Novel Agonist Bioisosteres and Common Structure-Activity Relationships for The Orphan G Protein-Coupled Receptor GPR139

GPR139 is an orphan class A G protein-coupled receptor found mainly in the central nervous system. It has its highest expression levels in the hypothalamus and striatum, regions regulating metabolism and locomotion, respectively, and has therefore been suggested as a potential target for obesity and Parkinson’s disease. The two aromatic amino acids L-Trp and L-Phe have been proposed as putative endogenous agonists, and three structurally related benzohydrazide, glycine benzamide, and benzotriazine surrogate agonist series have been published. Herein, we assayed 158 new analogues selected from a pharmacophore model, and identified 12 new GPR139 agonists, containing previously untested bioisosteres. Furthermore, we present the first combined structure-activity relationships, and a refined pharmacophore model to serve as a rationale for future ligand identification and optimization.

range. However, it cannot be excluded that the physiological activation of GPR139 is mediated by another more potent endogenous ligand.
Surrogate small molecule ligands for GPR139 have been reported by Hu et al. 9 Tables 1, 2 and 3). Herein, we have screened our pharmacophore model 10 against ~6 million drug-like compounds to identify more diverse agonist analogues, and conducted a joint structure-activity relationship (SAR) study together with the three published series.

Results
New agonist ligands and bioisosteres. Our search for GPR139 ligands featured an initial pharmacophore screening followed by three cycles of assaying in the Fluo-4 Ca 2+ -assay and analogue purchases, and covered in total 158 compounds (Supplementary Chart 1). This yielded 12 novel GPR139 agonists with efficacies similar to compound 1a (EC 50 = 39 nM) and potencies ranging from 364 nM to 4.7 μM (Fig. 2 and Table 4). These contain previously untested bioisosteres of the terminal aromatic systems (Table 5), as well as variations of the linker compared to the published reference agonists. The 12 novel agonists did not show activity in the CHO-M1 receptor cell line in the Fluo-4 Ca 2+ -assay, indicating specific interactions with the GPR139.
Characterisation of the preferred signalling pathway. Due to the different signalling pathway reported in the literature 6,10-12 , we tested two of the most potent known surrogate agonists; 7c and 1a, the L -Trp and L -Phe and the submicromolar compounds (DL43, DL126, DL24, DL22, DL132 and DL130) in the cAMP dynamic 2 assay (CisBio) and IP-One HTRF ® assay (CisBio) to check for biased signalling in the G s , G i and G q coupled assay. None of the tested compounds induced G s response in CHO-139 or in CHO-M1 cells ( Supplementary Fig. 1), which was included as a control. Furthermore none of the tested compounds were able to inhibit the cAMP response induced by 5 μ M forskolin in the CHO-139 cells by activating G i , except L -Trp, which inhibits the 5 μ M forskolin induced cAMP response by more than 10% in high concentrations, indicating G i activation ( Supplementary Fig. 2). However, this response was seen in both CHO-139 and CHO-M1 and is therefore an unspecific response. Hence, based on our data the compounds described herein activate GPR139 in CHO-k1 cells through the G q pathway, which is shown with both the Fluo-4 Ca 2+ -assay ( Fig. 2 and Table 4) and the IP-One assay ( Fig. 3A and Table 6). Furthermore, we show that the compounds activate CHO-GPR139 specifically through GPR139, as no response is seen in CHO-M1 cells (Fig. 3B).
Overall there was a good correlation between potencies observed in the Fluo-4 Ca 2+ -assay and the IP-One assay with compounds generally being 10-40 fold more potent in the former. This has been observed previously and has been ascribed to spare receptors where only a small proportion of receptors need to be activated to elicit a full calcium response 16 . Interestingly, the DL-compounds tested in both assays behave as full agonists in the Fluo-4 Ca 2+ -assay but as partial agonists in the IP-One assay. Moreover, racemic 7c and L -Trp behave as super-agonists compared to compound 1a ( Table 6). The IP-One assay is thus superior in differentiating the efficacy of the compounds.
Collective SAR -A common motif of two terminal phenyl rings, R1 and R2, separated by a 6-atom linker. The lowest energy conformations of the three most potent published reference agonists 1a, 7c and 39 demonstrated a near perfect overlay of their terminal phenyl rings, herein designated R1 and R2 (Tables 1,  2 and 3), and a six atom linker with multiple polar groups ( Supplementary Fig. 3). Below we present the first collective SAR analysis of the three series, and our new analogues, which are summarized in the form of a common pharmacophore model (Fig. 4b)  Linker length. The linker in 1a has a central nitrogen flanked by two amide groups (Table 1). 7c and 39 differ by having a carbon in the third position, in effect making the central portion of the linker a glycine moiety, and by having a methylated aliphatic carbon in the 6 th position (Tables 2 and 3) instead of an aromatic carbon in 1a.
Our new agonists ( Table 4) display variations of either carbon or nitrogen atoms in the positions 3, 5 and 6, specifically: CNC (DL130-132), NCC (DL146), NCN (DL22 and DL144), and NNC (DL148). Removal of the 5-nitrogen (DL96) in 1a resulted in loss of activity (Fig. 5), indicating that a reduction of the linker length cannot be tolerated. This is consistent with the presence of a shortened linker in many of the inactive analogues, such as DL3, DL5-8, DL15, DL47-48, DL51, DL55, DL59, DL61, DL95, and DL153 (Supplementary Chart 1). However, DL24, in which the 5-position and R2 are fused with a cyclohexyl, makes up an exception maintaining activity (EC 50 = 431 nM) despite a shorter linker. Interestingly, DL24 matched the pharmacophore elements well by adopting an atypical out-of-plane conformation between the acceptor (A4) and the aromatic (R1) pharmacophore elements (Supplementary Fig. 4 and Supplementary Chart 1). This may suggest that the cyclization rescues activity by locking the orientation of the R2 phenyl, which is free to rotate in the closest inactive analogues, DL6-8. We also found that generally compounds with linkers longer than 6-atoms, has decreased  15  Linker conformational restraints. The series by Hitchcock et al. 15       and/or the linker 1-2 atoms in a favourable conformation. This is proved by the generally improved potencies in the Hitchcock analogues (e.g. 39 vs. 1a and 7c). In the series by Dvorak et al. 13,14 , the 3-carbon has been substituted to change the central Gly residue into Ala, Ser, and Phe (10-13). The S- (11), but not R-methylation (10), strongly reduced the potency (compared to 7h). In the Hitchcock et al. 15 set, the same effect is seen in compounds 50 and 51 (compared to 46). Furthermore, even with the favourable enantiomer, hydroxymethyl substitution (12) also leads to a significantly reduced potency. Thus, although part of the effect could be mediated by the action of the two flanking sides of the ligands, there is no room for bulk around the 3-position and exclusion volumes were placed accordingly (Fig. 4b). Moreover, the 1-(1p and 8) and 4-carbonyl (9) oxygen atom removal that abrogated  Table 3. Chemical structures and potencies of selected compounds from Hitchcock et al. 15 confirming the SAR analysis for GPR139 analogues. activity may have done so partly because of the loss of the sp2 atom configuration, in addition to the loss of the hydrogen bond acceptor character. DL132 (EC 50 = 505 nM) and DL83 (EC 50 > 10 μ M -very weak agonist) are cyclized between the 5-and 6-positions. This has not been tested before, and it would be intriguing to investigate this cyclization on the two published series. As shown below, the 6-position is often fused with the R2 phenyl, but can also be aliphatic and methylated (7)(8)(9)(10)(11)(12)(13). Of note, both the 6-position (S) and (R) enantiomers of 7c 13,14 as well as compounds 5 and 6 in the Hitchcock et al. set 15 are potent, suggesting no preferred geometry in the binding pocket around this position.    . 1a), but has no effect in the series by Dvorak et al. (7h vs. 7q), which contains several additional equipotent single or double substituted analogues: methyl (7e), trifluoromethyl (7i), methoxy (7h, 7q) trifluoromethoxy (7j), cyanonitrile (7f), and several halogens (F: 7g; Cl/F: 7o, Br/Cl: 7p). This is also seen in the Hitchcock et al. 15 series where R1 has a 3-methoxy (e.g. 18 and 60), methyl (e.g. 28), triflouromethyl (e.g. 43), or several halogens (e.g. F: 14; Cl: 19). Together, this suggests the combined 3,5-substitution is not needed to maintain potency. Interestingly, the alternative pattern of 2,3-(7k) and 2,5-dichloro-substitution (7m) is equipotent to the 3,5-pattern (7n), but unlike the single 3-chloro (7c), the 2-chlorosubstituent (7b) alone displays a slightly reduced potency (~3-fold). This is in line with the series by Shi et al. 11 and Hitchcock et al. 15 Fig. 4). In contrast, the 4-position can accommodate the large 1-ethyl-2-methyl-benzoimidazole bioisostere found in multiple active compounds (DL43, DL85, DL164 and  DL148). Finally, all active surrogate agonists present an R1 terminal aromatic ring in our series, Shi et al. 11 , Dvorak et al. 13,14 , and Hitchcock et al. 15 , which likely contributes to the potency (feature R1 in Fig. 4b).
Common pharmacophore model. Pharmacophore composition. The above collective SAR was used to construct the updated pharmacophore model in Fig. 4B, and the detailed mapping of analogues into features is listed in the last columns of Tables 1-3. The pharmacophore contains two terminal aromatic features (R1 and R2) separated by a linker containing two hydrogen bond acceptors (A3 and A4), and one hydrogen bond donor (D5). An additional dual hydrophobic/halogen or acceptor element, H6/A2, is placed at position 3 of R1. Finally, exclusion volumes (yellow) have been incorporated to block substituents rendering a reduction or loss of activity.
Pharmacophore Validation. The updated pharmacophore model was matched against all 126 agonists described herein, and 6300 property-matched decoys generated by DUD-E 17 (50 decoys/ligand). The pharmacophore matched all known agonists, including the new analogues described herein. A receiver-operating characteristic (ROC) analysis displayed an area under the curve of 0.87 (Fig. 6). The early enrichment (top 1% of matches) was 72% actives, and strikingly 80% of all actives were retrieved already in the first 22% of the entire set, displaying both a good selectivity and sensitivity of the pharmacophore model. Matching of the same set of compounds against our original pharmacophore model (only based on the set by Shi et al. 11 ) displayed a significant improvement ( Supplementary Fig. 7). Our refined pharmacophore model successfully filtered out inactive analogues due to the added exclusion volumes, which did not hinder matching of the new series by Hitchcock et al. 15 ( Supplementary Fig. 6). Collectively, this suggests that the new pharmacophore model can be used to find potentially active analogues in future prospective screenings.

Discussion
The new bioisosteres and the common SAR analysis present features that could be useful in future medicinal chemistry and optimisation studies. For example, a 7c naphthyl to benzyl substitution would be an interesting combination of two of the three published series. From the new linker variants, cyclization of the linker 5and 6-positions, as in DL132, on the three published series could explore the conformational restraining of the R2-naphthyl and benzyl systems. Additionally, the necessity of the linker 1-position carbonyl oxygen's lone pair as a hydrogen bond acceptor is clear, but a carbonyl bioisostere, such as a spiro-oxetane, should provide the same electronic function while also testing a slight increase in steric bulk in a region devoid of exclusion volumes. On the other hand, constraining the linker 2-position via cyclization with a 4-oxo-3,4-dihydro-benzotriazene ring as in 39 and the rest of the Hitchcock series, showed a potential beneficial effect for the activity. Furthermore, some of the new bioisosteres might be able to give equivalent or close potencies if introduced to otherwise identical analogues of the reference ligands, while offering slightly modified solubility and polar surface areas (PSA) as described below.
As mentioned by Shi et al. it would be beneficial for brain penetration to reduce the polar surface area of GPR139 agonists 11 . Blood-brain barrier penetration is achieved when PSA is lower than 90 Å 2 18 , and the PSA of 1a is just short of this, 88.70 Å 2 . 1s displays an improved PSA of 76.7 Å 2 , but failed to increase brain exposure due to low plasma levels 11 . In contrast, 7c displays good cellular permeability (and no efflux potential), and a brain to plasma ration of 1.2 14 . Some of our bioisosteres could offer slightly more beneficial PSA values (Table 5). For R1, the PSA of 1-ethyl-2-methyl-benzoimidazole is slightly less than that of 3,5-diMeO-phenyl in Figure 6. ROC -Plot of the pharmacophore matching of all published, new compounds, and set of DUD-E generated decoys 17 . The area under the curve is 0.87, and the top 1% includes 72% actives, and 80% of the actives are matched in the first 22% of the screening set, demonstrating a very high discrimination of the active from inactive compounds.

Conclusions
In the present study we identified 12 GPR139 agonists (EC 50 = 364 nM to 4.7 μ M) containing previously untested aromatic bioisosteres, as well as novel linker variants. Our compounds were incorporated in the first combined SAR study of the three most potent series published by Lundbeck A/S, Janssen R&D and Takeda Ltd. This SAR was used to suggest a refined common pharmacophore model, which was able to discriminate between active and inactive compounds. The two proposed endogenous agonists L -Trp and L -Phe overlay with R1 and R2, respectively as well as additional linker functionalities ( Supplementary Fig. 8) showing that they too are likely to be accommodated by the common pharmacophore and the same binding site. This study could serve to guide the future ligand identification and optimization efforts. Studies to characterize the pharmacology and function of GPR139, as well as to identify antagonist tool compounds, are ongoing. Pharmacophore screening. The screening database, eMolecules plus 24 (~6 million drug-like compounds), was prepared with LigPrep 25 to desalt, add hydrogen atoms and generate tautomers, stereoisomers (max 32) and 3D conformations (max 10 ring conformations). Epik and the OPLS 2005 force field 26 were applied to generate charge states at pH: 7.0 ± 1.0 22 . LigFilter was used to remove structures with reactive functional groups. The Phase database was prepared with 100 maximum conformers, up to 10 conformations per rotatable bond, thorough conformational sampling, conformational variation of amide bonds and a maximum relative energy difference of 6.0 kcal/mol. Hit and analog purchases. A minimum of four matching pharmacophore 10 elements was required and a preference was set for partial matches involving more sites. Hits were sorted by fitness score and clustered with Canvas 27 to select diverse representative structures. After the first assaying round small structure-activity relationship analyses were conducted and the compounds sorted into lead ligand series. The selections of analogs were based on substructures drawn in MarvinSketch and queried in the eMolecules database loaded into Instant JChem (Marvin 5.12.3, 2014 and Instant JChem 6.2.0, 2014, ChemAxon, www.chemaxon.com). Compounds were purchased from Enamine (Kiev, Ukraine).

Compounds and buffers.
Compound 1a 11 was kindly provided by H. Lundbeck A/S, Denmark. Compound 7c (Enamine no: Z31449867) was tested as a racemate, as the (S)-form described by Janssen et al. was not commercially available. All compounds were dissolved to 20 mM in DMSO (Sigma, D2650) and subsequently diluted in a HEPES buffer (HBSS (Invitrogen, 14025) supplemented with 20 mM HEPES + 1 mM MgCl 2 + 1 mM CaCl 2 , pH = 7.4) to a final concentration 0.5% DMSO in the assay. The DMSO level was kept constant for all concentrations of all compounds. DMSO was confirmed not to have any activity by itself at this concentration 10 . L -Tryptophan (T0254) and L -Phenylalanine (P2126) were obtained from Sigma-Aldrich and dissolved directly in buffer.
IP-One assay. The IP-One assay was performed as described in the work by Thomsen et al. 28 , with few modifications. Breifly, CHO-139 and CHO-M1 were detached and re-suspended to a concentration of 10 million cells/ mL. 5 μ L 2x concentrated compound (+ 40 mM LiCl) and 5 μ L cell suspension (50,000 cells/well) was mixed. The plate was sealed and incubated at 37 °C for 1 hour. Next, 10 μ L detection reagents (lysis buffer containing 2.5% Eu 3+ -anti-IP1 antibody and 2.5% IP1-d2) was added and the plate was incubated for 1 hour at room temperature. The plate was read on an Envision (PerkinElmer, Waltham, MA, USA). The time resolved-fluorescence resonance energy transfer (TR-FRET) 665 nm/615 nm ratio, which is inversely proportional to the inositol monophosphate (IP 1 ) accumulation, was used to determine the concentration of the IP 1 response. cAMP assay. The cAMP assay was performed as described in the work by Thomsen et al. 28 with few modifications. Breifly, CHO-139 and CHO-M1 was detached and resuspended to a concentration of 1 million cells/ mL. 5 μ L 2x concentrated compound (for G s : + 100 μ M IBMX and for G i : + 100 μ M IBMX + 10 μ M forskolin (Fsk)) and 5 μ L cell suspension (5,000 cells/well) was mixed. The plate was sealed and incubated at room temperature for 30 min. Next, 10 μ L detection reagents (lysis buffer containing 2.5% Eu 3+ -anti-cAMP antibody and 2.5% cAMP-d2) was added and the plate was incubated for 1 hour at room temperature. The plate was read on an Envision (PerkinElmer, Waltham, MA, USA). The TR-FRET 665 nm/615 nm ratio, which is inversely proportional with the cAMP production, was used to determination the concentration of the cAMP response.
Generation of a common pharmacophore model. Phase version 4.5 29 was used to build the new pharmacophore model based on compounds from all GPR139 agonist series (Tables 1-4). Ligand conformations were generated with the thorough sampling option and hypotheses matching the variant AAADHRR (A hydrogen bond acceptor, D hydrogen bond donor, H hydrophobic group, and R aromatic ring) were taken into consideration. The EC 50 cut-off was set to 1 μ M and the hypothesis matching the defined elements was selected after scoring active and inactive compounds. Exclusion volumes were added to exclude inactive ligands volume sizes were defined based on the size of disfavoured substituents. 2D structures were drawn in MarvinSketch 30 and 3D structures were visualized in PyMOL 31 .