Structure-Activity Investigations and Optimisations of Non-metabolite Agonists for the Succinate Receptor 1

The succinate receptor 1 (SUCNR1) is a receptor for the metabolite succinate, which functions as a metabolic stress signal in the liver, kidney, adipose tissue and the retina. However, potent non-metabolite tool compounds are needed to reveal the physiological role and pharmacological potential of SUCNR1. Recently, we published the discovery of a computationally receptor-structure derived non-metabolite SUCNR1 agonist series with high target selectivity. We here report our structure-activity exploration and optimisation that has resulted in the development of agonists with nanomolar potency and excellent solubility and stability properties in a number of in vitro assays. Ligand-guided receptor models with high discriminative power between binding of active and inactive compounds were developed for design of novel chemotypes.

). Initially, the aromatic part was truncated by exchanging the fluorophenyl with a bromine (4), which resulted in more than one order of magnitude decrease in potency. Exchanging the bromofurane with the halogenated phenyls 5-7 revealed weak activity on both receptor orthologues of the ortho-chlorine and para-bromine analogs. Only the larger meta-iodine substituted compound 6 showed micromolar activity on both receptor orthologues. Inserting a methylene-linker between the amide and the bromophenyl (8) resulted in an inactive compound and exchanging the bromine for the fluorophenyl (9) only gave a weakly active compound, far less potent than the lead 3.
Since the smaller compounds did not show sufficient activity we moved the focus to explorations on the terminal phenyl ring (Table 2). Initially, the original 4-fluoro substituent was removed (10). This only affected the potency marginally on both receptor orthologues, but exposes a potential metabolically labile site. Next, a methyl scan of the ring was performed to investigate the binding site for additional space (11)(12)(13). Whereas both the ortho-and meta-methyl were tolerated only the para-methyl 13 led to a more potent compound on both human and mouse SUCNR1. Attempts to pick up hydrogen bond interactions revealed that 2-methoxy (16) was equivalent to the 2-methyl in potency and that 3-hydroxymethyl (14) was less tolerated than the 3-methyl (12) with a 3-fold decrease in potency. Again, the para-position was favoured with the 4-hydroxy (15) being almost equipotent with the 4-methyl compound. Further exploration of the para-position indicated that this part of the binding pocket is able to accommodate polar substituents such as methoxy (17) and nitrile (19) as well as nonpolar substituents such as trifluoromethyl (18), all showing similar potencies in the submicromolar range on the human receptor orthologue and with 17 slightly favoured on the mouse orthologue.
To further explore the 4-methoxy compound, the most potent analogue on the mouse receptor, a sub series of para-alkoxy analogues were investigated ( Table 3). The larger and more electronegative trifluoromethoxy (20) showed a 2-fold increase on hSUCNR1 but was accompanied by >4-fold decrease on mSUCNR1, rendering the compound approximately equipotent on both orthologues. Expanding to either ethoxy (21) or i-propoxy (22) resulted in a small increase in potency, indicating that the binding pocket can accommodate more elongated and bulky substituents. To explore whether or not the binding site could also accommodate larger hydrophilic substituents the oxatane 23 was investigated and found tolerable, but led to >2-fold reduction of potency. Finally, the mesylpropoxy analogue 24 was explored. This appendage, which lowers the lipophilicity an order of magnitude, has previously been applied to lower lipophilicity of ligands for the free fatty acid receptor FFA1 and has proven to be a metabolically stable substituent 13,14 . Lipophilicity of this compound is in the low end of the desired range but it is interesting to note that the compound exhibited a ligand lipophilicity efficiency (LLE) >5 based on ClogP and almost sustained potency on both mSUCNR1 and hSUCNR1.
Subsequently, the attention was directed towards the central aromatic ring (Table 4). Replacing the furane of 3 with the 1,3-substituted phenyl 25 improved the potency somewhat on the human orthologue whereas the potency on the mouse orthologue decreased 10-fold. The 1,4-substituted phenyl 26 clearly led to an unfavoured geometry with only trace activity on SUCNR1. The corresponding 2,6-substituted pyridine 27 sustained the potency on hSUCNR1 and almost regained the potency on mSUCNR1.
Next, a small selection of compounds was synthesised to combine the observed SAR of the alkoxy-substituents and alterations of the central ring ( Table 5). The 4-methoxy and 4-trifluoromethoxy substituents, being the most potent alkoxy-substituents on the mouse and human receptor orthologues, respectively, were attached to compounds bearing the 1,3-substituted phenyl and the 2,6-substituted pyridine as the central ring. No improvement was observed for the methoxy analogues (28)(29). In contrast, the trifluoromethoxy analogues (30)(31) were found to be the most potent compounds on hSUCNR1, but unfortunately without high potency on mSUCNR1, especially for 30. The elongated trifluoroethoxy analogue 32 showed improved potency on the mouse orthologue. Still, 31 remained the most potent agonist on hSUCNR1. Finally, the 4-propoxy analogue 33 was found to be approximately twice as potent as the corresponding methoxy analogue 29 and almost equipotent on both orthologues.
To get a better overview of the observed SAR on the human and mouse receptor orthologues a scatterplot of all active compounds was made (Fig. 2a). The compounds were color-coded according to the central or terminal ring, which clearly indicated that compounds with a central meta-substituted phenyl were better tolerated on hSUCNR1. The 2,6-substituted pyridines were in general more potent and especially 31 was favoured on hSUCNR1. Compounds with a furane as central ring were in general equipotent on the two receptor orthologues, but with addition of a phenalkoxy moiety, the compounds became more potent on mSUCNR1 with 21 being the most potent. Furthermore, dose-response curves of succinate, 3, 21, and 31 clearly showed that the non-metabolite compounds were partial agonists, with clear species differences on potency but not efficacy (Fig. 2b,c). The partial agonism was observed for all agonistic compounds (Tables 1-5, Supplementary Fig. S1).
Molecular modelling and unbiased ligand guided refinement of SUCNR1 receptor-ligand complexes. To describe the detailed molecular mechanism of action and to rationalise the structure-activity data,  we docked the compounds in Tables 1-5 to a model of SUCNR1 in complex with the lead compound (3) supported by a dozen of site directed mutagenesis data as presented earlier 4 . Despite that we observed binding conformations similar to the lead and with favourable scores, the model -as perhaps could be expected -was unable to rank the compounds in reasonable agreement with the activity data. The interpretation of SAR and structure based design reaches its maximum potential when the receptor displays the structural changes needed for ligand binding, as it has previously been shown in a blind prediction assessment of adenosine A2a receptor complexes (GPCR Dock 2008) [15][16][17][18] . We therefore applied an iterative Automated Ligand-guided Backbone Ensemble Receptor Optimisation protocol (ALiBERO) 19 , which samples full receptor and ligand flexibility guided by the ligand information gained in this study to validate and build confidence in the model. In brief, starting from the homology model supported by mutagenesis data 4 , ALiBERO introduced receptor flexibility via Normal Mode Analysis and Monte Carlo sampling, to generate a small subset of receptor models (pockets). All compounds tested in this study were grouped into an active (EC 50 ≤ 10 µM) and a decoy set (EC 50 > 10 µM) consisting of 25 compounds each. For a list of all compounds used in the optimisation protocol, see Table S1. Receptor structures were then chosen based on their ability to discriminate actives from inactives  in a retrospective virtual screening using the docking protocol and scoring function in ICM (Molsoft L.L.C., San Diego, CA, USA) 20 , as measured by the normalised square root area under the curve (NSQ_AUC). The best-performing structures from the first generation were consequently selected for the next generation and the steps were repeated in an iterative fashion until maximum docking performance of receptor structures to enrich active compounds was reached. The ALiBERO-optimised receptor ensemble was subsequently validated in a virtual ligand screening using an external test set with a higher active:decoy ratio (~1:50). Notably, the ligand-guided ALiBERO-based mSUCNR1 models demonstrated a dramatic improvement in retrospective virtual screening performance of the developed compounds compared with the initial homology models and proved successful in separating the majority of active from inactive ligands in docking screens (Fig. 3a, Supplementary Fig. S2). It is interesting to note that the active ligands bind to an extended binding cavity in a very consistent pose compared to the lead compound (3) supported by both loss-of-function but also gain-of-function mutagenesis data as we have reported previously 4 . For example, the receptor mutants R251:6.58 L and R276:7.39 F showed a potency decrease of more than 100-fold. The binding cavity of SUCNR1 is characterised by a polar network consisting of residues in TM-II (Y79:2.64), TM-III (R95:3.29), TM-VI (R248:6.55, R251:6.58), TM-VII (K269:7.32, Y272:7.35, R276:7.39), ECL2 (D174) as well as a relatively hydrophobic subpocket spanning between TM-I, -II, ECL1 and −2 up towards the extracellular surface of the receptor (residue numbering according to     shape and direction of the subpocket accommodating ring A, compounds with a bent conformation between the amide and the terminal ring, e.g. compounds having 2,5-substitued furan or meta-substituted phenyl as central ring, are sterically better tolerated than e.g. the para-substituted phenyl 26 and elongated para-substituted phenyl 9. Furthermore, compounds with a pyridine (e.g. 27, 29, 31), or furan as central ring, can stabilise the favoured bent conformation by intramolecular hydrogen bonding between the pyridine nitrogen or furan oxygen and the amide N-H, thereby inducing an optimal low energy conformation. The subpocket that is occupied by ring A, contains W10:1.42 which allows for parallel displaced aromatic stacking interactions that are present in all binding poses for the active ligands. As this pocket spans all the way to the extracellular tips of TM-I and -II, it is able to accomodate longer substituents, such as the mesylate 24 (Fig. 3b), which likely interacts with solvent water molecules on the extracellular surface of the receptor cavity. Compounds that contain O-alkyl, hydroxyl, or nitrile groups in the para-position on ring A, such as 15, 17, 19-24 and 28-31, can form hydrogen bond interactions with K83 (Asn87 in hSUCNR1) and backbone NH of Glu170 in ECL1 and −2b (Fig. 3c). Due to the geometry of this interaction, ortho-and meta-subtitutions (14 + 16) are less favorable.
In conclusion, the developed ligand-guided SUCNR1 models are consistent with the SAR, and they are sufficiently accurate to separate actives from decoys, suggesting that the models will be valuable in prospective studies to support structure-based drug design of additional chemotypes directly related to drug discovery applications.
Physicochemical and in vitro stability properties. Based on the SAR exploration, the original lead (3) and the most potent agonists on hSUCNR1 (31) and mSUCNR1 (21) were selected for further property investigations. All compounds showed excellent chemical stability (>98% after 70 hours) and kinetic solubility (>200 μM) in 10 mM PBS 7.4 . However, the high aqueous solubility made logD 7.4 determinations very challenging and only the most lipophilic compound 31 could be quantified (logD 7.4 = −2.13). The stability of the compounds was further examined in a selection of simulated gastrointestinal fluids (FaSSGF, FaSSIF and FeSSIF) and all compounds were found to be stable for 2 hours, with exception of 31 in FeSSIF, which could not be determined due to overlapping UV-absorption of the compound and media. Finally, the stability in mouse liver microsomes was investigated and all compounds were found to be stable for 1 hour, possibly partly due to the hydrophilic nature of the compounds.

Conclusion
We here report the structure-activity investigations of a series of non-metabolite SUCNR1 agonists that was originally identified from a computational, receptor-structure derived agonistic lead. ALiBERO optimised homology models of the mouse SUCNR1 were developed based on the x-ray structure of the closely related P2Y1 receptor and found to be sufficiently accurate to discriminate between actives and inactives and could explain the majority of the SAR observations. The exploration led to development of potent drug-like, non-metabolite tool compounds with nanomolar potency on both the human and murine receptor orthologues and excellent physicochemical and in vitro stability properties. We believe these compounds will be useful for further investigations of SUCNR1 as a potential therapeutic target and pharmacokinetic studies in mice are currently ongoing and will reveal if absorption of the compounds might be challenged by their hydrophilic nature and if the excellent stability properties are conserved in vivo.

Methods
General. Commercial starting materials and solvents were used without further purification, unless otherwise stated. THF was freshly distilled from sodium/benzophenone. DCM was distilled and stored over 3 Å sieves. MeCN and N,N-diisopropylethylamine were dried over 3 Å sieves. K 2 CO 3 was dried and stored in an oven. TLC was performed on TLC silica gel 60 F 254 plates and visualised at 254 or 365 nm or by staining with phosphomolybdic acid, ninhydrin, or KMnO 4 stains. Purification by flash chromatography was carried out using silica gel 60 (0.040-0.063 mm, Merck). 1 H and 13 C NMR spectra were recorded at 400 and 101 MHz, respectively, on a Bruker Avance III 400 at 300 K. Spectra were calibrated relative to the internal standard TMS or residual solvent peak: CDCl 3 (δ C = 77.16 ppm, δ H = 7.26 ppm), DMSO-d 6 (δ C = 39.52 ppm, δ H = 2.50 ppm) and acetone-d 6 (δ C = 29.84 ppm, δ H = 2.05 ppm).
Amide coupling. An oven dried microwave vial under argon atmosphere was charged with the acid (1.3 equiv), dry DCM (2 mL/mmol), N,N-diisopropylethylamine (5.5 equiv) and BTFFH (1.5 equiv). The reaction mixture was stirred at rt for 30 min before the HCl salt of the amine (1 equiv) was added. After addition, the vial was capped and heated to 80 °C overnight. The reaction was cooled to rt and diluted with water and extracted with EtOAc (x3). The organic phases were combined, washed with brine, dried over Na 2 SO 4 and concentrated in vacuo. The residue was purified by flash column chromatography (SiO 2 , EtOAc:petroleum ether).
Suzuki coupling. A schlenck flask under argon was charged with boronic acid (1.1 equiv), aryl/pyridyl halide (1 equiv) and Pd-XPhos-G4 (2 mol%). The flask was evacuated and backfilled with argon (x3). THF (5 mL/mmol) and aqueous 0.5 M K 3 PO 4 (2 equiv) was added, and the reaction was stirred at rt. After completion, the reaction mixture was diluted with water and extracted with EtOAc (x3). The organic phases were combined, washed with brine, dried over Na 2 SO 4 and concentrated in vacuo. The residue was purified by flash column chromatography (SiO 2 , EtOAc:petroleum ether).
Alkylation. The phenol (1 equiv) was dissolved in dry MeCN (~6 mL/mmol) in a dry vial under argon atmosphere. The alkyl halide/tosylate (2-7 equiv) and dry K 2 CO 3 (2 equiv) were added and the reaction was stirred at 50-55 °C until consumption of the phenol as monitored by TLC. After completion, the reaction was diluted with water and extracted with EtOAc (x3). The organic phases were combined, washed with brine, dried over Na 2 SO 4 , and concentrated in vacuo. The residue was purified by flash column chromatography (SiO 2 , EtOAc:petroleum ether).
Ester hydrolysis. The ester (1 equiv) was dissolved in THF (~6 mL/mmol), and aqueous 0.6 M LiOH (3 equiv) was added. The reaction was stirred at rt until consumption of the ester as monitored by TLC. After completion, the reaction was diluted with water, acidified with aqueous 1 M HCl and extracted with EtOAc (x3). The organic phases were combined, washed with brine, and dried over Na 2 SO 4 . The residue was concentrated in vacuo to give the pure title compounds.                    17 was synthesised from 17e (75 mg, 0.21 mmol) according to the general ester hydrolysis procedure to give 66 mg (96%) of a white solid: t R = 9.48 min (HPLC); 1 H NMR (400 MHz, Acetone-d 6 ) δ 8.05 (d, J = 8.2 Hz, 1H), 7.75 (d, J = 8.8 Hz, 2H), 7.19 (d, J = 3.6 Hz, 1H), 7.00 (d, J = 8.8 Hz, 2H), 6.84 (d, J = 3.6 Hz, 1H), 5.09-5.00 (m, 1H), 3.83 (s, 3H), 3.11-2.98 (m, 2H); 13         shaken at 700 rpm using an IKA ® KS 125 basic shaker for 24 h at room temperature. The parafilm was removed and the sample was allowed to equilibrate for 1 h before analysis. 100 μL of the octanol phase was withdrawn and diluted 1:10 with MeOH(+0.1% formic acid)/MilliQ water (4:1, v/v) and analysed by HPLC/UPLC. The interface was removed and the PBS 7.4 phase analysed directly by HPLC/UPLC. All analysis was performed in duplicates and logD values were calculated from the peak areas (mAU*min) and adjusted for difference in injection volume and concentration-absorption effects from the solvents, using two calibration points per compound per solvent, and dilution of the octanol phase. All compounds were analysed in triplicates.
Metabolic stability. Microsomal stability was studied in mouse liver microsomes (0.5 mg/mL) at a final test compound concentration of 1 μM and performed in triplicates in accordance to the published protocol 22 . In short: Prewarmed (37 °C) 0.1 M PBS 7.4, 10 mM NADPH in PBS 7.4 and test compound (1 mM in DMSO) were added to an Eppendorf ® Tube. The samples were incubated for 5 min at 37 °C before addition of newly thawned microsomes. The samples were mixed by gentle vortexing and incubated for 1 h at 37 °C, 300 rpm in an Eppendorf ® Thermomixer. Samples were quenched by addition of ice-cold MeOH/MeCN (1:1) and centrifuged for 5 min at 10,000 g. The supernatant was transferred to HPLC vials and stored in the freezer until analysis by HPLC/UPLC. The metabolic stability was calculated based on a 0 min sample. All compounds were analysed in triplicates.

Molecular biology, cell culture, and transfection. Receptor constructs for mSUCNR1 and hSUCNR1
were bought from Origene and cloned into the eukaryotic expression vector pCMV-Tag(2B) (Stratagene).
HEK-293 cells were cultured in Dulbecco's modified Eagle's medium 1885 (DMEM) supplemented with 10% fetal calf serum, 100 units/mL penicillin, and 100 μg/mL streptomycin. Transient transfection of the HEK-293 cells was done with Lipofectamine-2000 according to manufacturer's protocol. Cells were supplemented with fresh medium after 5 h. ALiBERO is an iterative sampling-selection protocol for receptor optimisation that relies on the use of ligand information for selecting the best-performing receptor conformations 19 . Homology models of the mouse SUCNR1 receptor were constructed according to Trauelsen et al. 4 and loaded into ICM (Molsoft L.L.C., San Diego, CA, USA). The structure was converted into an ICM object, thereby assigning protein atom types, optimising hydrogens and His, Pro, Asn, Gly and Cys side-chain conformations. The explored chemical compounds were used as a training set by dividing all compounds into an active (EC 50 ≤ 10 µM) and an inactive (EC 50 > 10 µM) group, consisting of 25 compounds each. For a list of all compounds used in the optimisation protocol, see Table S1. ALiBERO was performed using the prepared receptor structure and ligand training set as input. Binding site residues were manually selected based on proximity to the position of MRS2500 in the superimposed structure of the P 2 Y 1 receptor (PDB 4XNW). 100 elastic network normal mode analysis derived conformers were built in order to recreate backbone and side-chain flexibility (T = 300 K). Next, a flexible-ligand static-receptor small-scale virtual screening was performed on each of the receptor conformers, from which several pockets were selected for the following generation. The ligand and decoy molecules were docked into mSUCNR1, represented as pre-calculated potential grids and then sorted according to their ICM VLS scores. The maximum number of complementary pockets for each generation was set to 5 with a maximum of 10 generations. Receptor models were selected based on their combined screening performance, as determined by the normalised square root area under the curve, NSQ_AUC. After each round of virtual screening, an all-atom Monte Carlo side-chain refinement was performed to account for induced-by-ligand changes. NSQ_AUC values were calculated according to Katritch et al. 16 .
The optimised model ensemble was validated in a virtual ligand screening with an external test set, consisting of the 25 active compounds used in the ALiBERO optimisation and 1247 decoy molecules that were selected in a similarity search with a Daylight-type fingerprint threshold of T c > 0.6 with the active compound 3 as query molecule and subsequent structural clustering with a T c threshold of 0.15. The test set was docked into the best-performing receptor ensemble from generation 10 of the ALiBERO refinement, using ICM 4D docking. The ROC-plot for the combined performance of the optimised receptor ensemble was computed by taking the best ICL VLS docking score of the receptor models for each docked compound.
Data availability. The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.