Adenosine, a major constituent of nucleic acids, which consists of the purine base adenine linked to the ribose moiety, has important and diverse effects on many biological processes. Some of the physiological actions of adenosine include effects on heart rate and atrial contractility, vascular smooth muscle tone, release of neurotransmitters, lipolysis, renal function and white blood cell functions [1, 2].

Four adenosine receptor (AR) subtypes are known, A1, A2A, A2B and A3, all of which were cloned and pharmacologically characterized. The A2A and A2B receptors are positively coupled to adenylyl cyclase, while A1 and A3 adenosine receptors cause inhibition of cAMP formation. Adenosine acts via these different receptor subtypes, the affinity to which ranges from nanomolar (“high affinity” A1, 3–30 nM; A2A 1–20 nM) to micromolar (“low affinity” A2B, 5–20 μM; A3 > 1 μM) [3, 4]. These receptors belong to the large superfamily of G protein-coupled receptors [2].

Adenosine A1 receptors are ubiquitously expressed, e.g. in the central nervous system (CNS), especially in the brain, with high levels being expressed in many regions. The distribution of adenosine A2A receptors is wide ranging but restricted, including lymphocytes, platelets, brain striatum, vascular smooth muscle and endothelium [2].

The prototypical antagonists of the A1 adenosine receptor are the xanthines: theophylline and caffeine. Natural xanthines are non-specific adenosine antagonists. They are not selective for any of the adenosine receptor subtypes and have low affinity for the A1 receptor. Due to their CNS-stimulating effects, A1 adenosine receptor selective antagonists have been proposed as cognition enhancers for the treatment of dementias, such as Alzheimer’s disease. These receptors have been shown to be involved in sedative, antiseizure and anxiolytic effects. New potential indications are being discovered and investigated: in heart (for the treatment of cardiac arrhythmias and oedemas and as positive inotropic and cardiac protectants), in kidney (for oedemas and nephritis treatment), in lung (for asthma, oedema and lung protection) and in CNS (for depression, stress and coma) diseases. A1 AR antagonists are being investigated as antihypertensives and potassium-saving diuretics with kidney protective effects, for the treatment of depression and asthma and for the prevention of ischemia-induced injuries [57].

In the last years, numerous studies have confirmed the ability of A2A adenosine receptor antagonists to prevent neurodegenerative diseases such as Parkinson’s and Alzheimer’s diseases, ischemic brain damage and, recently, epilepsy and sensorimotor disorders (restless legs syndrome—RLS) [814]. Methylxanthines such as theophylline and caffeine have been known to enhance locomotor activity; however, these compounds are non-selective antagonists and have weak affinity for A2A AR.

Our efforts were directed towards the development of new selective xanthine adenosine receptor ligands. Our main interest has focussed on the investigation of tricyclic xanthine derivatives [1519]. The so far most active compounds are shown in Fig. 1. The most potent A1 AR ligands were found among 1,3-dipropyl-substituted benzylpyrimidopurinediones (I, II) [17], while A2A adenosine receptor ligands were 1,3-dimethyl-substituted aryl- (III, IV) [16] cycloalkyl- (V) [19] and phenalkylpyrimidopurinediones (VI) [18] (Fig. 1). Several of the most active ligands at adenosine A2A AR (e.g. III, IV and VI) [20] were demonstrated to exhibit antiparkinsonian effects. Among pyrimidopurinediones, compounds were found which displayed anticonvulsant properties—protective activity in subcutaneous pentylenetetrazole (ScMet) or in maximal electroshock (MES) and ScMet test; however, the mechanism of this action was not clear [16, 19].

Fig. 1
figure 1

Structures and adenosine receptor binding affinities of tricyclic xanthine derivatives (I–VI); Ki values are given in micromolar; h human, r rat

As a continuation of our search for potent adenosine A1 and especially A2A receptor ligands among cycloalkyl annelated xanthines, we have developed a new series of imidazo-, pyrimido- and diazepino[2,1-f]purinedione derivatives possessing aliphatic substituents (open congeners of cycloalkyl derivatives) in the annelated ring, e.g. alkyl, alkynyl and alkenyl chains. The compounds turned out to be selective A1 or A2A AR antagonists with moderate, submicromolar affinity as shown by radioligand binding studies at native rat receptors. Investigated compounds were examined also for A2B and A3 AR affinity at human recombinant AR. Additionally, the most active compounds were evaluated for affinity at human recombinant A1 and A2A AR affinity. Molecular modelling studies were performed to discuss affinity of the compounds through the docking onto the active sites of the A2A adenosine receptor model. Investigated compounds were examined for their anticonvulsant activity as well.

Results and discussion


The synthesis of tricyclic imidazo-, pyrimido- and diazepinopurinediones (Table 1) was accomplished as shown in Fig. 2. As starting material for 1,3-dimethyl-imidazo[2,1-f]purinediones (33, 34), 7-(2-bromoethyl)-8-bromotheophylline (1) was used, the preparation of which had been described by Caccacae [21]. In our laboratory, a modified procedure was developed using a two-phase catalysis method [16]. The other starting compounds, 7-(3-chloropropyl)-8-bromotheophylline (2) for 1,3-dimethyl-pyrimido[2,1-f]purinediones (524) and 7-(4-bromobutyl)-8-bromo-theophylline (3) for 1,3-dimethyl-diazepino[2,1-f]purinediones (2532), were obtained as previously described [22, 23]. 1,3-Dipropyl-7-(3-chloropropyl)-8-bromoxanthine (4) [16, 17] was used as starting material for the synthesis of pyrimido[2,1-f]purinedione (35). The cyclization reaction with amines possessing straight, branched or unsaturated chains was carried out under various conditions (excess of amine, solvent and different reaction time). The synthesis of compounds 8, 9 and 23 was described previously [24], but their structures had been confirmed only by UV spectra, and pharmacological tests had not been performed. Unsubstituted compounds 5, 25 and 33 were previously synthesized in our group [25, 26] and were now subjected to pharmacological tests to compare them with substituted derivatives. Compound 22 with an ethenyl moiety was obtained by dehydrohalogenation of 9-(2-bromoethyl)-1,3-dimethyl-6,7,8,9-tetrahydropyrimido[2,1-f]purinedione (21) [27] (formed from the appropriate hydroxy ethyl derivative 20 [28]) with ethanolic potassium hydroxide (Fig. 2). The structures of the synthesized compounds were confirmed by UV, IR and 1H NMR spectra: UV spectra showed a bathochromic shift typical for 8-aminoxanthine derivatives with λ max of about 300 nm [29]. The IR absorption bands were typical of xanthine derivatives [30], and in the 1H NMR spectra, the expected chemical shifts were observed. All compounds were purified by recrystallization.

Table 1 Structures of the tested imidazo-, pyrimido- and diazepino[2,1-f]purinediones (535)
Fig. 2
figure 2

Synthesis of substituted imidazo-, pyrimido- and diazepino[2,1-f]purinediones. Reagents and conditions: i appropriate primary amines, solvent (EtOH, n-PrOH, n-BuOH, DMF or without solvent), reflux; ii aminoethanol, reflux; iii PBr3, CHCl3, reflux; iv KOH, EtOH, reflux

X-ray structure analysis

Among derivatives with various alkyl substituents at N(9) or (10) (Table 1), monocrystals of three of them—10, 11 and 30—suitable for X-ray structure analysis could be selected. The structures represent pyrimido- (10 and 11) and diazepino- (30) purinediones with N1,N3-dimethyl substituents at the xanthine nitrogen atoms. They feature linear (pentyl and hexyl) or branched (iso-pentyl) substituents at N(9) or (10). The description and discussion of 10, 11 and 30 spatial properties are attached in the supplementary section.


All compounds were tested in vitro in radioligand binding assays for affinity to A1 and A2A adenosine receptors in rat cortical membrane and rat striatal membrane preparations, respectively. The results are presented in Table 2. The non-selective AR ligand caffeine, selective A2A AR antagonist Preladenant (SCH420814) [31] and selective A1 AR antagonist PSB-36 [31, 32] were included for comparison. Examined compounds were additionally tested for affinity at human recombinant A1, A2A, A2B and A3 receptors stably expressed in Chinese hamster ovary (CHO) cells (Table 3). The following radioligands were used: [3H]2-chloro-N6-cyclopentyladenosine ([3H]CCPA) for A1, [3H]1-propargyl-3-(3-hydroxypropyl)-7-methyl-8-(m-methoxystyryl)xanthine ([3H]MSX-2) [33, 34] for A2A, [3H]PSB-603 [3H]8-(4-(4-(4-chlorophenyl)piperazine-1-sulfonyl)phenyl)-1-propylxanthine [35] for A2B and [3H]2-phenyl-8-ethyl-4-methyl-(8R)-4,5,7,8-tetrahydro-1H-imidazo[2,1-i]purine-5-one ([3H]PSB-11) [36] for A3 binding studies.

Table 2 Affinities of imidazo-, pyrimido- and diazepino[2,1-f]purinediones at rat A1 and A2A adenosine receptors
Table 3 Affinities of imidazo-, pyrimido- and diazepino[2,1-f]purinediones at human recombinant A1, A2A, A2B and A3 adenosine receptors

Tricyclic xanthine derivatives, previously obtained in our laboratory, showed anticonvulsant activity [1518]. Therefore, compounds 5–35 were evaluated in vivo as potential anticonvulsants by the ADP (Antiepileptic Drug Development Program of the National Institute of Neurological Disorders and Stroke NINDS) according to the Antiepileptic Screening Project. Compounds were injected intraperitoneally as a suspension in 0.5 % methylcellulose into the mice and evaluated in the preliminary screenings with at least three dose levels (30, 100 and 300 mg/kg at 0.5- and 4-h time periods). Phase I of the evaluation included three tests: MES, ScMet seizure tests and the rotorod test for neurological toxicity (TOX). The tests were described in detail by Stables and Kupferberg [3739].

The MES test is a model for generalized tonic–clonic seizures and identifies compounds which prevent seizure spread. The ScMet is a model to test compounds that raise seizure threshold. The minimal motor impairment was measured by the rotorod test. The results are given in Table 4.

Table 4 Anticonvulsant activity and neurotoxicity of imidazo-, pyrimido- and diazepino[2,1-f]purinediones in mice (i.p.)

Some compounds (9, 11, 13, 15, 16, 22 and 30) were also administered orally to rats and examined in the MES, ScMet screen and TOX test (Table 5). Compound 9 was tested also in the hippocampal binding model in rats to evaluate its ability to prevent or modify both the expression and acquisition of focal seizures [39] (Table 6). For two compounds (9 and 13), a quantitative test in mice (i.p.) was made (ED50 and TD50 determination). The results of these experiments compared with literature data for valproate [40] are collected in Table 7.

Table 5 Anticonvulsant activity and neurotoxicity of selected compounds after oral administration (30 or 50 mg/kg) to rats
Table 6 Test results of compound 9—preliminary hippocampal binding screen in rats (i.p)
Table 7 Quantitative anticonvulsant activity and neurotoxicity of 9 and 13 and valproate in mice i.p

In vitro tests

Synthesized tricyclic xanthine derivatives displayed affinity towards both A1 and A2A ARs in radioligand binding studies performed at rat brain membranes (Table 2). The most active but not selective A1 AR ligand was 1,3-dipropylpyrimidopurinedione (35). In the group of diazepinopurinediones were found compounds which displayed submicromolar affinity towards adenosine A2A receptors. Diazepinopurinedione 28 (K i = 0.28 μM) showed moderate A2A AR selectivity (about sixfold). A2A AR ligands with submicromolar affinity were also found among pyrimidopurinediones: compounds 11 and 13 displayed K i values of 0.82 and 0.87 μM, respectively. N-substituted derivatives displayed higher affinity than unsubstituted analogues (5, 25 and 33). Elongation of the straight N-alkyl substituents led to an increase in both A1 and A2A AR affinities in the group of diazepinopurinediones. For pyrimidopurinediones, this tendency was not so obvious; it was only observed at A1 AR. In the case of A2A AR, the most active ones were compounds with 6 (comp. 11) and 3 (comp. 8) atom chains. Branched N-alkyl, especially α-branched substituents, generally were not favourable for AR affinity. In case of A1 AR ligands, such modifications led to a loss of affinity (compounds 12, 14, 15, 16 and 18) or its decrease (compound 19). A2A AR ligands were also sensitive to these modifications; however, compounds with branched N-alkyl substituents still showed micromolar affinity, and β-branched derivative 13 was the most active one with submicromolar affinity. Enlargement of the annelated ring was favourable for both A1 and A2A AR activities. Diazepinopurinediones with the same N-alkyl substituents as pyrimidopurinediones were up to sixfold more potent (compounds 9 and 28, and 8 and 27). An exception here was the N-isopropyl derivative 34 which possesses a five-membered annelated ring; however, there are only two members of this group for comparison. Introduction of N1,N3-dipropyl substituents into the xanthine core definitely has favourable influence on A1 AR affinity of N-alkylpyrimidopurinediones: compound 35 was 15-fold more potent than its N1,N3-dimethyl analogue 9. Ligands with unsaturated N-substituents (2224) displayed only moderate A1 and A2A AR affinity.

Selected compounds are mainly those that were the most active ones at the A2A AR (11, 13, 22, 27, 28 and 35) and were also tested for affinity to human A1 and A2A receptors (Table 3). Generally, the affinity at human recombinant A1 as well as A2A receptors was worse than that for native rat receptors (in the range of 1.5—threefold lower affinity). Only in the case of isobutylpyrimidopurinedione 13, the affinity for the human A1 AR was better than that for the rat receptor and almost equal for A2A AR.

Affinity to the human A2B and A3 receptors of all of the investigated compounds was very weak; only the dipropyl derivative 35 showed micromolar affinity at A2B AR and submicromolar affinity to the human A3 AR. Only one compound with a double-branched, long chain (19) from the group of dimethyl derivatives showed affinity for A3 AR in the micromolar range.

Two of the most potent rat and human A2A receptor ligands (compounds 28 and 35) were investigated for their functional properties using cAMP accumulation assay. They were investigated for their potency to inhibit NECA-induced cAMP accumulation in CHO cells expressing the human A2A receptor (Fig. 3). The compounds clearly behaved as competitive antagonists as the concentration–response curve of NECA was shifted to the right in a parallel fashion in their presence. K b values determined in living CHO cells expressing the human adenosine A2A receptor were well in accordance with K i values determined in radioligand binding studies at membrane preparations of the same cell line. Owing to the structural similarity of all compounds in this series, we suppose that they are all antagonists.

Fig. 3
figure 3

cAMP accumulation studies in CHO cells expressing the human adenosine A2A receptor. The dose–response curves for the NECA-induced stimulation of cAMP accumulation were generated with NECA in the absence or in the presence of two different concentrations of 28 (a) or 35 (b). Graphs from two independent experiments performed in duplicates with mean values ± SEM are shown. Both investigated compounds shifted the concentration–response curve for NECA in a parallel manner to the right, indicating competitive antagonism. Apparent K b values were as follows: 1,510 ± 20 nM (28) and 1,210 ± 130 nM (35)

Molecular modelling studies

In our previous molecular modelling studies on A1 and A2A adenosine receptors, the comparison of rhodopsin- and β2-adrenergic-based homology models through the docking studies was performed [41]. Since that time, a few X-ray structures of A2A AR in complex with various ligands have been reported in the Protein Data Bank [42, 43], among which those co-crystallized with xanthines: XAC (PDB code: 3REY) and caffeine (PDB code: 3RFM) are of great importance for our research. Analysis of the ligand binding mode, observed in the crystals 3REY and 3RFM, compared to the binding mode of an inverse agonist ZM241385 bound to A2A AR (PDB codes: 3PWH [42] and 3EML [43]) indicates flexibility of some amino acid side chains within the receptor binding cleft. In particular, the side chain of Asn253 (6.55), described as a crucial residue for ligand binding [4446], in 3REY is rotated relative to other structures, e.g. 3RFM with caffeine as a ligand. Nevertheless, in both crystal structures of A2A AR co-crystallized with xanthines, the terminal amino group of Asn253 (6.55) forms the hydrogen bond with the same carbonyl group present in purinedione core of ligands.

Due to the structure and the size of synthesized molecules, A2A AR-XAC (3REY) crystal seems to be the best choice as a template for docking. To validate the utilized molecular docking methods, XAC ligand was redocked to its X-ray structure of A2A receptor. In case of simulation without any constraints, the obtained highest ranked pose was distinct from the one in the crystal; superposition of both conformations gave a high RMSD value of 9.68 Å. In this pose, an imidazol nitrogen atom of the ligand plays a role of an H-bond acceptor involved in the contact with the distal amino group of Asn253 (6.55), while a phenyl ring is situated in the proximity of TM6 and ECL2. The docking was then repeated, setting the H-bond interaction between Asn253 (6.55) and one out of carbonyl groups of the ligand as a constraint. This time, the overlay with the experimental binding mode was better (Fig. 4, RMSD = 4.53 Å). The superposition of phenylpurinedione cores was almost perfect, giving an RMSD value of 0.42 Å, but calculated and experimental poses differ with orientation of a polar tail of XAC. However, in the crystal structure, the position of this flexible chain is not strictly fixed, as its electron density is not complete [42]. The protocol including constrained H-bond between a ligand and Asn253 (6.55) was chosen for further docking simulations.

Fig. 4
figure 4

Superposition of the XAC ligand from 3REY crystal structure (purple) and redocked pose of XAC (orange)

As an additional validation to the tested series of xanthines, the structure of caffeine was added and docked together with the rest of the compounds. Docking simulations of the set of all 30 compounds to both templates gave results that can be grouped in three clusters according to the calculated docking score values.

In the two first, highest ranked clusters, the obtained poses adopt a reflected or rotated orientation of the heterocyclic core compared to the XAC conformation from the crystal. In this position, in both cases, the carbonyl group C2 = O2 of the purinedione interacts with a side chain amino group of Asn253 (6.55), while C4 = O4 either corresponds to the 2-oxo fragment of XAC or points to the top of the receptor, between side chains of Glu169 and Leu267 (7.32).

However, it can be noticed that in both cases of published X-ray structures co-crystallized with xanthines, purinedione core interacts with Asn253 (6.55) in a similar way, with the same carbonyl oxygen atom (C4 = O4) of the ligand, even if Asn253 in this crystal acid adopts two different conformations of the side chain. Taking into account positions of both XAC and caffeine in the binding site, we can presume that the synthesized tricyclic derivatives of xanthine would bind into the A2A receptor in a similar mode (shown in Fig. 5a, b) despite lower docking scores. For this reason, the conformations belonging to the first two clusters were rejected.

Fig. 5
figure 5

a The model of compound 9 (orange) docked to 3REY crystal structure (chosen residues in green and XAC ligand in purple). b The model of compound 35 (orange) docked to 3REY crystal structure (chosen residues in green and XAC ligand in purple). c Superposition of XAC ligand (purple) and compound 9 (orange) in the binding site of A2A receptor

The third cluster is created by poses predicted for most of the compounds including caffeine (Fig. 5a, b). Although docking scores are lower than those for the two first sets, this pose is in agreement with the position of the ligand XAC in the crystal structure. The carbonyl group C4 = O4 interacts here with the terminal amino group of Asn253 (6.55), and the whole purinedione core of the modelled xanthines overlies the heterocyclic part of XAC with very good RMSD values (from 0.39 to 0.93 Å, for caffeine 0.58 Å), making π–π stacking with the aromatic part of Phe168. Two N-methyl (or N-propyl) substituents point towards the bottom of the binding site, likewise in case of XAC. The third annelated ring is situated between Phe168 and Glu169 of ECL2 from one side and Ile274 (7.39) and Met270 (7.35) from the other. The alkyl chain is buried in a similar pocket of the binding site to the phenoxy tail of XAC, in the long narrow cleft limited by residues from top fragments of TM2 (Ala63 (2.61), Ile66 (2.64) and Ser67 (2.65)) and TM7 (Leu267 (7.32) and Tyr271 (7.36)) as well as Tyr9 (1.35) (Fig. 5c).

The compounds 25 and 33, unsubstituted in the annelated ring, were not successfully docked in this position, while long alkyl chains in this position, branched or not, are easily adapted in the groove created by TM2 and TM7. Similarly, longer propyl fragments at 1- and 3-positions of 35 make hydrophobic contacts with residues Leu85 (3.33), Leu249 (6.51) and Met177 (5.38) from one side and Ile66 (2.64), Ala81 (3.29) and Val84 (3.32) from the other, increasing affinity to the receptor relative to short methyl substituents (35 vs. 9). Compounds 24 and 30 were not found among poses from this cluster either—the most probable explanation is that the linear structure of the triple bond in 24 or a double-branched chain of 30 causes a steric clash with residues forming the cleft.

The performed docking experiments did not explain in detail the correlation between structure and A2A AR affinity of the obtained xanthines. Nevertheless, it can be stated that both the third annelated ring and alkyl chains as N-substituents fit well to the A2A binding pocket, forming additional interactions, and their presence has a big influence on the affinity to the adenosine receptors compared to caffeine—non-selective weak A2A AR antagonist. The similar effect is observed for two N-propyl substituents and the phenyl ring of XAC, very potent adenosine receptor antagonist (35, Fig. 5b).

In vivo tests

Unsubstituted pyrimido- (5) and diazepinopurinedione (25) did not show protective activity in both electric and chemical seizures (Table 4). Introduction of N-alkyl substituents resulted in anticonvulsant activity in the pyrimidopurinediones, whereas these modifications did not affect in the same way the group of diazepinopurinediones as only compounds 31 and 32 showed protective activity in a dose of 300 mg/kg, but after 4 h caused death of tested animals. Probably, enlargement of the annelated ring in the xanthine derivatives is not favourable for anticonvulsant activity. Unsaturated substituents introduced into the tricyclic xanthine derivatives (23 and 24) were also disadvantageous. Generally, N-alkyl derivatives showed good protective activity in mice (i.p.), especially in ScMet test in short time (0.5 h). The best compound was N-hexylpyrimidopurinedione 11, with 100 % protection at a dose of 100 mg/kg with no symptoms of neurotoxicity. The other two compounds with long lipophilic chains 18 and 19 also showed good protection: derivative 19 was active in both tests, and ligand 18 displayed protection in ScMet test. Both compounds did not show neurotoxicity. The N-butyl and N-pentyl derivatives 9 and 10 displayed protective properties in MES and ScMet tests but at the same time showed neurotoxicity. Compounds with short chains (6, 7, 8 and 14) caused death of test animals. The most toxic substance was N-propargylpyrimidopurinedione 24, causing death following clonic seizures and tonic extension.

The most active compounds in mice were also examined for activity in rats after oral administration, showing higher seizure protection (Table 5). The best compound was the butyl derivative 9, showing 100 % protection in ScMet test in the dose of 50 mg/kg in short time (0.25 and 0.5 h) and 50 % protection in dose of 10 mg/kg after 1 2 and 4 h. Ligand 16 was active in MES test at 30 mg/kg and in ScMet test at 50 mg/kg. No symptoms of neurotoxicity were observed in both tested compounds. Ligand 9 was also tested in a hippocampal kindled seizure screen (Table 6). Results suggest the ability of 9 to prevent or modify fully kindled seizures. The N-butyl and N-isobutyl derivatives 9 and 13 were advanced for phase II of evaluation for quantification of activities (ED50 and TD50) against MES- and ScMet-induced seizures in mice (i.p.) (Table 7). These pharmacological parameters were compared with data for valproate [40, 47]. Both tested substances were characterized by a PIScMet higher than that of valproate, but a lower PIMES.

The anticonvulsant activity of the examined compounds was analysed for correlation with AR affinity. Some coincidence of adenosine A2A affinity and anticonvulsant activity was observed. Compounds showing anticonvulsant activity in ScMet test (9, 11, 13, 14, 15, 16 and 18) were among the pyrimidinepurinedione derivatives showing affinity towards A2A AR. However, there were also compounds showing anticonvulsant activity but were devoid of A2A AR affinity. The most active A2A AR ligand, the diazepine derivative 28 and the dipropyl derivative 35 were not active as anticonvulsants. Comparing in vitro and in vivo activities of the studied tricyclic xanthine derivatives, there is no clear correlation between A1/A2A AR affinity and anticonvulsant activity of the investigated ligands. So our previous observation that A2A selectivity of AR antagonists may be important for anticonvulsant activity has not been confirmed in the present study [19].


Tricyclic pyrimido- and diazepinopurinediones are moderately potent A1 and A2A AR antagonists. Enlargement of the annelated ring caused an increase in affinity but not selectivity. Pyrimidopurinodiones showed anticonvulsant activity in mice (i.p.) and rats (p.o.), but lower homologs were toxic. Enlargement of the annelated ring was unprofitable for anticonvulsant activity and led practically to a lack of the activity. Although there is no apparent correlation between anticonvulsant activity and adenosine receptor affinity, it seems that a lipophilic substituent is necessary for both activities.