The role of 5-arylalkylamino- and 5-piperazino- moieties on the 7-aminopyrazolo[4,3-d]pyrimidine core in affecting adenosine A1 and A2A receptor affinity and selectivity profiles

Abstract New 7-amino-2-phenylpyrazolo[4,3-d]pyrimidine derivatives, substituted at the 5-position with aryl(alkyl)amino- and 4-substituted-piperazin-1-yl- moieties, were synthesized with the aim of targeting human (h) adenosine A1 and/or A2A receptor subtypes. On the whole, the novel derivatives 1–24 shared scarce or no affinities for the off-target hA2B and hA3 ARs. The 5-(4-hydroxyphenethylamino)- derivative 12 showed both good affinity (Ki = 150 nM) and the best selectivity for the hA2A AR while the 5-benzylamino-substituted 5 displayed the best combined hA2A (Ki = 123 nM) and A1 AR affinity (Ki = 25 nM). The 5-phenethylamino moiety (compound 6) achieved nanomolar affinity (Ki = 11 nM) and good selectivity for the hA1 AR. The 5-(N4-substituted-piperazin-1-yl) derivatives 15–24 bind the hA1 AR subtype with affinities falling in the high nanomolar range. A structure-based molecular modeling study was conducted to rationalize the experimental binding data from a molecular point of view using both molecular docking studies and Interaction Energy Fingerprints (IEFs) analysis.


Introduction
Adenosine receptors (ARs) are classified as A 1 , A 2A , A 2B and A 3 subtypes 1,2 and typically inhibit (A 1 and A 3 ) or activate (A 2A and A 2B ) adenylyl cyclase. A 1 receptor is highly expressed in brain areas, such as the hippocampus and prefrontal cortex 3,4 , implicated in the control of emotions and cognition functions. Therefore, A 1 AR antagonists are investigated as therapeutic agents for mental dysfunctions, such as dementia and anxiety [3][4][5] . The A 2A AR subtype is present in the brain with the highest concentration in the striatum, nucleus accumbens, hippocampus and cortex, and its blockade has proven to be effective in neurodegenerative pathologies such as Parkinson's disease (PD) [6][7][8] . The A 2A AR antagonist istradefylline has been recently approved for marketing in Japan for the treatment of PD patients 9 . In preclinical studies, dual A 1 /A 2A antagonists have also turned out to be useful for PD therapy because they reduce both motor (A 2A ) and cognitive (A 1 ) impairment associated with this pathology 5,10-12 . Recent studies have highlighted new therapeutic applications of A 2A AR antagonists 12 . If topically administered, they diminish scar size and promote restoration of skin integrity 13 . A 2A AR antagonists have also demonstrated efficacy in enhancing immunologic response, especially by markedly improving anti-tumor immunity in mouse models, thus promoting tumor regression. A 2A AR antagonists have been shown to improve the effect of tumor vaccines during T-cell activation, and may work in concert with other immune checkpoint inhibitors in cancer immunotherapy 12,14 .
In our laboratory, much research has been addressed to the study of AR antagonists belonging to different classes [15][16][17][18][19][20][21][22][23][24][25][26] , including the 2-arylpyrazolo [4,3-d]pyrimidine derivatives 20,22,24,26 which display a broad range of affinity for the various AR subtypes, depending on the nature of the substituents at the 5-and 7-positions of the bicyclic scaffold. One recent study aimed at targeting the A 1 and A 2A ARs highlighted that the presence of a free 7amino group, combined with a benzyl or, even better, a 3-phenylpropyl chain at the 5-position (Figure 1, compounds A and C) shifted affinity toward these two AR subtypes 24 .
Hence, to further explore the structural requirements for addressing affinity toward the A 1 and/or A 2A ARs, various aryl(alkyl)amino-and 4-substituted-piperazin-1-yl-moieties were appended at the 5-position of the scaffold (compounds 1-24, Figure 2). These substituents were selected since they are a common feature of potent A 1 and/or A 2A AR antagonists structurally correlated to our pyrazolopyrimidine derivatives 12,27,28 (such as the triazolotriazines ZM-241385 and D, Figure 1). The pyrazolopyrimidines 1-24 were tested in binding assays to evaluate their affinity at cloned hA 1 , hA 2A and hA 3 ARs, stably expressed in CHO cells. Compounds were also tested at the hA 2B receptor by measuring their inhibitory effects on NECA-stimulated cAMP levels in CHO cells.
A structure-based molecular modeling study was performed on the new derivatives to rationalize the experimental binding data from a molecular point of view, using molecular docking studies in tandem with Interaction Energy Fingerprints (IEFs) analysis.
Allowing the 1-phenylpyrazolo [4,3-d]pyrimidine-5,7-dione 54 20 to react with phosphorus oxychloride and N,N-dimethylaniline under microwave irradiation, the 5,7-dichloro-derivative 55 was prepared, which was reacted with 33% aqueous ammonia solution under microwave irradiation at 100 C to give the 7-amino-5-chloro-pyrazolopyrimidine 56 as the only regioisomer. The 7-amino structure of 56 was expected on the basis of the well-known different mobility of the two chlorine atoms in the pyrimidine ring, also condensed with diverse heterocyclic systems [32][33][34] . To confirm the structure, derivative 56 was treated with benzylamine in tert-butanol, in the presence of diisopropylethylamine, and the 7-amino-5-benzylaminopyrazole derivative 5, already synthesized through the unambiguous synthesis as depicted in Scheme 1, was obtained. This reaction was carried out under prolonged microwave irradiation (about 1 h at 200 C) but conversion of derivative 56 into 5 occurred with unsatisfactory yields. The 1 H NMR spectrum of the crude reaction (data not shown) displayed the presence of both the 5-benzylamino derivative 5 and the starting material 56 (ratio about 3.5:1), besides degradation compounds, thus indicating the poor reactivity of the C5 atom toward the primary benzyl ammine group. Instead, microwave-assisted reaction of the 5-chloro derivative 56 with the N-substituted piperazines 57-63, in N-methylpyrrolidone and in the presence of diisopropylethylamine, proceeded to completion, thus giving the desired pyrazolopyrimidine derivatives 15-20 with good yields (48-85%). The piperazine derivatives 57, 58, 62 and 63 were commercially available, while derivatives 59 and 61 were prepared as previously described 35,36 . The piperazine derivative 60 was synthesized starting from the reductive alkylation of N-Boc-piperazine 63 with 2,4,6trifluorobenzaldeyde and triacetoxy sodium borohydride. The obtained tert-butyl 4-(2,4,6-trifluorobenzyl)piperazine-1-carboxylate was hydrolyzed with trifluoroacetic acid to give the 1-(2,4,6-trifluorobenzyl)piperazine 60, isolated as trifluoroacetate salt.
Reduction of the 2-furoyl carbonyl group of compound 20 with LiAlH 4 in anhydrous THF provided derivative 22. Finally, the Scheme 3. Reagents and conditions: (a) CF 3 COOH, CH 2 Cl 2 , reflux; (b) RCOCl, NEt 3 , anhydrous THF, room temperature. pyrazolopyrimidines 23-24, bearing an acyl moiety on the piperazine nitrogen, were synthesized as depicted in Scheme 3. Treatment of the N-Boc derivative 21 with trifluoroacetic acid furnished compound 64 which was reacted with suitable acyl chlorides, in the presence of triethylamine in anhydrous tetrahydrofuran, to provide the desired 23-24.

Structure-affinity relationship studies
The results of binding experiments and cAMP assays carried out on the new 5-substituted-pyrazolopyrimidines 1-14 and 15-24 are displayed, respectively, in Tables 1 and 2. Table 1 also includes the affinity data of the pyrazolopyrimidines A-C and of ZM-241385 reported as references.
As expected, the new derivatives 1-24 shared scarce or no affinities for the off-target hA 2B and hA 3 ARs, except the 5-anilinoand 5-benzylamino derivatives 1-3 and 5, respectively which displayed nanomolar affinity for the hA 3 subtype (K i ¼ 13-61 nM). In particular, compounds 2 and 3 are worth noting, being also highly hA 3 selective.
Since the purpose of the work was to target hA 1 and hA 2A ARs, SAR discussion was focused on hA 1 and hA 2A binding data. In this respect, results of some interest have been obtained from the 5arylalkylamino-pyrazolopyrimidines 1-14. In fact, compound 12 showed both good affinity and the best selectivity for the hA 2A AR, while compounds 1, 5, 13 and 14 were able to bind both the hA 1 and hA 2A ARs. Moreover, a derivative having nanomolar affinity and high selectivity for the hA 1 AR subtype was identified (compound 6).
The new 5-phenyl(alkyl)amino derivatives 1, 5 and 6 were designed as analogs of our previously reported antagonists 5-phenyl(alkyl) derivatives A, B and C 24 whose methylene linker at the 5-position of the bicyclic core was replaced with an NH. This modification, suggested by the structure of potent A 2A antagonists bearing arylalkylamino moieties as key substituents 12 , was thought to change the flexibility of the 5-lateral chain and, hopefully, to increase the affinity for the targeted ARs. Actually, the NH linker enhanced the hA 1 AR affinity (compare 1 and 5 to A and B, respectively) or maintained it in the nanomolar range (compare 6 to C). Instead, the hA 2A AR binding was ameliorated in one case, i.e. the 5-benzylamino derivative 5 which was more active than the corresponding phenylalkyl-derivative B.
Analyzing the hA 1 and hA 2A AR binding data of 1-6 in detail, it can be observed that 5-phenylamino derivative 1 binds to the hA 2A and hA 1 AR subtypes with scarce (K i ¼ 412 nM) and good affinity (K i ¼ 67 nM), respectively. Introduction of either a 4methoxy group or 2,4-dichloro substituents on the 5-aniline moiety of 1 (compounds 2 and 3) dropped affinity for hA 1 and hA 2A ARs. Instead, the presence of a 4-hydroxy residue (compound 4) reduced the hA 2A affinity while conserving some ability to bind the hA 1 receptor (K i ¼ 481 nM). Homologation of the 5-phenylamino moiety (derivative 1) to the 5-benzylamino group (derivative 5) produced some improvement in the binding activity at both hA 1 (K i ¼ 25 nM) and hA 2A ARs (K i ¼ 123 nM). Quite unexpectedly, homologation of the alkyl chain of compound 5, to obtain the 5phenethylamino-and the 5-phenylpropylamino derivatives 6 and 7, caused a drastic reduction of the hA 2A AR affinity and, in the former, it increased the hA 1 one, thus affording a selective hA 1 receptor ligand (K i ¼ 11.5 nM).
Replacement of the 2-phenyl group of derivatives 1 and 5 with a methyl residue, to give compounds 8 and 9, was performed to verify whether a reduction in the volume of the molecule might permit a better accommodation inside the recognition site of the targeted hARs. This modification, instead, annulled the capability to bind the target hARs. The same detrimental effect was obtained when the 2-phenyl ring of 5 was replaced with the more flexible benzyl moiety (derivative 10).
Insertion of the para hydroxy substituent on the 5-phenethylamino moiety of derivative 6, to give compound 12, was based on the structure of the well-known potent and selective hA 2A AR antagonist ZM-241385 5,12 ( Figure 1). Accordingly, we also thought it would be interesting to evaluate the 3,4-dihydroxy substitution (compound 14), as well as the 4-methoxy-and the 3,4-dimethoxysubstituents (derivatives 11 and 13). As expected, the presence of the 4-hydroxy group was able to shift the affinity toward the hA 2A AR. In fact, the 4-hydroxy-substituted derivative 12 showed good hA 2A affinity (K i ¼150 nM) and the best selectivity among all the ligands reported here. In contrast, reversed selectivity was demonstrated by the 4-methoxy derivative 11, which displayed good affinity for the hA 1 AR but not for the hA 2A subtype. Instead, the 3,4-dimethoxy substituted derivative 13 bound both hA 1 and hA 2A receptors and also the 3,4-dihydroxy derivative 14 showed quite good affinity for both the receptors, but especially for the hA 1 one.
Finally, to further explore the SARs in this class of AR ligands, various N-substituted piperazine moieties were appended at the 5-position (derivatives 15-24, Table 2), in accordance with the structure of known potent and selective hA 2A AR antagonists 27,28 .
In contrast to our expectations, none of the 5-(N 4 -R-piperazin-1-yl) derivatives 15-24 were able to bind effectively the A 2A AR while they possessed affinity for the hA 1 AR subtype, falling in the high nanomolar range. The most active compounds proved to be 22 (K i ¼ 92 nM) and 16 (K i ¼162 nM) which bear, respectively, the (2-furyl)-methyl and 2-benzyl pendant on the N4-piperazine moiety. Introduction of halogen atoms on the benzyl moiety of 16 left almost unchanged the hA 1 AR affinity (compounds 18 and 19) while elongation of the benzyl chain decreased it (compound 17). Also the other substituents evaluated on the piperazine ring, i.e. acyl moieties (derivatives 20, 23, 24) and the tert-butoxycarbonyl group (derivative 21) did not ameliorate the hA 1 AR affinities.

Molecular modeling studies
A structure-based molecular modeling study was conducted to rationalize the experimental binding data from a molecular point of view. Minor attention was devoted to the hA 2B AR subtype, since no significant binding affinity has been estimated for any of the compounds under investigation. Docking was performed on hA 1 , hA 2A and hA 3 AR subtypes, and the resulting poses were evaluated according to the van der Waals and electrostatic interactions, as previously reported 37,38 and described in detail in the "Experimental" section. Positive electrostatic and van der Waals values were used as filters to reject unfavorable docking poses. One pose for each ligand was selected on the basis of the Interaction Energy Fingerprints (IEFs) and by visual inspection.
An overview of the most favorable poses of all compounds on hA 1 , hA 2A and hA 3 ARs is reported in video SM1-SM2-SM3, included in Supplementary Material. The heat map depicted in the background reports the electrostatic and hydrophobic contributions of the residues mainly involved in binding ("ele" and "hyd" labels identify the major contribution type of the residue) by a colorimetric scale going from blue to green for negative to positive values. These crucial residues are mainly positioned on the superior half of TM6 and TM7 and EL2, and the overall binding modes of the compounds under examination are very consistent among them. Here, we describe in detail the poses of compound 1 as an example, because of its high binding affinity for all three AR subtypes taken into consideration (K i ¼ 67 nM for hA 1  With regard to the hA 1 AR, Glu172 (EL2) and Asn254 (6.55), represented by blue bars on electrostatic IEFs ( Figure 3, panel A on the left), emerge as important residues for electrostatic contribution, together with a slight contribution of Trp247 (6.48) and His251 (6.52). Asn254 (6.55) and Glu172 (EL2) are engaged in a three hydrogen bond pattern with N1 of pyrazole and with the exocyclic amine group at position 7 of compound 1, as shown in Figure 4, panel A. The aromatic pyrazolopyrimidine scaffold is involved in a p-p stacking interaction with Phe171 (EL2), which is one of the residues appearing to have the strongest hydrophobic interaction on the hydrophobic IEFs (green bars in Figure 3, panel A on the right). Val87 (3.32), Leu88 (3.33), Trp247 (6.48), Leu250 (6.51), Tyr271 (7.36) and Ile274 (7.39) are also involved in significant hydrophobic contacts, with Val87 (3.32), Leu88 (3.33), Trp247 (6.48) defining the bottom of the binding pocket.
The binding of compound 1 to the hA 3 subtype mainly engages Trp243 (6.48), Ser247 (6.52) and Asn250 (6.55) for electrostatic interactions, and Leu91 (3.33), Phe168 (EL2), Val169 (EL2), Trp243 (6.48), Leu246 (6.51), Leu264 (7.35), Tyr265 (7.36), Ile268 (7.39) for hydrophobic interactions, as can be seen in Figure 3, panel C. In this case only Asn250 can be involved in the hydrogen bond network (Figure 4, panel C), since in the A 3 AR the position equivalent to Glu172 of the hA 1 and Glu169 of the A 2A AR is occupied by Val169, which cannot establish a hydrogen bond with the amino group at position 5 of compound 1.
Most of the poses resemble the conformation that ZM-241385 assumes in the binding site of the hA 2A AR crystal structure and of hA 1 and hA 3 AR models. The benzene ring at position 2 occupies the position of the furan ring of ZM-241385, the 7-amino-pyrazolopyrimidine scaffold is well superimposed on the reference 7amino-triazolotriazine and the arylalkylamino group at position 5 points in the same direction as the para-hydroxyphenyl-ethylamino fragment. The similarity of the binding modes confirms the expectation provided by the IEFs comparison between 1 and ZM-241385 ( Figure 3, panels A, B and C). To quantitatively compare the calculated IEFs profiles, two novel analyses have been proposed called RMSD and RMSD trend analysis (see the Experimental Section for more details). In the case of derivative 1 both RMSD and RMSD trend between electrostatic and hydrophobic IEFs on each hAR subtype are quite low. However, the electrostatic RMSD (2.18 kcal/mol) and the electrostatic RMSD trend (0.47 kcal/mol) for the hA 1 subtype are higher than the values observed for hA 2A and hA 3 ARs. This does not seem to fit with the low K i (67 nM) for the hA 1 receptor; however, it appears that the major unfavorable contribution is provided by Glu170, which may probably be corrected by a slight rotation of the phenyl group of the compound.
Subsequently, we compared the binding behavior of compound 1 to that of its analog derivative A (K i ¼ 150 nM for hA 1 , K i ¼ 110 nM for hA 2A and I% ¼ 39 at 1 mM for hA 3 ), having a methylene instead of the NH linker at the 5-position. The IEFs comparison did not allow a complete rationalization of the different selectivity profiles of compounds 1 and A ( Figure SM1). Electrostatic RMSD and RMSD trend values on the hA 3 receptor (1.10 and 0.18 kcal/mol, respectively) are higher than those of compound 1 (0.60 and 0.06 kcal/mol, respectively), in accordance with the lower potency of derivative A (I ¼ 39% at 1 mM) compared with 1 (K i ¼ 13 nM). On the other hand, we have to honestly observe that also compound A presents higher RMSD and RMSD trend values (1.49 and 0.18 kcal/mol, respectively) on the hA 2A receptor as compared with compound 1 (0.96 and 0.08 kcal/mol, respectively), but in this case the affinity of the former (110 nM) is higher than that of the latter (412 nM). The result of the IEFs comparison is confirmed by the similarity of the binding modes of derivatives 1 and A at all receptor binding sites, as reported in Figure 4 (panels A, B and C). In this case docking is not sufficient to rationalize the difference in binding affinities. In fact, the mere examination of the final state of the binding process may not be sufficient to explain differences in the activity or selectivity profiles. The presence of water molecules and the entropic effect are only two among the pool of binding contributions that we are not taking into consideration during our docking simulations.
Similar considerations can be made observing the results of IEFs comparison for all the dataset compounds on the different AR subtypes ( Figure 5). We would have expected to find blue and red rectangles associated with good and bad binders, respectively, but this prevision was not satisfied: a major similarity of the IEFs between the target and the reference compounds are not always related to good binding affinity of the ligand. However, an interesting example is provided by compounds 8, 9 and 10, which have no affinity for any of the receptors. Red rectangles cross horizontally almost the whole hydrophobic RMSD and RMSD trend table, meaning that there is a considerable loss in the binding hydrophobic contribution in comparison with the reference. As a control experiment, ZM-241385 has been docked into the three AR subtypes, the IEFs have been computed for the selected poses and compared with that of the reference pose of ZM241385: as expected, the electrostatic and hydrophobic RMSD and RMSD trend values are close to zero ( Figure 5).
The 5-(N 4 -R-piperazin-1-yl) compounds 15-24 are hA 1 AR selective. These derivatives find a steric hindrance in the hA 3 binding site and the van der Waals values of the selected poses are positive (as indicated by exclamation points in Figure 5). However, from the IEFs comparison analysis (Figure 5), we would have predicted a hA 2A versus hA 1 selectivity (blue versus red rectangles). In fact, while at the hA 2A binding site the predicted poses of these compounds behave like ZM-241385, at the hA 1 binding site they deviate a little from the reference position, losing some of the canonical interactions (Video SM1-SM2). Interestingly, this diversion results in a gain for compounds 16, 17, 18, 19 and 22: the protonated amine at position 4 of the piperazine moiety is involved in an ionic interaction with Glu170 (EL2), which is confirmed by a highly negative electrostatic contribution reported on the heat map in the background of Video SM1. The absence of a negatively charged residue at a position equivalent to Glu170 on hA 2A (Leu167) and hA 3 (Gln167) receptors may be associated with the hA 1 selectivity of these compounds.

Conclusion
The herein reported structural investigation was carried out to identify new antagonists targeting the hA 2A AR or both the hA 1 / hA 2A ARs. Hence, various arylalkylamino-and 4-substituted-piperazin-1-yl-moieties were appended at the 5-position of the pyrazolo[4,3-d]pyrimidine scaffold. The 4-hydroxyphenylethylamino group was the most profitable, since the ZM-241385-based compound 12 showed both good hA 2A affinity (K i ¼ 150 nM) and the highest selectivity among all the ligands reported here. The 5-benzylamino moiety (compound 5) achieved the best combined hA 2A (K i ¼ 123 nM) and hA 1 affinity (K i ¼ 25 nM) while the 5-phenethylamino pendant (compound 6) afforded nanomolar affinity (K i ¼ 11 nM) and good selectivity for the hA 1 AR. The 5-(N 4 -substituted-piperazin-1-yl) derivatives 15-24 were inactive at the hA 2A AR while the hA 1 affinities spanned the high nanomolar range. These outcomes provide new insights about the structural requirements of our pyrazolopyrimidine series for hA 2A -and hA 1 -receptor ligand interaction. Nevertheless, the obtained results do not prompt us to synthesize further derivatives of this series featured by 5-arylalkyamino-and 5-piperazino-moieties.
A structure-based molecular modeling study was conducted to rationalize the experimental binding data from a molecular point of view using molecular docking studies in tandem with Interaction Energy Fingerprints (IEFs) analysis. Moreover, to quantitatively compare IEFs profiles and, consequently, to address the similarity of the binding modes of different compounds in different receptor subtypes, two novel analyses have been proposed, called RMSD and RMSD trend analyses. Even if, we are conscious that the simple inspection of the final state of the binding process may not be sufficient to explain differences in the activity or selectivity profiles, these novel tools can facilitate the mode of representation and interpretation of the docking data obtained by analyzing simultaneously several compounds against different receptor subtypes.

Chemistry
The microwave-assisted syntheses were performed using an Initiator EXP Microwave Biotage instrument (frequency of irradiation: 2.45 GHz). Analytical silica gel plates (Merck F254), preparative silica gel plates (Merck F254, 2 mm) and silica gel 60 (Merck, 70-230 mesh) were used for analytical and preparative TLC, and for column chromatography, respectively. All melting points were determined on a Gallenkamp melting point apparatus and are uncorrected. Elemental analyses were performed with a Flash E1112 Thermofinnigan elemental analyzer for C, H, N and the results were within ±0.4% of the theoretical values. All final compounds revealed a purity not less than 95%. The IR spectra were recorded with a Perkin-Elmer Spectrum RX I spectrometer in Nujol mulls and are expressed in cm À1 . The 1 H NMR spectra were obtained with a Bruker Avance 400 MHz instrument. The chemical shifts are reported in d (ppm) and are relative to the central peak of the solvent which was CDCl 3 or DMSO-d 6 . The assignment of exchangeable protons (OH, and NH) was confirmed by addition of D 2 O. The following abbreviations are used: s ¼ singlet, d ¼ doublet, t ¼ triplet, m ¼ multiplet, br ¼ broad and ar ¼ aromatic protons.

General procedure for the synthesis of 3-substituted-1-(3-cyano-1-R 2-1 H-pyrazol-4-yl)thioureas 32-42
The commercially available phenyl-, 4-methoxyphenyl-, 2,4-dichlorophenyl-and benzyl-isothiocyanates or the suitably synthesized phenethyl-29 , 4-methoxyphenethyl-31 , 3,4-dimethoxyphenethyl-30 , phenylpropyl-isothiocyanates 29  The obtained dark slurry was treated with water (20 mL) and, in the case of compounds 33-35 and 39, a solid precipitated which was collected by filtration. For derivatives 32, 36-38, 40-42, the aqueous mixture was extracted with EtOAc (30 mL Â3). The combined organic extracts were anhydrified (Na 2 SO 4 ) and the solvent evaporated at reduced pressure. The obtained solid was treated with Et 2 O (5-10 mL) and isolated by filtration. Crude compound 42 was purified by column chromatography (eluent:cyclohexane/ EtOAc/MeOH 6:4:1). Derivatives 32, 38-40, as well as 42, were unstable upon recrystallization, hence they were used as such for the next step.  33        General procedure for the synthesis of S-methylisothiourea derivatives 43-53 A mixture of the suitable thiourea derivatives 32-42 (0.92 mmol) and iodomethane (3.69 mmol) in 0.1 N NaOH solution (11.8 mL) was stirred at room temperature until the disappearance of the starting material (12-24 h). Then, glacial acetic acid was added until pH 6. The solid which precipitated was collected by filtration and dried, except compounds 46 and 51 which were isolated from the reaction mixture by extraction with EtOAc (30 mL Â3). Evaporation of the anhydrified (Na 2 SO 4 ) organic phase gave a solid which was collected by filtration. These S-methylisothiourea derivatives were unstable upon recrystallization, thus they were used for the next step without further purification. It was observed that derivatives 43-45 and 51 exist in two tautomeric forms in DMSO solution. In fact, in their 1 H NMR spectra there are two signals assignable to the SCH 3 and to the pyrazole proton. Compounds 44 and 51 also display, two signals assignable to the OCH 3 and SCH 3 substituents, respectively (see below for details).        General procedure for the synthesis of 5-aryl(alkyl)amino-7-amino-2H-pyrazolo[4,3-d]pyrimidine derivatives 1-3, 5-12 A mixture of the suitable S-methylisothioureas 43-53 (1 mmol) and NH 4 Cl (20 mmol) in formamide (2 mL) was microwave irradiated at 110 C for 20 min (compounds 9, 10), at 130 C for 40 min (compound 12) and for 2 h (compounds 2, 3), at 150 C for 15 min (compounds 1, 5, 8) and for 20 min (compounds 6, 7, 11). The suspension was then treated with NaHCO 3 saturated solution until pH 7 and the obtained solid was collected by filtration to give compounds 1-3. To isolate derivatives 5-12, the mixture was extracted with CHCl 3 (15 mL Â3), the organic phase was washed with water (15 mL Â2) and anhydrified (Na 2 SO 4 ). Evaporation of the solvent at reduced pressure afforded a residue which was taken up with diethyl ether (2-3 mL) and collected by filtration. The crude derivatives were purified by recrystallization, except compounds 1, 6, 7, 10, 11 which were first purified by column chromatography or preparative TLC (see below for details).
Molecular modeling studies have been performed on a 8 CPU (Intel V R Xeon V R CPU E5-1620 3.70 GHz) linux workstation.
Since to date there are no crystallographic structures available for hA 3 and hA 1 ARs, we retrieved from the Adenosiland web-platform 43,44 previously developed by our research group, their homology models constructed using 4EIY structure as template. Those models were constructed in the presence of ZM-241385 as environment for induced fit, so the resulting structures consist in complexes between each AR subtype and the antagonist ZM-241385.
The residues are identified according to the generic Ballesteros Weinstein numbering system 45 .

Molecular docking
Three-dimensional structures of ligands were built taking advantage of the MOE-Builder tool and ionization states were predicted using the MOE-Protonate-3D tool 46 . Ligand structures were subjected to MMFF94Â energy minimization until the root mean square (rms) gradient fell below 0.05 kcal mol À 1 Å À1 . GOLD docking tool 40 was selected as conformational search program and GoldScore as scoring function, thanks to a docking benchmark study previously carried out in our laboratory 38,47 . For each compound, 10 docking runs were performed on each receptor subtype, searching in a sphere of 20 Å radius centered on the coordinates of the center of mass of ZM-241385 in complex with the receptor. Along with the compounds under investigation, docking simulations were conducted also for ZM-241385 as a reference example.
After computing atomic partial charges both of ligand poses, using PM3/ESP method, and receptors, using Amber10EHT force field, electrostatic and van der Waals contributions to the binding energy were calculated with MOE.

Interaction energy fingerprints (IEFs)
Individual electrostatic and hydrophobic interactions, hereinafter identified as IEele and IEhyd, respectively, were computed between ligand poses and each protein residue involved in binding 37,38 . Both these contributions were computed using MOE and, in particular, IEele were calculated as non-bonded electrostatic interactions energy term of the force field, so they are expressed in kcal/mol. Instead, IEhyd were computed as contact hydrophobic surfaces and are associated to an adimensional score (the higher the better). The data obtained by this analysis were reported in a graphic, called Interaction Energy Fingerprints (IEFs), representing residues (x-axis) in the form of equally high rectangles rendered according to a colorimetric scale. As regards IEele, colors from blue to red represent energy values ranging from negative to positive values; for IEhyd, colors from white to dark green depict scores going from 0 to positive values. More precisely, we retrieved the coordinates of the center of mass of ZM-241385 in the structure of each AR subtype complex. Only residues within 10 Å from this point were retained as belonging to the binding site, and plotted in the IEFs.

Interaction Energy Fingerprints comparison (IEFs comparison)
A new method has been introduced to evaluate docking results, which rests on the observation that ligands able to bind the same site of a protein often share a similar interaction pattern, too. The new method consists in the comparison of the IEFs of the pose of a candidate ligand (hereinafter called "docked") with the IEFs of a ligand whose bound conformation is considered known (hereinafter called "reference").
A quantitative estimation of the similarity of IEFs is computed as root mean square deviation (RMSD) between per residue interaction energies of the docked and the reference poses, both for electrostatic and hydrophobic interactions. This would inform about the average divergence of the docked from the reference: in particular a high RMSD value corresponds to large differences.
So far, there is no information about the direction of the divergence thus, along with RMSD, another analysis, named RMSD trend , has been proposed. This consists of the sum of differences between per residue interaction energies of the docked and the reference, weighted by the number of residues of the binding site. A more favorable interaction energy profile would correspond to a negative RMSD trend in the case of electrostatic interactions, while to a positive one in the case of hydrophobic interactions.
In summary, low RMSD values, along with negative electrostatic RMSD trend and high hydrophobic RMSD trend could be interpreted as an indication of a higher "stability" of the docked pose respect the reference in the orthosteric binding state.
Moreover, this approach could be expanded to compare the behavior of the same ligand on different receptor subtypes, in order to have a preliminary "selectivity" profile based on the stabilities of the docked poses in their corresponding orthosteric binding states. In that case, RMSD and RMSD trend are computed for a docked compound against a reference on each receptor subtype. The reference compound should be a known good binder for each subtype and, at best, the crystallographic structure of the complex should be known.
In our case, ZM-241385 was chosen as reference compound, since it is a ligand for all ARs, having a K i of 774 nM for hA 1 AR, of 1.6 nM for hA 2A AR and of 743 nM for hA 3 AR 5 . As regards the hA 2A receptor, 4EIY crystallographic complex could be employed, while, for hA 1 and hA 3 ARs, we used the homology models, that are receptor-ZM-241385 complexes, since they were constructed considering ZM-241385 as environment for induced fit.
An additional graph was added, which allows to compare electrostatic and hydrophobic IEFs RMSDs and RMSD trend for different ligands on the different AR subtypes. RMSD and RMSD trend for the ligands (y-axis) on the various receptors (x-axis) were reported on a heat map, where they are represented by a colorimetric scale going from red to blue from unfavorable to favorable values. Finally, if a ligand presents blue rectangles on all receptors, it is expected to be "non-selective", otherwise red and blue rectangles should describe lower and higher stability values, respectively, among the different receptor subtypes.

MMsDocking video maker
To facilitate the visualization and analysis of data obtained from the docking simulations, we have implemented a in-house tool, named MMsDocking video maker, for the automated production of a video that shows the most relevant docking data, such as docking poses, per residue IEhyd and IEele data, experimental binding data and scoring values. Videos were mounted using MEncoder 48 starting from images obtained with the following procedure: the heat maps in the background were drawn with GNUPLOT 4.6 49 starting from per residue IEhyd and IEele data computed with MOE. 2d depictions of compounds were generated using the open-source cheminformatics toolkit RDKit 50 . Representations of docking poses within the binding site were constructed using CHIMERA 51 .

Pharmacological assays
Human cloned A 1 , A 2A and A 3 AR binding assay All synthesized compounds were tested to evaluate their affinity at human A 1 , A 2A and A 3 ARs. Displacement experiments of [ 3 H]DPCPX (1 nM) to hA 1 CHO membranes (50 mg of protein/assay) and at least 6-8 different concentrations of antagonists for 120 min at 25 C in 50 mM Tris-HCl buffer pH 7.4 were performed 52 . Non-specific binding was determined in the presence 1 mM of DPCPX ( 10% of the total binding). Binding of [ 3 H]ZM-241385 (1 nM) to hA 2A CHO membranes (50 mg of protein/assay) was performed by using 50 mM Tris-HCl buffer, 10 mM MgCl 2 pH 7.4 and at least 6-8 different concentrations of antagonists studied for an incubation time of 60 min at 4 C 53 . Non-specific binding was determined in the presence of 1 mM ZM-241385 and was about 20% of total binding. Competition binding experiments to hA 3 CHO membranes (50 mg of protein/assay) were performed incubating 0.5 nM [ 125 I]AB-MECA, 50 mM Tris-HCl buffer, 10 mM MgCl 2 , 1 mM EDTA, pH 7.4 and at least 6-8 different concentrations of examined ligands for 60 min at 37 C 54 . Non-specific binding was defined as binding in the presence of 1 mM AB-MECA and was about 20% of total binding. Bound and free radioactivity were separated by filtering the assay mixture through Whatman GF/B glass fiber filters by using a Brandel cell harvester. The filter bound radioactivity was counted by Scintillation Counter Packard Tri Carb 2810 TR with an efficiency of 58%.
Measurement of cyclic AMP levels in CHO cells transfected with hA 2B AR CHO cells transfected with hA 2B AR subtypes were washed with phosphate-buffered saline, diluted trypsin and centrifuged for 10 min at 200 g. The cells (1 Â 10 6 cells/assay) were suspended in 0.5 ml of incubation mixture (mM): NaCl 15, KCl 0.27, NaH 2 PO 4 0.037, MgSO 4 0.1, CaCl 2 0.1, Hepes 0.01, MgCl 2 1, glucose 0.5, pH 7.4 at 37 C, 2 IU/ml adenosine deaminase and 4-(3-butoxy-4methoxybenzyl)-2-imidazolidinone (Ro 20-1724) as phosphodiesterase inhibitor and preincubated for 10 min in a shaking bath at 37 C. The potency of antagonists to the A 2B AR was determined by the inhibition of NECA (200 nM)-induced cyclic AMP production 55 . The reaction was terminated by the addition of cold 6% trichloroacetic acid (TCA). The TCA suspension was centrifuged at 2000 g for 10 min at 4 C and the supernatant was extracted four times with water saturated diethyl ether. The final aqueous solution was tested for cyclic AMP levels by a competition protein binding assay. Samples of cyclic AMP standard (0-10 pmoles) were added to each test tube containing [ 3 H] cyclic AMP and incubation buffer (trizma base 0.1 M, aminophylline 8.0 mM, 2-mercaptoethanol 6.0 mM, pH 7.4). The binding protein prepared from beef adrenals was added to the samples previously incubated at 4 C for 150 min, and, after the addition of charcoal, was centrifuged at 2000 g for 10 min. The clear supernatant was counted in a Scintillation Counter Packard Tri Carb 2810 TR with an efficiency of 58%.

Data analysis
The protein concentration was determined according to a Bio-Rad method 56 with bovine albumin as a standard reference. Inhibitory binding constant (K i ) values were calculated from those of IC 50 according to Cheng & Prusoff equation where [C Ã ] is the concentration of the radioligand and K D Ã its dissociation constant 57 . A weighted non-linear least-squares curve fitting program LIGAND 58 was used for computer analysis of inhibition experiments. IC 50 values obtained in cyclic AMP assay were calculated by non-linear regression analysis using the equation for a sigmoid concentration-response curve (Graph-PAD Prism, San Diego, CA).