Discovery of a novel class of D-amino acid oxidase (DAO) inhibitors with the Schrödinger computational platform

D-Serine is a co-agonist of the N-methyl D-aspartate (NMDA) receptor, a key excitatory neurotransmitter receptor. In the brain, D-Serine is synthesized from its L-isomer by serine racemase and is metabolized by the D-amino acid oxidase (DAO, DAAO), a flavoenzyme that catalyzes the oxidative degradation of D-amino acids including D-serine to the corresponding α-keto acids. Many studies have linked decreased D-serine concentration and/or increased DAO expression and enzyme activity to NMDA dysfunction and schizophrenia. Thus, many companies have explored the possibility of employing DAO inhibitors for the treatment of schizophrenia and other indications. Powered by the Schrödinger computational modeling platform, we initiated a research program to identify novel DAO inhibitors with best-in-class properties. The program execution leveraged an hDAO FEP+ model to prospectively predict compound hDAO inhibitory potency and prioritize design ideas from both human design and computer enumeration by our AutoDesigner algorithm. A novel class of DAO inhibitors with desirable pharmacokinetic and brain penetration properties was discovered from this effort. In an in vivo mouse PK/PD model, tool compound 37 demonstrated modulation of D-serine concentrations in the plasma and brain through inhibition of DAO function. Continued SAR work has led to significant potency improvement in both DAO biochemical and cell assays. Our modeling technology on this program has not only enhanced the efficiency of medicinal chemistry execution, it has also helped to identify a previously unexplored subpocket for further SAR development.


INTRODUCTION
D-serine is a physiological co-agonist of the N-methyl D-aspartate (NMDA) type of glutamate receptor, a key excitatory neurotransmitter receptor in the brain. D-Serine in the brain is synthesized from its L-isomer by serine racemase and is metabolized by the D-amino acid oxidase (DAO, DAAO), a flavoenzyme that catalyzes the oxidative degradation of D-amino acids including D-serine to the corresponding α-keto acids. The function of the NMDA receptor requires the presence of both the agonist (glutamate) and the co-agonist (D-serine, glycine, and/or D-alanine). Importantly, D-serine has been reported to be the predominant NMDA co-agonist in the forebrain and linked directly to schizophrenia. 1 D-serine concentrations in serum and cerebrospinal fluid have been reported to be decreased in schizophrenia patients, 2 and oral administration of D-serine improved symptoms of schizophrenia when used as an adjuvant to typical and atypical antipsychotics. 3 Thus, it is plausible to explore pharmaceutical inhibition of DAO function as putative novel therapeutics to treat the positive (psychotic), negative and cognitive symptoms in schizophrenia.
The simplest DAO inhibitor benzoic acid (1) was reported in 1956. 4 Since the early 2000s, many small molecule DAO inhibitors have been reported in the literature ( Figure 1). 5 They all mimic the substrate D-serine and bind to the catalytic site of DAO. The early inhibitors (1)(2)(3)(4)(5)(6) 6 can all be characterized as aryl carboxylic acids or corresponding acid-bioisosteres with low molecular weight. Although they are potent and highly ligand efficient, they lack the vectors that are needed for optimization of potency and physicochemical properties. To that end in 2013, Astellas reported a new class of DAO inhibitors (7) which contain a tail group reaching into a hydrophobic pocket perpendicular to the head group. 7 Takeda also worked on a similar chemical series which culminated in the discovery of their clinical candidate TAK-831 (8). The kojic acid derivatives (9) were also explored by a Johns Hopkins research group. 8 In addition, Sunovion reported a new class of DAO inhibitors (10) that stabilize an activesite lid-open conformation, although the lead compounds suffer from poor pharmacokinetic and brain penetration properties. 9

Figure 1. Previously reported inhibitors of DAO
A few DAO inhibitors have entered into clinical development. Currently, only SyneuRx is actively developing NaBen® (sodium salt of 1) in a phase II/III clinical trial for refractory schizophrenia in adults. 10 Sepracor was developing SEP-227900 for neuropathic pain around 2010. Takeda was developing 8) in phase 2 clinical trials 11 for the treatment of schizophrenia, which was the subject of a license agreement with Neurocrine in 2020. 12 In March 2021, Neurocrine reported topline data from the Phase II INTERACT study in adults with negative symptoms of schizophrenia treated with luvadaxistat (NBI-1065844/TAK-831). Although luvadaxistat did not meet its primary endpoint in the study, as measured by the change from baseline on the PANSS NSFS at Day 84, Luvadaxistat met secondary endpoints of cognitive assessment, which merit further clinical evaluation. 13 The improvement of cognitive function for TAK-831 in schizophrenic patients is consistent with improvement of cognitive performance in rodent models. For example, another DAO inhibitor SEP-227900 increased D-serine in the cerebellum of rats in a dose dependent manner, and pretreatment of rats with this DAO inhibitor increased memory of the test object in the novel object recognition model in rats, suggesting improved cognitive function. 14 There are many published DAO co-crystal structures in the literature. Figure 2 shows the co-crystal structure of human DAO enzyme with a hydroxy pyridazinone ligand (11), which was one of the most potent DAO inhibitors reported by both Takeda and Astellas. 7 Overall, the ligand adopts an L-shaped conformation in the binding site. The hydroxy pyridazinone head group is stacked between the flavin ring of FAD (flavin adenine dinucleotide) and Tyr224. The hydroxy-carbonyl moiety forms a salt bridge with Arg283, and the N-H forms an H-bond with Gly313. On the other side, the phenyl ring sticks into a relatively hydrophobic pocket and stacks with Tyr224 to form a πedge interaction. Intrigued by the target biology and therapeutic potential for treatment of cognitive impairment in schizophrenia or other neurological disorders, we initiated a program to identify novel DAO inhibitors with best-in-class properties. The program leveraged the Schrödinger physics-based modeling technology, specifically, a human DAO Free Energy Perturbation (FEP+) model which was developed on the basis of published SAR data. 15 Key protein-ligand interactions presented in the co-crystal structures were taken into account as novel ligands were designed by multiple internal medicinal and computational chemists. The designs were further evaluated with the hDAO FEP+ model, and the top ideas were prioritized for synthesis. Among them (Table 1), 12, 13 and 14 16 showed good hDAO biochemical potency, which is consistent with the FEP+ model prediction. 17 Thanks to their low molecular weights, all three compounds have good ligand efficiency (LE) and lipophilic ligand efficiency (LLE). Initially SAR work indicated that various substituents can be tolerated on the phenyl ring of both dihydropyrazine dione (12, DHP dione) and N-hydroxyl pyrimidine dione (13, NHP dione) hit classes. For instance, the CF3 group of 12 can be replaced with a chlorine to yield 15 which shares similar DAO potency. Similarly, a chlorine can be incorporated at the para-position of 13 to afford 16, which is slightly more potent than 13. In order to understand the in vivo pharmacokinetic (PK) properties of the hits, especially their ability to cross the blood-brain barrier (BBB), compounds 14, 15, and 16 were dosed as a cassette in mice along with 17 18 (Table 3) as a reference compound. To understand the binding interactions of the DHP dione chemical series, a co-crystal structure of 12 was obtained via a soaking experiment with the hDAO apo crystal. As shown in Figure 3, compound 12 binds to the hDAO enzyme in a fashion very similar to ligand 11. The dihydropyrazine dione head group is stacked between the flavin ring and Tyr224. The hydroxy-carbonyl moiety forms a salt bridge with Arg283, and the NH forms H-bond with Gly313. On the tail side, the 4-trifluoromethyl phenyl sticks into the hydrophobic pocket. Unlike the acidic hydroxy pyridazinone head group in compound 11, the pKa of 12 was measured at 9.7, 20 which would imply a pKa penalty in binding to DAO, as only the anionic form can actively bind to DAO. In addition, the head group of 12 is pseudosymmetric with two possible anionic tautomers, and substitution on the head group can impact the tautomer distribution. Fortunately, quantum mechanics (QM) calculations suggested that the active tautomer is strongly favored for 12, by 0.8 kcal/mol. Although the DHP dione chemical series is relatively weaker than the hydroxy pyridazinone 21 chemical series due to the higher pKa of the head group, it may benefit from other properties such as pharmacokinetics and brain penetration.

SAR DEVELOPMENT
Initial SAR exploration was focused on the aromatic tail region of compound 12. Both rational design by medicinal chemists and computational enumeration by Schrödinger's AutoDesigner algorithm were applied to generate a diverse set of design ideas. The large number of designs were filtered by molecular properties, a CNS MPO, a druglikeness MPO and synthetic tractability, etc. The top scoring designs were progressed into FEP+ calculations to predict hDAO inhibitory potency. The compounds with favorable predicted hDAO potency were selected for synthesis at Charles River Laboratories (CRL). Additionally, active compounds were tested in the MDCK-MDR1 assays to assess cell permeability and efflux ratio (ER).
Compound 17 was also included in Table 3 as a reference compound, which was measured 17 nM in the hDAO biochemical assay. It is slightly right shifted in the hDAO cell assay, but about 4-fold left shifted in the mouse DAO cell assay. In the DHP dione chemical series, para-substitution on the phenyl ring is beneficial to potency, as the unsubstituted analog 18 is much less active. At the para-position, CN substitution (19) can also be tolerated in addition to Cl, while the methoxy analog 20 is less potent. From the mono Cl-substituted analogs 15, 21, and 22, para-substitution is the most preferred, while ortho-substitution is not tolerated. Compound 19 can be substituted with a fluorine ortho to the cyano group as in analog 23, while 3,5-dichloro substituted analog 24 is less active when compared to the mono-substituted analog 21. The tail region tolerates other hetero aromatic rings such as pyridine (25) and bicyclic aromatic rings such as quinoline (26) with some loss of potency. Polar groups can also be tolerated in this region as exemplified by compound 27 and 28. It is worth noting that both compound 27 and 28 were designed by the AutoDesigner algorithm featuring uncommon yet drug-like functionalities. In terms of hDAO FEP+ model performance, the majority of the prediction is within 1 log unit of the experimental IC50 value. Compared to 17, most analogs showed lower but moderate cell permeability and low efflux ratio in the MDCK-MDR1 assay, which may partially account for the near 10-fold shift in the hDAO cell assay. To ensure that the compound activity is not an artifact from their redox potential, the horseradish peroxidase assay (HRP) was developed as a counter screen. All compounds in Table 3 were shown to be clean up to 10 µM in the HRP assay.   (Table 5). In terms of the linker length, the 2-carbon linker (18) is superior to the 3-carbon linker (33) according to the FEP+ predictions. 22 It is also better than the 1-carbon linker based on the matched pair of 15 and 34. Analogs with fluorinated linker (i.e. 35) are also interesting, as fluorinesubstituted linkers were predicted to lower the pKa of the head group. However, compound 35 failed in the synthesis due to chemical stability issues. Linkers with hetero atoms were also explored. Although analogs with an oxygen linker failed in synthesis, the sulfur-linked analogs are stable enough for further SAR development. Encouragingly, both 36 and 37 are about 3-fold more potent than 12 in the hDAO biochemical assay. The enhancement in biochemical potency may partly be attributed to lower pKa's of the head groups in 36 and 37. Compound 36 has a measured pKa of 8.5, and compound 37 9.2. Compared to 12 with a pKa of 9.7, 36 and 37 are more favored to form the bio-active anionic structures. In addition, they both show a lower cell shift when compared to 12, possibly due to moderately higher cell permeability as measured in the MDCK-MDR1 assay. The tail SAR of the sulfur-linked analogs 36 and 37 largely resembles that of 12 (Table 6). Further exploration of the tail SAR of sulfur-linked analogs led to significant potency improvement in the hDAO biochemical and cell assays. In addition to Cl (38 & 39) and CN (40)  In order to further enhance compound inhibitory potency against hDAO, fused ring designs in the linker region were also assessed by the FEP+ model. One of the ideas that stood out was 54, which was designed by cyclizing the linker of 37 to form a fused 1,4-oxathiane ring. It was predicted that cyclization would lead to a gain in potency resulting at least in part from stabilization of the linker and tail piece. FEP+ predicted this compound to be a 3 nM inhibitor in the hDAO assay ( Figure 4a). To our delight, the compound showed an IC50 of 25 nM in the assay, a 3fold improvement from 37. Thus, cyclized analogs with the best substituents from the chemical series were prepared. Most of these analogs showed significant improvement when compared to their acyclic counterparts in the hDAO biochemical assay. However, there was less improvement in the human and mouse cell DAO assays due to larger cell shift for the cyclized analogs. The binding mode of the cyclized analogs was confirmed by X-ray cocrystal structure of 59 (Figure 4b), which very much resembles the FEP+ snapshot of compound 54 binding to hDAO. The biggest changes are in the tail region due to different substituents at the para-position.  Another interesting design on the cyclized analogs is quaternary methyl adduct 61. While this methyl addition was not initially predicted by our FEP+ model to lead to any gain in potency when compared to the des-methyl analog 55, the racemic quaternary methyl 61-rac 23 was tested to be about 5-fold more potent than 55-rac. That is because we did not have more closely related starting references for our FEP+ model at the time of the original prediction for 61. Subsequently we troubleshooted the FEP+ model by using the more closely related des-methyl analog 55 as the starting reference and observed that the methyl group displaces a high-energy water molecule concurrent with a predicted gain of potency ( Figure 5). The quaternary methyl group was incorporated into other analogs and resulted in roughly 2-fold improvement in DAO biochemical and cellular assays. Notably, compound 63 showed inhibition potency near 100 nM in both human and mouse DAO cell assays.  In an effort to explore new opportunities for potency enhancement, careful examination of the DAO catalytic site revealed a subpocket just beyond the tail region, which was not explored by other groups (Figure 6a). In order to design into this subpocket, we employed our AutoDesigner algorithm to enumerate novel design ideas using compound 54 as a template. Initially over 198 million design ideas were generated by the algorithm, which were filtered by an array of criteria such as molecular properties, CNS and drug-like MPO's, and synthetic complexity. After GLIDE docking into the hDAO crystal structure, the surviving compounds were evaluated by the hDAO FEP+ model for potency. Only three top compounds were selected for synthesis, among which compound 66 stood out as a single digit nM hDAO inhibitor on the project. With just one round of synthesis, we were able to confirm that the subpocket is a viable design space to further enhance compound binding potency to the hDAO enzyme, which opens up much needed new SAR space for this target.

ADME/PK PROPERTIES
In order to demonstrate the therapeutic potential of the DHP dione chemical class, the team next tried to identify a suitable candidate to probe PK/PD relationship in vivo. As mentioned earlier in the SAR, most analogs showed moderate permeability and low efflux ratio in the MDCK-MDR1 cell line. They showed excellent stability in the human and mouse liver microsome assay. The compounds have also shown good stability in human and mouse hepatocytes, as no significant turnover was observed for most compounds under the assay conditions employed. The in vivo drug metabolism and pharmacokinetic (DMPK) properties were assessed in cassettes of five compounds each, including 17 as the reference. Cassette administration is an extremely useful approach to generate in vivo PK data quickly in a cost effective and animal sparing fashion. A cassette dosing strategy also enabled direct comparison of drug brain penetrability among a set of compounds within the same set of animals. In practice, cassette doses were prepared for both intravenous (IV) and oral (PO) administration utilizing a standard dose formulation for each route throughout the project. Table 10 shows mouse plasma PK of a few compounds in the chemical series. Most analogs showed low to moderate clearance and normal volume of distribution in mice, which resulted in good half-life values. They are also well absorbed when dosed orally with oral bioavailability generally over 40%. Not surprisingly, compound 66 showed reduced and less favorable oral bioavailability, possibly due to multiple rotatable bonds in the structure. The brain PK parameters of the same set of compounds are presented in Table 11. Compared to 17, the DHP dione analogs are generally more brain penetrant, which can be a key advantage. The best analogs achieved Kp,uu around 0.5. Overall, compounds 37, 42 and 46 gave the highest brain drug exposure. Surprisingly, compound 36, a close analog to 37, showed very low brain exposure and Kp,uu despite very nice plasma PK. Disappointingly, both 63 and 66 showed little to no brain exposure in mice, clearly suggesting that the more potent analogs still require much improvement in PK/brain penetration.

IN VIVO PK/PD MODEL
Based on compound potency and brain exposure, both 37 and 42 were considered as potential PK/PD candidates.
Compound 46 was deprioritized due to lower free drug fraction in the brain. To enable selection of a PK/PD candidate, high dose oral PK studies were carried out at 10 and 100 mg/kg for both compounds. Compound 37 demonstrated good dose linearity in brain, while 42 showed sub dose proportionality at 100 mg/kg ( Figure 7).
A B Ratio AUC0-inf = 11.6 Ratio AUC0-inf = 5.2 Modeling of the PK and theoretical enzyme occupancy (Equation 1) after a single 100 mg/kg dose identified 37 to be the optimal compound to progress into a PK/PD study with a 150 mg/kg BID, Q4hr dosing regimen. This study design, in conjunction with the measured mouse cell IC50, the concentration of 37 in the cerebellum and the corresponding free fraction in this tissue was predicted to provide enzyme occupancy and coverage commensurate with an in vivo biomarker response (see Figure 7). Projected tissue concentrations at 15, 50 and 150 mg/kg were calculated following a linear extrapolation of the measured values obtained from the 100 mg/kg dosing cohort illustrated in Figure 7. These data were used in Equation 1 to generate the %tEO profiles in Figure 8. Both 37 and 42 have been extensively screened in vitro for potential off-targets. In the Eurofins Safety/Diversity panel (Table 12), COX2 is the only off-target for both 37 and 42, representing about 93-fold in vitro selectivity for 37 and 132-fold for 42. In addition, the compounds have also been screened against six additional CNS targets at Eurofins, and none of them showed significant activity at 10 µM on the six off-targets. No significant inhibition of the major human CYP enzymes (<40%, 3A4, 2D6, 2C9, 2C19, 2C8, 1A2) was observed for either compound at 10 µM. In addition, there was a complete absence of any cytotoxicity signal for either compound when they were tested at 100 µM in a HepG2 assay that measured 72-hour ATP production and 24-hour Glu/Gal mitotoxicity. Following ethical review and approval of the study protocol, the PK/PD assessment was undertaken to measure the modulation of D-serine levels in the cerebella of mice following administration of the test compound at one dose using the regimen described above. Two cohorts of animals were tested (compound and vehicle) using 33 animals in total (n=8/group for 37 and n=3/timepoint for vehicle). In both cases plasma and cerebella samples were collected following animal "take-down" at 4-hour, prior to 2nd dosing, 6-hour, and 10-hour after the initial dosing. The levels of D-serine in plasma and brain tissue were quantitatively determined using a chiral LC-MS/MS method, ensuring both adequate sensitivity and selectivity. In addition, CSF was sampled from the animals at the 10-hour timepoint to determine the free, unbound levels of 37.
The bioanalytical results obtained from the PK/PD study are shown in Figure 9a. As can be seen based on a mouse cell EC50 of ~150 ng/mL and brain tissue binding of 94.8%, free drug exposures exceeding the mouse cell EC50 were observed at 10 hours in the plasma, cerebellum and CSF. A significant increase of D-serine levels compared to vehicle was also observed in both the plasma and cerebellum at all three time points measured (4, 6 and 10-hour) ( Figure 9b). In addition, a parallel study was run to assess the receptor occupancy (RO) in the cerebellum with the Takeda tracer compound PGM019260 following the protocol published in Neurochemistry Research 2017 (Ref. 18). As shown in Figure 10, the study confirmed significant RO of compound 37 in the PK/PD study, as projected by PK modeling ( Figure 8). The results of the PK/PD study are summarized in Table 13. Based upon the data, the PK/PD study with compound 37 has successfully demonstrated the pharmacological potential of hDAO inhibitors from the DHP dione chemical series.

.4 59
In parallel with the PK/PD study, compound 37 was also assessed in a catalepsy model using the same dosing regimen (150 mg/kg p.o. BID, Q4hr) that had generated the positive response in the PK/PD study. During this study, plasma samples were taken and used to assess the prolactin levels at 6 hours post the first dose, which was predicted to be around Cmax. As shown in Figure 11, no catalepsy or increase in prolactin levels was observed in this study. Plasma and brain concentrations of 37 were determined indicating that levels were similar to those achieved in the PK/PD study (data not shown). This study confirms that 37 is well tolerated in vivo at exposure levels required to evoke the desired PD responses.

CHEMICAL SYNTHESIS
Due to the diversity of the SAR, a wide variety of chemistry was attempted to synthesize the compounds on this project. A few typical procedures applied in the syntheses were described below. Please refer to the experimental section for detailed synthesis of the individual analogs.
The synthesis of compound 12 is illustrated in Scheme A. Treatment of commercial material I-A with 4-(trifluoromethyl)styryl boronic acid A-1 under Suzuki-Miyaura cross coupling conditions afforded A-2, which was subsequently reduced to the corresponding alkane A-3 by catalytic phase-transfer hydrogenation. Finally, compound 12 was obtained by refluxing A-3 in a 1:1 mixture of dioxane and 2N aqueous HCl.
Scheme A. Synthesis of compound 12.
Synthesis of the α-thioether analogs follows the general procedure in Scheme B, unless otherwise noted. Treatment of B-1 with sodium methanethiosulfone gave rise to B-2, which was reacted with the lithium salt of 2,3dichloropyrazine to afford B-3. Subsequently B-3 was converted to B-4 by reacting with methanol under basic conditions. Finally, hydrolysis of B-4 with 2N HCl produced compound 36.
Scheme B. General Procedure for the synthesis of the α-thioether compounds.
Scheme C. Synthesis of compound 37. Palladium catalyzed coupling of I-A to 2-Isopropenyl-4,4,5,5-tetramethyl-1,3,2-dioxaborolane gave rise to E-1, which was treated with NBS and NaOH to produce epoxide E-2. Reacting E-2 to thiol I-B under activation by InCl3 furnished thiol ether E-3, which was subsequently converted to the cyclized ether E-4 with palladium catalysis. Final hydrolysis of E-4 under acidic conditions followed by chiral resolution yielded the quaternary methyl compound 62.
Scheme E. Synthesis of the quaternary methyl analogs.

CONCLUSIONS
In summary, we have discovered a novel class of small molecule inhibitors against the human D-amino Acid Oxidase (DAO). Different from the earlier lead compounds, this chemical class features a non-acidic dihydropyrazine dione head moiety. Starting from hit compound 12, SAR work in the linker region led to the discovery of thioether linker analogs which showed enhanced DAO potency with desirable PK and brain penetration properties. With tool compound 37, we were able to demonstrate PK/PD in an in vivo mouse model at drug exposure levels devoid of any adverse events. Continued SAR work has led to compounds with significant improvement in both DAO biochemical and cellular potency.
We have leveraged Schrödinger's computational modeling technology extensively to accelerate the program execution. Free energy perturbation (FEP+) technology was applied to prioritize compounds based on prospective binding potency predictions. Overall, the FEP+ models have performed well in predicting compounds binding potency to the hDAO enzyme. As shown in Figure 12, compound experimental hDAO inhibitory potency correlates well with prospectively predicted potency across the three chemical series. 24 Of the ~11000 ideas designed and profiled in silico, we synthesized 208 compounds and only 20 of these were unexpectedly inactive (>10 uM), demonstrating that the physics-based methods allowed us to quickly prioritize compounds of interest and deprioritize compounds that did not meet project objectives. In addition to structure-based design by seasoned medicinal chemists and modelers, we have also applied computational enumeration with our AutoDesigner algorithm to generate novel design ideas. Most notably, this effort has helped to identify a novel subpocket for further SAR development on the project. As is common for CNS programs, the challenge is to balance compound potency with desirable PK/brain penetration properties. While a working model to predict PK/brain penetration has been elusive on this project, we will continue to apply the Schrödinger computational modeling technology along with drug-likeness and CNS MPO filters to prioritize compounds for synthesis. Further optimization work toward a development candidate will be reported in due course.

In vitro assay protocols
The D-amino acid oxidase (DAO) assays are fluorescence-based assays, in which the hydrogen peroxide (H2O2) generated from the reaction of D-serine with DAO and Flavine Adenine Dinucleotide (FAD), is linked to oxidation of Amplex Red in the presence of horseradish peroxidase (HRP). The Amplex Red reagent reacts with H2O2 in a 1:1 stoichiometry to produce the red-fluorescent oxidation product, resorufin, which is measured fluorometrically.

Preparation of Assay Ready Plates
100-fold concentrated, 12-point serial dilutions of test compounds (250 nL) in duplicate were prepared using DMSO in 384 well black plates (262260, Nunc) using the Echo555 Acoustic Dispenser (Labcyte). These assay ready plates were employed in the biochemical and cellular screening assays described below.

Human DAO Biochemical Assay
The human DAO biochemical assay was performed using reagents at the following final assay concentrations: 1 nM recombinant full-length human DAO protein, D-Serine at Km concentration (10 mM), 50 µM FAD (excess), 50 µM Amplex Red and 0.1 U/mL HRP in the presence of compound or DMSO vehicle (1%). All reagents were made up in assay buffer containing 20 mM Tris, pH 7.4 + 0.1% BSA. The final assay volume was 25 µL/well.
Briefly, 10 µL of a working solution containing 2.5 nM hDAO (TECC-1280-14AA, Takeda) and 125 µM FAD (F6625, Sigma) in assay buffer was added to all the wells in the assay ready plate (containing 250 nL compound / DMSO vehicle per well) except for the negative control wells. 10 µL of 125 µM FAD (working solution) was added to the negative control wells containing 250 nL of DMSO vehicle. The plates were incubated at 25˚C for 20 minutes (pre-incubation of compound with human DAO).
10 µL of a working solution containing 125 µM Amplex Red and 0.25 U/mL HRP (A22188, ThermoFisher Scientific) in assay buffer was then added to all the wells. The reaction was initiated by the addition of 5 µL of 50 mM D-Serine (S4250, Sigma-Aldrich) to all the wells. The plate was incubated for 4 hours in the dark at 25˚C before measuring fluorescence in each well using the Envision plate reader with excitation at 530 nm and emission at 595 nm. Concentration response curves were generated using ActivityBase (IDBS). IC50 values were determined by plotting % Inhibition vs Log10 compound concentration using a sigmoidal fit with a variable slope (four parameter fit).
Briefly, 5 µL of 250 µM FAD and 5µl of 50mM D-Serine were added to all the wells in the assay ready plate (containing 250 nL compound / DMSO vehicle per well). 5 µL of a working solution containing 250 µM Amplex Red and 0.5 U/mL HRP was added to all the wells except for the negative control wells. 5 µL of 250 µM Amplex Red was added to the negative control wells containing 250 nL of DMSO vehicle. 10 µL of 5 µM H2O2 was added to all the wells. The plate was incubated for 10 minutes in the dark at 25˚C before measuring fluorescence in each well using the Envision plate reader with excitation at 530 nm and emission at 595 nm.
Dose response curves were generated using ActivityBase (IDBS). IC50 values were determined by plotting % Inhibition vs Log10 compound concentration using a sigmoidal fit with a variable slope (four parameter fit).

Human DAO Cell Assay
The human DAO cell assay routinely employed a CHO-K1 clone, which was stably transfected with a mammalian expression plasmid containing the human DAO nucleotide encoding the full-length human DAO protein. This cell line was originally generated as described in Howley et al, 2017 (Supplementary Information). The human DAO CHO-K1 stable cell line was routinely cultured in Gibco Ham's F-12 Nutrient Mix (31765-027, ThermoFisher Scientific) containing 10% FBS (10082-147, ThermoFisher Scientific) and 500 µg/mL Geneticin™ Selective Antibiotic (10131-027, ThermoFisher Scientific).
The human DAO cell assay was performed using the following final assay conditions: 25,000 human DAO CHO-K1 stable cells/well, 50 mM D-Serine, 50 µM Amplex Red and 0.125 U/mL HRP. All cells and reagents were made up in 10 mM HEPES buffer (15630-056, ThermoFisher Scientific). The final assay volume was 25 µL/well. The human DAO CHO-K1 stable cells were trypsinised, resuspended in complete medium and centrifuged at 1200 rpm for 4 minutes at room temperature. The cell pellet was then washed in 10 mM HEPES buffer and centrifuged at 1200 rpm for 4 minutes at room temperature. The resulting cell pellet was resuspended in 10 mM HEPES buffer at 1.25x10 6 cells/mL. 25,000 human DAO CHO-K1 stable cells (20 µL in 10 mM HEPES buffer) were added to all the wells in the assay ready plate (containing 250 nL compound / DMSO vehicle per well). 5µl of a working solution containing 250 mM D-Serine, 250 µM Amplex Red and 0.625 U/mL HRP in assay buffer was added to all the wells except for the negative control wells. 5 µL of a working solution of 250 µM Amplex Red and 0.625 U/mL HRP in assay buffer was added to the negative control wells. The plate was incubated for 30 minutes in the dark at 25˚C before measuring fluorescence in each well using the Envision plate reader with excitation at 530 nm and emission at 595 nm. Dose response curves were generated using ActivityBase (IDBS). IC50 (Point of Inflection) values were determined by plotting % Inhibition vs Log10 compound concentration using a sigmoidal fit with a variable slope (four parameter fit).

Mouse DAO Cell Assay
The mouse DAO cell assay routinely employed CHO-K1 cells, which were transiently transfected with an expression plasmid containing the mouse DAO nucleotide encoding the full-length mouse DAO protein. The T175 flasks, which were seeded with CHO-K1 cells were transfected with mouse DAO/ pcDNA3.1+C_(K)-DYK expression plasmid as follows. A transfection mix for each T175 flask was made up containing 43.75 µL mouse DAO/ pcDNA3.1+C_(K)-DYK expression plasmid (OMu05394D_pcDNA3.1+ C_(K)-DYK endotoxin free (maxiprep, 1 mg/mL, Genscript), 8575 µL of Optimem (31985-062, ThermoFisher Scientific) and 131.25 µL of Lipofectamine LTX (15338-100, ThermoFisher Scientific). The transfection mix was incubated at room temperature for 25 minutes. During this incubation period the complete growth medium was removed by aspiration from the CHO-K1 cells, which were seeded 24 hours previously and replaced with 35ml of fresh Gibco Ham's F-12 Nutrient Mix containing 10% FBS. The transfection mix was then added to each flask containing CHO-K1 cells and incubated for a further 24 hours at 37 °C/5% CO2.
The mouse DAO transiently transfected CHO-K1 cells were trypsinised, resuspended in complete medium and centrifuged at 1200 rpm for 4 minutes at room temperature. The cell pellet was then washed in 10mM HEPES buffer and centrifuged at 1200 rpm for 4 minutes at room temperature. The resulting cell pellet was resuspended in 10 mM HEPES buffer at 1.75x10 6 cells/mL. 35,000 mouse DAO CHO-K1 transiently transfected cells (20 µL in 10 mM HEPES buffer) were added to all the wells in the assay ready plate (containing 250 nL compound / DMSO vehicle per well). 5 µL of a working solution containing 250 mM D-Serine, 250 µM Amplex Red and 0.625 U/mL HRP in assay buffer was added to all the wells except for the negative control wells. 5 µL of a working solution of 250 µM Amplex Red and 0.625 U/mL HRP in assay buffer was added to the negative control wells. The plate was incubated for 30 minutes in the dark at 25˚C before measuring fluorescence in each well using the Envision plate reader with excitation at 530 nm and emission at 595 nm.
Dose response curves were generated using ActivityBase (IDBS). IC50 (Point of Inflection) values were determined by plotting % Inhibition vs Log10 compound concentration using a sigmoidal fit with a variable slope (four parameter fit).

Mouse pharmacokinetics
Male C57Bl/6NCrl mice, Inbred, SPF-Quality, Charles River, Germany between 8 and 10 weeks of age, ranging from 20 to 40 grams were used to study the pharmacokinetics of test compounds. On arrival and following randomization animals were housed individually in polycarbonate cages equipped with water bottles, unless contraindicated by study procedures (such as pharmacokinetic blood sampling) or clinical signs. Pelleted rodent diet (SM R/M-Z from SSNIFF® Spezialdiäten GmbH, Soest, Germany) was provided ad libitum throughout the study, except during designated procedures. The compounds were administered to the mice via a single intravenous (slow bolus) injection to the tail vein using a vehicle comprising DMSO, PEG200 and Water . Terminal blood samples were collected via aorta puncture following inhalation anaesthesia into K2EDTA tubes and stored on wet ice. Oral cohorts were dosed by gavage using a vehicle of 0.5% (w/v) methylcellulose and 0.1% (v/v) Tween80 in water with bloods collected using a similar procedure. Whole blood was processed to plasma by centrifugation (3000g for 10 minutes at 5°C) within 30 minutes of collection. Plasma samples were transferred into 96 well plates (matrix tubes) and stored at < -75°C. Following termination, brains were collected from the animals and the cerebella separated. Both tissues were rinsed with saline, weighed and stored at ≤ -75 oC prior to analysis using LC-MS-MS.
Plasma and brain samples were extracted by protein precipitation using acetonitrile containing an appropriate internal standard. Specific reaction monitoring transitions were identified using automated instrumental optimization procedures for each compound studied, to ensure adequate linearity of response and define the upper and lower limits of quantitation. Samples were injected (SIL-30AC Autosampler, Schimadzu, Kyoto, Japan) onto a reverse phase chromatography system (A: 0.1% formic acid in ultrapure water; B: 0.1% formic acid in acetonitrile, Waters Corporation Acquity® UPLC column HSS T3 1.8μ). Analysis was performed using an API 5000 triple quadrupole mass spectrometer fitted with an electrospray ionisation source (AB Sciex, Ontario, ON, Canada). Pharmacokinetic analysis was performed with IDBS E-WorkBook v10 using mean data, non-compartmental analysis and the nominal dose of test item administered to the study animals.

Hepatic microsomal stability
The stability of the test compounds (1 µM) was measured following incubation at 37 °C with hepatic microsomes (0.5 mg protein/mL for all species) in the presence of the cofactor, NADPH. Incubates were prepared in duplicate, with aliquots removed at 0, 5, 10, 20 and 40 minutes and reactions terminated and compound extracted by the addition of acetonitrile containing an analytical internal standard. The disappearance of parent compound was monitored by LC-MS/MS and the half-life determined over the time-course of incubation. The half-life values were used to calculate their in vitro intrinsic clearance expressed as µL/min/mg protein.

Cryopreserved hepatocyte stability
The stability of test compounds (1 µM) were measured following incubation at 37 °C with cryopreserved hepatocytes in suspension at a cell density of 0.5 million cells per mL. Incubates were prepared in duplicate with aliquots removed at seven time points over a period of 120 minutes and reactions terminated and compound extracted by the addition of acetonitrile containing an analytical internal standard. The disappearance of the parent compounds were monitored by LC-MS/MS and half-life values determined over the course of the incubation. The half-life values obtained were used to calculate their in vitro intrinsic clearance expressed as µL/min/million cells.

MDCK assay protocol
MDR1-MDCK cells were seeded into 24 well Transwell plates and cultured for 3 days to form monolayers. The test compounds were prepared at 10 µM in Hanks' Balanced Salt Solution containing 25 mM HEPES and loaded into the donor compartments of Transwell plates bearing the cell monolayers (pH 7.4 for both donor and receiver compartments). Lucifer Yellow was added to the apical buffer in all wells to assess integrity of the cell monolayer. Duplicate wells were prepared and incubated at 37°C in a CO2 incubator. Samples were removed at time zero and 60 minutes and test compound analysed by LC-MS/MS. Concentrations of Lucifer Yellow in the samples were measured using a fluorescence plate reader. The apparent permeability (Papp) values of test compound were determined for both the apical to basal (A>B) and basal to apical (B>A) permeation and the efflux ratio (B>A: A>B) determined.

Animal Models
In vivo studies were performed at Charles River Laboratories (South San Francisco, CA, USA) under animal welfare protocols approved by the Institutional Animal Care and Use Committee of Charles River Laboratories, South San Francisco, and they adhere to the ACS Ethical Guidelines for animal studies.
In the PK/PD study, adult male C57Bl/6 mice with 7-8 weeks of age were dosed orally with compound 37 as a suspension in 1% Tween 80 in 0.5% methylcellulose at 150 mg/kg BID q4h. Terminal tissue collection was conducted at 4, 6, and 10 hours after treatment (11 mice/timepoint). Mice being euthanized for the 4-hour group were euthanized before 2nd dosing. At each collection timepoint, mice were euthanized by CO2 asphyxiation and blood was collected via cardiac puncture into vials containing K + EDTA anticoagulant. Then, brains were extracted, and cerebellum dissected, separated into 2 equal parts then placed into pre-weighed 1.5ml tubes. Terminal CSF was collected for 10h treatment group only. Upon collection, all tissue samples and CSF were weighed, snap frozen in liquid nitrogen and stored at -80 °C for analysis.
The receptor occupancy study followed very similar protocol as the PK/PD study. In addition to treatment group with compound 37 and the vehicle group, a third group of C57Bl/6 mice (n = 12, 4 at each timepoint) were dosed IV with tracer compound PGM019260 at 60 μg/kg in 10% DMSO in 0.5% 90% HP-β-CD, 20 minutes prior to the defined takedown time. Terminal tissue collection was conducted at 4, 6, and 10 hours after treatment (14 mice/timepoint). At each collection timepoint, mice were euthanized by CO2 asphyxiation and brains were extracted and dissected in cerebellum and prefrontal cortex tissue samples and placed into pre-weighed 2 ml tubes. Upon collection, all tissue samples were weighed, snap frozen in liquid nitrogen and stored at -80 °C for analysis.
before being cooled to 0 °C. A solution of S- (4-(trifluoromethyl)benzyl) methanesulfonothioate (665 mg, 2.46 mmol) in dry THF (2 mL) was added dropwise over 10 min. The reaction mixture was stirred at 0 °C for 30 minutes, allowed to warm to room temperature and stirred for 3 hrs. Saturated aqueous ammonium chloride solution (15 mL) was added followed by water (10 mL). The mixture was extracted with ethyl acetate (100 mL, 20 mL, 20 mL) and the combined organic layers were washed with brine (20 mL), dried over magnesium sulfate, filtered and concentrated under reduced pressure. The crude material was purified by flash column chromatography (10 -20% dichloromethane in cyclohexane) to yield the title compound as a yellow oil (269 mg, 39%). ¹H NMR ( Step C: 2,3-Dimethoxy-5-((4-(trifluoromethyl)benzyl)thio)pyrazine To a suspension of sodium hydride (60% in mineral oil, 188 mg, 7.81 mmol) in dry dioxane (3 mL) under nitrogen at room temperature was added dry methanol (0.32 mL, 7.81 mmol) dropwise over 10 min. The reaction mixture was stirred at room temperature for 1 hour. A solution of 2,3-dichloro-5-((4-(trifluoromethyl)benzyl)thio)pyrazine (265 mg, 0.781 mmol) in dry dioxane (2 mL) was added over 5 min. and the mixture was stirred at room temperature for 18 hrs. Saturated aqueous ammonium chloride solution (15 mL) was added followed by water (10 mL). The mixture was extracted with ethyl acetate and the combined organic layers were washed with brine (20 mL), dried over magnesium sulfate, filtered and concentrated under reduced pressure. The crude material was purified by flash column chromatography (10 -20% dichloromethane in cyclohexane) to yield the title compound as a pale-yellow oil (138 mg, 53%). ¹H NMR ( Step The following compounds were synthesized following the same procedure as 36. Step A: 2-((3,4-Difluorobenzyl)thio)-6-methoxypyrazine Step B: