A Ligand-Triggered Receptor Conformation Enables the Design of Selective Agonists for the Dopamine 3 Receptor (D3R) Using a Bitopic Strategy Authors: Sandra Arroyo-Urea, Antonina L. Nazarova, Alexander Knieb, Hawau Abdulsalam, Khorshada Jahan, Ting Du, Song Gao, Amy Hauck Newman, Vsevolod Katritch, Javier García-Nafría, Alessandro Bonifazi Published: 2026 ABSTRACT While G protein-coupled receptors (GPCRs) represent the largest drug target family, designing subtype-selective molecules is still a challenge, especially to distinguish among closely related subtypes. One of the most challenging cases is the distinction between dopamine D2R and D3R, pivotal receptors in motor functions and cognition, and targets of Parkinson’s disease treatments, schizophrenia, or substance use disorders. Attempts to design D3R-selective molecules with ligands binding toward the first transmembrane helix (the most sequence-diverse and conformationally flexible segment in GPCRs but rarely participating in ligand binding) allowed us to discover a ligand-induced ordering of TM1 unique to D3R, yielding an unexploited selectivity site for drug development. Using rational bitopic drug design and the ligand-triggered conformation of the D3R we designed, synthesized, and characterized the most selective D3R agonists to date, >100,000-fold more selective than available ligands. More specifically, we report D3R partial agonists AB12-82 (6d) and AB13-73A (11), with >575,000- and >750,000-fold subtype selectivity, picomolar potency, and 85% and 49% efficacy, respectively. We also present the most selective full agonists reported to date, AB13-08 (4b) and AB13-46A (9), presenting low and subnanomolar potencies with >2,800- and 6,300-fold selectivity for D3R. Overall, we introduce a first-in-class pharmacological toolbox to dissect the (patho)-physiology of D3R, open new avenues for the design of improved neurotherapeutics, and show that using ligand-induced TM1 reorganizations might represent a promising strategy for the design of subtype-selective molecules in other GPCRs. Introduction G-protein-coupled receptors (GPCRs) represent the largest family of targets for FDA-approved drugs. The development of subtype-selective ligands targeting aminergic GPCRs belonging to the same subfamilies (e.g., dopamine receptors) represents a persistent challenge in medicinal chemistry. This is mostly due to the high level of sequence and structural homology existing between receptors of the same subfamilies in their orthosteric binding sites (OBS), where the endogenous neurotransmitters bind. Particularly relevant is the case of dopamine D2-like receptors (D2R, D3R, and D4R subtypes), where D2R and D3R share 100% sequence identity in the OBS, and 78% in their transmembrane domains. , D2R and D3R regulate motor functions, learning, memory, and overall cognitive processes, and are among the most abundant clinical targets being explored for the treatment of Parkinson’s disease (PD), impulsivity, and schizophrenia. ,− Still, there is a lack of subtype-selective agonists that can be used as research tools and therapeutics. High levels of D3R selectivity are needed to study its (patho)­physiological relevance, considering its overall lower expression compared to D2R, which makes selective in vivo targeting of the D3R even more challenging. One of the key approaches to achieving high ligand selectivity is bitopic drug design. , A bitopic drug consists of a primary pharmacophore (PP), targeting the OBS, covalently tethered to a secondary pharmacophore (SP) via a properly designed and functionalized linker. The SP would bind allosteric sites, affording selectivity to the bitopic molecule harboring an agonist moiety at the PP. Using this strategy we recently identified FOB02-04A (1), a 50-fold D3R/D2R selective agonist. Agonist 1 (tested as a trans-cyclopropyl linker mixture) was demonstrated to significantly decrease impulsive choice decision making in a dose-dependent manner in high-impulsive rats, unlike nonselective ligands such as PD128,907, pramipexole, or 7-OH-DPAT. Since animals with high impulsivity display a significant decrease in D3R expression in the nucleus accumbens shell when compared to D2R, this shows the benefits of targeting D3R selectively. Additionally, rac-trans-1 did not alter omissions in the cognitive processes, suggesting no altering effects on motivation, reward, and attention. A further increase in D3R selectivity, affinity, and potency might provide improved behavioral outcomes with lower doses, with the concomitant advantage of completely avoiding undesired overactivation of D2R, thus minimizing the risk of behavioral, cognitive, and motivational side effects. Moreover, highly selective agonists might provide new pharmacological tools to evaluate D3R cell- and receptor-specific mechanisms toward translational neurotherapeutic applications and medication development. To understand the binding mode and selectivity mechanism of 1, we recently reported the cryo-EM structure of the human D3R coupled to a GαOβγ heterotrimer and bound to compound 1. In this structure we identified a new TM2-ECL1-TM1 region, which required the N-terminal ordering of TM1 upon agonist binding (disordered in other structures) (Figure ). TM1 is the most sequence-diverse and flexible transmembrane helix in GPCRs but rarely interacts with modulators. Hence, the presence of 1 identified a region in the binding pocket that differed significantly in sequence and structure from D2R, forming a new, unexploited selectivity site in the D3R (Figure ). The key residues of this D3R site are G94ECL1 and H291.32 (Ballesteros–Weinstein numbering), which correspond to a gap and Y1.32 in D2R, respectively (Figure S1). However, there is not only sequence diversity, TM1 is positioned further away from the core of the receptor in D2R, making such a site unique in D3R. We hypothesized that performing rational structure-based bitopic drug design to strengthen interactions with the newly reorganized TM2-ECL1-TM1 pocket could yield highly selective D3R agonists, providing a new experimental approach potentially applicable across other aminergic GPCRs. Reorganization of TM1 at the hD3R upon compound 1 binding and structures of reference compounds 1 and 2. (A) Structure of the hD3R (violet, cartoons) with 1 bound (cyan carbons, sticks) (PDB 9F33). (B–C) Binding profile of 1 in the hD3R pocket, displaying interacting residues (sticks) and the TM1 reorganization when compared to D2R (green, cartoons) (PDB 6VMS) and D3R bound to pramipexole (yellow, cartoons) (PDB 7CMU). (D) 2D structure of compounds 1 and 2 highlighting the positions for structural modifications of the indole SP. HBD = Hydrogen-Bond Donor, HBA = Hydrogen-Bond Acceptor. Results and Discussion Rational Design and Synthesis of D3R-Selective Bitopic Agonists Bitopic drug design provides the rationale for modifying the structure of a parent molecule one component at a time (PP, linker, and SP). ,, Herein, we synthesized bitopic molecules retaining the optimized structural and stereochemical composition of the substituted (2S,5S)-morpholine PP, but introduced modifications at the SP from compound 1, aiming to promote stronger H291.32 interactions by introducing H-bond donor (HBD) and/or H-bond acceptor (HBA) substituents at either position 5 or 6 of the 2-indole amide nucleus (Figure ). We inserted combinations of three different linkers: (1R,2S)- and (1S,2R)-cyclopropyl stereoisomers from compound 1, as well as an n-butyl linker (from AB04-88 (2)) (Figure ; Schemes – and Supporting Information). We also aimed to retain optimal central nervous system multiparameter optimization scores (CNS-MPO) to predict drugs’ ability to cross the blood–brain barrier as potential neurotherapeutics/neuroprobes. All the newly synthesized compounds are within, or above, our established threshold − for expected brain distribution (CNS-MPO score > 3.0 represents compounds with relevant brain permeability, values between 2.5 and 3.0 are worthy of further evaluation) (Table , Tables S1 and S2). Molecules with trans-cyclopropyl linker were first synthesized as racemates and the most promising analogs after initial characterization were resolved into their respective enantiomers. a) Ar-COOH, HOBt, EDC·HCl, DIPEA, DCM, or THF, or DMF, RT; b) Dess–Martin Periodinane (DMP), DCM, 0 °C; c) Cat. AcOH, Na­(AcO)3BH, DCE, from 0 °C to RT; d) TFA, DCM, RT a) Ar-COOH, HOBt, EDC·HCl, DIPEA, DCM, or THF, or DMF, RT; b) DMP, DCM, 0 °C; c) Cat. AcOH, Na­(AcO)3BH, DCE, from 0 °C to RT; d) TFA, DCM, RT a) LiAlH4, THF, from 0 °C to RT; b) Ar-COOH, HOBt, EDC·HCl, DIPEA, DCM, or THF, or DMF, RT; c) i) DMP, DCM, 0 °C, ii) AcOH, Na­(AcO)3BH, DCE, from 0 °C to RT aPreparative HPLC resolution using Chiralpak AD-H for 9–10, 11–12, and 13–14, and Chiral OZ-H for intermediates 15–16. Preparative HPLC resolution using Chiralpak AD-H for 9–10, 11–12, and 13–14, and Chiral OZ-H for intermediates 15–16. 1Pharmacological Characterization hD2RhD3R Compound K i ± SEM (nM) (n)pEC50 ± SEM [nM] E max ± SEM (%) K i ± SEM (nM) (n)pEC50 ± SEM [nM] E max ± SEM (%)Binding D3R SelectivityPotency D3R SelectivityCNS-MPO Scores 2 (AB04-88)84.6 ± 15.9 (3)7.15 ± 0.39 [70.8]96.63 ± 0.294.77 ± 1.18 (3)8.69 ± 0.16 [2.04]99.30 ± 10.6517.734.74.70 4b (AB13-08)2690 ± 1390 (4)<5.5 [>3,160]63.69 ± 14.38# 53.9 ± 6.24 (5)8.96 ± 0.027 [1.10]94.81 ± 13.1749.9>2,8004.15 1 (FOB02-04A)91.3 ± 52.7 (5)7.48 ± 0.21a [33.1]74.95 ± 3.45a 1.75 ± 0.805 (6)9.45 ± 0.35a [0.355]100.13 ± 1552.293.24.52 6d (AB12-82)41.0 ± 4.62 (3)<6 [>1,000]73.15 ± 10.22# 1.88 ± 0.227 (3)11.76 ± 0.34** [0.00174]84.57 ± 10.221.8>575,0003.99 6f (AB13-13)302 ± 29.3 (4)<7 [>100]79.45 ± 13.92# 4.26 ± 0.815 (4)11.91 ± 0.67* [0.001]58.77 ± 6.770.9>100,0003.55 7 (AB13-31)165 ± 44.0 (3)<7 [>100]80.73 ± 10# 4.07 ± 0.442 (4)8.34 ± 0.35 [4.57]73.91 ± 24.7140.5>223.37 6e (AB13-46)156 ± 21.3 (7)6.14 ± 0.5 [724]74.38 ± 14.320.879 ± 0.107 (7)10.28 ± 0.47 [0.0525]61.28 ± 13.3217713,8003.99 9 (AB13-46A)104 ±6.42 (4)6.12 ± 0.35 [759]91.02 ± 13.480.505 ± 0.0932 (5)9.92 ± 0.21 [0.120]130 ± 2.582066,3003.99 6g (AB13-59)84.8 ± 21.8 (5)<6 [>1,000]80.37 ± 17.8# 1.60 ± 0.294 (5)9.54± 0.25 [0.288]49.53 ± 11.8353>3,4003.55 8 (AB13-61)171 ± 29.3 (3)<6.5 [>316]92.65 ± 12.54# 3.25 ± 0.430 (3)9.76 ± 0.56 [0.174]70.9 ± 24.7352.6>1,8003.37 6i (AB13-75)111 ± 17.3 (3)<5 [>10,000]52.83 ± 2.08# 3.29 ± 0.506 (3)8.08 ± 0.14* [8.32]63.32 ± 18.8833.7>1,2002.590 6j (AB13-73)191 ± 40.5 (6)<6 [>1,000]77.95 ± 21.99# 1.32 ± 0.226 (6)12.01 ± 0.32** [0.000977]60.28 ± 11.89145>1,000,0002.59 11 (AB13-73A)166 ± 18.2 (3)<6 [>1,000]80.06 ± 2.8# 0.803 ± 0.209 (4)11.88 ± 0.52* [0.00132]48.9 ± 10.24207>750,0002.59 6n (AB13-74)244 ± 35.0 (6)6.63 ± 0.34 [234]87.05 ± 17.431.84 ± 0.226 (6)10.67 ± 0.54 [0.0214]80.92 ± 14.313310,9003.31 13 (AB13-74A)79.6 ± 27.3 (5)<6 [>1,000]77.29 ± 5.33# 0.443 ± 0.0479 (6)10.08 ± 0.54 [0.0832]52.96 ± 9.08180>12,0003.31 23 (AB14-30)653 ± 147 (4)7.9 ± 0.33 [12.6]105.42 ± 20.053.21 ± 0.282 (4)8.9 ± 0.51 [1.26]61.28 ± 16.06203104.40i K i obtained from radioligand competition vs [3H]-(R)-(+)-7-OH-DPAT. pEC50 and E max values measured using the TRUPATH assay. All data are mean ± SEM of at least three independent experiments (n > 3) performed in technical triplicates. E max values are normalized to the efficacy of quinpirole at hD3R for D3R and at hD2R for D2R (Quinpirole data from refs , ). Potency estimates for curves not reaching saturation were obtained by assuming full agonist efficacy, conservatively determining pEC50 and stating more/less than values. #Mean efficacy values for hD2R at the maximum concentration. aData from our previous report. *p < 0.05, **p < 0.01 (Welch’s unpaired t-test) for compounds tested at D3R compared with their parent compound. Pharmacological Characterization K i obtained from radioligand competition vs [3H]-(R)-(+)-7-OH-DPAT. pEC50 and E max values measured using the TRUPATH assay. All data are mean ± SEM of at least three independent experiments (n > 3) performed in technical triplicates. E max values are normalized to the efficacy of quinpirole at hD3R for D3R and at hD2R for D2R (Quinpirole data from refs , ). Potency estimates for curves not reaching saturation were obtained by assuming full agonist efficacy, conservatively determining pEC50 and stating more/less than values. #Mean efficacy values for hD2R at the maximum concentration. aData from our previous report. *p < 0.05, **p < 0.01 (Welch’s unpaired t-test) for compounds tested at D3R compared with their parent compound. Binding and Pharmacological Characterization The D3R/D2R binding selectivity of the new bitopic molecules was measured by competition radioligand binding assays versus [3H]-(R)-(+)-7-OH-DPAT, the optimal radioligand to characterize agonists’ affinity at D2-like receptors (Table and Table S1). , The data show that modifications at the SP harboring an n-butyl or cyclopropyl linker result in a substantial increase in D3R/D2R selectivity, with the latter yielding significantly higher selective agonists. The best bitopic molecule containing the n-butyl linker (derivative of (2)) was compound 4b, which displayed a 50-fold D3R/D2R selectivity. Interestingly, 4b achieved higher selectivity by decreasing its binding to D2R, rather than enhancing binding to D3R. With the D2R being more plastic than D3R and hence harder to eliminate binding promiscuity, the 5-methyl-2-(2H-1,2,3-triazol-2-yl)­benzamide group at 4b SP opens new routes to hinder D2R binding while preserving D3R activation. This effect achieved by this SP was however only observed in the bitopic ligand containing the n-butyl linker but not the cyclopropyl-containing linker (6a). Remarkably, indole substitutions in the cyclopropyl-containing agonists demonstrated a significant increase in the D3R/D2R selectivity, especially for molecules containing the (1R,2S)-cyclopropyl stereochemistry and HBAs in position 6 of the indole, which supports interactions with H291.32 of the D3R. In particular, compound 9, the eutomer of the 6-OMe derivative 6e, showed subnanomolar affinity for D3R (K i = 0.505 nM), 3.5-fold higher than 1, and >200-fold selectivity over D2R (in comparison to the 52-fold selectivity of 1). Similarly, both eutomers 11 (6-NHCOOtBu carbamate) and 13 (6-COOMe) showed subnanomolar affinities for D3R (K i = 0.803 nM and 0.443 nM, respectively) and subtype selectivity (207- and 180-fold over D2R, respectively). These three new analogues represent the most selective D3R agonists reported to date using radioligand binding assays. Parent compound 1 (as trans-linker mixture, as well as the eutomer) and selected cyclopropyl-containing lead compounds 6d, 9, and 11 were tested for off-target interactions with structurally and physiologically related GPCRs, including dopamine D1R, serotonin 5HT1AR 5HT2AR and 5HT2CR (trans-1), as well as dopamine D4R, orexin OX1R and OX2R, and μ-opioid receptors (MOR) (1, 6d, 9, and 11), showing very low to completely absent binding (>1–10 μM, across all off-targets; Tables S3, S4 and S5) hence highlighting the D3R-selective nature of this bitopic structural template. The most selective bitopic agonists from radioligand binding assays were subjected to pharmacological characterization (10 compounds 1-derived from the cyclopropyl linker and one compound 2-derived from the butyl linker). For this purpose, we determined potency (pEC50) and efficacy (E max) using cellular signaling assays that monitor the Gαβγ heterotrimer dissociation upon agonist application in transfected HEK293T cells with either hD2R or hD3R (Table , Figure S2). In pharmacological assays, 1 and 2 displayed 100- and 34-fold D3R/D2R selectivity respectively, closely matching the selectivity profiles in the radioligand binding assays. In the case of the newly synthesized molecules, 9 of 11 bitopic compounds displayed a remarkable increase in selectivity and different levels of efficacy ranging from 50% to 126% when compared to their reference molecule. In our previous work, we reported changes in the efficacy of 1, depending on the binding mode of its SP, hence, such variation in efficacy is expected when adding substituents at this position. Overall, 9 molecules displayed >1,000-fold D3R/D2R selectivity, of which 5 molecules were >10,000-fold selective. Lead partial agonists with the most selectivity include 6j (60.3% efficacy) and 6d (84.6% efficacy), with 1,000,000-fold and >575,000-fold selectivity for D3R versus D2R (Table , Figure ), respectively. The pure enantiomers displayed similar profiles as the racemic mixtures. Lead full agonists with the highest D3R/D2R selectivity include 4b (2-derived with an n-butyl linker) with >2,800-fold selectivity and the enantiomer 9 (1-derived with a cyclopropyl linker) with 6,300-fold selectivity. Pharmacological characterization of the most selective D3R agonists. Concentration–response curves for reference and lead agonists for hD2R (blue) and hD3R (yellow) for full (A) and partial (B) agonists. The reference curve for each lead agonist is the one from its parent compound. E max values are normalized to the efficacy of quinpirole at hD3R for D3R and at hD2R for D2R. The racemic mixture of 9 (6e) displayed almost 14,000-fold selectivity and 61.3% efficacy, also becoming a lead partial agonist. To ensure such high selectivity was not due to a kinetic effect, where bitopic agonists with a slower association rate for the D2R than to the D3R would suggest an artifactual high selectivity for D3R, if experimental measures are taken too early, we measured concentration–response curves for 9 and 11 for D2R and D3R at different time points. The results showed that efficacy was lost over time; however, selectivity ratios for both receptors were maintained across the different time points (Figure S3). To our knowledge, these are the most D3R-selective and potent full and partial agonists reported to date, greatly enhancing the pharmacological profiles of currently available molecules. There are however certain limitations to these compounds, which tend to display low Hill coefficients (Table S7), narrowing down the useful window of concentrations where D3R (but not D2R) would be activated. This is especially true for compound 6d, where activation of D2R seems to occur readily at low concentrations. Compounds 4b, 9, and 11 have a wider concentration window to selectively activate D3R. The changes in the Hill coefficient might originate in the fact that the secondary binding site is a known allosteric site with both positive and negative effects, hence yielding complex pharmacological outcomes and likely contributing to some differences between radioligand binding and potency in in vitro pharmacology (changes in magnitude but not rank-order). Interaction Modes at the Secondary Pocket While based on the pharmacological data the selectivity of 4b and 9/6e originates mainly from a decrease in D2R binding (in line with radioligand binding studies), the selectivity of 6j/11 and 6d originates from an enhanced D3R potency when compared to 1, suggesting strengthened interaction with H291.32 and in line with our rational design. To confirm the binding mode of the lead compounds we aimed to determine the cryo-EM structures of D3R coupled to a GαOβγ heterotrimer and bound to the full agonist 9 and the partial agonist 11. While we were able to determine the structure of D3R bound to 9 (full agonist) (Figure , Table S6, Figures S4 and S5) attempts to determine the structure of 11 were unsuccessful, in line with previous evidence showing that agonist efficacy directly correlates with GPCR-G protein stability in cryo-EM structure determination. The structure of D3R bound to 9 showed a binding mode similar to that of the parent compound 1. However, TM1 did not appear further stabilized when compared to the structure of D3R bound to compound 1, nor did the bitopic molecule contain H-bonding interactions to the imidazole group of H291.32. However, this is in line with the selectivity of this compound arising from a diminished binding to D2R rather than a significant enhancement in D3R binding, which displayed just a 3-fold increase in potency. In order to understand whether H291.32 is involved in bitopic binding we first mutated it to different residues including alanine, lysine, and phenylalanine. In line with the parent compound 1 binding mode, mutation of H291.32 decreased agonist efficacy for 9 (Figure D), however, with a higher impact in 9 vs 1 (WT100%, H291.32A58.56%, H291.32K39.44%, H291.32F49.37%). Such a difference in impact agrees with 9 having a slightly bulkier SP, which is more sensitive to variations in the H291.32. Since mutation of H291.32 had a major impact on efficacy and not on potency, these results support the role of H291.32 in promoting an optimal binding mode of both 1 and 9 (previous studies highlighted that an optimal binding mode at the orthosteric binding site is required for full efficacy at D3R , ). In contrast, compounds 6j/11 and 6d displayed a significant increase in potency at the D3R, suggesting new strong interactions with H291.32. In line with this hypothesis, alanine mutation of H291.32 decreased the potency of the lead compound 11 over 100-fold (WTpEC50 11.58 ± 0.52, H291.32ApEC50 9.92 ± 0.25), bringing its potency down to a magnitude comparable to that of compound 1 (Figure E). These results confirm the criticality of H291.32 in the bitopic agonist design. Structural analysis of D3R bound to Compound 9. (A) D3R:GαOβγ:Compound 9 structure (cartoon, colored by chain) with the ligand-binding pocket highlighted in violet and the cryo-EM density for Compound 9 as green mesh. (B) Binding mode of Compound 9 (green carbons, sticks) at the D3R (violet, sticks, and cartoon) with cryo-EM density as green mesh. (C) Comparison of the binding mode between Compound 1 (yellow carbons, sticks) and 9 (green carbons, sticks). (D) Concentration–response curves in D3R WT (yellow), H291.32A (blue), H291.32F (maroon), and H291.32K (orange), upon GOA activation by Compound 9 (shown as net BRET). (E) Concentration–response curves in D3R WT (yellow) and H291.32A (blue) upon GOA activation by Compound 11 (shown as net BRET). All data are mean ± SEM derived from three independent experiments (n = 3), each performed in technical triplicate. All source data within this figure are provided as a Source Data file. To confirm the binding mode of the other lead compounds we performed computational docking on the D3R structure. For each binding pose, the minimized ligand internal energy (E min), RTCNN, and Docking Scores were evaluated (see Methods). All bitopic molecules consistently converged to two binding modes: mode A, where the SP fragment engages the TM2-ECL1-TM1 subpocket including H291.32 and forms additional interactions with E902.65 and Y321.35, and mode B, where the indole moiety docks into an alternative hydrophobic pocket formed by V2716.53, Y2766.58, and C18145.50, with a key anchoring π–π interaction to Y2766.58 (Figure ). For each ligand, the top consensus poses corresponding to modes A and B across independent docking runs were retained and refined by Monte Carlo (MC)-guided local minimization with side-chain relaxation within 5 Å of the ligand. After refinement, the ICM Docking Score and RTCNN machine-learning affinity score (https://www.molsoft.com/RTCNN.html) were recalculated, together with the E min obtained by energy minimization of the ligand in the receptor pocket in the ICM force field. , A similar pair of distinct conformations was computationally and experimentally found for 1 in our previous report, where conformation A was mainly responsible for D3R activation, while B did not contribute to activation. We hypothesized a similar pattern in the newly synthesized analogs, affecting both their selectivity and efficacy. Indeed, although we do not observe a discrete secondary conformation for compound 9 in the cryo-EM structure, some dynamism at the SP was observed during cryo-EM data processing. Docking of lead agonists at the D3R pocket. D3R (cartoon, yellow) with a surface representation of the pocket; selected pocket residues are shown as sticks and labeled. Lead compounds as red (A conformation) and green (B conformation) sticks. Intermolecular hydrogen bonds and salt bridges are indicated by black dashed lines. The binding-pocket property mesh highlights regions favorable for specific interactions (red, hydrogen-bond acceptors; blue, hydrogen-bond donors; green, hydrophobic contacts). The conformational transition from A to B is indicated by bold arrows. Overall, docking results suggest SP binding to the TM2-ECL1-TM1 subpocket, while the predicted stability of A vs B conformations correlated with the level of bitopic selectivity and linker chirality, e.g., for the enantiomeric pairs of 6j with a tert-butyl carbamate addition at the SP, where 11 (>750,000-fold selectivity) strongly prefers mode A, whereas 12 (16-fold selectivity, not selected for pharmacological characterization) shifts toward mode B (Figure ). The results show that the correct linker chirality favors interactions with H291.32 and hence increased selectivity. In Vitro Hepatic Metabolism To understand whether the lead compounds had drug-like properties, we conducted in vitro phase I and II studies using human liver microsomes at a concentration that is typically tested in most of the preliminary ADME studies. The results showed that for phase I stability, when 6e, 6j, and 6n were incubated with human liver microsomes, all of these compounds degraded in a time-dependent manner (Figure ). However, all compounds retained more than 50% of the parent concentration after 6 h, indicating relatively good metabolic stability under the tested conditions. UPLC analysis with a PDA detector showed that extra peaks with UV absorption similar to those of the parent compounds were observed (Figures S9–S12, Table S8), suggesting the formation of phase I metabolites. Collectively, these results indicate that these lead compounds possess moderate metabolic stability while undergoing phase I metabolism mediated by CYPs. Metabolic stability of the tested compounds in a Phase I human liver microsomes (HLM) reaction system. The percentage of parent drug remaining for 6e, 6j, and 6n was monitored over time to assess enzymatic stability. Compounds (20 μM) were incubated with 0.1 mg/mL HLM (blue lines) or with a blank potassium phosphate (KPI) buffer control (red lines). Data are expressed as the mean ± SD (n = 3). Apparent half-lives of the tested compounds (table) were derived using first-order kinetic modeling; however, these values represent projections beyond the 6hr study duration. Because the parent compound concentrations remained above the 50% threshold throughout the assay, the resulting values are sensitive to minor analytical variations and should be interpreted as mathematical approximations of metabolic stability. The results also showed that for the phase II glucuronidation reaction, when 6e, 6j, and 6n were incubated with human liver microsomes, no significant degradation was observed for these three compounds (Figure ), suggesting that under the current experimental conditions, the chance for glucuronidation for these compounds is low. Metabolic stability of the tested compounds in a Phase II glucuronidation reaction with human liver microsome (HLM) reaction system. The percentage of parent drug remaining for 6e, 6j, and 6n was monitored over time to assess enzymatic stability. Compounds (20 μM) were incubated with 0.1 mg/mL HLM (blue lines) or with a blank potassium phosphate (KPI) buffer control (red lines). Wogonin (20 μM) was used as the positive control to demonstrate the glucuronidation activity. Data are expressed as the mean ± SD (n = 3). Caco-2 Cell Permeability Studies Compounds 6e, 6j, and 6n were evaluated using the Caco-2 cell culture model to determine their bidirectional permeability and potential for active efflux. A single concentration of 10 μM, typical for preliminary ADME screening, was utilized for this study. The results showed that the uptake permeabilities (P a–b ) of 6e, 6j, and 6n were 1.14 ± 0.19 × 10–6, 1.20 ± 0.13 × 10–6, and 2.31 ± 0.20 × 10–6, respectively (Figure ). The permeability rank order is 6n > 6j > 6e. Based on the reference, these values suggest low to moderate gastrointestinal absorption in humans, as values >10–5 cm/s typically predict >85% human absorption. The results also showed that the efflux permeabilities (P b–a ) of these three compounds are higher than their uptake permeabilities (P a–b ), suggesting the involvement of active efflux transporter(s). The efflux ratios of 6e, 6j, and 6n were 1.79, 2.23, and 1.23, respectively. According to FDA guidance, an efflux ratio >2 indicates the involvement of active efflux transporter(s). Given that both permeability and efflux ratios are concentration-dependent in the presence of transporters, future additional dose–response studies might be required to characterize the saturation kinetics and provide a more definitive conclusion for these compounds. Intracellular exposure was also evaluated. For all three compounds, intracellular exposure was consistently higher in the A to B direction compared to those in the B to A direction (Figure ). Overall, these results indicate that the three compounds have low to moderate permeability, which may predict a low to moderate GI absorption. The consistently higher intracellular exposure observed in the A to B direction supports asymmetric transport, potentially impacted by apical efflux transporter(s). The absorption and efflux permeabilities of rutin, a control compound with known low permeability, were on the order of 10–7 cm/s, consistent with expected values. This supports the integrity of the cell monolayer and the validity of the experimental system. Permeability and transport amount of 6e, 6j, 6n and rutin in the receiver side of the Caco-2 cell monolayer. Amount of the 6e, 6j, and 6n in the insert of the Caco-2 cell monolayer. Conclusions In conclusion, using a ligand-triggered conformation of the D3R we were able to exploit a new binding site involving TM1 (the most sequence-diverse region in GPCRs) to develop the most selective full and partial D3R agonists reported to date. These bitopic molecules are first-in-class picomolar potency and absolute subtype-selectivity D3R agonists, which tackle a long-standing pharmacological distinction challenge between two of the most targeted receptors in the clinic. We show that structure determination and design of bitopic ligands occupying binding sites near the TM1 region can be a highly translational strategy for the development of subtype-selective molecules in GPCRs, the major drug target family in humans. The molecules represent unique tools for dissecting the (patho)­physiological role of D3R in neurological/neuropsychiatric processes and lay the foundation to develop improved treatments for locomotor disorders and high impulsivity. The in vitro hepatic metabolism and cell permeability studies conducted on the isomeric mixtures 6e, 6j, and 6n support the translational potential toward future in vivo studies for these compounds, and especially for their lead eutomers 9 and 11, given the high Phase I and Phase II metabolic stability, moderate cellular permeability (1 × 10–6 < Papp < 10 × 10–6) and efflux ratio between 1.23 and 2.23, predictive of moderate GI absorption and possibly BBB permeability. Methods Radioligand Binding Studies[3H]-(R)-(+)-7-OH-DPT HEK293 cells stably expressing human D2LR or D3R were grown in a 50:50 mix of DMEM and Ham’s F12 culture media, supplemented with 20 mM HEPES, 2 mM L-glutamine, 0.1 mM nonessential amino acids, 1X antibiotic/antimycotic, 10% heat-inactivated fetal bovine serum, and 200 μg/mL hygromycin (Life Technologies, Grand Island, NY), and kept in an incubator at 37 °C and 5% CO2. Upon reaching 80–90% confluence, cells were harvested using premixed Earle’s Balanced Salt Solution (EBSS) with 5 mM EDTA (Life Technologies) and centrifuged at 3,000 rpm for 10 min at 21 °C. The supernatant was removed, and the pellet was resuspended in 10 mL hypotonic lysis buffer (5 mM MgCl2, 5 mM Tris, pH 7.4 at 4 °C) and centrifuged at 14,500 rpm (∼25,000 g) for 30 min at 4 °C. The pellet was then resuspended in fresh binding buffer. A Bradford protein assay (Bio-Rad, Hercules, CA) was used to determine the protein concentration. For [3H]-(R)-(+)-7-OH-DPAT binding studies, membranes were harvested fresh; the binding buffer was made from 50 mM Tris, 10 mM MgCl2, 1 mM EDTA, pH 7.4. On test day, each test compound was diluted into half-log serial dilutions using 30% DMSO vehicle. Membranes were diluted in fresh binding buffer. Radioligand competition experiments were conducted in 96-well plates containing 300 μL fresh binding buffer, 50 μL of diluted test compound, 100 μL of membranes (40–80 μg/well and 20–40 μg/well for hD2LR and hD3R, respectively), and 50 μL of radioligand diluted in binding buffer ([3H]-(R)-(+)-7-OH-DPAT: 1.5 nM final concentration for hD2L and 0.5 nM final concentration for hD3R, Novandi Chemistry AB, Södertälje, Sweden). Aliquots of [3H]-(R)-(+)-7-OH-DPAT solution were also quantified accurately to determine how much radioactivity was added, taking into account the experimentally determined counter efficiency. Nonspecific binding was determined using 10 μM (+)-butaclamol (Sigma-Aldrich, St. Louis, MO) and total binding was determined with a 30% DMSO vehicle (3% final concentration). All compound dilutions were tested in triplicate and the reaction was incubated for 90 min at RT. The reaction was terminated by filtration through Perkin-Elmer UniFilter-96 GF/C, presoaked for 90 min in 0.5% polyethylenimine (PEI), using a Brandel 96-Well Plates Harvester Manifold (Brandel Instruments, Gaithersburg, MD) or a Revvity Unifilter-96 Cell Harvester (Revvity, Waltham, MA). The filters were washed 3 times with 3 mL (3 × 1 mL/well) of ice-cold binding buffer. 65 μL Perkin-Elmer MicroScint 20 Scintillation Cocktail was added to each well and filters were counted using a Perkin-Elmer MicroBeta Microplate Counter (Revvity, Waltham, MA, USA). IC50 values for each compound were determined from dose–response curves and K i values were calculated using the Cheng-Prusoff equation; K d values for [3H]-(R)-(+)-7-OH-DPAT were determined via separate homologous competitive binding experiments. When a complete inhibition could not be achieved at the highest tested concentrations, K i values have been extrapolated by constraining the bottom of the dose–response curves (= 0% residual specific binding) in the nonlinear regression analysis. These analyses were performed using GraphPad Prism version 10.0 for Macintosh (GraphPad Software, San Diego, CA). All the results were rounded to the third significant figure. K i values were determined from at least 3 independent experiments, performed in technical triplicate, and are reported as mean ± SEM. Constructs and Molecular Cloning All molecular cloning was performed using the in vivo DNA assembly methodology and the IVA Prime server. , The cDNAs of the hD2LR (HASS-FLAG-EGFP-3Cprotease-D2R) and hD3R (HASS-FLAG-EGFP-3Cprotease-D3R) were obtained through gene synthesis (Gene Fragments, Twist Bioscience) and cloned into the pcDNA4. The GαOA-RLuc8 (pcDNA5), Gβ3 (pcDNA3.1), and Gγ9-GFP2 (pcDNA3.1) constructs were a gift from Bryan Roth’s lab (Addgene plasmid kit #1000000163). The human dominant-negative GαOA subunit (S47NG.H1.02, G204AG.s3h2.02, E246AG.H3.04, M249KG.H3.07, and A326SG.s6h5.03) was generated by gene synthesis (Gene Fragments, Twist Bioscience) and the K54EG.H1.09 mutation was subsequently introduced into the GαOA sequence during subcloning into the pBacPak8 vector to enhance GPCR:G protein stability. Rat His8-Gβ1 (pBacPak8) and human Gγ2 (pBacPak8) were used for virus generation. Protein Production and Purification For the production of the D3R:Gαβγ protein complex, recombinant baculoviruses encoding D3R, GαOA, Gβ1, and Gγ2 were generated using the FlashBAC ULTRA system (Oxford Expression Technologies). Tni cells were cultured in suspension in ESF921 media to a density of 2–3 × 106 cells/mL, coinfected with D3R, GαOA, 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 compound 9 for 30 min 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, 5% glycerol, 0.007:0.001% LMNG/CHS, 10 μM of compound 9 and 5 mM imidazole followed by 20 CV of the same buffer with 2.5% glycerol and 20 mM imidazole. The sample was eluted with 20 mM HEPES pH 7.5, 150 mM NaCl, LMNG 0.003%, 10 μM of compound 9 and 250 mM imidazole. The resulting complex was incubated for 1 h with anti-FLAG resin (Sigma-Aldrich, A2220) pre-equilibrated with 20 mM HEPES pH 7.5, 25 mM NaCl, 0.007:0.001% LMNG/CHS, and 10 μM of compound 9. After incubation, the resin was washed with 20 CV of buffer containing 20 mM HEPES pH 7.5, 150 mM NaCl, 0.007:0.001% LMNG/CHS, and 10 μM of compound 9. The complex was subsequently eluted in the same buffer supplemented with 0.15 mg/mL of FLAG peptide. The purified protein was concentrated to 2.8 mg/mL, and compound 9 was supplemented to a final concentration of 50 μM. Cryo-EM Vitrification and Data Collection Three μL of D3R:GαOβγ:Compound 9 at 2.8 mg/mL were applied to 300 mesh Quantifoil 1.0/1.0 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 Basque Resource for Electron Microscopy (BREM). A total of 13,180 movies were recorded at a magnification yielding 0.82 Å/pixel with a dose rate of 12.6 e–/pixel/s and a defocus range between −1 and −3 μm using the Smart EPU Software (ThermoFisher Scientific). Movies were split into 50 frames each and exposed to a total dose of 52.3 e–/Å2 (1e–/Å2 per frame) using a total exposure time of 2.8 s. Cryo-EM Data Processing and Model Building RELION-5.0 was used for all data processing unless otherwise specified. Drift and beam-induced motion correction (5 × 5 patches) were performed using MotionCor2 along with dose weighting. Contrast transfer function (CTF) estimation and determination of the defocus range were performed with CTFFIND-4.1. Automated particle picking was carried out with Topaz and 11,587,210 particles were extracted and subjected to 3 rounds of 2D classification. Classes showing a clear GPCR complex were selected and used for template picking, then re-extracted and subjected to 3 rounds of 2D classification, resulting in 2,507,892 particles. All particles (derived from Topaz picking and template picking) were subsequently merged and duplicates were removed, yielding 3,666,677 particles, which were transferred to CryoSPARC. 9 rounds of heterogeneous refinement with one class designated as “junk” to filter out low-quality particles while retaining the true positives were performed. A total of 403,865 particles were then selected, recentered on the orthosteric site and re-extracted using a box size of 320 pixels. After one round of heterogeneous refinement, the particles underwent 2D classification to remove low-quality particles, leaving 270,679 particles for further heterogeneous refinement. To improve the map quality at the receptor binding site, a 3D classification without alignment, focused on the receptor, was performed. This was followed by a final 2D classification, from which 120,394 particles were selected. These particles were then subjected to a final round of nonuniform refinement, yielding a map at 3.47 Å resolution based on the gold-standard FSC criterion of 0.143. Model building and refinement were carried out using the CCP-EM software and Phenix. The D3R, Gβ1, and Gγ2, starting coordinates were taken from the Gαo-coupled D3R structure (PDB 9F33). D3R was modeled from residue H291.32 to I2235.73 and from R3236.29 to C400. GαO was modeled from T4G.HN.10 to E54G.H1.09, T182G.hfs2.05 to V234G.s4h3.07, and N242 G.s4h3.15 to Y354G.H5.26. Manual modification and restrained real-space refinement were performed using Coot and Phenix. A dictionary describing the compound 9 and its coordinates was created using AceDRG and manually fitted into the density for its later refinement in real space using Coot and Phenix. B factors were reset to 40 Å2 prior to refinement. The final model achieved good geometry (Table S6) with validation performed in Coot, EMRinger, and MolProbity. The goodness of fit of the model to the map was carried out using Phenix, using a global model-vs-map FSC correlation (Table S6). Final coordinates and the cryo-EM map have been deposited in the PDB and EMDB respectively with accession codes 31BG and EMD-58252. Cellular BRET Assays Human embryonic kidney HEK293T cells (ATCC, not authenticated in-house) were grown in DMEM supplemented with 100 U/mL streptomycin, 100 mg/mL penicillin, and 10% (v/v) fetal bovine serum. Manipulation and maintenance were done in a biological safety cabinet class 1 and in an incubator at 37 °C, 5% CO2, and 90% relative humidity. G protein dissociation BRET assays were performed following the TRUPATH system in HEK293T cells. 50,000 cells/well were seeded in poly-lysine-coated 96-well white plates. The following day, cells were transfected with either D2R or D3R and GαOA-RLuc8:Gβ3:Gγ9-GFP at a 2:1:1:1 ratio using PEI MAX (Polysciences Inc.) at a 6:1 PEI:DNA ratio. After 48 h, the medium was replaced with 90 μL/well of 1× Hank’s balanced salt solution (HBSS) with 20 mM HEPES pH 7.4 and 7.5 μM Coelenterazine 400a. For each ligand tested, 10 μL of varying ligand concentrations were added and BRET signals were measured using a CLARIOstar (BMG Labtech) with 400 nm (RLuc8) 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. Data analysis was performed using GraphPad Prism 8.0.1, with data normalized and fit to a four-parameter logistic curve. Results are presented as mean ± SEM from at least three independent experiments (each with technical triplicates). Molecular Docking The active-state structure of the D3 receptor complex (PDB ID: 9F33) was retrieved from the RCSB Protein Data Bank. This structure represents the human receptor in an agonist-bound state, complexed with the selective bitopic agonist 1. The GαOβγ heterotrimer fragment was removed, retaining the receptor subunit, and compound 1. Hydrogen atoms were added and optimized, and the side-chain conformations were refined to improve structural accuracy. Prior to docking, the analogs 4b, 6d, 10, 11, 12, 13, 14, and 23 underwent chiral definition and formal charge assignment. Two-dimensional molecular representations were converted into three-dimensional models, which were energy optimized using the MMFF94 force field. A biased probability Monte Carlo (BPMC) optimization approach was applied for docking simulations, where compound conformations in internal coordinates were sampled using precalculated grid energy potentials of the receptor. The grid potentials accounted for receptor-limited flexibility through “soft” van der Waals potentials, while maintaining the receptor’s overall conformational state. All-atom docking simulations were conducted with an effort value of 5. The docking box was defined to encompass the extracellular half of the receptor. A minimum of 15 independent docking runs, each generating three conformations, were performed starting from random ligand conformations. The best ten docking results were analyzed for consistency by comparing ligand poses. Top-scoring docking solutions corresponding to the observed conformations A and B were refined iteratively using energy minimization and Monte Carlo sampling. Refinements focused on ligand conformations and receptor side chains within a 5 Å radius of the binding site. Docking and RTCNN scores, as well as energies, have been recalculated for the minimized structures and reported. All of the above-mentioned molecular modeling operations were performed in the ICM-Pro v3.9–2b molecular modeling and drug discovery suite (Molsoft LLC). UPLC-UV and UPLC-MS for Compounds Quantification in In Vitro Hepatic Stability and Cellular Permeability Assays LC-MS grade solvents (acetonitrile, methanol, and water) were purchased from VWR Chemicals (Chicago, IL, USA). LC-MS grade formic acid was obtained from Fisher Scientific (Pittsburgh, PA, USA). All other chemicals were of analytical reagent grade and were sourced from commercial suppliers. The compounds were analyzed using a Waters Acquity UPLC system (Waters Co., Milford, MA, USA) equipped with a Waters C18 column (2.1 × 50 mm, BEH C18, 1.7 μm). The flow rate was set at 0.5 mL/min. The mobile phase consisted of water containing 0.1% formic acid (mobile phase A, MPA) and acetonitrile containing 0.1% formic acid (mobile phase B, MPB). The gradient elution program was as follows: 0–0.5 min, 5% B; 0.5–1.0 min, 5–20% B; 1.0–2.0 min, 20–50% B; 2.0–2.5 min, 50–95% B; 2.5–3.5 min, 95% B; 3.5–4.0 min, 95–5% B; and 4.0–5.0 min, 5% B. The column temperature was maintained at 40 °C, and the injection volume was 10 μL. Rutin was used as the internal standard. Standard curve samples were prepared by serial dilution of stock solutions (10 mM in DMSO) of the analytes with KPI buffer (pH 7.4). The resulting solutions were centrifuged at 15,000 rpm at 4 °C for 15 min prior to UPLC injection. The UPLC-UV chromatograms for the analytes are shown in Figures S9–S12. The quantification results show that the standard curves for these four compounds were linear at least from 100.00 to 0.32 μM. Subsequent analysis revealed that 6e exhibited two peaks in the UPLC chromatogram under a more complex gradient elution method. Both peaks showed identical UV spectra. Given the presence of a chiral centers, these peaks are attributable to stereoisomers that can be separated on the C18 column. All quantifications were performed under the original chromatographic conditions, with a single peak integrated for total quantification. In Vitro Hepatic Metabolism Wogonin (positive control for glucuronidation reaction) was purchased from AvaChem Scientific (San Antonio, Cat# 1120, CAS#632-85-9); All solvents, including water (J.T. Baker; CAS Number: 7732-18-5), acetonitrile (J.T. Baker; Catalog JT9012–3), and formic acid (CAS# 64-18-6), were of HPLC grade to maintain consistency in all experimental procedures and were obtained from Avantor (PA, USA). Human liver microsomes (50-donor, mixed gender) were obtained from BioIVT Elevating Science (Westbury, NY, USA). Other chemicals were used as received. Phase I reaction system: The phase I reaction was carried out using a standard established protocol. Briefly, mix microsomes (final concentration = 0.1 mg protein/mL), NADP (2.61 mM), 6-DP (6.6 mM), MgCl2 (6.6 mM), different concentrations of substrates in a 50 mM potassium phosphate buffer (pH 7.4), and 6-DPD in 5 mM sodium citrate solution (40 U/mL, added last) and incubate the mixture (final volume = 200 μL) at 37 °C for up to 6 h. The reaction was then terminated by adding 100 μL acetonitrile solution containing 0.6% formic acid. The samples were vortexed for 0.5 min, centrifuged at 15,000 rpm for 15 min, and 10 μL of supernatant was injected into the UPLC system for analysis. Phase II glucuronidation system: The glucuronidation reaction was carried out using a standard protocol established in our lab for more than a decade. Briefly, tissue S9 fractions, MgCl2 (5 mM), saccharolactone (4.4 mM), alamethicin (0.022 mg/mL), and UDPGA (3.5 mM) were mixed in a 50 mM potassium phosphate buffer (pH 7.94, total volume 170 μL). Then tested compounds were incubated in the above mixture for up to 24 h at 37 °C. The reaction was then terminated by adding 100 μL acetonitrile solution containing 0.6% formic acid. Samples were then vortexed and centrifuged at 15,000 rpm for 15 min at 4 °C for injection. Caco-2 Cell Permeability Phosphate-buffered saline (PBS), HBSS, and HEPES were obtained from Sigma-Aldrich (St. Louis, MO, USA). DMEM medium and 6-well cell insert plates were purchased from Corning (Glendale, AZ, USA). Caco-2 cell culture: The cell culture was following standard protocol. Briefly, the Caco-2 TC7 cells were maintained at 37 °C, 90% humidity, and 5% CO2. The culture media (DMEM medium supplemented with 10% regular FBS) will be changed every other day. To seed the cells, we used 0.4 μm porous polyester cell culture inserts from Nunc, which has a surface area of 4.2 cm2. The seeding density will be 100,000 cells/cm2. The cells will be fed every other day, and the cell monolayer will be ready to use in approximately 19–23 days postseeding and 24-h postfeeding. The quality of the cell monolayer will be measuring the transepithelial electrical resistance (TEER, normally >400 ohms·cm2). Transport and steady-state accumulation experiments. The transport permeability and intracellular accumulation were conducted using the established conditions. Briefly, a testing solution with appropriate concentrations of the compound in an HBSS buffer (pH 7.4) was loaded onto the donor side. Three donor (200 μL) and three receiver samples (200 μL) were taken at 0, 0.5, 1, 2, and 4 h followed by the addition of 200 μL of fresh donor solution to the donor side or 200 μL of fresh buffer to the receiver side. Immediately after the transport experiment, the incubation buffer was removed, and the Caco-2 cell layer was gently washed 3 times with ice-cold saline. Then, the cells will be collected carefully by cutting the insert membrane (with cells on it) and transferred the insert into a 1.5 mL centrifuge tube, which will be extracted with 1.0 mL of HBSS (pH 7.4) by sonication for 1 h at 4 °C using an Aquasonic 150D sonicator (VWR Scientific, Bristol, CT) at the maximum power (135 average watts). The tubes were centrifuged for 15 min at 15,000 rpm and 200 μL of the supernatant was collected to determine the intracellular concentration. Data analysis: Donor samples, receiver samples, intracellular samples, along with standard curve samples were analyzed using LC-MS. The apparent permeability coefficient (P) was determined by the equation: P=(dQ/dt)/(A×C0) where dQ/dt is the drug permeation rate (ng/mol/min), A is the surface area of the epithelium (cm2) and C 0 is the initial concentration in the donor compartment at time zero (μM or ng/mol/mL). Statistical analysis: Data are analyzed using one-way ANOVA or Student’s t-test. A prior level of significance is set at 5% or p < 0.05. The software package used is the GraphPad statistical analysis tool. Supplementary Material