Computer-Aided Design and Biological Evaluation of Diazaspirocyclic D4R Antagonists

Parkinson’s disease (PD) is a neurodegenerative disorder characterized by the progressive loss of dopaminergic neurons in the substantia nigra, resulting in motor dysfunction. Current treatments are primarily centered around enhancing dopamine signaling or providing dopamine replacement therapy and face limitations such as reduced efficacy over time and adverse side effects. To address these challenges, we identified selective dopamine receptor subtype 4 (D4R) antagonists not previously reported as potential adjuvants for PD management. In this study, a library screening and artificial neural network quantitative structure–activity relationship (QSAR) modeling with experimentally driven library design resulted in a class of spirocyclic compounds to identify candidate D4R antagonists. However, developing selective D4R antagonists suitable for clinical translation remains a challenge.


■ INTRODUCTION
Parkinson's disease (PD) is a debilitating neurodegenerative disorder characterized by progressive motor dysfunction resulting from the degeneration of dopaminergic neurons in the substantia nigra. 1,2The resulting dopamine deficiency leads to the classic motor symptoms of PD, including bradykinesia, resting tremors, and rigidity. 1While current treatments, such as enhancing dopamine signaling and providing dopamine replacement therapy, have been effective in alleviating motor symptoms in the early stages of PD, the need for innovative therapeutic approaches is underscored by the challenges of maintaining their long-term efficacy and minimizing the risk of side effects, including medication-induced dyskinesias. 2,3−6 Dopamine receptors are divided into two families based on structural similarities, function, and pharmacological properties: the D 1 -like receptor family, which includes primarily the D 1 R and D 5 R subtypes, and the D 2 -like receptor family, which includes D 2 R, D 3 R, and D 4 R. 7−9 Functionally, these two families have opposing mechanisms, with D 1 -like receptors stimulating adenyl cyclase through G αs signaling and D 2 -like receptors inhibiting adenyl cyclase through G αi/o signaling. 7urther receptor subtype heterogeneity can be found at the level of genetic polymorphisms.−25 D 4 R has garnered increasing attention in recent years due to its distinctive expression pattern within the central nervous system and its potential role in modulating dopamine signaling. 5,6,26−48 Consequentially, interest in the development of selective D 4 R antagonists has increased in recent decades, selected examples of which can be seen in Figure 1.The approved antipsychotics clozapine and haloperidol have also been included for reference due to their historical significance, though these are not selective for D 4 R. 9,30,49−59 The central challenge in designing D 4 R antagonists as an adjuvant therapy for PD lies in obtaining selectivity for D 4 R over the other dopamine receptor subtypes, action at which could produce undesired side effects.−64 Therefore, the pursuit of D 4 R antagonists for PD  therapy demands meticulous attention to the selectivity and efficacy of the designed compounds.−69 Building off of these rich structure−activity relationship data, we disclose herein the development of a novel class of potent, selective D 4 R antagonists suitable for further preclinical optimization.

■ RESULTS
Ligand-Based Ultralarge Library Screening to Identify Candidate D 4 R Antagonists.To identify new D 4 R antagonists, we first performed ligand-based ultralarge library screening using multitask classification artificial neural network (ANN) quantitative structure−activity relationship (QSAR) models (see Computational Methods and Materials in the Supporting Information).We trained four unique QSAR models on publicly available confirmatory screening data (molecules had reported IC 50 and/or K i /K d values) from PubChem, one each for D 2 R, D 3 R, D 4 R, and D 5 R.Each model was trained to predict the likelihood that a molecule is active at or below the following thresholds: 1, 10, 100, 1000, and 10,000 nM.Two primary metrics guided our analysis: (1) the probability that a molecule is active against D 4 R at or below 10 nM and (2) the predicted selectivity for D 4 R, where selectivity is given by the equation below.where P Dd 4 R,10nM is the QSAR-predicted probability of a molecule to be active at or below 10 nM, P Dd 4 R,1000nM is the same metric for D 2 R at or below 1000 nM, etc.Our formulation of selectivity specifically evaluates the likelihood of a molecule being selective for D 4 R at 2 orders of magnitude (active at 10 nM D 4 R vs 1000 nM D 2 , D 3 , and D 5 ).We applied our QSAR models to screen over 1 billion molecules sourced from LifeChemicals and the Enamine REAL database (Figure 2A).Compounds with 10 nM D 4 R activity prediction scores at or above 0.8 were moved forward for further analysis.Preference was given to compounds also exhibiting a selectivity score exceeding 0.4.We performed property-based flexible alignment 70 of a subset of 500 molecules to the crystallographically bound pose of the D 4 Rselective antagonist L-745,870, 68 followed by visual inspection.Ultimately, we chose 89 molecules to acquire from Enamine and LifeChemicals for experimental screening at Eurofins Discovery.
Table 1.Southern Region SAR a Values were obtained from Eurofins Discovery.See Supporting Information for more details.
Our screening efforts yielded notable outcomes, with 38 of the selected molecules displaying inhibitory activity exceeding 50% at 10 μM and 17 (see Supporting Information for structures) showing greater than 85% inhibition at 10 μM for D 4 R (Figure 2B,C).Our success for identifying selective molecules was much lower.This is not unexpected as the selectivity metric is built from multiple independent predictions (eq S1), and thus, error from each prediction accumulates in the final score.Frequently, molecules predicted to be D 4 R selective were only selective against a single offtarget subtype.Nonetheless, a subset of D 4 R-active compounds exhibited varying degrees of selectivity relative to at least one other dopamine receptor subtype (Figure 2B).
Identification of a Spirocyclic Core for D 4 R Antagonists.From our initial screen, we identified compound VU6052469, which is structurally similar to a previously published D 4 R antagonist by Carato et al. bearing a piperidine core with a naphthamide substituent that exhibits high potency and selectivity for D 4 R over D 2 R; 71 however, VU6052469 itself is nonselective (Figure 2B,D).We docked VU6052469 and the Carato compound into D 4 R (PDB ID: 6IQL) 68 to investigate the potential binding mode of our hit (Figure 2E).One challenge with designing D 4 R antagonists is the topological pseudosymmetry of D 4 R-active compounds, which in the case of VU6052469 and the Carato compound entails two distal aryl rings linked to a piperidine core (Figure 2D).In principle, this symmetry could enable the molecules to bind such that the halogen-substituted phenyl ring interacts with either transmembrane helices 2 (TM2) and TM3 (Figure S103A) or alternatively with TM4/5/6 (Figure S103B).In either binding pose, for example, VU6052469 hydrogen bonds with the conserved D3.32 side chain, and V3.33 can stack with its aromatic rings (Figure S103).The pocket formed by TM2/3 is hydrophobic and has previously been implicated in ligand selectivity. 68,69Indeed, the TM2/3 interface differs between D 4 R and D 2 R in that D 2 R contains aromatic ring side chains, while in D 4 R, there are aliphatic chains (Figure S104).In contrast, the amino acid composition of TM4/5/6 is a mixture of polar and hydrophobic residues.Notably, a cluster of serine residues engaged in internal backbone hydrogen bonds in TM5/6 renders this portion of the pocket more sterically accessible.
We reasoned that the latter pose is less likely as it induces a greater loss of planarity of the amide linker within the docked pose, which is supported by density functional theory (DFT) conformational stability calculations and molecular orbital analysis performed at the wB97X-D/6-311G(d,p) level of theory 72 (Figure S103C,D).We estimate that the first pose of VU6052469 (Figure S103A) is 11.3 kcal/mol more energetically favorable, and it follows that the Carato compound adopts a similar binding conformation (Figure 2E).Despite being nonselective, our docked poses suggest that VU6052469 could readily be made selective through extending the amide bond via a methylene linker and truncating the arene without altering the orientation of the ligand within the binding pocket.To that end, we replaced the secondary amide with an azetidine amide to give a 2,7-diazaspiro [3.5]nonane core, resulting in compound 4, which displayed selectivity for D 4 R with only a partial loss of on-target activity (Table 1).
To better understand the mechanism of selectivity imparted by the spirocyclic core, we docked 4 into D 4 R and D 2 R (see Supporting Information) (Figure 3A−C).We verified the binding mode by running molecular dynamics (MD) simulations and analyzing ligand root-mean-square-deviation (rmsd) over time (Figure S105).Our docked poses suggest that the difluorophenyl of 4 differentially engages the TM2/3 hydrophobic pocket in D 4 R versus D 2 R. Compared to its complex with D 4 R, in the D 2 R complex 4 is shifted deeper into the TM2/3 pocket such that the hydrogen bond geometry between the orthosteric pocket aspartate D3.32 and the protonated piperidine is suboptimal (Figure S106).We confirmed that the D 2 R electrostatic interactions are less favorable than D 4 R by performing geometry optimization and subsequent interface energy calculations of the complexes using the semiempirical quantum mechanics (QM) tightbinding density functional theory (DFTB) method with dispersion corrections, DFTB3-D3(BJ) (Figure 3D) (see Supporting Information). 73,74The interaction energies of 4 with respect to the conserved central aspartate D3.32 and TM2/3 hydrophobic pocket in D 4 R and D 2 R are estimated to be −24.46 and −18.53 kcal/mol, respectively (Figure 3D).
Optimization of Spirocyclic D 4 R Antagonist Potency and Selectivity.We sought to improve upon the potency and selectivity of 4 by screening analogues with differing polar aromatic or heteroaromatic groups on the southern end of the compound, installing methyl groups at the 2 or 3 position of the piperidine and probing the effect of the substitution pattern and substituent type on the northern phenyl ring on activity (Tables 1 and 2).The general synthetic scheme for this class of compounds is shown in Scheme 1, and detailed experimental procedures are provided in the Supporting Information for all intermediates and final compounds as well as compound 1.Briefly, compound 1 underwent TFA-mediated boc-deprotection followed by HATU amide coupling to afford intermediate 2. Subsequent benzyloxycarbonyl removal via hydrogen over palladium reduction gave key intermediate 3, which was subjected to either reductive amination with assorted aryl aldehydes to afford compounds 5−12 or an S N 2 reaction with 3,4-difluorobenzyl bromide to provide compound 4. To obtain azetidine amides 17−33, commercially available tert-butyl 2,7-diazaspiro [3.5]nonane-2-carboxylate was subjected to reductive amination with 6-fluoro-1Hindole-3-carbaldehyde to give intermediate 15.Boc-deprotection with TFA afforded 16, which then underwent HATU amide coupling with assorted aryl carboxylic acids to give compounds 17−34.
Overall, this focused collection of spirocyclic antagonists provided a number of valuable SAR insights.With respect to the southern region, replacing the difluorophenyl moiety with the analogous dichlorophenyl substituent (9) resulted in significantly increased activity; however, a significant decrease in selectivity between the DR subtypes was also observed.Incorporation of other substituted arenes, such as fluorophenol (8), benzodioxole (11), and fluoropyridine (10), resulted in a steep decrease in inhibition (Table 1).By installing a 6fluoroindole heterocycle (5) as we used previously in our morpholine core D 4 R antagonist (VU6004432, Figure 1), 58 we observed drastically improved activity over 4, though the overall selectivity was mildly decreased.Exchanging the indole for an indazole 6 resulted in an improvement in the selectivity against all subtypes, with a mild improvement in activity at D 4.4 R.This is in stark contrast to the incorporation of benzisoxazole (7), which essentially abolishes activity.Modifications to the spirocyclic core were not favorable as the addition of methyl groups to the 2 or 3 position of the piperidine ring (compounds 14 and 13, respectively) significantly reduced the potency and affinity of the compound compared to that observed with 5 (Table 1), while expansion Scheme 1. Synthesis of Diazaspiro [3.5]nonane D 4 R Antagonists of the azetidine to a pyrrolidine led to a substantial decrease in inhibitory activity (compound 42; see Supporting Information).
To better understand the differences in activity between 5, 6, and 7, we first docked 5 to D 4 R. Once again, pseudosymmetry within 5 rendered two flipped binding modes plausible.The first binding mode (Figure 3E) follows from the predicted poses of VU6052469 and 4. Interestingly, however, an alternative binding mode in which the indole ring of 5 adopts a pose mimicking the experimentally determined bound pose of L745,870 68 is also possible.To determine which pose is more likely, we performed MD simulations starting from each docked pose.We observed that the pose consistent with 4 (Figure 3E) is more likely to remain near the docked binding pose (Figure S107A,B) and adopt favorable hydrogen bond geometry with D3.32 (Figure S107C,D).Furthermore, the interaction energy rankings for this binding mode (Figure 3E) are consistent with the experimental results and demonstrate the activity cliff in 7 (Table 1 and Figure 4A,B).In contrast, the binding mode mimicking the L745,870 pose yields interaction energy estimates inconsistent with experiment (data not shown).Visualization of the surface electrostatic potentials of D 4 R complexed with 5, 6, or 7 at the DFT wB97X-D/6-31G(d) level of theory 72 suggests that this activity cliff is due to loss of complementary electrostatic interactions and an abundance of anionic charge near TM2 (Figure 4C−  E).
While indazole antagonist 6 provided the best potency and selectivity profile thus far, we proceeded with the combination of the 6-fluoroindole southern ring and the unmodified 2,7diazaspiro [3.5]nonane core for exploration of the northern region SAR as 5 performed similarly and was more costeffective for library synthesis.Therefore, we employed 5 as a starting point for pursuing a focused library of aryl amides on the northern end of the scaffold for further improvement of DR subtype selectivity (Table 2).Overall, alkyl and chloro substituents were well-tolerated, with the sole exception of the 3,5-dichlorophenyl analogue (28), which demonstrated drastically reduced inhibitory activity (49%).The 2,4dichlorophenyl regioisomer (19) retained activity, however, indicating that D 4.4 R inhibition is sensitive to subtle changes in substitution pattern in this region.In contrast to alkyl and chloro groups, incorporation of alkoxy groups generally led to a significant reduction in activity against D 4.4 R (23−26), with the sole exception being 20 (Figure 3F), which bears a benzodioxole heterocycle (D 4.4 R IC 50 = 84 nM; K i = 23 nM).The potency of benzodioxole-bearing compound 20 suggests that the lack of activity observed in compounds 23−26 is a result of unfavorable steric interactions facilitated by their freely rotating alkyl groups rather than ring electronics.In addition to the benzodioxole example (20), increasing the size of the aryl amide from a monocycle to a fused bicycle in other instances was also well tolerated (27, 31), with naphthalene 27 exhibiting particularly potent activity (D 4.4 R IC 50 = 28 nM; K i = 7.6 nM).With respect to selectivity, a strong sensitivity to regioisomerism was observed, which was most clearly demonstrated in compounds 29, 32, and 33, which bear para-, meta-, and ortho-toluamides, respectively.Of these, compound 29 demonstrates the highest D 4.4 R activity (D 4.4 R IC 50 = 62 nM; K i = 17 nM), and it exhibits a moderately improved selectivity profile over 5.Both meta and ortho isomers (compounds 32 and 33, respectively) display reduced activity compared to para isomer 29.Compound 33 (Figure 3G), however, exhibited the best selectivity profile of all compounds disclosed herein, with a notable 0% activity against D 2S .It was also observed that replacement of the paratoluamide of 29 with a tosylamide (34) mildly reduced the D 4.4 R activity but notably increased the inhibitory activity at all other tested DR subtypes, possibly due to the reduced planarity of the sulfonamide.

In Vitro and In Vivo DMPK Analysis of Selected
Compounds.A subset of compounds that demonstrated high potency and excellent selectivity were selected for pharmacokinetic characterization (Table 3).In vitro stability experiments in rat and human microsomes returned high clearance (>70% Q h ) across all compounds, except for 20, which exhibited moderate hepatic clearance (CL H of 13.7 and 36.2 in human and rat microsomes, respectively).The free fraction in plasma ranged from 1 to 19% in rat and 3−26% in human.Notably, both compounds 33 and 20 exhibited increased free fractions compared to the original hit (4).Three compounds (4, 20, and 33) were selected to assess in vivo pharmacokinetics (compounds 29 and 32 were excluded as they exhibited worse free fraction in plasma compared to 20 and 33).Upon intravenous dosing in rats, all three compounds demonstrated superhepatic clearance (>100% Q h ).This result is consistent with these compounds experiencing high hepatic metabolic clearance and may also indicate contribution to clearance through a different route, such as extrahepatic metabolism or active direct excretion.Despite high clearance, compounds 4, 20, and 33 exhibited moderate to high distribution into tissues (volume of distributions of 5.52, 44.4, and 36.9L/kg, respectively), explaining the reasonable half-lives for these compounds (1.05, 4.55, and 4.02 h, respectively).
Compound 33 was subjected to metabolite profiling in human and rat hepatocytes to provide insight into potential clearance mechanisms and metabolic liabilities (Figure 5).After incubation for 4 h, 33 exhibited low turnover in human hepatocytes and moderate turnover in rat hepatocytes, with 87.9 and 65.8% of parent compound (33) remaining postincubation, respectively.In both species, only two major metabolites were observed: mono-oxidation of the benzylic methyl group and piperidine N-dealkylation.The latter means of metabolism was elevated in rats (32.4%) compared to that in humans (6.3%).

■ DISCUSSION
−78 One of the central findings of the present study was the discovery of 2,7-diazaspiro [3.5]nonane as an applicable core motif for selective D 4 R antagonists.While we initially identified the highly potent antagonist VU6052469, which exhibited a high degree of structural similarity to a previously reported selective D 4 R antagonist, 71 it notably lacked selectivity (Figure 2B,D,E).We postulated that this lack of selectivity arose from the difference in length between these two compounds, with the naphthalene and 4-chlorobenzyl moieties of the Carato compound potentially leading to poorer steric interactions within the TM2/3 pocket of D 2 R than the dimethylphenyl and 3,4-difluorobenzyl moieties of VU6052469 (Figure 2E).By replacing the core piperidine of VU6052469 with 2,7-diazaspiro [3.5]nonane, the dimethylphenyl ring is extended further into the TM4/5/6 pocket, affording potent and selective activity against D 4 R (Table 1).While there have been reported examples of substituted diazaspirocycles bearing D 4 R activity, this activity was not the desired mode of action (i.e., the intent was to target σ receptors) nor did the more potent compounds exhibit DR subtype selectivity. 79Therefore, to the best of our knowledge, this is the first report of the use of diazaspirocycles in pursuit of selective D 4 R antagonists.Interestingly, our investigation revealed an activity cliff when comparing the indole/indazole-vs benzisoxazole-substituted compounds (compounds 5/6 and 7, respectively).Activity cliffs are subtle structural changes leading to significant alterations in inhibitory activity.In this case, the subtle difference in ligand interaction energies with the receptor went undetected by the docking score function.It was only after performing geometry optimization and interaction energy calculations with the more computationally demanding semiempirical QM method DFTB3-D3(BJ) that we understood the case of the reduction in binding affinity, which was a result of an accumulation of anionic charge near TM2 with no available hydrogen bond donors.This example emphasizes the continued importance of developing force fields and/or deep learning algorithms for binding affinity prediction that can be used during rapid screening protocols.
A key challenge in the rational design of selective D 4 R antagonists is the topological pseudosymmetry displayed by most antagonists.This challenge is 2-fold: (1) highly similar antagonists may be oriented in conformations 180°opposed to one another, and (2) the internal pseudosymmetry of many D 4 R antagonists renders it difficult to ascertain their appropriate binding modes.Despite extensive computational validation, it is possible that our putative binding modes are inaccurate, which may lead to false structure−activity relationships.Further experimental structural evidence, such as crystal structures of these spirocyclic antagonists bound to D 4 R, will be valuable in the design of future D 4 R antagonists with similar potencies and selectivity.
Modifications to the northern aryl amide of this scaffold demonstrated the sensitivity of D 4 R potency and selectivity to ring substituent choice and regioisomerism.Overall, compound 33, which bears an ortho-toluamide northern substituent, displayed the best selectivity profile of the tested compounds while retaining potent D 4.4 R antagonism and affinity (IC 50 = 210 nM; K i = 59 nM).Though our study has yielded promising D 4 R antagonists such as this, an ongoing challenge in the design of this class of compounds is the optimization of pharmacokinetic properties.While this class of compounds exhibited excellent aqueous solubility (see Supporting Information), both in vitro and in vivo pharmacokinetic analysis of selected compounds demonstrated a key limitation of the present class: high metabolic clearance.The findings of these assays underscore the need for continued efforts to improve the pharmacokinetic profiles of potential D 4 R antagonist drug candidates, most likely via design changes to remove metabolic hotspots within this chemical series.
Altogether, our study has unveiled a spirocyclic core for D 4 R selective antagonists, providing a foundation for further drug development efforts in the context of PD.Our insight into DR subtype selectivity and activity cliffs offers valuable guidance for future research in this area.The improvement of spirocyclic D 4 R antagonist DMPK properties, however, remains requisite for the development of a suitable preclinical lead within this class as a potential adjuvant therapy for PD.

Figure 2 .
Figure 2. Virtual high-throughput screening for D 4 R antagonists.(A) Predicted D 4 R activity vs selectivity from the ligand-based multitask ANN QSAR model ultralarge library virtual high-throughput screening.Dashed lines indicate QSAR-predicted active classification probabilities at or greater than 80% (horizontal) and 40% (vertical) for D 4 R 10 nM activity and overall selectivity, respectively.Plot color is contoured by the density of molecules, with higher-density regions appearing blue and lower-density regions appearing red.(B) Sample molecules identified during the virtual high-throughput screening.(C) D 4 R hit-rate for experimentally validated molecules.(D) 2D structures of Carato et al.: compound 22 71 and VU6052469.(E) Overlay of docked poses of Carato et al.: compound 22 71 and VU6052469 within D 4 R.

Figure 4 .
Figure 4. Surface electrostatics analysis of D 4 R selective antagonists in complex with D 4 R. (A) Schematic southern aryl substitution on compound 4. (B) Interaction energies for the model systems containing compounds 5, 6, or 7. Surface electrostatic potential analysis of (C) compound 5, (D) compound 6, and (E) compound 7. Electrostatic potentials are calculated for model systems (C−D) at the wB97X-D/6-31G(d) level of theory with solvation model density (SMD) aqueous implicit solvent following geometry optimization of the receptor pocket and ligand in complex utilizing DFTB3-D3(BJ) with SMD solvent water.

Table 2 .
Northern Ring SARStructure for this compound is a sulfonamide bound to the azetidine nitrogen of the spirocycle.b Values were obtained from Eurofins Discovery.See Supporting Information for more details. a

Table 3 .
In Vivo and In Vitro Results of Selected Compounds Figure 5. Metabolite analysis of compound 33 in human and rat hepatocytes.Parent compound incubated in human or rat hepatocytes for 4 h.Percentages (determined via LC/MS) indicate the relative percentage of compounds present postincubation.See Supporting Information for details.