Optimization of 2-Amino-4,6-diarylpyrimidine-5-carbonitriles as Potent and Selective A1 Antagonists

We herein document a large collection of 108 2-amino-4,6-disubstituted-pyrimidine derivatives as potent, structurally simple, and highly selective A1AR ligands. The most attractive ligands were confirmed as antagonists of the canonical cyclic adenosine monophosphate pathway, and some pharmacokinetic parameters were preliminarilly evaluated. The library, built through a reliable and efficient three-component reaction, comprehensively explored the chemical space allowing the identification of the most prominent features of the structure–activity and structure–selectivity relationships around this scaffold. These included the influence on the selectivity profile of the aromatic residues at positions R4 and R6 of the pyrimidine core but most importantly the prominent role to the unprecedented A1AR selectivity profile exerted by the methyl group introduced at the exocyclic amino group. The structure–activity relationship trends on both A1 and A2AARs were conveniently interpreted with rigorous free energy perturbation simulations, which started from the receptor-driven docking model that guided the design of these series.


■ INTRODUCTION
The endogenous nucleoside adenosine is essential for the proper functioning of every cell in mammalian species. 1,2 Adenosine is produced intra-and extracellularly (both in the brain and in the periphery) under diverse physiological and pathophysiological conditions, and its effects are mediated through activation of four membrane adenosine receptors (ARs), namely, A 1 AR, A 2A AR, A 2B AR, and A 3 AR. 1,3 ARs are expressed ubiquitously and play critical roles in the regulation of cardiac muscles, 4 neuronal function, 5,6 pain, 2,3 and sleep. 3,7,8 In addition to its cytoprotective mission, there are instances in which a chronic overproduction of adenosine becomes pathological (e.g., cancer, diabetes, colitis, fibrosis, hepatic steatosis, or asthma). 2,3,9,10 A large body of evidence supports that the regulation of the adenosinergic signaling pathways by compounds that modulate the different ARs [e.g., (full or partial) agonists, antagonists/inverse agonists, and allosteric modulators] constitutes innovative approaches to address challenging medical needs. 11−14 Since its early discovery and cloning, 15 the A 1 AR has been considered an attractive target for therapeutic intervention. 16 It is highly abundant not only in the central nervous system (cortex, hippocampus, cerebellum, astrocytes, oligodendrocytes, and microglia) but also in peripheral tissues (heart, kidney, airway smooth muscles, skeletal muscles, liver, or pancreas), thus emphasizing its pivotal role in a diverse physiological process. 3,14 The A 1 AR is implicated not only in the central excitatory system, participating within the development of several neurological and neurodegenerative disorders (e.g., epilepsy, depression, or Parkinson's), but also in cognitive functions. 3,7 Recent evidence supports that the A 1 AR blockade increases extracellular levels of acetylcholine, a neurotransmitter highly decreased in Alzheimer's disease. 17 On the other hand, peripheral A 1 AR has been targeted in the search of novel drugs for hypertension, heart failure, allergy, or asthma. 18 In particular, A 1 AR antagonists have been proposed as effective potassium-sparing diuretic agents with kidney protecting properties. 19 Currently, the only A 1 AR antagonist in clinical studies is PBF-680 (structure not disclosed), which is undergoing phase II as a peripheral selective oral treatment for respiratory diseases (asthma and COPD). 20,21 The increasing availability of crystallographic and cryo-EM AR structures, complemented with homology models and decades of site-directed mutagenesis studies, 22 allowed us to improve our molecular understanding of ligand recognition and receptor signaling within the AR family, thus providing solid foundations for the rational design of AR modulators. 23,24 In particular, we now have A 1 AR structures in both the inactive and active states. 25,26 Moreover, the X-ray crystal structures of both of A 1 AR and A 2A AR in complex with the A 1 selective antagonist PSB36 provided structural insight into receptor selectivity, 27 further explored with computational methods. 28 The therapeutic applications emerging from A 1 AR modulation stimulated the development of several series of small molecule A 1 AR ligands. 18−21 From these, A 1 AR antagonists can be classified in two structural families: xanthines and nonxanthines ( Figure 1). The discovery that naturally occurring alkylxanthines (e.g., caffeine, theophylline, and theobromine) are micromolar (non-selective) AR antagonists inspired extensive pharmacomodulation of the xanthine moiety, thus culminating with the identification of potent and selective A 1 AR, A 2A AR, and A 2B AR antagonists. Xanthine-based A 1 AR antagonists generally contain a bulky hydrophobic group at position 8 and alkyl chains at positions 1 and 3 ( Figure 1, Cpds 1−6). 25,29−35 Despite possessing excellent affinity and subtype selectivity, the advancement of xanthine-based A 1 AR antagonists as drug candidates has been hampered by their poor bioavailability and low water solubility, narrow efficacy, and off-target effects. 18−21 Efforts to identify non-xanthine A 1 AR antagonists mostly focused on bicyclic scaffolds that somehow mimic the adenine core present in the endogenous ligand (adenosine) and, to a lesser extent, tricyclic (Figure 1, Cpds 7 and 8) 36,37 and monocyclic systems (Figure 1, Cpds 9− 11). 38−40 However, the high structural homology between the A 1 AR and A 2A AR, particularly in the orthosteric site, has limited the development of A 1 AR antagonists exhibiting both high affinity and selectivity against the A 2A AR. Thus, only a few truly selective monocyclic A 1 AR antagonists have been described so far, with representative examples based on the thiazole and pyrimidine cores ( Figure 1, . 41,42 It follows that the identification of highly potent and selective structurally simple A 1 AR antagonists remains a challenging goal. As part of a program aimed at the development of adenosine receptor antagonists, we here report the discovery, optimization, pharmacological profiling, and structure-based SAR of potent, structurally simple, and highly selective non-xanthine A 1 AR antagonists. A large library, consisting of 108 ligands derived of the 2-amino-4,6-disubstitutedpyrimidin-5-carbonitrile chemotype, was obtained by using a novel, succinct, and efficient three-component synthetic strategy. The interpretation of the main structure−activity relationship trends within the series was supported by free energy perturbation (FEP) simulations based on the crystal structure of the human A 1 receptor. A preliminary exploration of the pharmacokinetic profile of the most attractive ligands identified (19l, 19v, and 19ao) was carried out by determining its microsomal stability and solubility. Finally, to explore the potential of the designed ligands as CNS agents, we investigated their ability to be substrates of P-glycoprotein (P-gp), the efflux pump present at the blood brain barrier, which represents the first line of defense of the CNS.
Pharmacomodulation of the 4-aminopyrimidine core successfully afforded a novel series of selective A 2A AR antagonists, achieved by substitution at position 5 of the heterocycle and adequate decoration of positions 2 and 6 ( Figure 2, Cpds 16). Conversely, this same scaffold provided A 1 AR antagonists by introducing a cyano group at position 5 and transforming the amino group in substituted amides ( Figure 2, Cpd 15c). In a clear contrast, most ligands derived from the 2-aminopyrimidine scaffold retained the dual A 1 AR/A 2A AR antagonistic profile (Figure 2, Cpds 17a−d), showing that achieving a selective profile for this scaffold is somehow more challenging. Two remarkable exceptions are compounds 17e and 17f ( Figure 2). 42 The former was developed by van Veldhoven et al. 42 by the introduction of analogous substitutions used in their A 1 -selective 4-amidopyrimidine 15c, while 17f was obtained by moving the aromatic ring (naphthyl group) to position 5 and the introduction of the cycloalkyl fragment present in SLV320 ( Figure 1) in position 2. Although compounds 15c and 17e showed high A 1 AR potency and selectivity, most of their congeners could not escape the dual A 2A AR/A 1 AR profile observed in the early series, thus suggesting that amide formation is not enough to achieve A 1 AR selectivity. The beneficial effects of dual A 1 AR/A 2A AR antagonism in Parkinson animal models, observed for Cpds 17c−e 43 (Figure 2), supported prioritization of dual ligands, resulting in a decay of the interest in the identification of selective ligands between these two subtypes of the ARs.
We herein developed a novel series of 2-aminodiarylpyrimidine derivatives (Figure 2, Cpds 18−20), eliciting excellent A 1 AR affinity and selectivity, which distinctively combine chemical decorations inspired by the SAR data discussed above: (i) a cyano group at position 5, (ii) diverse aryl groups at positions 4 and 6, and (iii) free, mono-, or disubstituted amino groups at position 2. The hypothesis behind this design relies on the effect of the cyano group at position 5, which increases the acidity of the exocyclic (substituted) amino group, leading to stronger binding to the ARs by reinforcing the double-hydrogen bond with Asn 6.55 , while R 4 , R 6 , and particularly R 2 would control the selectivity profile.
Biological Evaluation. The adenosinergic profile (affinity and selectivity) of the 108 synthesized derivatives of the 2aminopyrimidine-5-carbonitrile scaffold was evaluated in vitro using radioligand binding assays at the four human AR subtypes. 46 3 AR were employed as radioligands in binding assays. The biological data are expressed as K i (nM, n = 3) or as percentage inhibition of specific binding at 1 μM (n = 2, average) for those compounds that did not fully displace specific radioligand binding. K i values were obtained by fitting the data with non-linear regression using Prism 2.1 software (GraphPad, San Diego, CA). For comparative purposes, the binding affinities obtained for three representative AR ligands (XAC, ZM241385, and DPCPX), using the binding protocols herein employed, are included in Tables 1−3. The whole set of ligands (18−20) was in silico evaluated, using the PAINS filter in the RDkit, 50 to rule out these ligands being promiscuous pan-assay interference compounds (PAINS).
Functional Experiments and Preliminary ADME Determinations. A representative set of the obtained A 1 AR ligands (19ao, 19l, and 19v) was evaluated in cAMP assays to determinate their ability to reverse the inhibitory effect of NECA (100 nM) on forskolin-stimulated (3 μM) cAMP production. The log concentration-response curves of cAMP accumulation for selected antagonists to hA 1 ARs are presented in Figure 3. These experiments demonstrated that selected compounds (19ao, 19l, and 19v) and XAC reverse the inhibitory effect of NECA on FSK-induced cAMP accumulation, unequivocally validating their antagonism at hA 1 ARs. The K B values obtained during the functional experiments at hA 1 ARs show low nanomolar range data (K B = 3.90, 6.21, 9.72, and 14.50 nM). As a complement of these experiments, the functional data of selected compounds (19ao, 19l, and 19v) was investigated at the other three adenosine receptor subtypes (hA 2A ARs, hA 2B ARs, and hA 3 ARs). This study (Supporting Information, Table S2) confirmed that the excellent selectivity profile observed in the binding studies (Tables 1 and 2) is reproduced when evaluating the functional behavior of the new ligands documented here.
Some preliminary ADME experiments were performed to gain insight into the pharmacological profile of representative ligands (19ao, 19l, and 19v). A solubility assay was performed to evaluate the aqueous solubilities of 19l, 19v, and 19ao. The solubilities, phosphate buffered saline (pH 7.4), were determined to be 75.4, 18.0, and 5.4 μM, respectively. The low solubility observed for 19ao may be attributed to the lipophilic piperonyl group at R 4 . The stability of selected compounds in human microsomes was also studied (Supporting Information, Table S1). After 60 min of incubation in human microsomes, the remanent ligand ranged from 3.7 to 15%, so further structural optimization should be performed to improve the microsomal stability profile within the series. According to McNaney et al.'s classification, 51  ·min −1 ). P-Glycoprotein Interaction Assays. An exploratory cellular-based assay was performed to evaluate the potential of the A 1 AR antagonists here described as active agents at the Scheme 1. Three-Component Assembly of the Novel 2-Amino-4,6-diaryl-5-carbonitriles (18−20) Journal of Medicinal Chemistry pubs.acs.org/jmc Article CNS level. P-Glycoprotein (P-gp) 52 is an ATPase representing a first line of defense of our brain toward toxins and drugs. Pgp uses the hydrolysis of ATP to efflux drugs out from the brain parenchyma. Therefore, the P-gp interaction profile of drug candidates constitutes an in vitro assay informative of the ability of drugs to hit the central targets. For this purpose, we studied the ability of selected compounds (19l, 19v, 19ao, 19af, and 19aj) to compete with the transport of a profluorescent probe, Calcein-AM, that is also a P-gp substrate, in a cell line overexpressing P-gp (MDCK-MDR1 cell line) mimicking the BBB that was measured. Briefly, in MDCK-MDR1 cells, the pro-fluorescent Calcein-AM is not able to enter the cell membrane as effluxed by P-gp; in the presence of an agent able to interact with the pump (as a substrate),      Table 4. As observed, any of the tested compounds showed a significant interaction in MDCK-MDR1 cells with the Calcein-AM transport with respect to the P-gp reference substrate verapamil (EC 50 = 0.50 μM). 56 This preliminary (cellular) data suggest that, herein, described ligands should not be effluxed by the pump, thus showing a potential ability to overcome the BBB. Structure−Activity Relationship. Examination of the binding data reveals the identification of eight A 1 AR ligands that combine high affinity (K i < 50 nM) and outstanding selectivity (>1000-fold; see Table 2,compounds 19l,19v,19z,19af,19aj,19ao,19az,and 19bd). Although this project focuses on the identification of A 1 AR antagonists, during the pharmacological screening of the obtained library, we identified three A 2A AR selective ligands eliciting high (19n, K i = 46.3 nM) or moderate (19ba, K i = 102.0 nM; 19ay, K i = 368.1 nM) affinity for this receptor and negligible affinities for        (Table 2). Moreover, three of the pyrimidine-5-carbonitriles prepared exhibited an attractive dual A 2A AR/A 2B AR profile ( Table 2, compounds 19ab, 19an, and 19bf), which are now being investigated within the context of our anticancer programs.
For a more immediate and efficient analysis of the variation of both affinity and selectivity, the binding data of the main series 18 and 19 (Tables 1 and 2) is presented as a function of the pKi A 2A AR (Y axis) vs pKi A 1 AR (X axis) (Figures 4 and   5). Compounds lining around the diagonal of this square plot will bear equal affinities at both receptors, whereas A 1 AR or A 2A AR selective compounds will cluster on regions below or above the diagonal, respectively, with the distance from the diagonal being directly correlated with their degree of selectivity. In this work, the emphasis is put on compounds with high A 1 AR affinity and high A 1 AR/A 2A AR selectivity, so we will thus focus on the lower right-hand side corner of the plots.
Series I (compounds 18) always maintained a nonsubstituted exocyclic amino group and was designed in two subsets: in the first one (Table 1, compounds 18a−18ag), a phenyl ring was maintained invariable at position 6 to explore the effect of diverse substituents at position 4. In the second subset, the chemical groups at positions 4 and 6 were simultaneously modified (Table 1, compounds 18ah−18am). Inspection of the pharmacological data obtained for the whole series of 2-aminopyrimidine-5-carbonitriles (18; Figure 4, black circles, and Table 1) reveals that, regardless of the aryl or heteroaryl group present at R 4 and R 6 , most compounds exhibit a rather dual A 1 AR/A 2A AR affinity profile. Collectively, these ligands elicit superior (low nanomolar) affinity at A 1 AR but generally exhibiting low (3-to 10-fold) selectivity toward A 2A AR. This trend can be visualized in Figure 4 (black circles), with most derivatives appearing only slightly under the diagonal in the right part of the plot. The only exception to this trend is observed for 18al, which elicits modest A 1 AR affinity (K i = 74.3 nM) and a noticeable selective profile (Table  3).
Interestingly, accompanying their A 1 AR/A 2A AR profile, a relevant A 2B AR binding affinity is observed for compounds bearing 3-substituted phenyl groups or heterocyclic moieties at R 4 (e.g., 18e, 18f, 18g, 18h, 18t, 18u, 18x, and 18y) and R 6 (18ah and 18am). The absence of any AR affinity in ligands bearing cyclopentyl or cyclohexyl groups at R 4 ( Table 1, compounds 18ad and 18ae) confirms the importance of the (hetero)aromatic substituents at these positions. Moreover, the data presented in Table 1 is coherent with the SAR trends observed for a structurally related series (Figure 2), 42−44 thus indicating that the aromatic moieties at 4 and 6 are critical contributors for recognition and binding at both A 1 AR and A 2A AR.
Inspection of the data presented in Table 2 (series II) reveals the significant effect of the substituent on the amino group (R 2 ) in the adenosinergic profile of these series. While these derivatives retain the excellent A 1 AR affinity observed in series I discussed above, alkylation at the amino group substantially affects the observed selectivity profile (Table 2 and Figure 5). Thus, pyrimidine-5-carbonitriles bearing a methylamino group at position 2 generally combine high A 1 AR affinity and outstanding selectivity toward A 2A AR (Table 2, compounds 19j, 19l, 19v, 19z, 19af, 19aj, 19ao, 19az, and 19bd), which appear consequently clustered in the low-right corner of the corresponding selectivity plot (Figure 4, orange squares, and Figure 5). In a clear contrast, the introduction of a    Figure 4 (blue triangles), together with compounds 19ac, 19an, and 19bf which, as mentioned before, exhibit an attractive dual A 2A AR/A 2B AR profile (inverted triangles in Figure 4). In particular, ligand 19bf has similar affinity at A 2A AR (K i = 27.1 nM) and A 2B AR (K i = 29.4 nM), constituting a highly attractive pharmacological tool to explore the effect of simultaneous blockage of A 2A AR and A 2B AR in A 2 AR-responsive cancer cell lines.
As part of the SAR study, it was decided to briefly explore the effect of introducing an ethyl group at the exocyclic amine (R 2 = Et; Table 2). These derivatives elicit excellent to satisfactory A 1 AR affinity (K i = 2.46−135.9 nM), though the selectivity profile toward A 2A AR is rather moderate (30-to 65fold, green stars in Figure 4), thus suggesting a very specific effect of the methyl group in the exocyclic amine.
With 66 pyrimidine-5-carbonitrile derivatives, series II constitutes the most interesting subset for further exploration. Ligands illustrative of the observed SAR trends were selected for graphical representation ( Figure 5) as a function of the pKi A 2A AR (Y axis) vs pKi A 1 AR (X axis). The observation in series I where the introduction of two phenyl-substituted residues at R 4 and R 6 of the pyrimidine core (Table 1, compounds 18ah− 18al) does not improve the affinity/selectivity profile aimed us to (mostly) maintain invariable a phenyl group at R 6 in series II and instead focus on an exhaustive exploration of the substitution patterns at R 4 with phenyl or heteroaryl groups. For comparative reasons, some derivatives bearing phenylsubstituted residues at positions 4 and 6 were synthesized and tested ( Table 2, compounds 19bi−19bn) and are represented in Figure 5. In a clear contrast with their analogues in series I, which exhibited a non-selective profile, all derivatives of series II bearing two (identical or different) phenyl substituted residues at R 4 or R 6 proved to be inactive, irrespectively of the group contained in the exocyclic amino group.
The data on Table 2 and Figure 5 evidence that the substituent at the phenyl group has a clear impact on both affinity and selectivity. Thus, 2-substituted derivatives (19d− 19i) were either non-selective (2-F) or inactive (2-Cl and 2-OMe), while pyrimidine-5-carbonitriles bearing a 3-substituted phenyl group at R 4 (19j−19u) generally reproduced the nonselective profile observed in series I. Within the 3-phenyl substituted derivatives ( Figure 5), a 3-chlorophenyl residue led to attractive ligands (19l and 19n). Interestingly, while the 2methylamino derivative 19l is a potent (K i = 15.7 nM) and selective A 1 AR antagonist, its 2-phenylamino analogue (19n) exhibits moderate and selective A 2A AR affinity (K i = 46.3 nM). The introduction of substituents at position 4 of the R 4 phenyl group afforded several ligands with excellent (19v) to moderate (19w, 19z, 19ab, 19ad, and 19ae) A 1 AR affinity and selectivity ( Figure 5). However, as observed early in this series, only ligands bearing a methylamino group at position 2 of the heterocyclic core combined the desired affinity and selectivity profile (e. g., 19v, 19z, and 19ab).
Thirteen derivatives were selected to explore the effect of different disubstituted patterns on the phenyl group at R 4 ( Table 2, compounds 19af−19aq). The SAR trends discussed above for monosubstituted phenyl groups were generally reproduced within this subset ( Figure 5), with three pyrimidine-5-carbonitriles (e.g., 19af, 19aj, and 19ao) eliciting excellent A 1 AR affinity and selectivity ( Figure 5, blue and purple squares). Among these ligands, 19ao ( Figure 5, purple square) stands out as the most attractive A 1 AR antagonist identified during this study, combining high potency (K i = 6.11 nM) with excellent selectivity toward the rest of the ARs (Table 2). It should be noticed that, in addition to its 2methylamino group at 2, 19ao contains a piperonyl group at R 4 , a relatively frequent motif within A 1 AR antagonists. 18−21 Further introduction of pentagonal or hexagonal heterocyclic moieties at R 4 enabled the identification of three potent and selective A 1 AR ligands (19az, 19bb, and 19bd) that combine 3-thienyl, 4-pyridyl, or 3-pyridyl groups at R 4 with an exocyclic methylamino group in R 2 (K i = 42.8, 56.9, and 19.3 nM, respectively). Interestingly, pyrimidine derivatives bearing 3thienyl or 3-furyl substituents at R 4 and an N-phenylamino at R 2 exhibit moderate affinity (K i = 368 and 102 nM, respectively) and complete selectivity toward the A 2A AR. This data is coherent with the potent and selective A 2A AR profile observed for 19n (Table 2 and Figure 5), which contains a 3-chlorophenyl group at R 4 . Finally, the introduction of a cyclohexyl group at R 4 or of two substituted phenyl rings at both R 4 and R 6 ( Table 1, compounds 19bg− 19bn) afforded inactive compounds, irrespectively of the substitution pattern at any of the three points of diversity explored (i.e., R 2 , R 4 , and R 6 ). Similarly, pyrimidine-5carbonitriles bearing an N,N-dimethylamino group at position 2 showed to be inactive (Table 3).
Molecular Modeling. Taking advantage of the AR experimental crystal structures, we carried on a structurebased analysis of the binding mode of these compound series, addressed to further interpret the SAR observations discussed above. The study consisted of a first phase, where all compounds with measured A 1 AR affinity were docked on both this receptor and the A 2A AR, leading to two alternative binding models. Each of these binding mode proposals was the bases of extensive free energy perturbation (FEP) simulations on the A 1 AR, which univocally selected one binding mode and allowed a quantitative interpretation of the observed SAR, setting the grounds to further structure-based design optimizations.
The two alternative binding modes arose as a consequence of the asymmetric substitution pattern of these compounds. In both orientations, the central heterocycle and exocyclic amino group maintain the two key hydrogen bonds with the side chain of Asn 6.55 , totally conserved within the ARs ( Figure 6A). The two binding modes essentially differ on the orientation of the bulkiest substituent (at R 4 or R 6 ), which is either located at the extracellular loop region ( Figure 6A, conformation A, orange) or within the deep TM cavity of the receptor ( Figure  6B, conformation B, magenta). To identify the most probable binding mode, each binding mode was the starting point of a series of FEP calculations performed on a selection of compounds from series II. The criteria of selection were to cover a wide span of experimental affinities on the A 1 AR and sufficient structural diversity while retaining the most interesting scaffolds from the medicinal chemistry perspective, resulting in an initial subset of 21 A 1 AR antagonists where R 2 = Me. From these, we further retained those compounds where a change in R 2 would lead to a substantial change in their experimental selectivity profile, leading to a final selection of 18 compounds (19a, 19d, 19j, 19l, 19s, 19v, 19z, 19ab, 19ad, 19af, 19aj, 19am, 19ao, 19at, 19ax, 19az, 19bb, and 19bd). The dataset was studied on each binding pose through 28 FEP pair comparisons, see Figure S1. Each FEP cycle was performed with the QligFEP protocol, 57 leading to estimated relative affinities between each compound pair (ΔΔG bind ). The absolute binding affinity (ΔG bind , kcal/mol) was then calculated with a cycle closure correction approach following the idea presented by Wang et al. 58 (see Experimental Section). The results clearly favor conformation A ( Figure  6B, MUE = 0.87 ± 0.17 kcal/mol, RMSE = 1.13 ± 0.18 kcal/ mol, and SEM = 0.34 ± 0.03 kcal/mol) as it shows better predictivity and convergence than the alternative conformation B (MUE = 1.68 ± 0.39, RMSE = 2.26 ± 0.59, and SEM 1.12 ± 0.1, Figure S2). Consequently, conformation A was retained for further analysis. The structural binding model "conformation A" shows a pattern of ligand−receptor interactions that is compatible with the binding mode of monocyclic compounds predicted in the A 3 and A 2B ARs as a part of our ligand design programs on these receptors. 47,59,60 A qualitative structural analysis of all docked compounds with measured affinity, on both A 1 AR and A 2A AR, allowed us to rationalize the experimentally observed differences in selectivity. The series of free-amine compounds (series I, A 1 /A 2A non-selective profile) invariably shows a conserved interaction pattern with both A 1 AR and A 2A AR, consisting of the double-hydrogen bond with Asn 6.55 (Asn254/ 253 in A 1 /A 2A AR, respectively) and a π-stacking between the aromatic core and the conserved phenylalanine in EL2 (Phe171/168 EL2 ).
The substituents at R 2 explored within series II, however, show different behaviors at A 1 AR and A 2A AR (Figure 7). Generally speaking, N-alkylation (series II) causes a decrease in affinity compared with the free-amine compounds (series I) in both the AR subtypes. It is indeed the relative loss in affinity for one or another receptor that drives that gain in A 1 AR selectivity, with N-methyl and N-ethyl derivatives showing a much smaller loss of affinity in A 1 AR than A 2A AR, a pattern that is somehow inverted in N-phenyl substituted compounds. The binding model resulting from our computational study ( Figure 7) offers a structural interpretation of these tendencies. N-Methyl ( Figure 7A) and N-ethyl ( Figure 7B) compounds can maintain the dual H-bond in the A 1 AR with Asn254 6.55 observed on the free-amine compounds in series I, though showing some difficulty in accommodating the new substituents within the pocket defined by Thr259 6.58 and Met177 5.35 . A bulky phenyl group at R 2 , however, has a greater impact in obstructing the double-hydrogen bond formation ( Figure 7C), explaining a greater loss in affinity than the methyl and ethyl substituted compounds. The difference between the two receptors in responding to these substitutions resides on the two possible conformations of the subpocket accommodating the R 2 substituent in the A 2A AR: open ( Figure 7D−F, gray) and closed ( Figure 7D, red), as defined by the absence or presence, respectively, of the salt bridge between His264 EL3 and Glu169 EL2 connecting EL3 and EL2. N-Methyl and N-ethyl bearing compounds can be hardly accommodated in the closed conformation of the A 2A AR (Figure 7D), neither they can stabilize the A 2A AR open conformation ( Figure 7E). N-Phenyl compounds, on the other side, clearly stabilize this A 2A AR open conformation by hydrophobic interactions (Figure 7F), providing a rationale for their increased selectivity for this receptor. A further look into the complex of A 2A AR with the congeneric series formed by compounds 19l (R 2 = Me), 19l (R 2 = Et), and 19n (R 2 = Ph) illustrates this idea ( Figure S3). To further confirm this hypothesis, we conducted unbiased MD simulations of the A 2A AR in complex with the methyl (19l) and phenyl (19n) derivatives and monitored the distance between His264 EL3 and Glu169 EL2 . The results ( Figure S4) show how the latter is incompatible with a closed conformation without really stabilizing the open alternative, while the N-methyl derivative promotes a closing of the loops.

■ CONCLUSIONS
In summary, we have disclosed a large collection of 2-amino-4,6-disubstituted-pyrimidine derivatives as potent, structurally simple, and highly selective A 1 AR ligands. The pharmacological characterization of the most attractive A 1 AR ligands identified during this study confirmed its antagonistic behavior (through cAMP assays). Further studies to complete the bioavailability profile and in vivo BBB permeation of lead compounds are currently in progress. The reliable and efficient three-component reaction facilitated the rapid assembly of a large library, thus enabling to comprehensively examinate the most prominent features of the SAR and SSR in this series. This building-block scheme is an asset in our lab to further grow the chemical library, guided by the rationale derived from this work. The SSR studies highlighted the influence of the aromatic residues at R 4 and R 6 of the pyrimidine core to the selectivity profile but most importantly the prominent role exerted by the methylation of the 2-amino group as the main contributor to the unprecedented A 1 AR selectivity profile observed in these series. The SAR trends herein disclosed were complemented and interpreted with a comprehensive computational modeling analysis based on rigorous FEP simulations, starting from the receptor-driven docking model that initially guided the design of these series. Particularly revealing was the orientation of the new asymmetrically substituted scaffold, for which the binding mode on the A 1 AR was herein supported by first-principle binding affinity calculations, which can be therefore used in the next stage of ligand optimization.

■ EXPERIMENTAL SECTION
Chemistry. Unless otherwise indicated, all starting materials, reagents, and solvents were purchased and used without further purification. After extraction from aqueous phases, the organic solvents were dried over anhydrous sodium sulfate. The reactions were monitored by thin-layer chromatography (TLC) on 2.5 mm Merck silica gel GF 254 strips, and the purified compounds each showed a single spot; unless stated otherwise, UV light and/or iodine vapor were used to detect compounds. The Biginelli reactions were performed in coated Kimble vials on a PLS (6 × 4) organic synthesizer with orbital stirring. The purity and identity of all tested compounds were established by a combination of high-performance liquid chromatography (HPLC), elemental analysis, mass spectrometry, and NMR spectroscopy as described below. Purification of isolated products was carried out by column chromatography (Kieselgel 0.040−0.063 mm, E. Merck) or medium pressure liquid chromatography (MPLC) on a CombiFlash Companion (Teledyne ISCO) with RediSep pre-packed normal-phase silica gel (35−60 μm) columns followed by recrystallization. Melting points were determined on a Gallenkamp melting point apparatus and were uncorrected. The NMR spectra were recorded on Bruker AM300 and XM500 spectrometers. Chemical shifts were given as δ values against tetramethylsilane as the internal standard, and J values were given in Hz. Mass spectra were obtained on a Varian MAT-711 instrument. High-resolution mass spectra were obtained on an Autospec Micromass spectrometer. Analytical HPLC was performed on an Agilent 1100 system using an Agilent Zorbax SB-Phenyl, 2.1 mm × 150 mm, 5 μm column with gradient elution using the mobile phases (A) H 2 O containing 0.1% CF 3 COOH and (B) MeCN and a flow rate of 1 mL/min. All reported compounds are >95% pure by HPLC analysis. HPLC traces obtained for representative lead compounds herein identified are provided in the Supporting Information. The structural and spectroscopic data obtained for all compounds described are provided in the Supporting Information.
General Procedure for the Three-Component Synthesis of 2-Amino-4,6-diarylpyrimidin-5-carbonitriles (18−20). A mixture of α-cyanoketone 21a−j (1 mmol), aldehyde 22a−j (1 mmol), the guanidine salt 23a−d (1.2 mmol), and Na 2 CO 3 (3 mmol) in 3 mL of THF in coated Kimble vials was stirred with orbital stirring at 80°C for 12 h. After completion of the reaction (controlled by TLC), the solvent was evaporated to dryness and the resulting residue was resuspended in water and extracted with ethyl acetate. The organic phase was dried with Na 2 SO 4 and evaporated to dryness, when the oily residue was resuspended with methanol the product generally precipitates, was filtered, and purified by recrystallization or column chromatography (silica gel) generally using hexane/AcOEt mixtures as the eluent.
Pharmacological Characterization. Radioligand binding competition assays were performed in vitro using human ARs expressed in Solubility Determinations. The stock solutions (10 −2 M) of the selected ligands were diluted to decreased molarity, from 300 to 0.1 μM, in a 384-well transparent plate (Greiner 781801) with 1% DMSO:99% PBS buffer. These were incubated at 37°C and read after 2 h in a NEPHELOstar Plus (BMG LABTECH). The results were adjusted to a segmented regression to obtain the maximum concentration in which compounds are soluble.
Human Microsomal Stability. The human microsomes employed were purchased from Tebu-Xenotech. The compound was incubated with microsomes at 37°C in a 50 mM phosphate buffer (pH = 7.4) containing 30 mM MgCl 2 , 10 mM NADP, 100 mM glucose-6-phosphate, and 40 U/mL glucose-6-phosphate dehydrogenase. Samples (75 μL) were taken from each well at 0, 10, 20, 40, and 60 min and transferred to a plate containing 75 μL of acetonitrile (4°C ), and 30 μL of 0.5% formic acid in water was added for improving the chromatographic conditions. The plate was centrifuged (4000g, 60 min), and supernatants were taken and analyzed in a UPLC-MS/MS (Xevo-TQD, Waters) by employing a BEH C18 column and an isocratic gradient of 0.1% formic acid in water:0.1% formic acid acetonitrile (60:40). The metabolic stability of the compounds was calculated from the logarithm of the remaining compounds at each of the time points studied.
Functional Experiments. cAMP assays were performed at human A 1 ARs using a cAMP enzyme immunoassay kit (Amersham Biosciences). CHO cells were seeded (10,000 cells per well) in 96well culture plates and incubated at 37°C in an atmosphere with 5% CO 2 in Nutrient Mixture F-12 Ham (Ham's F-12) containing 10% fetal bovine serum dialyzed (FBS), penicillin/streptomycin (1%), amphotericin B (2.5 μg/mL), and Geneticin (400 μg/mL). Cells were washed 2× with 200 μL of the assay medium (Ham's F-12 and 25 mM HEPES pH = 7.4) and pre-incubated with the assay medium containing 20 μM rolipram and test compounds at 37°C for 15 min. Stimulation was carried out by the addition of 0.1 μM NECA incubated for 10 min and 3 μM forskolin incubated for 5 min at 37°C (total incubation time, 30 min). Reaction was stopped with lysis buffer supplied in the kit, and the enzyme immunoassay was carried out for detection of intracellular cAMP at 450 nm in an Ultra Evolution detector (Tecan). For data analysis, IC 50 values were obtained by fitting the data with non-linear regression using Prism 5.0 software (GraphPad, San Diego, CA). For those compounds that showed either little affinity or poor solubility, a percentage inhibition of specific binding was reported. Results are the mean of three experiments (n = 3) each performed in duplicate.
Calcein-AM Experiments. Calcein cell accumulation was evaluated by following a previously described method. 53−55 The MDCK-MDR1 cell line (30,000 cells per well) was seeded into a 96well black culture plate with 100 μL of the medium and allowed to become confluent overnight. Test compounds (100 μL) were solubilized in the culture medium and added to monolayers, with final concentrations ranging from 0.1 to 100 μM. The 96-well plate was incubated at 37°C for 30 min. Calcein-AM was added in 100 μL of phosphate buffered saline (PBS) to yield a final concentration of 2.5 μM, and the plate was incubated for 30 min. Each well was washed three times with ice-cold PBS. Saline buffer was added to each well, and the plate was read with Victor 3 (PerkinElmer) at excitation and emission wavelengths of 485 and 535 nm, respectively. In these experimental conditions, Calcein cell accumulation in the absence and in the presence of tested compounds was evaluated and the fluorescence basal level was estimated with untreated cells. In treated wells, the increase in fluorescence with respect to the basal level was measured. EC 50 values were determined by fitting the fluorescence increase percentage versus log [dose].
Protein Preparation and Ligand Docking. Receptor structures were retrieved from the PDB with codes 5N2S (hA 1 AR), 4EIY (hA 2A AR-closed), and 3UZC (hA 2A AR-open) and prepared for ligand docking and MD simulations. The initial preparation steps were performed with the Schrodinger suite (protein preparation wizard) and included modeling of the missing loop segments, reverting the protein construct to the wt sequence, addition of protons, and assessment of Asn/Gln/His rotamers and protonation states (in all cases, Asp, Glu, Lys, and Arg residues were assigned in their default charged state). All His residues in both receptors were modeled as neutral with the proton on Nδ except for His 6.52 , protonated on Nε, and His264 in A 2A AR that is positively charged. Each receptor structure was then inserted in the membrane and equilibrated under periodic boundary conditions (PBC) using the PyMemDyn protocol described elsewhere. 61 Shortly, the receptor was embedded in a preequilibrated membrane consisting of POPC (1-palmitoyl-2-oleoyl phosphatidylcholine) lipids, with the TM bundle aligned to its vertical axis. An hexagonal prism-shaped box was then built and soaked with bulk water; thereafter, the system was energy-minimized with GROMACS 4.6. 62 using the OPLS-AA force field 63 for combination with the Berger parameters for lipids. 64 An energy minimization of the system (50,000 conjugate gradient steps, convergence criteria of 1000 kJ/mol) precedes a short (2.5 ns length) MD equilibration, where initial restraints imposed on protein heavy atoms are gradually released as described in detail in our original protocol. 61 The final receptor structure was energy-minimized with similar settings as above.
An automated docking exploration was performed with GlideXP, 65 applying default parameters, for ligands 18a, 19a, 19b, and 19c as model compounds of free-amine, methylamine, ethylamine, and phenylamine derivatives, respectively. These ligands were initially built in their 2D structures, and the SD file generated was the input for the ligand preparation wizard in Schrodinger, which generated the most probable protonation state and an energy-minimized 3D conformer with the OPLS3 force field. The search box was defined by the co-crystallized ligand in each case, resulting in very similar boxes since all ligands occupy the same orthosteric site. We used the results of this automated docking exploration to build the corresponding complexes with an expanded dataset of 60 ligands, consisting of the compounds from series II that have measurable K i affinity values for either A 1 AR or A 2A AR, plus the analogous of these on series I (Tables 1 and 2). Each of these compounds was directly built from the structurally closest ligand−receptor complex, from those generated by automated docking (i. e., 18a, 19a, 19b, and 19c), and energy minimization of the resulting complex followed (default parameters within the Schrodinger suite).
MD and FEP Calculations. Selected receptor−ligand complexes were grouped in a set of pair comparisons for free energy perturbation (FEP) calculations using the QligFEP protocol 57 and the MD software Q. 66,67 The so-called FEP pathway (see Figure S1) was designed based on maximal compound similarity, computed upon Morgan Fingerprint descriptors, with a series of corrections to ensure a cycle closure correction. This approach allows the estimation of absolute binding free energies (ΔG bind ) using the experimental value of one compound in the series as a reference, together with the associated statistical figures of merit: the mean unassigned error (MUE) and root mean squared error (RMSE), between calculated and experimental binding affinities, together with the convergence obtained along the calculations (expressed as standard error of the mean, SEM), in all cases in kcal/mol. Confidence intervals for the regression metrics were estimated using bootstrap sampling.
A 25 Å sphere centered on the center of geometry of the ligand is considered for MD simulations under spherical boundary conditions. Protein atoms in the boundary of the sphere (22−25 Å outer shell) had a positional restraint of 20 kcal/mol/Å 2 , while solvent atoms were subject to polarization and radial restrains using the surface constrained all-atom solvent (SCAAS) 66,68 model to mimic the properties of bulk water at the sphere surface. Atoms lying outside the simulation sphere were tightly constrained (200 kcal/mol/Å 2 force constant) and excluded from the calculation of non-bonded interactions. Long-range electrostatic interactions beyond a 10 Å cutoff were treated with the local reaction field method, 69 except for the atoms undergoing the FEP transformation where no cutoff was applied. Solvent bonds and angles were constrained using the SHAKE algorithm. 70 All titratable residues outside the sphere were neutralized, and histidine residues were assigned a hydrogen atom on the δ nitrogen. Residue parameters were translated from the OPLS-AA/M force field, 71 and the ligand parameters were generated using the ffld server as implemented in the Schrodinger suite and the lipid parameters as described above. The simulation sphere was warmed up from 0.1 to 298 K, during a first equilibration period of 0.61 ns, where an initial restraint of 25 kcal/mol/Å 2 imposed on all heavy atoms was slowly released for all complexes. Thereafter, the system was subject to 10 parallel replicates of unrestrained MD, where the following FEP protocol was applied for each ligand transformation: an initial 0.25 ns unbiased equilibration period, with different initial velocities for each replica, was followed by 101 FEP λwindows, consisting of 10 ps each, distributed using a sigmoidal sampling schedule. During the FEP transformation, the potentials of the two ligands involved were combined using a double topology scheme. 57 To fulfill a thermodynamic cycle and calculate relative binding free energies, analogous FEP transformations were run for the same ligand pair in a sphere of water, maintaining the same MD parameters (i.e., sphere size, simulation time, etc.). The relative binding free energy difference was then estimated by solving the thermodynamic cycle using the Bennett