Probing Substituents in the 1- and 3-Position: Tetrahydropyrazino-Annelated Water-Soluble Xanthine Derivatives as Multi-Target Drugs With Potent Adenosine Receptor Antagonistic Activity

Tetrahydropyrazino-annelated theophylline (1,3-dimethylxanthine) derivatives have previously been shown to display increased water-solubility as compared to the parent xanthines due to their basic character. In the present study, we modified this promising scaffold by replacing the 1,3-dimethyl residues by a variety of alkyl groups including combinations of different substituents in both positions. Substituted benzyl or phenethyl residues were attached to the N8 of the resulting 1,3-dialkyl-tetrahydropyrazino[2,1-f ]purinediones with the aim to obtain multi-target drugs that block human A1 and A2A adenosine receptors (ARs) and monoaminoxidase B (MAO-B). 1,3-Diethyl-substituted derivatives showed high affinity for A1 ARs, e.g., 15d (PSB-18339, 8-m-bromobenzyl-substituted) displayed a Ki value of 13.6 nM combined with high selectivity. 1-Ethyl-3-propargyl-substituted derivatives exhibited increased A2A AR affinity. The 8-phenethyl derivative 20h was selective for the A2A AR (Ki 149 nM), while the corresponding 8-benzyl-substituted compound 20e (PSB-1869) blocked A1 and A2A ARs with equal potency (Ki A1, 180 nM; A2A, 282 nM). The 1-ethyl-3-methyl-substituted derivative 16a (PSB-18405) bearing a m,p-dichlorobenzyl residue at N8 blocked all three targets, A1 ARs (Ki 396 nM), A2A ARs (Ki 1,620 nM), and MAO-B (IC50 106 nM) with high selectivity vs. the other subtypes (A2B and A3 ARs, MAO-A), and can thus be considered as a multi-target drug. Our findings were rationalized by molecular docking studies based on previously published X-ray structures of the protein targets. The new drugs have potential for the treatment of neurodegenerative diseases, in particular Parkinson's disease.


INTRODUCTION
Adenosine receptors (ARs), specifically those of the A 2A subtype, have emerged as new targets for neurodegenerative diseases, in particular for Parkinson's (PD) and Alzheimer's disease (AD). Several A 2A -selective AR antagonists have been evaluated in preclinical and clinical trials. The 8-stryrylxanthine derivative istradefylline (Nouriast R , 1, Figure 1) was approved in Japan as adjunctive treatment of PD in combination with levodopa (Dungo and Deeks, 2013). The consumption of caffeine (2), which is a weakly potent and non-selective AR antagonist (Figure 1), was found to protect from PD and AD as demonstrated in a number of animal models as well as in large epidemiological studies in humans (Chen and Chern, 2011;Flaten et al., 2014).
The concept of multi-target drugs interacting simultaneously with two or more pharmacological targets was proposed as a strategy for the treatment of complex diseases such as cancer, psychiatric disorders and neurodegenerative diseases (Geldenhuys and Van Der Schyf, 2013). Multi-target drugs may exhibit high efficacy due to synergistic effects, show a reduced risk of side effects, and result in improved compliance, especially in elderly patients, as compared to combination therapies of two or more different drugs.
In 2009, Petzer et al. suggested that simultaneous targeting of the dopamine-metabolizing, H 2 O 2 -producing enzyme monoamine oxidase B (MAO-B) by inhibitors, and of A 2A ARs by antagonists, may be advantageous for the treatment of PD due to their dopamine-enhancing effects. While MAO-B inhibition directly inhibits the degradation of dopamine, A 2A AR blockade enhances dopamine-induced D 2 receptor signaling in the A 2A -D 2 heteromeric receptor (Navarro et al., 2018). In addition, A 2A AR antagonists reduce cAMP production by blocking A 2A AR-induced activation of adenylate cyclase (AC); thus they exert the same intracellular effect as dopamine receptor agonists activating G i protein-coupled receptors, thereby also inhibiting AC. Moreover, both, MAO-B inhibition and A 2A AR blockade, are expected to show additional neuroprotective activities, MAO-B inhibitors by reducing hydrogen peroxide production, and A 2A AR antagonists by various mechanisms (Fišar, 2016;Xu et al., 2016). Therefore, such a dual target-directed approach may result in synergistic or at least additive effects thereby possibly halting or reducing the devastating progression of neurodegenerative diseases.
Several studies focused on the design of caffeine derivatives that display A 2A AR antagonistic as well as MAO-B inhibitory activity have been published . 8-m-Chlorostyrylcaffeine (CSC, 3) was the first reported example of an A 2A AR antagonist that also showed high MAO-B inhibitory activity (Figure 2; Chen et al., 2002). Petzer and coworkers reported on a series of (E,E)-8-(4-phenylbutadien-1yl)xanthines, among which caffeine derivative 4 showed potent A 2A AR/MAO-B inhibitory activity (K i rat A 2A AR: 59.1 nM; IC 50 human MAO-B: 37.9 nM) (Pretorius et al., 2008). Recently, Wang et al. published another series of xanthine-based dual A 2A AR antagonists/MAO-B inhibitors. The most potent example of this series was PX-D-P6 (5) (K i human A 2A AR: 330 nM; IC 50 human MAO-B: 260 nM), which showed anti-cataleptic effects in a haloperidol model in rat (Wang et al., 2017). The first nonxanthine-derived dual A 2A AR antagonists/MAO-B inhibitors were reported by our group: N-(4-oxo-4H-3,1-benzothiazin-2-yl)-4-phenylbutanamide (6, Figure 2) was the most potent compound in that series of benzothiazines displaying a K i value of 39.5 nM at the human A 2A AR, and an IC 50 value of 34.9 nM at human MAO-B combined with excellent selectivity vs. other AR subtypes as well as vs. MAO-A (Stössel et al., 2013).
As a further dual-target drug approach for the treatment of PD, the combination of A 2A and A 1 AR blockade was suggested. The dual A 1 /A 2A AR antagonist ASP-5854 (7) (K i human A 1 9.03 nM; K i human A 2A 1.76 nM) was extensively characterized in several animal models of PD, as well as for its effects on cognition. Compound 7 reversed haloperidol-induced catalepsy in monkeys. Moreover, it produced positive results in rats in the passive avoidance test, a model of cognition, in which the A 2A -selective antagonist istradefylline (1) had been inactive (Mihara et al., 2007). The aminopyrimidine-based dual A 1 /A 2A AR antagonist 8, which displays high affinity for both AR subtypes (K i rat A 1 6.34 nM; K i rat A 2A 9.54 nM), showed in vivo efficacy in a rat model of haloperidol-induced catalepsy (Robinson et al., 2015). These results support the hypothesis that a dual A 1 /A 2A AR antagonist may provide additional benefit to PD patients as compared to antagonists that selectively block A 2A ARs, due to their positive effects on cognitive impairment often associated with the disease.
In the present study, we report on the synthesis of a series of 64 novel tetrahydropyrazino[2,1-f ]purinedione derivatives 11-20. The final products 13-20 were evaluated as antagonists at all four AR subtypes (A 1 , A 2A , A 2B , A 3 ) and as inhibitors of both MAO isoenzymes (MAO-A and MAO-B). Substituents on positions N1, N3 and N8 were varied in order to modulate the biological activities of the compounds (Figure 4). Differently substituted benzyl and phenethyl residues were introduced at position 8 keeping the nitrogen atom N8 basic to allow for protonation. In order to study the effects of substituents at nitrogen atoms N1 and N3 on the biological activity of the compounds, methyl groups found in caffeine derivatives and in the majority of published tetrahydropyrazino[2,1f ]purinediones were replaced by ethyl, propyl, cyclopropyl, or propargyl (prop-2-yn-1-yl) moieties, or remained unsubstituted. Within the series of 1-ethyl-3-propargyl-tetrahydropyrazino[2,1f ]purine-2,4(1H,3H)-diones 20 phenyl residues bearing different substituents were additionally introduced in position 8 for comparison with benzyl-and phenethyl-substituted derivatives.

General Information
All commercially available reagents and solvents were used without further purification. The reactions were monitored by thin layer chromatography (TLC) using aluminum sheets coated with silica gel 60 F 254 (Merck). Melting points were determined on a Büchi 530 melting point apparatus and are uncorrected. Column chromatography was carried out on silica gel 0.040-0.063 mm using a Sepacore flash chromatography system (Büchi). 1 H NMR and 13 C NMR data were recorded on a Bruker Avance spectrometer at 500 MHz for proton and 125 MHz for carbon at ambient temperature (for 13 C NMR data, see Supplementary Materials). Shifts are given in ppm relative to the remaining protons of the deuterated solvents. The purity of the tested compounds was determined by HPLC-UV obtained on an LC-MS instrument (Applied Biosystems API 2000 LC-MS/MS, FIGURE 3 | Structures and K i /IC 50 values of tetrahydropyrimido[2,1-f]purinediones (9a,b) and tetrahydropyrazino[2,1-f]purinediones (10a,b,c) as dual-and multi-target drugs (r, rat; h, human).
HPLC Agilent 1100) using the procedure as follows: dissolving of the compounds at a concentration of 1.0 mg/mL in methanol and if necessary sonication to complete dissolution. Then, 10 µL of the substance solution was injected into a Phenomenex Luna C18 HPLC column (50 × 2.00 mm, particle size 3 µm) and elution was performed for 30 min at a flow rate of 250 µL/min with a gradient of water: methanol either containing 2 mM ammonium acetate from 90:10 up to 0:100, starting the gradient after 10 min (system A) or containing 2 mM ammonium acetate and 0.1% formic acid from 90:10 up to 0:100, starting the gradient after 10 min (system B). UV absorption was detected from 220 to 400 nm using a diode array detector. Mass spectra were recorded on an API 2000 mass spectrometer (electron spray ion source, Applied Biosystems, Darmstadt, Germany) coupled with an Agilent 1100 HPLC system.

Radioligand Binding Assays at Adenosine Receptors
The following highly (>100-fold) selective radioligands were employed: A 1 ARs, Klotz et al., 1989, 1 nM, K D  Membrane preparations and radioligand binding assays at rat A 1 (rat brain cortex) and rat A 2A ARs (rat brain striatum) were performed as previously described (Ozola et al., 2003;Alnouri et al., 2015). For assays at human A 1 , A 2A, A 2B, and A 3 ARs, CHO cell membranes expressing one of the human AR subtypes were used as previously reported (Klotz et al., 1998;Alnouri et al., 2015;De Filippo et al., 2016).

Monoamine Oxidase Assay
The determination of MAO-A and MAO-B inhibition was performed using commercially available recombinant human MAO-A and MAO-B enzymes expressed in baculovirus-infected insects sells (Sigma-Aldrich, M7316 and M7441) applying the commercially available Amplex R Red monoamine oxidase assay kit (Invitrogen A12214). The assays were performed as previously described. The determination of rat MAO-B inhibition was performed using mitochondrial-enriched fractions from male Sprague Dawley rat livers. The assays were conducted as previously described (Stössel et al., 2013).

Molecular Docking Studies at Adenosine Receptors
The recent co-crystal structures of the human A 1 AR (5N2S.pdb, Cheng et al., 2017) and the A 2A AR (5N2R.pdb, Cheng et al., 2017) with the antagonist PSB-36 was obtained from the RCSB (Research Collaboratory for Structural Bioinformatics) Protein Data Bank (PDB) (Berman et al., 2000). The downloaded crystal structures were prepared by means of the Molecular Operation Enviroment (MOE 2016.08), chemical computing group. Montreal, Quebec, Canada, 2014) protein structure preparation tool. The hydrogen atoms were assigned according to Protonate-3D implemented in MOE 2016.08. The crystal structures of the human A 1 AR and A 2A AR were applied for flexible ligand docking using AutoDock 4.2 (Morris et al., 2009). During the docking simulations, the ligands were fully flexible while the residues of the receptor were treated as rigid. Selected compounds were docked into the active site of the receptors to predict the binding modes of the compounds. The atomic partial charges were added using AutoDockTools (Sanner, 1999;Morris et al., 2009). Three-dimensional energy scoring grids for a box of 60 × 60 × 60 points with a spacing of 0.375 Å were computed. The grids were centered based on the co-crystallized ligand, PSB-36. Fifty independent docking calculations using the varCPSO-ls algorithm from PSO@Autodock implemented in AutoDock4.2 were performed and terminated after 500,000 evaluation steps (Namasivayam and Günther, 2007). Parameters of varCPSO-ls algorithm, the cognitive and social coefficients c1 and c2 were set at 6.05 with 60 individual particles as swarm size. All the other parameters of the algorithm were set at their default values. Possible binding modes of the compounds were explored by visual inspection of the resulting docking poses.

Molecular Docking Studies at Monoamine Oxidase B
The following programs were used: LigPrep; Maestro; Schrödinger Suites; Schrödinger, LLC: New York, NY, USA, 2017. The X-ray structure of the complex of human MAO-B/safinamide with the accession code 2V5Z was downloaded from the PDB. For the X-ray model, the Protein Preparation Wizard (Schrödinger Inc.) was used in order to add hydrogen atoms, to assign partial charges, and to build missing atoms, side chains and loops. The resulting structures were submitted to energy optimization by using a specific workflow already reported in a previous study (Gidaro et al., 2015). The cocrystallized ligand, safinamide, was used to generate the docking grid box and to check the prediction of the binding affinity. Finally, re-docking simulations were carried out in order to get a protocol validation observing a good capability of the docking software to reproduce the experimental pose of the co-crystallized inhibitor. In standard virtual docking studies, ligands are docked into the binding site of a receptor held as rigid, and the ligand is free to move. However, the assumption of a rigid receptor can give misleading results, since in reality many proteins undergo side-chain or back-bone movements, or both, upon ligand binding. These changes allow the receptor to adapt its binding site to the presence of a certain ligand, a process that is often referred to as the induced fit docking (IFD). This is one of the main complicating factors in structure-based drug design. For that reason, docking studies were carried out by using a specific, previously described IFD workflow (Varela et al., 2012). An initial Glide SP docking of each ligand was performed by using a softened potential, a van der Waals radius scaling factor of 0.50 for receptor/ligand atoms, and a number of 20 poses per ligand to be energy minimized with the OPLS-2005 force field. The poses were saved for each ligand and submitted to the subsequent Prime side chain orientation prediction of residues with a distance cutoff of 5 Å around each ligand. After the Prime SCHEME 1 | Synthesis of 8-substituted 6,7,8,9-tetrahydropyrazino[2,1-f]purine-2,4(1H,3H)-diones 11-15. Reagents and conditions: (a) i. glycolic acid (1.2 eq.), 1 h, 100 • C, neat, ii. H 2 O, NaOH, pH 12-13, 100 • C, 4 h; (b) 1,2-dibromoethane (6 eq.), dimethylformamide (DMF), DIPEA, 70 • C, 16 h; (c) PBr 3 (4 eq.), CH 2 Cl 2 , 0 • C to rt, 1 h; (d) amine (2 eq.), dimethoxyethane, DIPEA, rt, 16 h. energy minimization of the residues and the ligand for each pose, a Glide SP re-docking of each protein/ligand complex structure within 30 kcal/mol above the global minimum was performed. Finally, each output pose was estimated by the binding energy (G-score) and visually examined.
The structures of all products were confirmed by nuclear magnetic resonance ( 1 H NMR, and in many cases additional 13 C NMR) and mass spectral analyses. Melting points were determined for all new compounds. The purity of the tested compounds was confirmed by high-performance liquid chromatography (HPLC) coupled to electrospray ionization mass spectrometry (ESI-MS) using two different methods (for details, see Experimental Section) and demonstrated to be generally greater than 95%, except for compounds 20b and 20k (purity > 94%).
compounds were further investigated for their affinity to human A 1 and A 2A ARs recombinantly expressed in Chinese hamster ovary (CHO) cells. All compounds were additionally investigated for their affinity to human A 2B and A 3 ARs recombinantly expressed in CHO cells to determine their AR subtype selectivity. The following radioligands were employed for radioligand binding studies: [ 3 H]2-Chloro-N 6cyclopentyladenosine ([ 3 H]CCPA, A 1 ) (Klotz et al., 1989), (Müller et al., 2000),  (Müller et al., 2002). It is well known that all xanthine derivatives lacking a ribose moiety, including tricyclic compounds, can only block ARs, but never act as AR agonists; therefore, additional functional studies were not required. All compounds were initially tested for inhibition of human MAO-B at a concentration of 10 µM. For compounds that showed an inhibition of greater than 70% full concentration-inhibition curves were recorded and IC 50 values were determined. Potent MAO-B inhibitors were additionally investigated for inhibition of human MAO-A to assess their selectivity. Results are presented in Tables 1-6, and data of standard ligands are included for comparison.
The N8-(2-bromobenzyl)-substituted compound 14a was found to be a potent dual A 1 /A 2A AR antagonist displaying >10fold selectivity for the human A 1 vs. the human A 2A AR (K i , A 1 AR = 41.7 nM; K i , A 2A AR = 497 nM). Species differences between the human and rat A 1 AR were most prominent in case of the para-substituted benzyl derivatives. Compounds 14c and 14h were significantly more potent at the rat A 1 AR as compared to the human A 1 AR. Further derivatives, 14j-14q, bearing a disubstituted benzyl moiety at position N8 were designed, which are lacking a substituent in the para-position. A 2,3-, 3,4-or 3,5-disubstitution pattern on the benzene ring was well tolerated by the A 1 AR and also improved in most cases the affinity for the A 2A AR. However, compound 14q bearing two larger CF 3 -groups in positions 3 and 5 was nearly inactive.
Compound 14m having a 2-Cl-5-CF 3 -benzyl moiety at position N8 of the tetrahydropyrazino[2,1-f ]purinedione core was found to be a potent A 1 AR antagonist at both rat and human ARs (K i human A 1 = 55.9 nM; K i rat A 1 = 76.2 nM) and displayed a 16-fold selectivity for the (human) A 1 vs. the A 2A AR.
None of the designed compounds lacking a substituent at the N3-position of the tricyclic purinedione core (series 13)-which corresponds to the xanthine N1 positiondisplayed any significant affinity for ARs ( Table 2). The presence of a substituent at that position appeared to be essential for blocking A 1 and A 2A ARs, e.g., compare 1methyltetrahydropyrazino[2,1-f ]-purinedione derivative 13h with its 1,3-dimethyltetrahydropyrazino[2,1-f ]-purinedione derivative 10b, or compare 13a and 13b with 14a and 14e, respectively.
Similar to series 14, all compounds of the N1,N3-diethylsubstituted tetrahydropyrazino[2,1-f ]purinedione series 15 displayed higher affinity for the A 1 than for the A 2A AR subtype ( Table 3). Comparison of 15b with 14g as well as 15c with 14i reveals that elongation of the substituent in the N1-position from methyl (series 14) to ethyl (series 15) resulted in an increase in affinity for the A 3 AR. The m-bromobenzyl derivative 15d was found to be a very potent and selective antagonist of the human and rat A 1 AR (K i , human A 1 AR = 13.6 nM; K i , rat A 1 AR = 21.5 nM). Introduction of a second substituent (fluorine atom) at the benzyl moiety of 15d (compounds 15k and 15l) resulted in a decrease in both A 1 AR affinity and selectivity vs. the A 2A AR. Within the examples having a disubstituted 8-benzyl moiety, compounds having a trifluoromethyl substituent in meta-position and second substituent in para-or meta-position (15i, 15p, 15q) show affinity for the rat A 2A AR. However, the A 1 AR tolerates a m-trifluoromethyl,p-chloro substitution pattern at the 8-benzyl moiety (15p).
Within the 1-ethyl-3-propargyltetrahydropyrazino[2,1f ]purinedione series 20, benzyl derivative 20e was found to be a balanced dual A 1 /A 2A AR antagonist with good potency in both species, rat and human ( Table 4). Introduction of a methoxy group in the ortho-position in 20e led to a similarly potent dual A 1 /A 2A antagonist in humans (20f, K i A 1 = 210 nM; K i A 2A = 311 nM) but showed larger species differences in rat. N8-Phenethyl-substituted tetrahydropyrazinopurinedione 20h was most potent and selective A 2A AR of this series in humans (K i hA 2A = 149 nM). In case of the rat receptor, the opposite results were observed. Compound 20h displayed a lower K i value for the A 1 AR than for A 2A AR (K i A 2A = 1,700 nM; K i A 1 = 117 nM). A methoxy group in the para-position of the phenethyl ring (20k, 20l) led to reduced affinities at both rat adenosine receptor subtypes.
In the last series of compounds (16)(17)(18)(19), the influence of ethyl and cyclopropyl at the N1-position as well as ethyl and  propargyl at the N3-position was probed ( Table 5). These were combined with a 3,4-dichlorobenzyl substituent at N8 as in lead structure 10b (1,3-dimethyl-substituted analog, see Figure 3). Compound 16a was found to be a potent dual A 1 /A 2A AR antagonists with ancillary MAO-B inhibitory activity, similarly to 10b. Switching from a 3,4-dichloro substitution pattern (16a) of the benzyl moiety to 3,5-dichloro substitution (in 16b) resulted in an improvement in affinity for both human and rat A 1 AR. Compared to 16a, a cyclopropyl moiety at the N1-position (compound 17) was less tolerated. A propargyl substituent at N3 combined with a 3,4-dichlorobenzyl at position N8 (19a) somewhat improved the affinity for the human A 3 AR. Changing the substitution pattern on the benzyl ring from 3,4-dichloro to 2-chloro-5-trifluoromethyl (in 19b) eliminated the affinity for the A 3 AR and increased the affinity for the A 1 /A 2A ARs.

Docking Studies at A 1 and A 2A Adenosine Receptors
As shown in Figures 5A,B, the purinedione core structure of the dual A 1 /A 2A AR antagonist 16a forms one of the key π-π stacking interactions with Phe171 and utilizes the hydrophobic surface provided by Leu250 in the human A 1 AR. The carbonyl group at position C4 which corresponds to the C6-carbonyl of xanthine forms another key hydrogen bond interaction with Asn254. The methyl substituent at N1 (corresponding to N3 of xanthine) of purinedione derivative 16a binds within the hydrophobic sub-pocket formed by the residues Leu88, Met180 and Leu250. Similarly, the ethyl substituent at N3 binds in another sub-pocket formed by Ala66, Ile69, Val87, Ile274, and His278. The 3,4-dichlorobenzyl substitution at N8 was found to occupy the pocket formed by the residues Tyr12, Ile69, Asn70, Glu170, Glu172, Ser267, and Tyr271. The tetrahydropyrazine ring which is annelated to the xanthine core, and the methylene group in the benzyl moiety direct the aromatic substituent into a specific binding pocket. A possible electrostatic interaction between the chloro substituent at the 3-position of the benzyl moiety and Glu170 may be beneficial for interaction with the human A 1 AR. This was supported by the observed high affinity of compound 16b with 3,5-dichloro substitution, a modification which possibly increases the chance to form interactions with Glu170. In comparison to classical 8substituted xanthine derivatives such as PSB-36 found in recently published X-ray structures (Cheng et al., 2017), compound 16a and related compounds featuring a tricyclic core structure are somewhat less potent possibly due to the loss of the free N7-H in xanthines, which forms interactions with Asn254 or watermediated interactions with Glu172. As shown in Figures 5C,D, compound 16a follows a similar interaction pattern in the human A 2A AR for the purinedione We additionally docked the A 1 -selective compound 15d (Figure 6A), which contains a m-bromobenzyl residue. Its selectivity for the A 1 vs. the A 2A AR can be explained by strong electrostatic interactions between bromine and Glu170.  33.5 ± 5.6 (r) 756 ± 46 (r) a n = 3. b n = 2. c % inhibition of radioligand binding at indicated concentration.
This may be the reason for the increased affinity of the bromo-substituted benzyl derivative in comparison to the fluoro-(15b) or the chloro-(15c) substituted analogs. The A 2Aselectivity of compound 20h (Figure 6B), a phenethyl derivative, is likely due to strong hydrophobic interaction with Met270, which controls the positioning of the compound toward the binding pocket. The obtained orientation of 20h may be further stabilized by hydrophobic interactions with the residues Leu167 and Leu267. The proposed hypothesis that the hydrophobic residue Met270 likely plays a role is supported by the results at the rat A 1 AR, in which that methionine is replaced by isoleucine (Ile270) showing better affinity in comparison to threonine (Thr270) present in the human A 1 AR (see Sequence Alignment in Figure S1, Supplementary Material).

Species Differences at Adenosine Receptors
The majority of compounds was investigated at both, rat as well as human A 1 and A 2A ARs. This was done because previous studies had revealed major species differences for some classes of AR antagonists between human and rodent  receptors (Maemoto et al., 1997;Burbiel et al., 2016;Szymanska et al., 2016). Rats or mice are typically used for preclinical studies. Therefore, it is important to determine the affinity of tool compounds and preclinical drug candidates in rodent species. The correlation of pK i values obtained at rat vs. human A 1 and A 2A ARs is depicted in Figure 7. Many of the compounds were somewhat more potent at rat than at human receptors, although for some compounds the opposite was true. The correlation was in general quite good (less than 3-5-fold difference in K i values), although some outliers were observed (see Figure 7) confirming subtle differences in the binding sites.

Structure-Activity Relationships at Monoamine Oxidases
The inhibitory activities of the tetrahydropyrazino[2,1f ]purinediones 13-20 at human MAO-A and MAO-B are listed in Table 6. All compounds, tested at a concentration of 10 µM, were found to be inactive at MAO-A. In case of MAO-B, compound 13g bearing a 3,4-dichlorophenethyl moiety at the N8-position was the only example within the series of 1-methyltetrahydropyrazino[2,1-f ]purinediones 13 displaying an IC 50 value in the submicromolar range. Alteration of the substitution pattern on the phenyl ring from 3,4-dichloro to 2,4-dichloro resulted in a 3-fold reduction in MAO-B inhibitory   Within the series of 1,3-diethyl-substituted tetrahydropyrazino[2,1-f ]purinediones 15, compounds having a di-substituted phenyl ring (15i-q) showed a higher MAO-B inhibitory potency as compared to the mono-substituted derivatives of this series (15a-h). In general, a 3,5-and a 3,4-disubstitution pattern of the phenyl ring was beneficial for MAO-B inhibition. The most potent MAO-B inhibitor of this series was derivative 15r (IC 50 = 524 nM) bearing an 3,4,5-trifluorobenzyl moiety at the N8-position. This is in good agreement with results observed within the reported 1,3-dimethyltetrahydropyrazino[2,1-f ]purine-2,4-dione series (Brunschweiger et al., 2014).
The only compound of the 3-ethyl-1methyltetrahydropyrazino[2,1-f ]purinediones series 20 displaying inhibition of more than 50% at a high test concentration of 10 µM was 20c bearing a 3-methoxyphenyl moiety at the N8-position, whereas all other derivatives of this series were inactive.

Docking Studies at Monoamine Oxidase B
In Figure 8, the main interactions between MAO-B and inhibitors 16a (A) and 16b (B), the two active compounds, are depicted. Stacking interactions appear to stabilize the tricyclic inhibitors 16a and 16b within the MAO-B binding site. In case of 20l, which was randomly selected from the group of inactive compounds, this kind of contacts are not detected, which may be the reason for its different biological profile. It is worth noting that Tyr326 in the active site plays an essential role in binding: it is likely involved in interactions with the two active inhibitors 16a and 16b, as with the standard inhibitor safinamide (Binda et al., 2007). Moreover, as we recently observed in case of linear ligands (Carradori et al., 2016;Meleddu et al., 2017), opposite head-tail orientations can easily occur according to docking results due to comparable energy levels. The presence of Cl at position 5 on the benzyl moiety of 16b induces a modest steric hindrance with Tyr188, sufficient to force its inverted orientation in the best pose with respect to 16a.
A comparative analysis performed by the induced-fit docking G-score estimation revealed that compound 16b can establish a better molecular recognition within the MAO-B isoform with respect to MAO-A. The G-score gap is in the order of about 6.7 kcal/mol and likely due to the different binding pocket volumes between those isoforms (Alcaro et al., 2010). The π-π interaction of 16b with Tyr326 in MAO-B ( Figure 8B) cannot be established in the isoform A, where this residue is replaced with and Ile residue.
Water Solubility 1,3,8-Substituted tetrahydropyrazino[2,1-f ]purine-2,4(1H,3H)diones have previously been shown to possess excellent watersolubility at pH 1 due to protonation of the nitrogen atom N8 (Brunschweiger et al., 2014). This is expected to facilitate dissolution in the stomach. Depending on the substitution pattern they may also be well soluble at higher pH values (Brunschweiger et al., 2014). In the present study we measured thermodynamic solubility of only very few compounds exemplarily (14a-e, 14g, 15f,g) and confirmed their high solubility at pH 1 (see Table S1 in Supplementary Material). Most of these compounds were still soluble at pH 7.4. The best soluble derivative was the 3-ethyl-1-methyl-8-m-fluorobenzyl derivative 14g with solubilities of >1.5 g/L (4.2 mM) at pH 1, 50 mg/L (0.14 mM) at pH 4, and 40 mg/L (0.084 mM) at pH 7.4. The protein is shown as gray cartoon while the ligands are shown as green carbon sticks; H-bonds and π-π interactions are depicted as dashed black and cyan lines respectively. Amino acid residues involved in the interactions are shown as gray sticks. The covalently bound cofactor flavine adenine dinucleotide (FAD) is depicted on the left hand side as green sticks. Nitrogen atoms are in blue, oxygen atoms in red, hydrogen atoms in white, phosphorus atoms in purple.

CONCLUSIONS
A large library of novel 1,3,8-substituted tetrahydropyrazino[2,1f ]purinediones was synthesized. For the first time we systematically and extensively studied the exchange of methyl groups in the 1-and 3-position of the theophylline-/caffeinederived tricyclic scaffold for a variety of alkyl residues including cyclic and unsaturated ones. Series of compounds with different 1-and 3-substituent were also obtained. The compounds were tested for antagonistic potency at all four ARs as well as for inhibitory potency at both MAO enzymes with the goal to improve potency at A 1 and A 2A ARs and MAO-B, which are (potential) targets for the treatment of neurodegenerative diseases, in particular for PD. The A 1 AR affinity was dramatically improved by 3-ethyl-1-methyl (e.g., 14a), 1,3-diethyl (15d) and 1-ethyl-3-propargyl (20e) substitution. Good A 2A affinity was obtained for 3-ethyl-1-propargyl derivatives 20, e.g., 20h and 20e. Compounds with a balanced A 1 /A 2A inhibitory potency were also obtained in this group (e.g., 20e). Significantly increased MAO-B inhibitory activity while keeping selectivity vs. the isoenzyme MAO-A, which is important to avoid sideeffects, was obtained by 1-ethyl-3-methyl substitution of the pyrazinopurinedione structure (16a, 16b). Molecular docking studies based on recently published X-ray structures of A 1 and A 2A ARs, and of MAO-B supported the results obtained by SAR analysis. Besides modulating affinities, replacement of the methyl groups in the 1-and 3-position of pyrazinopurinediones may also be useful for fine-tuning metabolic stability of the compounds since the methyl groups of caffeine and theophylline are known to be subject to oxidative demethylation (Fredholm et al., 1999). The present study provides new insights into the SARs of tricyclic xanthine derivatives, which are suitable scaffolds for multi-target drug development. As a next important step, such multi-target drugs will have to be tested in in vivo models in comparison to drugs selective for a single target to prove their potential superiority.

AUTHOR CONTRIBUTIONS
CM and AB designed the study inspired by KK-K, and CM supervised the experiments; AB, SU, and JHo performed the syntheses; BL, SH and PKü performed the biological evaluation of the compounds; PKo and CM analyzed the data and wrote the manuscript. VN, AM, SA, MW, and CM supervised and/or performed the molecular modeling studies. All authors read and contributed to the manuscript.