-04A (dark red carbon, sticks), pramipexole (green carbon, sticks), and rotigotine (cyan carbon, sticks) binding into the D₃R pocket (yellow, surface representation). Dashed circles indicate OBS, established SBP, and the SBP_2-ECL1-1 site. D Overall binding mode of the bitopic molecule to the D₃R and ordering of TM1 upon bitopic binding. FOB02-04A (dark red, sticks) is displayed on superposed structures of D₃R bound to eticlopride (cartoon, cyan) and FOB02-04A (cartoon, yellow) E Schematic of the FOB02-04A binding into the D₃R ligand binding pocket. F Binding details of FOB02-04A (dark red, sticks) at the D₃R (yellow sticks) with cryo-EM density as gray mesh. G Binding details of FOB02-04A (dark red, sticks) at the D₃R (yellow cartoons) with residues at the ligand binding pocket colored by functional effect when mutated to alanine: decreased efficacy – green carbons, decreased potency – blue carbon and non-detectable binding – red carbon. H pEC₅₀ values for alanine mutation of the residues at the ligand binding site in response to G_OA activation by FOB02-04A using the TRUPATH assay. All data are means ± SEM of four independent experiments (n = 4) performed in technical triplicates except for D110^3.32A, S196^5.46A, Y365^7.35A, T369^7.39A, W342^6.48A and Y373^7.43A for which there was n = 3, WT, V86^2.61A, L89^2.64A, E90^2.65A, ∆G94^ECL1, F346^6.52A for there was n = 5, and L119^3.41W, T115^3.37A for which there was n = 6. *p < 0.05 (one-way ANOVA with Dunnett post hoc analysis) for D₂R (p = 0.0081), V111^3.33A (p = 0.0049), T115^3.37A (p = 0.0074) and I183^45.82A (p = 0.0013) and nd - non-detectable. I Concentration-response curve of D₃R ΔG94^ECL1 upon G_OA activation by quinpirole (light blue, n = 4) and FOB02-04A (deep blue, n = 5) (shown as net BRET). All data are means ± SEM of the specified biological replicates, each performed in technical triplicates. J Emax values for alanine mutation of the residues at the ligand binding site in response to G_OA activation by FOB02-04A using the TRUPATH assay. Emax values have been normalized to D₃R WT. All data are means ± SEM of four independent experiments performed in technical triplicate (n = 4) except for D110^3.32A, S196^5.46A, Y365^7.35A, T369^7.39A, W342^6.48A, Y373^7.43A (n = 3), WT, V86^2.61A, L89^2.64A, E90^2.65A, ∆G94^ECL1, F346^6.52A (n = 5) and L119^3.41W, T115^3.37A (n = 6). *p < 0.05 (Holm-Sidak multiple comparisons tests two-tailed p value) for H29^1.32A (p = 0.016), V86^2.61A (p = 0.026), F106^3.28A (p = 0.003), F346^6.52A (p = 0.019). K Concentration-response curves of D₃R H29^1.32A (orange), H29^1.32F (pink), H29^1.32K (yellow), and H29^1.32R (green) upon G_OA activation by FOB02-04A (shown as net BRET). All data are means ± SEM derived from three independent experiments (n = 3), each performed in technical triplicate except for H29^1.32A (n = 4). All source data within this figure is provided as a Source Data file.

The PP pocket at the OBS is defined by strong salt bridge interactions with D110^3.32, and a cavity formed by S196^5.46, F345^6.51, F346^6.52, W342^6.48, V111^3.33, T115^3.37 and I183^45.52, with an additional weak H-bond with S192^5.42 (Fig. 3). To correlate structural information with functional activity, most of the residues involved in ligand binding were mutated to alanine, following quantification of their surface expression and measurement of their ligand-induced activation using functional BRET assays in HEK293T cells²⁶ (see Methods and Supplementary Fig. 7). At the OBS there were critical residues which showed no detectable activity when mutated to alanine such as the conserved D110^3.32, which forms a stable charge interaction with almost all agonists in aminergic receptors, and W342^6.48, the conserved toggle switch residue at the bottom of the OBS pocket that is essential for signaling. In addition, I183^45.52, which sandwiches the ligand from the extracellular side (ECL2), V111^3.33, and T115^3.37 had a significant impact on agonist potency when mutated (Fig. 3H, I). V111^3.33 is specifically relevant for FOB02-04A since its mutation does not have an impact on the D₃R-induced activation by pramipexole, rotigotine, and PD128907^30,31. In turn, T115^3.37 is relevant for FOB02-04A and pramipexole in contrast to PD128907 and rotigotine. Finally, an agonist interaction with S192^5.42 is found within most aminergic receptor-agonist pairs, however, it seems to be less important for FOB02-04A binding (Fig. 3 and Supplementary Fig. 7). This is in line with non-catechol agonists not relying heavily on S192^5.42 for binding and activation⁴⁴ (also observed for pramipexole³¹). A conserved hydrophobic pocket between T369^7.39 and F345^6.51 is efficiently occupied by the rotigotine, pramipexole and PD128907 propylamine group, while it is barely occupied by a methyl group by FOB02-04A (Supplementary Fig. 8). This may explain the lack of effect of F345^6.51A upon activation by FOB02-04A and suggests that a larger hydrophobic group at this position might improve its binding.

The linker component of the FOB02-04A, which connects PP and SP, interacts with residues at the established SBP in aminergic receptors^3,13, an unexploited region in pramipexole and PD128907 but occupied by the propylthiophene group in rotigotine³⁰. The pocket is formed by residues V86^2.61, F106^3.28, T369^7.39, and Y373^7.43 and has been proposed to have different plasticity among dopamine receptors, and hence a source for ligand specificity³⁰. In the case of FOB02-04A, three residues showed a significant reduction in activity when mutated to alanine: Y373^7.43, F106^3.28, and V86^2.61. Y373^7.43A showed non-detectable activity, and, although this residue is known to be relevant for maintaining the D110^3.32 geometry to make the conserved charged interactions with agonists, its mutation does not have such a pronounced effect on the activity of pramipexole, dopamine and PD128907 ³¹ as it has on the activity of rotigotine or FOB02-04A. This suggests a role for this residue in the binding and/or function of the bitopic molecule to the receptor, in addition to its known role with D110^3.32. In addition, the alanine mutation of F106^3.28 and V86^2.61 showed reduced efficacy. This is likely to be FOB02-04A specific since V86^2.61A did not reduce efficacy upon pramipexole activation³¹. Overall, the linker connecting the PP and SP has an active role in the D₃R selective binding and function of FOB02-04A and its related bitopic analogs³¹.

Finally, the FOB02-04A SP binds in a groove-shaped pocket at the receptor extracellular region, denoted as SBP_2-ECL1-1, and is formed by the tips of TM1, TM2, and ECL1. Remarkably, in contrast to prior D₃R structures – whether in active or inactive conformations – the outermost extracellular residues of TM1 undergo a rearrangement that positions H29^1.32’s imidazole group, situated between TM2 and TM7, to stack with the 1H-indole group of the ligand SP. Given the absence of H29^1.32 in preceding D₃R cryo-EM^30,31 and crystal structures²⁵, we sought to ascertain the orientation of the imidazole moiety of H29^1.32. For this purpose, we performed comparative MD simulations, involving two D₃R complexes coupled to Gα_Oβγ and bound to either FOB02-04A or pramipexole (PDB 7CMU), both within a membrane bilayer and aqueous milieu and executed across five parallel runs of 0.6 µs each. MD analysis elucidated a more consistent localization of H29^1.32 between TM2 and TM7 when complexed with FOB02-04A relative to pramipexole. In this conformation, the H29^1.32 side chain is directed towards the SBP_2-ECL1-1, engaging with the SP of FOB02-04A bitopic ligand (Supplementary Fig. 9). Although the protonated N(ε) atom of H29^1.32 imidazole and the carboxyl entity of E90^2.65 are too distant to support strong polar or ionic interactions, the D₃R complex with the bitopic ligand FOB02-04A exhibited a narrower distance distribution than in pramipexole complex (Supplementary Fig. 9). In addition, in the D₃R-FOB02-04A complex, the N(ɛ) atom of H29^1.32 consistently interacts with the backbone oxygen of E90^2.65. Conversely, when complexed with pramipexole, three of the five trajectories show this distance consistently surpassing 10 Å. This observation reinforces that, while in the FOB02-04A:D₃R complex, the H29^1.32 side chain is predominantly positioned in the SBP_2-ECL1-1 where it is stabilized by the ligand, in the pramipexole-bound complex H29^1.32 side chain points away, likely due to the absence of the allosteric pharmacophore in pramipexole (Supplementary Fig. 9).

To gain further insights into the SBP_2-ECL1-1 role, we mutated all residues within this site to alanine (except for G94^ECL1, which was deleted) and measured ligand-induced activation using BRET2 assays. These experiments revealed that the deletion of G94^ECL1, which prevents ECL1 from reaching the SP, is essential for FOB02-04A activity (Fig. 3H, I). A previous study identified G94^ECL1 as a key determinant for binding of a similar bitopic molecule, however, only a reduction in affinity was observed (using radioactive ligands and fluorescence)⁴⁵ while, in the current study, ligand-induced activity seemed to be fully ablated. This suggests that FOB02-04A could potentially still bind in the ΔG94^ECL1 variant (although with lower affinity) but triggers no detectable Gαβγ activation, hence G94^ECL1 is likely to determine affinity and efficacy. As a control, the ΔG94^ECL1 variant was activated by quinpirole, a ligand that does not reach ECL1, highlighting the specific effect of the mutation on the activation by the bitopic FOB02-04A (Fig. 3I). Further mutational analysis of residues within SBP_2-ECL1-1 identified H29^1.32 as a key residue, with only a slight reduction in potency (~ 3-fold reduction in EC₅₀) but a significant decrease in efficacy upon alanine mutation (Fig. 3J, K). Previous studies predicted how slight variations in the position of the PP at the D₃R OBS could modulate compound efficacy⁴⁶. Since several residues at the SBP modulate FOB02-04A efficacy, it is likely that the linker and SP conformation are currently optimal to position the PP for maximal efficacy at the OBS, and that mutations around the SBP restrict conformations of the FOB02-04A PP to less efficacious alternatives. This scheme yields a marked segregation of the functional roles of the protein residues for each bitopic component. While mutations significantly decreasing potency (> 100-fold the EC₅₀) are primarily found at the OBS, mutations at the SBP mainly decrease FOB02-04A efficacy (Fig. 3G). This suggests that the SP is not only involved in D₃R selectivity (see below) but also in optimally positioning the PP for activity. Such conclusions are in line with previous suggestions originating in computational and functional assays⁴⁷. In order to understand the role of H29^1.32 in SBP binding, we mutated it to residues with different physicochemical properties and sizes (aside from alanine), including H29^1.32F, H29^1.32K, and H29^1.32R. While the H29^1.32K variant displayed a reduced efficacy similar to H29^1.32A when activated with FOB02-04A, both H29^1.32F and H29^1.32R displayed a further reduction in efficacy, likely due to potential steric clashes with the SBP of the bitopic agonist (Fig. 3K). In contrast, the H29^1.32A and H29^1.32F variants did not produce such a marked effect when activated with quinpirole and pramipexole, respectively (Supplementary Fig. 7). Interestingly, the H29^1.32A variant had reduced efficacy when activated with pramipexole, an agonist that does not reach H29^1.32 (Supplementary Fig. 7). In the cryo-EM structure as well as in the MD simulation, H29^1.32 was seen to engage in an H-bond network with E90^2.65. Such a site might be structurally important for the SBP in the D₃R, and hence mutating H29^1.32 might disrupt the interaction with FOB02-04A but also distort the D₃R binding pocket. It is not unusual for residues at the most extracellular sites to have an impact on intrinsic receptor function⁴⁸. Hence, we cannot discard that the functional effects of H29^1.32 variants on bitopic binding might contain additional contributions not related to ligand binding.

Additional analysis of the MD trajectories with the D₃R-FOB02-04A complex suggested a more robust interaction of FOB02-04A with D₃R than pramipexole. This was observed by looking at the stable salt bridge interaction between the trans-cyclopropyl amine group of FOB02-04A and the carboxyl group of D110^3.32 in D₃R (which underscores the stable binding pose of the 6-(aminopyridin-3-yl)-5-methylmorpholine PP moiety) (Supplementary Fig. 6). However, for the pramipexole-bound D₃R complex, three out of five MD trajectories displayed substantial deviations in either the equivalent salt bridge interaction with pramipexole amino group, as well as the interactions distance between S196^5.46 in D₃R and the pramipexole’s amino group. Since pramipexole and FOB02-04A have similar binding affinities, the propensity of pramipexole towards dissociation observed during MD simulations suggests potential faster association and dissociation rates, in line with the larger bitopic molecule, requiring longer times for association and dissociation (Supplementary Fig. 6).

Overall, the bitopic agonist FOB02-04A uses all three components (PP, linker, and SP) to make critical interactions with the ligand binding pocket since each component contributes with one critical interaction which, if mutated, the ligand-induced activation as a whole, is eliminated. This highlights that selective bitopic molecules are required to bind en bloc and that the SP which contains the address component is required to contribute significantly to the overall ligand function, otherwise selectivity would be lost.

Structural basis of FOB02-04A D₃R/D₂R selectivity

The bitopic FOB02-04A ligand has been designed for its PP to carry the agonist message while the SP carries the address, and has been reported to be 50-fold more selective for D₃R over D₂R⁹. Since quantification of selectivity at the D₃R/D₂R is assay and condition-dependent^6,9,17, we measured the D₃R/D₂R selectivity using cellular BRET assays, which confirmed the 50-fold selectivity (Supplementary Fig. 7). D₃R and D₂R have 78% sequence similarity at the transmembrane region and, residues within interacting distance of FOB02-04A, showed high structural similarity and 100% sequence identity at the OBS and established SBP²⁴. However, FOB02-04A interactions with the G94^ECL1 and H29^1.32 within the SBP_2-ECL1-1 form a region that is structurally and sequence-diverse between D₃R/D₂R. The D₃R TM2-ECL1 harbors an extra glycine residue that is absent in D₂R (93GGV95 in D₃R vs 98GE99 in D₂R), which allows this region to interact with the SP in the D₃R and not in the D₂R (Fig. 4A). Deletion of the extra glycine G94^ECL1 in D₃R ablates ligand-induced activation by FOB02-04A (Fig. 3I and Supplementary Fig. 7). This is in line with previous studies where similar bitopic molecules showed reduced affinity in D₃R lacking G94^{ECL1 45}. This reduction in activity makes G94^ECL1 the most critical residue for D₃R/D₂R selectivity. In addition, H29^1.32 is positioned in TM1, the most sequence-diverse transmembrane helix in GPCRs, and that, within D₃R/D₂R, shows both sequence and structural diversity (Fig. 4A). Exploiting this unforeseen H29^1.32 conformation has potential for the development of selective D₃R agonists.

Fig. 4. Sequence and structural diversity of the SBP_2-ECL1-1 in aminergic GPCRs.

A Comparison of the D₃R (yellow cartoons with relevant residues as sticks) and D₂R (light blue cartoons with relevant residues as sticks) TM2-ECL1 and TM1 regions within reach of FOB02-04A (dark red, sticks). B Relative binding sites of other bitopic ligands bound to D₂R (haloperidol (PDB 6LUQ), spiperone (PDB 7DFP), risperidone (PDB 6CM4) and 5-HT₁AR (aripiprazole (PDB 7E2Z), IHCH-7179 (PDB 8JT6), vilazodone (PDB 8FYL) and buspirone (PDB 8FYX) as shown on the D₃R:FOB02-04A cryo-EM structure. C Sequence alignment of TM1 and TM2-ECL1 regions in aminergic GPCRs with residues around the SBP_2-ECL1-1 embedded in a box. Sequence conservation is color-coded above each residue position (gradient from dark red, conserved, to dark blue, non-conserved). Structural differences at the SBP_2-ECL1-1 site among closely related adrenergic receptors (D, E) and serotonin receptors (F, G). Receptors are shown as cartoons colored by receptors with relevant residues shown as sticks.

Diversity of the SBP_2-ECL1-1 in other aminergic receptors

There are 9 groups of (clinically relevant) closely-related aminergic receptors sub-types (M_1-5, ARα_1A-1D, ARα_2A-2C, ARβ_1-3, D₁, and D₅, D₂-D₄, H_3-4, 5-HT_1A-1F, 5-HT_2A-2C) for which sequence similarity poses problems to generate subtype selective ligands. Selectivity can arise from sequence diversity, structural divergence as well as differences in structural plasticity. Using sequence alignments and the recent explosion in GPCR structural information, we assessed whether the SBP_2-ECL1-1 is a novel site of high diversity that could be exploited to develop subtype-selective drugs in other aminergic GPCRs. We first compared the D₃R:FOB02-04A complex with structures of other aminergic receptors bound to bitopic ligands that sample different secondary binding sites within the different receptors. Some examples include the D₂R bound to spiperone (PDB 7DFP)⁴⁹, risperidone (PDB 6CM4)²⁹ or haloperidol (6LUQ)⁵⁰ and the serotonin 5HT_1A bound to aripiprazole (7E2Z)⁵¹, IHCH-7179 (8JT6)⁵², Vilazodone (8FYL)⁵³ or buspirone (8FYX)⁵³. A structural superposition of these complexes shows that no other ligand interacts with G94^ECL1 or H29^1.32 equivalent residues in the D₂R or the 5HT_1AR (Fig. 4B). The closest binding mode would be that of vilazodone binding on the serotonin 5HT_1A receptor, where the terminal amide group of vilazodone is close to N100 in the ECL1 (the G94 equivalent), but it is out of H-bonding distance as modeled in the cryo-EM structure.

A further analysis of this site showed that the SBP_2-ECL1-1 is variable either in sequence, structure, or both within most aminergic receptor subtypes (Fig. 4C–F). The amount of diversity at the SBP_2-ECL1-1 varies within each subfamily, with the least variable being the muscarinic receptors where TM1 is too far apart to contribute in all available structures and the equivalent G94 position is only different in M₃R (N131^ECL1 vs a glycine residue in M₁, M₂, M₄ and M₅). However, there are marked differences in several other subgroups. First, the serotonin 5-HT₁ and 5-HT₂ groups show variable sequence or structure at the G94^ECL1 equivalent position while TM1 is too far apart (Fig. 4C, D). In addition, the recent structural determination of all five dopamine receptors (D₁R-D₅R) highlighted the SBP_2-ECL1-1 as the most variable region between them³⁰. Finally, there are groups with marked differences at the SBP_2-ECL1-1 site, e.g., the ARα_2A-2C subgroup. ARα_2A-2C shows differences at the G94^ECL1 equivalent position, while they have an increasingly ordered TM1, which could potentially contribute to specific interaction in each receptor. While in ARα_2B TM1 is far apart, it is longer in ARα_2A, where it could contribute with main chain atoms of Y43, and in ADα_2C, where the N-terminus folds over the TM2-ECL1 site providing with additional specific residues (Fig. 4D). In ARβ_1-3, the TM2-ECL1 has structural and sequence divergence that could be used to design highly subtype selective bitopic molecules (Fig. 4C). Overall, the SBP_2-ECL1-1 site is a major specificity region that is underexploited for developing subtype-selective drugs. However, this site is far away from the canonical ligand binding site and might be better accessible with bitopic molecules.

Alternative FOB02-04A conformation at the ligand binding site

Docking of FOB02-04A to the D₃R reliably reproduced its binding mode when compared to the cryo-EM structure. Yet, a second conformation of FOB02-04A was revealed with comparable docking scores, suggesting a second plausible orientation (Fig. 5C). In the alternative binding mode, termed Conformation B, the 1H-indole-2-carboxamide SP of FOB02-04A is seen to interact with a less hydrophobic pocket defined by the polar side chains S182^41.51, Y365^7.35, as well as V360^ECL3 and P362^7.32 residues, termed hereafter SBP_ECL2-ECL3. Notably, π-π stacking interactions between the 1H-indole part of FOB02-04A and Y365^7.35 stabilize Conformation B (Fig. 5). MD simulations indicated that the indole SP of FOB02-04A oscillates between SBP_2-ECL1-1 (Conformation A) and the comparatively less hydrophobic SBP_ECL2-ECL3 (Conformation B). A detailed examination of the proximity between D₃R E90^2.65 and the FOB02-04A SP (accentuated with a red palette) juxtaposed with proximity measurements between D₃R Y365^7.35 and the FOB02-04A SP (illustrated in green pallet) provides insights into the temporal predominance of FOB02-04A’s Conformation A versus Conformation B (Fig. 5D). Subsequent frequency analyses showed that Conformation A, that engages SBP_2-ECL1-1, is predominant with an estimated 90% prevalence, in contrast to the 10% observed for Conformation B, targeting SBP_ECL2-ECL3 region (Fig. 5D and Supplementary Fig. 6). This information triggered a targeted search for Conformation B within the cryo-EM dataset, which resulted in a model at 3.09 Å resolution (Fig. 1 and Supplementary Figs. 2, 4). In this model, cryo-EM density supports the second conformation for the FOB02-04A SP so as to make π-π stacking interactions with Y365^7.35 in a similar manner as found in docking and MD simulations (Fig. 5A, B and Supplementary Fig. 6). Interestingly, in this cryo-EM map, the extracellular residues of TM1, including H29^1.32, are not resolved, reminiscent of the pramipexole, rotigotine and PD128907 bound D₃R structures (Fig. 5B). This suggests that binding of the SP to the SBP_2-ECL1-1 stabilizes the TM1 conformation described above (in agreement with our MD simulations). A comparison of particle numbers between cryo-EM models of Conformation A and B also supported a predominance of Conformation A over B (~ 60%). In order to understand the role of Conformation B on the function of FOB02-04A, we performed BRET assays on the Y365^7.35A variant, which exclusively affects Conformation B, resulting in wild-type pharmacological properties (Fig. 3 and Supplementary Fig. 7). This is in stark contrast to the ΔG94^ECL1 variant (only affects Conformation A) which fully ablated Gαβγ dissociation. There are at least two potential explanations for such results. On the first, the bitopic ligand behaves as an antagonist/weak partial agonist when in Conformation B. Such a hypothesis would be in line with our previous observation that residues at the SBP, as well as the position of the SP, are highly relevant for the optimal positioning and efficacy of the PP. In support of this, a minor twist of the PP at the OBS is observed in Conformation B with respect to Conformation A, and minor modifications at the position of the ligand at the D₃R OBS have been shown to regulate ligand efficacy. However, we cannot rule out that the slight difference in PP position is a consequence of the low map resolution. The second explanation would be that Conformation B is an agonist but a very low populated conformation (10% predicted by MD simulations) with too low affinity to be detected in functional assays. This conformation would not be as selective as Conformation A since all residues forming the binding site are conserved between D₂R and D₃R. We assessed whether Conformation B could be occurring at D₂R and could account for the activity of FOB02-04A at D₂R despite not having G94^ECL1 and an H29^1.32 equivalent (with G94^ECL1 being essential for the activity of FOB02-04A in D₃R). However, Y408^7.35A in D₂R did not result in a signaling loss in functional assays (Supplementary Fig. 7), suggesting that alternative binding modes are likely to exist in D₂R aside from Conformation A and B equivalents. Since D₂R is more plastic than D₃R, the binding mode of this bitopic molecule to D₂R might be hard to predict and structural studies would be required.

Fig. 5. Conformation A and B within the D₃R-G_O:FOB02-04A complex.

Coordinates of the D₃R (yellow cartoon, with relevant residues as sticks) are shown with FOB02-04A in Conformation A (dark red, sticks) superposed to Conformation B (green, sticks). Cryo-EM density is shown as gray mesh for Conformation A (A) and Conformation B (B) with both superposed FOB02-04A conformations. C Predicted binding poses of bitopic FOB02-04A with D₃R obtained by MD simulations showing Conformation A and Conformation B with intramolecular interactions shown as black dashed lines. Black arrows indicate distances for assessing bitopic FOB02-04A binding pose distribution between Conformations A and B with specified closest distances (E90^2.65 carboxyl group in D₃R to FOB02-04A indole atom N5 and from Y365^7.35 4-hydroxyphenyl moiety in D₃R to the phenyl ring of FOB02-04A 1H-indole-2-carboxamide SP). A semi-transparent skin reveals the receptor molecular surface, which is colored by residue properties (red (acidic), blue (basic), green (hydrophobic)). D Interaction dynamics between D₃R E90^2.65 and FOB02-04A SP (depicted in brown palette) compared with proximity distance between D₃R Y365^7.35 and the SP of FOB02-04A (shown in green palette) suggest that FOB02-04A predominantly adopts Conformation A over B. Data from five independent simulations of the D₃R-Gα_Oβγ heterotrimer complex are shown, spanning 0.6 μs of cumulative time per system, with a sampling rate of 10 frames per ns. Solid lines and same-color shadows represent moving average values and one standard deviation, respectively, from 50 frames in all cases.

Discussion

Aminergic receptors are highly relevant drug targets, but the high sequence and structural similarity within the family pose a great challenge to developing subtype-selective drugs. Here, we have reported the cryo-EM structure of the human D₃R in complex with the D₃R-selective bitopic agonist, FOB02-04A, and coupled to a G_O heterotrimer. FOB02-04A binds D₃R with all three components (PP, linker, and SP), fully exploiting the OBS, established SBP, and a new extended SBP_2-ECL1-1 that confers FOB02-04A with D₃R selectivity. This SBP_2-ECL1-1 is structurally and/or sequence diverse also in aminergic receptors and could potentially be used to develop subtype-selective ligands. Especially interesting is the TM1 contribution to ligand binding since it is the most sequence-diverse transmembrane region in GPCRs, rarely contributes to ligand binding, and could be exploited through the use of bitopic molecules with the required composition and length. Mutational profiling of the ligand binding site showed marked segregation in functional roles of the residues at the OBS and the SBP. While the majority of mutations that impaired potency were located mainly at the OBS, mutations that impaired efficacy were enriched at the SBP. This highlights the relevant role of the SP binding in optimally positioning the PP at the OBS for maximal activity. The computational design of bitopic molecules might benefit from taking such roles into consideration. In addition, the mutational analysis pointed to a mutually PP, linker, and SP-dependent binding mode, i.e., all components contribute with essential interactions for the en bloc binding of the bitopic molecule. This is likely required when higher selectivity is desired since independent binding might yield promiscuous PP binding. Therefore, the message and address components should not be treated as separate entities when developing specific bitopic molecules, but rather working together in tandem with the appropriate linker in between⁶.

A second antagonist/non-selective conformation of the FOB02-04A bitopic molecule is proposed, which suggests that care should be taken when developing subtype selective bitopic molecules since the position of the PP at the OBS seems to be altered easily (at least for the D₃R in the case of FOB02-04A) and bitopic molecules tend to be large and flexible, and alternative non-productive conformations might obscure highly specific and potent conformations in functional assays. Such problems likely contribute to the challenges associated with developing agonist bitopic molecules⁶. Including structural determination in the drug development pipeline is likely to accelerate future progress. Additional structural information on other bitopic-receptor complexes might shed light on this topic.

Regarding the D₃R/D₂R selectivity, a recent report describing the structures of the five dopamine receptors (D₁R-D₅R) pointed towards H^6.55 as a specificity determinant, since this residue changes conformation between D₂-like receptors in an agonist-dependent manner³⁰. While H^6.55 was located far away from the bitopic molecule under study, molecules with combined interactions at H^6.55 and the extended SBP_2-ECL1-1 site have the potential to yield highly specific molecules within D₂-like receptors. Such molecules could help to improve current treatments targeting the D₃R, a current target for Parkinson’s disease and other neurological disorders and neuropsychiatric disorders, including substance use disorders^54–56.

Overall, this work extends the usable SBP in aminergic receptors exploiting an extracellular region of high sequence and structural variability and highlights new insights and pitfalls into the development of highly selective subtype selective bitopic molecules with desired functional efficacies.

Methods

Construct design and molecular cloning

All mutagenesis and molecular cloning procedures were performed using the in vivo DNA assembly method^57,58. The cDNAs of the human D₃R (HASS-FLAG-EGFP-3C_protease-D₃R with the L^3.41W mutation) and human dominant-negative Gα_OA subunit (S47N^G.H1.02, G204A^G.s3h2.02, E246A ^G.H3.04, M249K ^G.H3.07 and A326S ^G.s6h5.03)²⁴ were obtained through gene synthesis (Gene Fragments, Twist Bioscience) and cloned into the pBacPak8. Rat His₈-Gβ₁ (pBacPak8), human Gγ₂ (C68S) (pBacPak8), and a baculovirus expressing the scFv16 with a gp67 secretion signal and a C-terminal His₈-tag were a gift from Christopher G. Tate’s laboratory. For BRET assays in HEK293T cells, the same human HASS-FLAG-EGFP-3C_protease-D₃R construct was sub-cloned into the pcDNA4/TO vector, upon which all mutants were made (including the wild-type D₃R). Constructs containing the Gα_ΟΑ-RLuc8, Gβ_3, and Gγ₉-GFP2 in pCDNA5 and pCDN3.1 were a gift from Bryan Roth´s lab (Addgene plasmid kit # 1000000163). The Gα_ΟΑ mutants to study the G_i/O selectivity of the D₃R were done in the Gα_ΟΑ-RLuc8 construct. The oligonucleotides used can be found in Supplementary Data 1.

D₃R:Gαβγ:scFv16 production and purification

The scFv16 was produced by infecting Trichoplusia ni (Tni, Expression Systems, not authenticated in-house) cells grown in ESF921 media (Expression Systems) at a density of 2-3 × 10⁶ cells/ml and incubated for 48 h at 29 °C. The supernatant was clarified by centrifugation, dialyzed to 20 mM Tris-base pH 8, 500 mM NaCl, and 20 mM imidazole, and loaded onto a pre-equilibrated HisTrap Excel column (Cytiva). The scFv16 was eluted with an imidazole linear gradient, concentrated, and loaded onto a Superdex 200 10/300 GL increase column (Cytiva) equilibrated in 20 mM HEPES pH 7.5 and 100 mM NaCl. Pure protein was concentrated to 4.2 mg/ml, flash-frozen, and stored at – 80 °C until further use.

For the production of the D₃R:Gαβγ protein complex, recombinant baculoviruses expressing D₃R, Gα_OA, Gβ₁ and Gγ₂ were prepared using the FlashBAC ULTRA® system (Oxford Expression Technologies). Tni cells were grown in suspension in ESF921 media to a density of 2-3 × 10⁶ cells/ml, co-infected with D₃R, Gα_OA, Gβ₁, and Gγ₂ baculoviruses (1:1:1:1 ratio) and shaken at 29 °C for 48 h. Cells were harvested, flash-frozen in liquid nitrogen, and stored at − 80 ˚C for further use. Cell pellets were thawed in 20 mM HEPES pH 7.5, 150 mM NaCl, 10% glycerol, 20 mU/mL apyrase and protease inhibitors cocktail (10 µM E-64, 0.05 µg/ml aprotinin, 0.02 µg/ml leupeptin, 10 µM benzamidine HCl, 0.01 µg/ml pepstatin, 10 µM bestatin and 10 µM PMSF) and incubated with 10 µM of FOB02-04A (compound 53a in ref. ⁹.) for 30 minutes at 4 °C. Cells were then solubilized with 0.5% (w/v) lauryl maltose neopentyl glycol (LMNG, Anatrace) supplemented with 0.071% (w/v) cholesterol hemisuccinate (CHS, MP Biomedicals ™) at 4 °C for 1 h. The sample was clarified by centrifugation and the supernatant was incubated with Talon Superflow (GE Healthcare Life Sciences) resin overnight at 4 °C. The resin was then washed with 20 column volumes of 20 mM HEPES pH 7.5, 150 mM NaCl, 10% glycerol, 0.007:0.001% LMNG/CHS, 10 µM FOB02-04A and 5 mM imidazole followed by 20 CV of the same buffer with 20 mM imidazole. The sample was eluted with 20 mM HEPES pH 7.5, 150 mM NaCl, 10% glycerol, LMNG 0.003%, 10 µM FOB02-04A and 250 mM imidazole. The complex was concentrated and incubated with pure scFv16 at a molar ratio of 1:1.1 (D₃R:Gαβγ:FOB02-04A to scFv16) at 4 °C for 30 min. The resulting complex was purified with a Superose 6 Increase 10/300 GL column (Cytiva) equilibrated with 20 mM HEPES pH 7.5, 25 mM NaCl, 0.003:0.0004% LMNG/CHS and 10 µM FOB02-04A. Pure protein was concentrated at 2.8 mg/ml and FOB02-04A ligand was added to a final concentration of 50 µM.

Cryo-grid vitrification and data collection

3 µl of D₃R:Gαβγ:FOB02-04A:scFv16 at 2.8 mg/ml were applied to 300 mesh Quantifoil 0.6/1 Au grids previously glow discharged with a Leica EM Ace200 Vacuum Coater at 15 mA for 60 s and vitrified with ethane using a Vitrobot Mark IV (FEI Company). Data collection was carried out in a Titan Krios at 300 kV using a K3 detector at the European Synchrotron Radiation Facility (ESRF). A total of 22,655 movies were recorded at a magnification yielding 0.84 Å/pixel with a dose rate of 17.6 e^-/pixel/s and a defocus range between − 1 to − 3 µm using the Smart EPU Software (ThermoFisher Scientific). Movies were split into 50 frames each and exposed to a total dose of 50 e^-/Ų (1e^-/Ų per frame) using a total exposure time of 2 s.

Cryo-EM Data processing

RELION-4.0⁵⁹ was used for all data processing unless otherwise specified. Drift and beam-induced motion correction (5 × 6 patches) were performed using MotionCor2⁶⁰ along with dose weighting. Contrast transfer function (CTF) estimation and determination of defocus range were performed with CTFFIND-4.1⁶¹. Automated particle picking was carried out with Topaz⁶². The initial particles were reduced to 475,951 after 2 rounds of 2D and 3D classifications (using an ab initio model). The best model was refined and subjected to CTF refinement and Bayesian polishing following a 3D classification focused on the receptor (with a mask around the receptor) that yielded 429,908 particles. Refinement of this set of particles yielded a model at 3.16 Å but poor cryo-EM density at the SBP. To improve map quality at the ligand binding site two parallel processing paths were pursued with the 429,908 particle set: 1) a recentering of the particles at the ligand binding site (re-extracted in a 320-pixel box) followed by 3D classification (resulting in 360,038 particles), and 2) 3D classifications with a mask at the extracellular half of the receptor followed by a recentering of the particles at the ligand binding site (as described before) which were further 3D classified (resulting in 176,315 particles). The two sets of particles were merged and duplicates removed, yielding 275,383 particles which were refined using for the last iteration a mask that precluded the Gα_O-helical domain and the micelle. Post-processing resulted in a cryo-EM map for Conformation A at 3.05 Å. Conformation B was obtained by performing a 3D classification on the 429,908 particles set with a mask on the extracellular half of the receptor, resulting in a model with 252,959 particles, which were subsequently re-centered at the ligand binding site and further 3D classified. Particles belonging to the best model, with 201,219 particles, were subjected to heterogeneous refinement and 159,184 particles were lastly refined through non-uniform refinement in CryoSPARC⁶³. This resulted in a cryo-EM map at 3.09 Å according to the gold-standard FSC of 0.143. Local resolution was calculated using CryoSPARC for both models.

Model building

Model building and refinement were carried out using the CCP-EM software⁶⁴ and Phenix⁶⁵. The D₃R, Gβ₁, Gγ_2, and scFv16 starting coordinates were taken from the Gα_i-coupled D₃R structure (PDB 7CMV)³¹. The Gα_O starting coordinates were taken from the Gα_O-coupled α_2β adrenoreceptor structure (PDB 6K41). D₃R was modeled from residue H29^1.32 to I223^5.73 and from R323^6.29 to C400 in Conformation A (Conformation B starts at Y32^1.35). Gα_O was modeled from T4^G.HN.10 to K54^G.H1.09, T182^G.hfs2.05 to V234^G.s4h3.07 and N242 ^G.s4h3.15 to Y354^G.H5.26. Jelly-body refinement was performed in REFMAC5⁶⁶ followed by manual modification and restraint real space refinement in Coot⁶⁷ and Phenix. A dictionary describing the ligand FOB02-04A and its coordinates was created using AceDRG⁶⁸ and manually fitted into the density for its latter refinement in real space using Coot and Phenix. B factors were reset to 40 Ų prior to refinement. The final model achieved good geometry (Supplementary Table 1) with validation performed in Coot, EMRinger⁶⁹, and Molprobity⁷⁰. The goodness of fit of the model to the map was carried out using Phenix, using a global model-vs-map FSC correlation (Supplementary Table 1).

Cellular BRET assays

pEC₅₀ values were determined using cellular BRET2 assays with the TRUPATH system²⁶ on HEK293T (ATCC, not authenticated in-house). 50,000 cells/well were seeded in previously poly-lysined 96-well white plates with clear bottoms. The following day, cells were transfected with TransIT-2020 (Mirus Biosciences) at the ratio of 2:1:1:1 of D₃R:Gα_OA-RLuc8:Gβ₃:Gγ₉-GFP2 (7:1:1:1 for D₃R-Y373^7.43A, D₃R-ΔG94^ECL1 and D₂R-Y408^7.35A) following manufacturer instructions. After 48 h, the medium was replaced by 90 µl/well of freshly prepared assay substrate buffer (1 × Hank’s balanced salt solution, 20 mM HEPES pH 7.4, Coelenterazine 400a 7.5 µM). 10 µl of each concentration of FOB02-04A was added and the plate was read using a CLARIOstar (BMG Labtech) with 400 nm (RLuc8-Coelenterazine) and 498.5 nm (GFP2) emission filters at integration times of 1.85 s. BRET ratios were calculated as the ratio of the GFP2 signal to the Rluc8 signal. Equivalent expression of the Gα_ΟΑ-RLuc8 variants was confirmed by monitoring luminescence at 400 nm. Data analysis was performed using GraphPad Prism 8.0.1. Data were normalized and a four-parameter logistic curve was fit into the data. Data are presented as mean ± SEM of at least three independent experiments performed in technical triplicate. Source data is provided as a Source Data File.

Surface expression quantification

HEK293T cells (ATCC, not authenticated in-house) were plated in previously poly-lysine 96-well white plates (50,000 cells/well) and transfected the next day with the D₃R and D₂R variants using PEI MAX® at a 2:1 ratio (PEI:DNA). After 48 h, cells were washed twice with 1X Phosphate Buffered Saline (PBS) and fixed with 4% paraformaldehyde for 20 min at RT. Cells were then washed three times with PBS for 5 min and 100 µl of 1X PBS with 5% BSA (w/v) was added to each well and incubated at RT for 30 min. Subsequently, media was replaced with 1X PBS-5% BSA with an anti-Flag HRP conjugate (1:10,000) and incubated at RT for 30 min. Cells were then rinsed twice with PBS and 50 µl of HRP substrate (Clarity Max™ Western ECL Substrate) was added to each well and incubated for 5 min prior to chemiluminescence detection using a CLARIOstar (BMG Labtech). Data analysis was performed using GraphPad Prism 8.0.1 Chemiluminescence values were normalized to D₃R WT and presented as a ratio of D₃R WT. Data are presented as mean ± SEM of three independent experiments performed in technical triplicate. Source data is provided as a Source Data File.

Molecular dynamic simulations

The Gromacs simulation engine (version 2020.3)⁷¹ was used to run all molecular dynamics simulations under the Charmm36 force field topologies and parameters^72,73. Charmm force field parameters and topologies for the ligands FOB02-04A and pramipexole were generated using Charmm-GUI’s “Ligand Reader & Modeler” tool⁷³. The loop grafting and optimization for modeling missing side chains and loops were performed in the ICM-Pro v3.9-2b molecular modeling and drug discovery suite (Molsoft LLC)⁷⁴. The structurally conserved H29^1.32, A30^1.33, and Y31^1.34 at the N-terminus in the pramipexole bound D₃R (PDB ID: 7CMU)³¹ were modeled using human FOB02-04A bound D₃R as the template structure. The lobe in Gα_O protein was modeled using a human agonist-bound CB2-Gα_i structure (PDB ID: 6PT0)⁷⁵. Structure regularization and torsion profile scanning were done using ICMFF force field⁷⁶. The FOB02-04A-bound and pramipexole-bound structures of D₃R complexes coupled to a Gα_Oβγ heterotrimer were then uploaded to the Charmm-GUI webserver^72,77, where the starting membrane coordinates were determined by the PPM⁷⁷ server using the Charmm-GUI interface. The complexes were then embedded in a lipid bilayer composed of 1,2-dipalmitoyl-sn-glycero-3-phosphatidylcholine (DPPC), 1,2-dioleoyl-sn-glycero-3-phosphatidylcholine (DOPC), and cholesterol (CHL1) following the recommended ratio of 0.55:0.15:0.30 respectively⁷⁸. The FOB02-04A bound D₃R coupled to a Gα_Oβγ heterotrimer contained 220 DPPC, 60 DOPC, and 120 CHL1 lipids, 38818 water molecules, 112 sodium and 104 chloride ions. The pramipexole-bound D₃R coupled to a Gα_Oβγ heterotrimer contained 220 DPPC, 60 DOPC, and 120 CHL1 lipids, 37934 water molecules, 108 sodium and 102 chloride ions. Both systems were first subjected to 50000 steps of initial energy minimizations, then 60 ns of equilibration, followed by production runs of up to 600 ns. The simulations were carried out on GPU clusters at the University of Southern California’s High-Performance Computing Center. Since the structural insights into the binding mode of the D₃ receptor bound to a bitopic agonist were efficiently achieved using standard MD simulations, without the need to explore rare events or surmount significant energy barriers, no enhanced sampling methods were required. The temperature of 310 K and the v-rescale thermostat algorithm were used during the production run⁷⁹. MD simulations were conducted using standard methods without the need for enhanced sampling techniques. The analyses of molecular dynamics trajectories were performed with MDTraj software package⁸⁰.

Molecular docking

The D₃R structure was taken from the current work. The protein-stabilizing single-chain antibody scFv16 was removed from the D₃R structure, leaving the receptor protein subunit. The protein was processed via the addition and optimization of hydrogens and optimization of the side chain residues. Prior to conducting molecular docking, pramipexole underwent chiral definition and formal charge assignment. The compounds’ molecular models were created from their two-dimensional representations, and their three-dimensional geometry was refined using the MMFF-94 force field⁸¹. For docking simulations, a biased probability Monte Carlo (BPMC) optimization approach was employed, adjusting the internal coordinates of the compound based on pre-calculated grid energy potentials of the receptor⁸². The grid potentials, while preserving the receptor’s conformational state, considered receptor flexibility through the utilization of “soft” Van der Waals potentials. All-atom docking was performed with the energy-minimized structure of FOB02-04A employing an effort value of 5. The ligand docking box was selected to encompass the extracellular half of the protein for potential grid docking. At least ten independent docking runs were conducted, with the three distinct lowest energy conformations being retained from each run. Consistency across the docking results was assessed by comparing the ligand conformations that achieved the best docking scores. The unbiased docking procedure did not rely on distance restraints or any predefined information regarding the ligand-receptor interactions. From these docking experiments, two top-scoring docking solutions, referred to as Conformation A and Conformation B, representing FOB02-04A bound to D₃R complexes, were further refined. This refinement involved successive rounds of minimization and Monte Carlo sampling, focusing on the ligand conformation and including sidechain residues within 5 Å of the binding site. All the above-mentioned molecular modeling operations were performed in the ICM-Pro v3.9-2b molecular modeling and drug discovery suite (Molsoft LLC)⁷⁴.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

J.G.N. and S.A.U. designed the experiments. S.A.U. performed molecular cloning and mutagenesis, receptor expression, purification, complex formation, cryo-EM data collection, cryo-EM data processing, model building, and functional BRET2 assays. A.C.A. performed scFv16 purification and functional BRET2 assays. J.G.N. and S.A.U. performed structure analysis. A.B. and F.O.B. synthesized FOB02-04A, supervised by A.H.N. A.L.N. performed molecular dynamics simulations and docking studies. JHL helped perform molecular dynamics simulations. V.K. supervised the docking studies, molecular dynamics simulations, and manuscript discussion. The manuscript was written mainly by J.G.N. and S.A.U. and included contributions from all authors.

Peer review

Peer review information

Nature Communications thanks Maria Marti-Solano, Sanduo Zheng, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The coordinates for the D₃R:Gαβγ:scFv16 with the FOB02-04A in conformation A and B have been deposited in the Protein Data Bank with accession codes 9F33 (Conformation A D₃R:FOB02-04A structure) and 9F34 (Conformation B D₃R:FOB02-04A structure). The cryo-EM maps generated in this study have been deposited in the Electron Microscopy Data Bank (EMDB) with accession codes 50168 (Conformation A D₃R:FOB02-04A cryo-EM map) and 50169 (Conformation B D₃R:FOB02-04A cryo-EM map). The following existing PDB entries were used in the course of this study: 7CMU, 7CMV, 8IRT, 7DFP, 6CM4, 6LUQ, 7E2Z, 8FYL, 8FYX, 8JT6. The trajectories for the Molecular Dynamics simulations have been deposited as an open-access repo on Zenodo⁸³: Nazarova, A. (2024). Molecular Dynamics Trajectories for the D3 receptor (D3R) complexes bound with a Gα_Oβγ heterotrimer and 1) FOB02-04A bitopic agonist; 2) pramipexole (Zenodo 10800784). Source data are provided in this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-024-51993-4.