Subtle Chemical Changes Cross the Boundary between Agonist and Antagonist: New A₃ Adenosine Receptor Homology Models and Structural Network Analysis Can Predict This Boundary

Abstract

Distinguishing compounds’ agonistic or antagonistic behavior would be of great utility for the rational discovery of selective modulators. We synthesized truncated nucleoside derivatives and discovered 6c (K_i = 2.40 nM) as a potent human A₃ adenosine receptor (hA₃AR) agonist, and subtle chemical modification induced a shift from antagonist to agonist. We elucidated this shift by developing new hA₃AR homology models that consider the pharmacological profiles of the ligands. Taken together with molecular dynamics (MD) simulation and three-dimensional (3D) structural network analysis of the receptor–ligand complex, the results indicated that the hydrogen bonding with Thr94^3.36 and His272^7.43 could make a stable interaction between the 3′-amino group with TM3 and TM7, and the corresponding induced-fit effects may play important roles in rendering the agonistic effect. Our results provide a more precise understanding of the compounds’ actions at the atomic level and a rationale for the design of new drugs with specific pharmacological profiles.

Graphical Abstrcat

INTRODUCTION

When an agonist binds to a G protein-coupled receptor (GPCR), a substantial conformational change occurs that leads to the recruitment of a G protein. However, an antagonist cannot induce such conformational changes and ultimately blocks the biological response.¹ Both agonists and antagonists can be used as drugs depending on the therapeutic application; therefore, distinguishing compounds’ agonistic or antagonistic behavior would be of great utility for the discovery of selective drugs. As GPCRs, adenosine receptors (ARs) bind with the endogenous agonist adenosine to regulate many physiological functions.² Among the four AR subtypes (A₁, A_2A, A_2B, and A₃), the most recently identified A₃AR is overexpressed in inflammatory and cancer cells,^2,3 making it a promising target for the development of new therapeutic agents.⁴

Many potent and selective ligands that activate or inactivate A₃AR have been reported, some of which are currently being evaluated in preclinical or clinical trials.^4–11 However, the underlying mechanism of the agonistic or antagonistic behavior at the atomic level still remains unclear. Especially, there have been continuous reported examples of small structural modifications to ligands leading to major changes in their functional activity, converting agonists into antagonists or vice versa.^12,13 If we have an improved understanding of the mechanism governing this “activity shift” for a GPCR-targeted ligand, it would be of great help for the design of more selective molecules.

In this study, we report the discovery of novel ligands for the A₃ adenosine receptor (A₃AR) and the finding that small chemical modifications to a compound can shift its pharmacological profile between antagonist and agonist. To elucidate the underlying cause of this shift, we have developed new hA₃AR homology models that consider the pharmacological profiles of the ligands. We also incorporated the molecular dynamics (MD) simulation and three-dimensional (3D) structural network analysis to predict the hot spot residues for intramolecular signal flow. Our models gave a rationale for this agonist–antagonist boundary, indicating that the H-bonding interactions between the ligands and two crucial residues, namely, Thr94^3.36 and His272^7.43 (the numbers in superscript follow the Ballesteros–Weinstein numbering system), and the corresponding induced-fit effects may play an important role in the agonistic effect on A₃AR. We observed that these interactions could be stably maintained only in the agonist-bound forms. These results contribute to a better understanding of their actions at the atomic level and can be used as a powerful tool for the rational design of new A₃AR ligands with specific pharmacological profiles.

RESULTS AND DISCUSSION

Design Strategy.

We have been engaging in discovering agonists and antagonists of A₃AR and trying to understand the correlation between their structures and pharmacological profiles (Scheme 1). The compounds 1a (IB-MECA, R¹ = H)¹⁴ and 1b (Cl-IB-MECA, R¹ = Cl)¹⁵ are representative A₃AR agonists in clinical trials. The NH of the 5′-uronamide in 1a and 1b serves as a H-bond donor essential for receptor activation.¹⁶ Their 4′-thio analogues 2^4,17 were reported to be potent A₃AR agonists with the same binding mode as 1, showing strong anticancer and anti-inflammatory activities.^18–21 Also, 3 was shown to be a highly potent and selective A₃AR agonist with improved aqueous solubility.²² Based on these observations, we designed 4 by appending a methyl group to the NH to eliminate the H-bond donor. As expected, 4 turned out to be potent A₃AR antagonists. The truncated analogues 5, in which we removed the 5′-uronamide groups of 1 and 2 to minimize steric repulsion at the binding site and abolish the H-bonding ability, were also potent and selective A₃AR antagonists.^23,24 These findings prompted us to design the truncated 3′-amino derivatives 6, which combined the properties of 3′-amino nucleoside 3 with that of the truncated nucleosides 5 and compare their binding affinities to hA₃AR.

Scheme 1.

Shift of the Agonist–Antagonist Boundary in A₃AR Induced by Small Chemical Changes

Chemistry.

We have accomplished the regio- and stereo-selective synthesis of 6a–c from a chiral template, d-mannose (Scheme 2), via the key acylinium ion intermediate, which was then regioselectively reacted with bromide (Scheme 3). The 3′-amino group was introduced by treating the bromide with lithium azide in an ionic liquid, followed by reduction. Synthesis of the truncated 3′-amino-4′-thioadenosine derivatives 6a–c is shown in Scheme 2. d-Mannose was converted to glycosyl donor 7, which was successively converted to known diols 12a–c according to our previously published procedure.^23,24 The glycosyl donor 7 was condensed with 2,6-dichloropurine and 6-chloropurine in the presence of trimethylsilyl trifluoro-methansulfonate (TMSOTf) as a Lewis acid to give the desired β-anomers 8 (79%) and 9 (90%) exclusively. The anomeric assignments were easily achieved by ¹H NOE experiments. Strong NOE effects between H-8 and 3′-H were observed which confirmed the β-orientation of 8 and 9. These on treatment with 2 N HCl gave the corresponding diols 10 and 11, respectively. The desired N⁶-substitution was achieved by treating 10 and 11 with 3-iodobenzylamine to yield the 2-Cl-N⁶-3-iodobenzylamino derivative 12a and 2-H-N⁶-3-iodobenzylamino derivative 12b, respectively. The treatment of 11 with methylamine gave 2-H-N⁶-3-methylamino derivative 12c in good yield.

Scheme 2.

Syntheses of 6a–c from Glycosyl Donor 7^a

^aReagents and conditions: (a) 2,6-dichloropurine or 6-chloropurine, ammonium sulfate, HMDS, 170 °C, 15 h, then TMSOTf, 1,2-dichloroethane (DCE), rt to 80 °C, 3 h; (b) 2 N HCl, rt, 15 h; (c) 3-iodobenzylamine or methylamine, Et₃N, EtOH, rt, 15 h; (d) AcOCMe₂COBr, MeCN, 0 °C, 2.5 h; (e) LiN₃, dimethylformamide (DMF)-[bmim]⁺BF₄⁻, 120 °C, 12 h; (f) NH₃/MeOH, rt, 15 h; (g) PPh₃, NH₄OH, tetrahydrofuran (THF), rt, 15 h.

Scheme 3.

Possible Reaction Pathways for the Formation of 13a–c

To functionalize the 3′-position, introduction of an azido group followed by its reduction seemed to be the best choice. For this purpose, the Mattocks dibromoacetylation^25,26 which converts the vicinal cis-diol into vicinal trans-bromoacetate, was utilized as a key reaction. The diols 12a–c were treated with 2-acetoxyisobutyryl bromide to give the corresponding bromoacetates 13a–c exclusively. As shown in Scheme 3, the diols 12a–c on treatment with 2-acetoxyisobutyryl bromide initially formed an acylinium ion, which then regioselectively reacted with the bromide at the 3′-position because of steric repulsion between the purine base and the bromide, giving 13a–c as exclusive regioisomers. The regioisomers were easily confirmed by the coupling pattern of the proton attached to the carbon bearing the acetoxy group in ¹H NMR. It is understood that in the desired regioisomers 13a–c, the 2′-H proton should split into a doublet of doublets (dd) because of the two vicinal protons. This situation (3′-H) cannot be the case for the alternative regioisomers 13a′–c′ because of the presence of three vicinal protons. Indeed, ¹H NMR spectrum of 13b indicated that 2′-H showed a doublet of doublets at δ 5.86 ppm, confirming that the reaction proceeded via route A. In addition, the stereochemistry of 2′-OAc and 3′-Br was confirmed by ¹H NOE experiments. Strong NOE effects were observed between H-8 and 2′-H, but none between H-8 and 3′-H, confirming the structure of desired compounds 13a–c.

The conversion of bromoacetates 13a–c to the corresponding azidoacetates 14a–c caused difficulties. The conventional method to react bromoacetates 13a–c with sodium azide or lithium azide in DMF gave desired azidoacetates 14a–c in very low yield due to an elimination reaction. However, the addition of an ionic liquid [bmim]⁺BF₄⁻ to DMF resulted in a dramatic increase in the formation of azidoacetates 14a–c. The crude reaction mixture was then subjected to deprotection conditions. The acetyl group was thus removed by the treatment of 14a–c with methanolic ammonia to give the corresponding hydroxyl derivatives 15a–c. Treatment of 15a–c with Ph₃P and NH₄OH in THF-H₂O afforded the desired amino derivatives 6a–c, respectively.

As shown in Table 1, 6c has the best binding affinity (K_i = 2.40 ± 0.20 nM) at hA₃AR with high selectivity in comparison to other subtypes, which is better than the reference compound 5 (K_i = 4.16 ± 0.50 nM). This result indicates that the replacement of 3′-OH with 3′-NH₂ can also contribute toward ligand binding to hA₃AR. Interestingly, the cAMP functional assay (see the Supporting Information (SI))²⁷ showed that 6c is a quite potent agonist with an EC₅₀ of 11.3 ± 1.3 nM (Figure 1). The maximal efficacy at 10 μM was 83.9 ± 3.2%, compared to the full A₃AR agonist, Cl-IB-MECA (100%). This result is surprising compared to our earlier work that discovered the derivatives lacking the 4′-hydroxymethyl moieties as A₃AR antagonists.^23,24 It indicates that many of these nucleosides are right at the agonist–antagonist boundary, and subtle chemical modifications can push them one way or the other. There have been continuous reports of small structural modifications to ligands leading to major changes in their functional activity, converting agonists into antagonists or vice versa.^5,13 A better understanding of this activity shift would be useful for the design of more selective GPCR-targeted ligands.

Table 1.

Binding Affinities of the A₃AR Antagonist 5 and 3′-Amino Derivatives 6a–c at Three Subtypes of hARs

	Ki (nM ± SEM or % inhibition)a
compound (X, R1, R2)	hA1	hA2	hA3
5 (X = OH, R1 = Cl, R2 = 3-iodobenzyl)b	2490 ± 940	341 ± 75	4.16 ± 0.50
6a (X = NH2, R1 = Cl, R2 = 3-iodobenzyl)	2200	2390 ± 70	380 ± 190
6b (X = NH2, R1 = H, R2 = 3-iodobenzyl)	2450 ± 740	2300	260 ± 80
6c (X = NH2, R1 = H, R2 = methyl)	1670 ± 200	20%	2.40 ± 0.20

^aValues are expressed as mean ± SEM, n = 3–4 (outliers eliminated) and normalized against a nonspecific binder, NECA (10 μM). The percentage values refer to the % inhibition of specific radioligand binding at 10 μM.

^bData from Jeong et al.^23,24

Figure 1.

Activity of 6c as an agonist at the human A₃AR expressed in CHO cells, compared to the maximal effect of Cl-IB-MECA (=100%). The assay measured G_i-coupled A₃AR-dependent inhibition of cAMP formation when stimulated using 10 μM forskolin. This is a representative curve of three separate determinations. The EC₅₀ values for 6c (11.3 ± 1.3 nM, maximal efficacy was 83.9 ± 3.2%) and reference full agonist Cl-IB-MECA (1.21 ± 0.35 nM) were determined.

Homology Modeling.

To investigate why such a subtle chemical change shifts this boundary, we performed molecular modeling studies (Figure 2). No X-ray crystal structure of hA₃AR has yet been reported; however, structures of the homologous subtype hA_2AAR co-crystallized with a known agonist and antagonist were recently released.^28,29 The hA_2A and hA₃AR sequences were aligned and refined, focusing on the highly conserved transmembrane (TM) regions (Figure S1). As the receptor structures showed significant conformational changes upon agonist binding, we built the hA₃AR homology models utilizing both agonist- and antagonist-bound hA_2AAR crystal structures^28,29 (see the SI).

Figure 2.

Model selection workflow of the antagonist- and agonist-based hA₃AR homology models for induced-fit docking and molecular dynamics (MD) simulation studies.

The dynamics of receptor conformation must be considered for an accurate docking study. Among the many implementations proposed to account for protein flexibility in molecular docking, the multiple receptor conformation (MRC) method is a straightforward, practical, and intuitive way to discretely mimic target plasticity.³⁰ We applied a very efficient sampling strategy using elastic network-normal mode analysis (EN-NMA)³¹ for the rapid generation of MRCs. The appropriateness of the models in diverse conformations was examined based on the docking results of the representative nucleoside ligands with high potency and ca. 50-fold higher selectivity for hA₃AR against other subtypes (Table S1). The best models that produced reasonable binding modes for all representative ligands were selected (Table S2) and further optimized through the MD simulation (Models S1 and S2).

The helical structures of the best agonist- and antagonist-based hA₃AR models showed quite different conformations (Figure 3A,B). The most significant movements upon agonist binding were observed in TM5, TM6, and TM7 (Figure 3C), i.e., an outward tilt and rotation of the cytoplasmic half of TM5–6, and an inward movement of TM7. These changes were consistent with those observed in A_2AAR, opsin, and β-adrenergic receptor.^28,32,33 Especially the movement of TM6 is reported to be important for G protein binding and GPCR activation.¹ The rearrangement of the helices also resulted in models with different binding cavity surfaces (Figure 3D).

Figure 3.

Human A₃AR models in the agonist-based (TM helices in rainbow colors; loops in white) and antagonist-based (light blue) forms. (a) Side view with the membrane (light gray). (b) Top view from the extracellular region. Representative agonist (20, magenta) and antagonist (24, sky blue) are docked and displayed in space-fill. (c) Root-mean-square deviation (RMSD) of the Cα atoms between two models. (d) Binding cavity surfaces of the agonist-bound (magenta) and antagonist-bound (sky blue) models.

MD Simulation and Binding Mode Analysis.

Based on the resulting homology models, induced-fit docking study was performed to get the complex structures of hA₃AR with the analogues designed in this work. The docking results showed that both the antagonist 5 and the agonist 6c bind well, explaining the high binding affinities to hA₃AR. To overcome the inaccuracies of the homology model, as well as to consider the dynamic aspects of the receptor–ligand binding, we conducted MD simulation of the complexes with the agonists (6c and 20) and the antagonists (5 and 24). After 300 ns simulations of each system, the lowest-energy conformations were acquired from the well-equilibrated trajectories (i.e., 100–300 ns trajectories, Figure S2) to generate the final binding modes shown in Figure 4.

Figure 4.

Binding modes of (a) agonist 6c, (b) antagonist 5, (c) agonist 20, and (d) antagonist 24 in hA₃AR. The poses were selected by the lowest-energy conformation during the MD simulation. The ligands are depicted in sticks with the carbon atoms in magenta (6c), teal (5), hotpink (20), and sky-blue (24). The agonist- and antagonist-based models are colored in pale pink and pale blue, respectively. The key interacting residues are displayed in sticks with their residue numbers marked. The H-bonds are shown as black dashed lines, and the nonpolar hydrogens are undisplayed for clarity.

The adenine parts of the analogues formed π–π stacking interaction with Phe168 and a H-bond with Asn250. The substituents at the N⁶ position of the adenine moiety were oriented toward the extracellular region, and the thio-sugar rings were located at the bottom of the binding site, forming H-bonds with Ser271 and/or His272. Surprisingly, the H-bond with Thr94, which was lost in 5 (Figure 4B), was recovered by the 3′-amino group in 6c (Figure 4A). The H-bond with Thr94^3.36 is important for AR activation¹⁶ and the intramolecular signal transduction of AR.³⁴ For 1 and 2, the H-bond with Thr94 was formed by the 5′-uronamide groups.¹⁶ As shown in Figure 4C, the corresponding 4′-thio-analogue, 20, also formed the H-bond interaction with Thr94 with its 5′-uronamide moiety. This interaction was broken in the case of the antagonist 24 because of the absence of the H-bond donor feature (Figure 4D). Although 6c is truncated at the 5′-position, its 3′-amino group preserved the key H-bond with Thr94. The partially negative side-chain oxygen (i.e., −0.683 in the OPLS force field) of Thr94 interacted favorably with the 3′-amino group of 6c, which is protonated under physiological conditions (Figure S3). Since the 3′-NH₃⁺ can make a bond via ionic interaction, it will be more advantageous to interact with Thr94 than the 3′-OH group, even if the distance is a little bit far.

To analyze the dynamic receptor–ligand interactions, the time traces of key interactions were monitored along the MD simulations. We observed that several key interactions are continuously formed only in the agonist-bound state (Figure 5). Thr94 has a stable H-bond interaction with the agonist ligands, while this interaction can be transiently broken in the antagonist-bound structures (see the red plots in Figure 5). The H-bond interaction with His272 seems to be also important for the agonist-bound states (green plots in Figure 5). Interestingly, the H-bond interactions with Ser271 and His272 located in TM7 can complement each other in the agonist(6c)-bound structures. When the interaction with His272 breaks, the interaction with Ser271 is formed alternatively. It may be difficult to explicitly conclude that one specific residue is important for all agonistic effects. Rather, we speculate that these three residues, i.e., Thr94, Ser271, and His272, at the bottom of the ligand-binding pocket work in a very cooperative way, and the network among the ligand and these key residues are essential for agonism.

Figure 5.

Time traces of the receptor–ligand binding during the MD simulation of hA₃AR complexed with (a) 6c, (b) 5, (c) 20, and (d) 24. The atom pairs utilized in distance measurement are depicted in red, blue, cyan, green, and dark green lines with the chemical structures at the upper right box. The hydrogen bonds (HB) formed by Thr94, Ser271, and His272 residues are displayed in red, blue, and green, respectively. The rmsd values of the ligand molecule itself (rmsd^lig) is marked in black. At the lower right box, the key interactions are marked with the z-clipped surface of the binding site cleft.

Additionally, the ligand-binding cavity is quite distinct in the two states (Figure 5, lower right box), as the TM regions are slightly rearranged upon agonist binding. The rearrangement resulted in a smaller cavity at the bottom of the ligand-binding site for the agonist(6c)-bound complex (Figure 5A, lower right box) than for the antagonist(5)-bound complex (Figure 5B, lower right box). It is possible that the binding of 6c causes the upper part of the TM helices to fit more tightly to the ligand, allowing key interactions for receptor activation, such as with Thr94 and His272. Taken together with the reported mutational and SAR data,^35,36 our simulation results indicate that the H-bond between the 3′-amino group and Thr94/His272 plays an important role in the agonistic effect of the 3′-amino derivatives.

In addition, both agonist ligands, 6c and 20, showed relatively stable binding modes during the simulation (see the lower right box in Figure 5; note that rmsd^lig, i.e., rmsd of the bound ligand, is always less than ~3 Å). However, rmsd^lig of the antagonist molecules, 5 and 24, increased up to ~6 Å, reflecting their degrees of freedom at the binding site. It seems that the agonist molecules can fit tightly at the bottom binding pocket, maintaining the fundamental interactions; while the antagonists cannot continuously maintain the key interactions and thus, do not transmit the signal.

Structural Network Analysis.

For a more detailed theoretical explanation, we adopted the network analysis based on graph theory to the 3D structure of hA₃AR. In the graph representation of the protein structure, each amino acid residue denotes one node; and if two residues are close enough to make contact, these two nodes are linked to each other (Figure 6A). Among several network properties which can be used to determine the characteristics of the network, betweenness centrality (C_B) is one of the most fundamental concepts in network analysis for identifying the mediators of information flow in a given network topology (eq 1 in the SI). For example, in the simplified network, as shown in Figure 6B, node X has the highest C_B value among all nodes. Upon the removal of node X, the whole graph would be split into two pieces. Therefore, node X is the most critical one for intranetwork communication or the flow of information.

Figure 6.

Structural network analysis of the antagonist- and agonist-bound hA₃AR. (a) Representation of the protein structure based on the residue interaction network. Small light-blue spheres represent the nodes in the network (i.e., Ca atom of each residue), and the light-blue dashed lines mean the links between residues. (b) Simplified exemplary network in which the node of highest betweenness centrality is marked. (c) Network analysis results representing the difference between agonist-bound (magenta) and antagonist-bound (teal) states. 〈C_B〉 is the average betweenness centrality during the simulation for a given residue. 〈C_B〉^agonist and 〈C_B〉^antagonist are plotted in magenta and teal, respectively. ΔC_B means the difference between two states (i.e., 〈C_B〉^agonist – 〈C_B〉^antagonist). Network models of hA₃AR bound with 6c and 5 are displayed at the left and right sides. The protein secondary structure is displayed in ribbon, and its network model is overlaid with their nodes in small spheres. The residues (i.e., nodes) with increased betweenness centrality values (|ΔC_B| > 0.02) in each state are represented in larger spheres together with their surfaces, and the bound ligand is depicted in sticks. In the agonist-bound structures, Thr94 and His272 are identified as a residue with high betweenness centrality and colored in green.

Interestingly, this C_B-based analysis showed quite different results for the agonist-bound and the antagonist-bound structures (Figure 6C). Some regions showed significantly increased betweenness centrality in the agonist-bound state during the simulation. To extract these regions, we calculated ΔC_B (= 〈C_B〉^agonist – 〈C_B〉^antagonist), where ⟨C_B⟩ is the average betweenness centrality of a given residue during the simulation (note that the equilibrated 100–300 ns trajectories were used for the analysis). The positive ΔC_B is for the residues with increased ⟨C_B⟩ in the agonist-bound state, whereas the negative ΔC_B is for the ones with increased ⟨C_B⟩ in the antagonist-bound state (see the magenta and teal boxes in Figure 6C). Mapping the residues with high ΔC_B values into the structure, Thr94 and His272 were identified as hot spot residues where the betweenness centrality is increased in the agonist-bound state. It reflects their influence on the signal flow in the entire structure. On the other hand, in the antagonist-bound form, these residues could not work as a mediator of information flow. Our previous work with hA_2AAR structure also identified the corresponding Thr residue, i.e., Thr88 in hA_2AAR, as one of the hot spot residues for intramolecular signal transduction.³⁴ Also, the residues with high ΔC_B in the agonist-bound state formed a network linking the ligand-binding site to the cytoplasmic G protein binding site through the middle channel of TM3, TM5, TM6, and TM7 (see the magenta surface in Figure 6C). We speculate that these regions may be important for agonistic signal flow within the GPCR by increasing network connectivity upon agonist binding.

CONCLUSIONS

In conclusion, we discovered that subtle chemical changes performed on A₃AR ligands could induce an activity shift from antagonist to agonist. We were able to explain this agonist–antagonist boundary using our hA₃AR homology models, MD simulation, and structural network analysis. Thr94 in the TM3 helix and His272 in TM7 that form the stable H-bonding with agonist, together with the induced-fit effect, might be important for sensing the agonist binding and transmitting signals to the intracellular G protein binding site. These results provide a more precise understanding of the compounds’ actions at the atomic level and a rationale for the design of new drugs with specific pharmacological profiles.

EXPERIMENTAL SECTION

Chemical Synthesis.

General Methods.

Proton (¹H) and carbon (¹³C) NMR spectra were obtained on a Bruker AV 400 (400/100 MHz), Bruker AMX 500 (500/125 MHz), Jeol JNM-ECA600 (600/150 MHz), or Bruker AVANCE III 800 (800/200 MHz) spectrometer. Chemical shifts are reported as parts per million (δ) relative to the solvent peak. Coupling constants (J) are reported in hertz (Hz). Mass spectra were recorded on a Thermo LCQ XP instrument. Optical rotations were determined on Jasco III in appropriate solvent. UV spectra were recorded on U-3000 made by Hitachi in methanol or water. Infrared spectra were recorded on Fourier transform infrared (FT-IR; FTS-135) made by Bio-Rad. Melting points were determined on a Buchan B-540 instrument and are uncorrected. The crude compounds were purified by column chromatography on a silica gel (Kieselgel 60, 70–230 mesh, Merck). Elemental analyses (C, H, and N) were used to determine the purity of all synthesized compounds, and the results were within ±0.4% of the calculated values, confirming ≥95% purity. High-performance liquid chromatography (HPLC) analysis was also performed to determine the purity of the final compounds 6a–c, confirming ≥95% purity.

General Procedure for the Synthesis of 13a–c. (2R,3S,4R)-4-Bromo-2-(2-chloro-6-((3-iodobenzyl)amino)-9H-purin-9-yl)-tetra-hydrothinophen-3-yl acetate (13a).

A solution of 2-acetoxyisobutyryl bromide (0.26 mL, 1.78 mmol) in freshly distilled acetonitrile (5 mL) was added dropwise to a well-stirred suspension of 12a (300 mg, 0.60 mmol) in acetonitrile (10 mL) under nitrogen at 0 °C. The reaction mixture was further stirred at 0 °C for 2.5 h. The mixture was quenched with saturated sodium bicarbonate solution at 0 °C until the pH of solution reached 8, and then the solution was neutralized with acetic acid to pH 5. The reaction mixture was extracted with ethyl acetate and washed with sodium bicarbonate (2 × 5 mL) and brine. The organic layer was dried over anhydrous MgSO₄, filtered, and evaporated to give the crude residue, which was purified by silica gel column chromatography (hexane/ethyl acetate = 3/1) to give 13a (360 mg, 70%) as a colorless syrup: [α]d²⁵ = −2.4 (c = 1.03, CH₃OH); ¹H NMR (400 MHz, CDCl₃) δ 8.08 (s, 1H), 7.74 (s, 1H); 7.61–7.63 (m, 1H), 7.35 (d, J = 7.2 Hz, 1H), 7.07 (t, J = 8.0 Hz, 1H), 6.89 (br s, 1H), 5.99 (d, J = 6.0 Hz, 1H), 5.68 (dd, J = 9.2, 6.0 Hz, 1H), 4.73–4.78 (m, 2H), 4.47–4.53 (m, 1H), 3.77 (pseudo t, J = 11.2 Hz, 1H), 3.42 (dd, J = 11.2, 6.0 Hz, 1H), 2.10 (s, 3H); ¹³C NMR (100 MHz, CDCl₃) δ 176.5, 169.9, 155.1, 140.4, 138.9, 137.2, 136.9, 130.6, 127.5, 94.7, 83.8, 60.3, 46.3, 46.5, 44.2, 36.0, 24.9, 20.7; UV (MeOH) λ_max 273 nm; MS (ESI+) [M + Na]⁺ m/z 629.8826; Anal. calcd for C₁₈H₁₆BrClIN₅O₂S: C, 35.52; H, 2.65; N, 11.51; found: C, 35.55; H, 2.99; N, 11.11.

(2R,3S,4R)-4-Bromo-2-(6-((3-iodobenzyl)amino)-9H-purin-9-yl)-tetrahydrothinophen-3-yl acetate (13b).

Compound 13b (82%) was obtained as a colorless syrup from 12b in a similar manner to that described for the preparation of 13a: [α]d²⁵ = −0.41 (c = 1.38, CH₂Cl₂); ¹H NMR (400 MHz, CDCl₃) δ 8.40 (s, 1H), 8.05 (s, 1H), 7.73–7.74 (m, 1H), 7.61 (d, J = 8.0 Hz, 1H), 7.35 (dd, J = 8.0, 0.8 Hz, 1H), 7.06 (t, J = 8.0 Hz, 1H), 6.26 (br s, 1H, D₂O exchangeable), 6.01 (d, J = 6.0 Hz, 1H), 5.86 (dd, J = 8.8, 6.0 Hz, 1H), 4.82–4.84 (m, 2H), 4.49–4.56 (m, 1H), 3.85 (pseudo t, J = 10.4 Hz, 1H), 3.43 (dd, J = 10.8, 6.0 Hz, 1H), 2.06 (s, 3H); ¹³C NMR (100 MHz, CDCl₃) δ 169.7, 154.8, 153.6, 141.1, 138.8, 136.8, 136.8, 130.6, 127.1, 94.8, 83.5, 60.6, 46.8, 36.2, 29.9, 20.8; UV (CHCl₃) λ_max 263.5 nm; MS (ESI+) [M + H]⁺ m/z 573.9410; Anal. calcd for C₁₈H₁₇BrIN₅O₂S: C, 37.65; H, 2.98; N, 12.20; found: C, 37.77; H, 2.58; N, 11.98.

(2R,3S,4R)-4-Bromo-2-(6-(methylamino)-9H-purin-9-yl)-tetrahy-drothiophen-3-yl acetate (13c).

Compound 13c (81%) was obtained as a colorless syrup from 12c in a similar manner to that described for the preparation of 13a: [α]_D²¹ = −6.3 (c = 0.24, CH₃OH); ¹H NMR (400 MHz, CDCl₃) δ 8.39 (s, 1H), 8.08 (s, 1H), 6.04 (d, J = 4.8 Hz, 1H, D₂O exchangeable), 6.01 (d, J = 6.0 Hz, 1H), 5.85 (dd, J = 9.2, 6.0 Hz, 1H), 4.48–4.54 (m, 1H), 3.83 (pseudo t, J = 10.4 Hz, 1H), 3.41 (dd, J = 10.8, 6.0 Hz, 1H), 3.18 (s, 3H), 2.04 (s, 3H); ¹³C NMR (100 MHz, CDCl₃) δ 169.4, 155.5, 153.3, 139.3, 138.2, 120.2, 83.2, 60.1, 46.5, 35.8, 20.5; UV (CH₂Cl₂) λ_max 266.0 nm; MS (ESI+) [M + H]⁺ m/z 372.0124; Anal. calcd for C₁₂H₁₄BrN₅O₂S: C, 38.72; H, 3.79; N, 18.81; found: C, 38.99; H, 3.98; N, 19.12.

General Procedure for the Synthesis of 15a–c. (2R,3R,4S)-4-Azido-2-(2-chloro-6-((3-iodobenzyl)amino)-9H-purin-9-yl)-tetrahydrothiophen-3-ol (15a).

To a well-stirred suspension of 13a (138 mg, 0.226 mmol) in DMF-[bmim]⁺BF₄⁻ (10:1 v/v) was added LiN₃ (55 mg, 1.13 mmol, 20 wt % in H₂O) at room temperature. The reaction mixture was heated to 120 °C, and the contents were stirred at the same temperature for 12 h. The reaction mixture was evaporated, and then the crude residue obtained was directly treated with methanolic ammonia. The reaction mixture was stirred at room temperature for 15 h. After evaporation, the residue was purified by silica gel column chromatography (CH₂Cl₂/MeOH = 20:1) to give 15a (38 mg, 32%) as a white foam: [α]_D²⁵ = −51.74 (c = 1.09, MeOH); ¹H NMR (400 MHz, CD₃OD) δ 8.31 (s, 1H), 7.77 (s, 1H); 7.59 (d, J = 8.0 Hz, 1H), 7.39 (dd, J = 8.0, 0.8 Hz, 1H), 7.08 (t, J = 8.0 Hz, 1H), 5.91 (d, J = 6.0 Hz, 1H), 4.88 (dd, J = 6.0, 4.0 Hz, 1H), 4.70 (br s, 2H), 4.31 (dd, J = 8.8, 4.8 Hz, 1H), 3.57 (dd, J = 11.2, 5.2 Hz, 1H), 3.01 (dd, J = 11.2, 4.8 Hz, 1H); ¹³C NMR (100 MHz, CDCl₃) δ 156.4, 155.6, 151.6, 142.7, 141.7, 137.9, 137.5, 131.4, 128.2, 120.0, 94.9, 80.8, 65.9, 64.6, 44.4, 32.4; IR (KBr) 2110 cm⁻¹; UV (MeOH) λ_max 273.5 nm; MS (ESI +) [M + H]⁺ m/z 528.9815; Anal. calcd for C₁₆H₁₄ClIN₈OS: C, 36.34; H, 2.67; N, 21.19. Found: C, 36.54; H, 2.68; N, 21.11.

(2R,3R,4S)-4-Azido-2-(6-((3-iodobenzyl)amino)-9H-purin-9-yl)-tetrahydrothiophen-3-ol (15b).

Compound 15b (26%) was obtained as a white foam from 13b in a similar manner to that described for the preparation of 15a: [α]d^24.7 = −0.71 (c = 0.07, CH₂Cl₂); ¹H NMR (400 MHz, CDCl₃) δ 8.36 (s, 1H); 8.05 (s, 1H), 7.73 (s, 1H), 7.62 (d, J = 10.4 Hz, 1H), 7.34 (d, J = 8.4 Hz, 1H), 7.07 (t, J = 8.0 Hz, 1H), 6.37 (d, J = 3.2 Hz, 1H, D₂O exchangeable), 6.21 (br s, 1H, D₂O exchangeable), 5.85 (d, J = 6.0 Hz, 1H), 4.83 (br s, 2H), 4.73–4.76 (m, 1H), 4.43 (dd, J = 7.2, 4.4 Hz, 1H), 3.47 (dd, J = 11.6, 4.0 Hz, 1H), 3.03 (dd, J = 11.6, 3.2 Hz, 1H); ¹³C NMR (100 MHz, CDCl₃) δ 171.7, 152.9, 140.9, 139.4, 136.7, 130.5, 126.9, 94.7, 79.9, 66.2, 60.7, 56.3, 35.6, 29.8, 22.8, 21.1; IR (KBr) 2107 cm⁻¹; UV (CHCl₃) λ_max 268.0 nm; MS (ESI+) [M + H]⁺ m/z 495.0202; Anal. calcd for C₁₆H₁₅IN₈OS: C, 38.88; H, 3.06; N, 22.67; found: C, 38.54; H, 3.09; N, 21.99.

(2R,3R,4S)-4-Azido-2-(6-(methylamino)-9H-purin-9-yl) tetrahydro-thiophen-3-ol (15c).

Compound 15c (27%) was obtained as a white foam from 13c in a similar manner to that described for the preparation of 15a: [α]d²⁶ = −35.16 (c = 0.091, CH₃OH); ¹H NMR (400 MHz, CD₃OD) δ 8.46 (s, 1H); 8.30 (s, 1H), 6.04 (d, J = 6.4 Hz, 1H), 4.67 (dd, J = 6.8, 3.6 Hz, 1H), 4.47 (dd, J = 7.6, 3.6 Hz, 1H), 3.55 (dd, J = 11.2, 4.4 Hz, 1H), 3.14 (s, 3H), 2.96 (dd, J = 10.8, 3.2 Hz, 1H); ¹³C NMR (100 MHz, CDCl₃) δ 173.0, 142.1, 80.9, 74.4, 64.0, 61.6, 35.3, 20.9, 14.5; IR (KBr) 2107 cm⁻¹; UV (CH₃OH) λ_max 267.0 nm; MS (ESI+) [M + H]⁺ m/z 293.0912; Anal. calcd for C₁₀H₁₂N₈OS: C, 41.09; H, 4.14; N, 38.33; found: C, 41.45; H, 3.98; N, 38.76.

General Procedure for the Synthesis of 6a–c. ((2R,3R,4S)-4-Amino)-2-(2-chloro-6-((3-iodobenzyl)amino)-9H-purin-9-yl)-tetrahydrothiophen-3-ol (6a).

A solution of 15a (78 mg, 0.15 mmol) in THF (5 mL) was treated with triphenylphosphine (77 mg, 0.30 mmol) at 0 °C. After 30 min, ammonium hydroxide solution (0.5 mL) and water (0.1 mL) were added, and the reaction mixture was stirred at room temperature for 15 h. After evaporation, the residue was purified by silica gel column chromatography (CH₂Cl₂/CH₃OH = 10:1) to give 6a (51 mg, 69%) as a white solid: mp 211–212 °C; [α]d²⁵ = −68.4 (c = 0.095, CH₃OH); ¹H NMR (400 MHz, DMSO-d₆) δ 8.91 (pseudo t, J = 6.0 Hz, 1H, D₂O exchangeable); 8.50 (s, 1H), 7.74 (s, 1H), 7.60 (d, J = 7.6 Hz, 1H), 7.34–7.36 (m, 1H), 7.13 (t, J = 7.6 Hz, 1H), 5.79 (d, J = 4.8 Hz, 1H), 5.78 (br s, 1H, D₂O exchangeable), 4.59–4.60 (m, 2H), 4.34–4.36 (m, 1H), 3.63–3.64 (m, 1H), 3.27–3.30 (m, 1H), 2.70 (dd, J = 10.4, 6.0 Hz, 1H), 1.74 (br s, 2H, D₂O exchangeable); ¹³C NMR (100 MHz, DMSO-d₆) δ 154.8, 153.0, 150.0, 141.9, 140.5, 136.0, 135.6, 130.5, 126.8, 118.5, 94.7, 79.0, 63.4, 55.9, 42.5, 34.8; UV (CH₃OH) λ_max 271.5 nm; MS (ESI+) [M + H]⁺ m/z 502.9911; Anal. calcd for C₁₆H₁₆ClIN₆OS: C, 38.22; H, 3.21; N 16.72; found: C, 37.90; H, 2.90; N 17.12; HPLC purity = 98.4%.

((2R,3R,4S)-4-Amino)-2-(6-((3-iodobenzyl)amino)-9H-purin-9-yl) tetrahydrothiophen-3-ol (6b).

Compound 6b (99%) was obtained as a white solid from 15b in a similar manner to that described for the preparation of 6a: mp 199–201 °C; [α]d²⁵ = −1.50 (c = 0.04, CH₃OH); ¹H NMR (400 MHz, CD₃OD) δ 8.45 (s, 1H), 8.22 (s, 1H), 7.72 (s, 1H), 7.58 (d, J = 8.0 Hz, 1H), 7.35 (d, J = 7.3 Hz, 1H), 7.11 (t, J = 15.2, 7.6 Hz, 1H), 5.88 (d, J = 4.8 Hz, 1H), 5.76 (s, 1H), 4.65 (s, 1H), 4.39–4.42 (m, 1H), 3.66–3.67 (m, 1H), 3.30 (s, 1H), 2.71 (dd, J = 16.0, 4.0 Hz, 1H); ¹³C NMR (100 MHz, CD₃OD) δ 152.9, 143.1, 140.4, 136.1, 135.9, 131.1, 127.2, 95.1, 79.1, 64.1, 56.2, 42.7, 38.7, 35.0; UV (CH₃OH) λ_max 268.0 nm; MS (ESI+) [M + H]⁺ m/z 469.0285; Anal. calcd for C₁₆H₁₇IN₆OS: C, 41.03; H, 3.66; N, 17.95; found: C, 41.43; H, 3.26; N, 18.12; HPLC purity = 97.6%.

((2R,3R,4S)-4-Amino)-2-(6-(methylamino)-9H-purin-9-yl)-tetrahydrothiophen-3-ol (6c).

Compound 6c (97%) was obtained as a white solid from 16c in a similar manner to that described for the preparation of 6a: mp 200 °C (decomp); [α]d²² = −8.36 (c = 0.64 in CH₃OH); ¹H NMR (400 MHz, CD₃OD) δ 8.34 (s, 1H), 8.24 (s, 1H), 6.00 (d, J = 6.4 Hz, 1H), 4.65 (dd, J = 9.6, 3.6, 1H), 4.45 (dd, J = 11.2, 4.0, 1H), 3.52 (dd, J = 11.2, 4.4, 1H), 3.10 (s, 3H), 2.94 (dd, J = 10.8, 3.6 Hz, 1H); ¹³C NMR (100 MHz, CD₃OD) δ 153.8, 141.2, 109.5, 80.8, 74.4, 63.9, 35.2; UV (CH₃OH) λ_max 266.5 nm; MS (ESI+) [M]⁺ m/z 468.0212; Anal. calcd for C₁₀H₁₄N₆OS: C, 45.10; H, 5.30; N, 31.56; found: C, 45.11; H, 5.67; N, 31.96; HPLC purity = 95.3%.

Pharmacology.

Radioligand binding assays were carried out using membrane preparations from Chinese hamster ovary (CHO) cells expressing the human A₁ or A₃AR, or from the human embryonic kidney cells (HEK-293) expressing the human A_2AAR.

Cell Culture and Membrane Preparation.

CHO cells stably expressing either the recombinant human A₁ or human A₃AR and HEK-293 cells stably expressing the human A_2AAR were cultured in Dulbecco’s modified Eagle’s medium (DMEM) and F12 (1:1) supplemented with 10% fetal bovine serum, 100 units/mL penicillin, 100 μg/mL streptomycin, and 2 μmol/mL glutamine. In addition, 800 μg/mL geneticin and 500 μg/mL hygromycin were added to the A_2AAR media and the A₁ and A₃ARs media, respectively. After harvesting the cells, they were homogenized for 10 s with an electric homogenizer, pipetted into 1 mL vials, and stored at −80 °C until the binding experiments were conducted. The concentration of protein was determined using a BCA Protein Assay Kit from Pierce Biotechnology (Rockford, IL).³⁷

Radioligand Membrane Binding Studies.

Radioligand binding assays with A₁, A_2A, and A₃ARs were performed according to the procedures described previously.^38–40 Briefly, each tube contained 100 μL of membrane suspension (20 μg of protein), 50 μL of a stock solution of agonist radioligand, and 50 μL of increasing concentrations of the test ligands in Tris-HCl buffer (50 mM, pH 7.5) containing 10 mM MgCl₂. Nonspecific binding was determined using a final concentration of 10 μM NECA (5′-N-ethylcarboxamidoadenosine), a nonspecific agonist, diluted with the buffer. The mixtures were incubated at 25 °C for 60 min. Binding reactions were terminated by filtration through Whatman GF/B filters under reduced pressure using an MT-24 cell harvester (Brandell, Gaithersburg, MD). The filters were washed three times with 5 mL of 50 mM ice-cold Tris-HCl buffer (pH 7.5). The radioactive agonists [³H]R-PIA (R-N⁶-phenylisopropyladenosine) and [³H]CGS21680 (2-p-(2-carboxyethyl)phenethylamino-5′-N-ethylcarboxamidoadenosine) were used for the A₁ and A_2AAR assays, respectively, while [¹²⁵I]AB-MECA (2-[p-(2-carboxyethyl)-phenyl-ethylamino]-5′-N-ethylcarboxamidoadenosine) was used for the A₃AR assay. All of the filters were washed three times with Tris-HCl, pH 7.5. Filters for A₁ and A_2AAR binding were placed in scintillation vials containing 5 mL of Hydrofluor scintillation buffer and counted using a PerkinElmer Tricarb 2810TR Liquid Scintillation Analyzer. Filters for A₃AR binding were counted using a PerkinElmer Cobra II γ-counter.

Cyclic Adenosine Monophosphate (AMP) Assay.

A cAMP assay at the human A₃AR expressed in Chinese hamster ovary (CHO) cells was performed according to the following procedure based on a competitive protein binding method.²⁷ After harvesting the CHO cells following trypsinization, the cells were centrifuged and resuspended in medium and then plated in 24-well plates using 0.5 mL medium/cell. Following 24 h, the medium was removed and the cells were washed three times with 1 mL of DMEM, containing 50 mM N-(2-hydroxyethyl)-piperazine-N′-2-ethanesulfonic acid (HEPES), pH 7.4. Phosphodies-terase inhibitor rolipram (10 μM), adenosine deaminase (3 units/mL), and finally a nucleoside derivative (test compound) were added. After 45 min, forskolin (10 μM) was added to the medium and the incubation was continued for an additional 15 min. The supernatant was removed to terminate the reaction, and the cells were lysed upon the addition of 200 μL of 0.1 M ice-cold HCl. The cell lysate was resuspended in aliquots and could be stored at −20 °C. For determination of cAMP production, protein kinase A (PKA) was incubated for 150 min at 4 °C with [³H]cAMP (2 nM) in K₂HPO₄/ethylenediaminetetraacetic acid (EDTA) buffer (K₂HPO₄, 150 mM; EDTA, 10 mM), 20 μL of the cell lysate, and 30 μL of 0.1 M HCl or 50 μL of cAMP solution (0–16 pmol/200 μL for standard curve). Bound radioactivity was separated by rapid filtration over a Whatman GF/C filter using a Brandel cell harvester (Brandel, Gaithersburg, MD). The filter was washed once with cold buffer. Bound radioactivity was measured by liquid scintillation spectrometry.

Data Analysis.

Binding and functional parameters were calculated using the Prism 5.0 software (GraphPad, San Diego, CA). IC₅₀ values obtained from competition curves were converted to K_i values using the Cheng–Prusoff equation.⁴¹ Data were expressed as mean ± standard error.

Computational Studies.

Based on the sequence alignment using CLUSTAL W, and manual refinement focused on the highly conserved TM regions (Figure S1), homology models were generated. The most appropriate structures with the lower probability density function (PDF), total energy, and larger binding pocket size were selected and refined by energy minimization. For more accurate docking simulation, we applied a novel sampling strategy using elastic network-normal mode analysis (EN-NMA), which is an effective method for fast generation of multiple receptor conformations.³¹ Using EN-NMA, 100 conformations were constructed for each homology model, and they were clustered based on their binding pocket RMSD with a cutoff of 0.6 Å. Three conformations for agonist-based model and 16 for antagonist-based model remained, and they were minimized to obtain the energetically relaxed structures (Figure 2).

We then examined the appropriateness of the models based on the docking results of the representative nucleoside agonists and antagonists which have high potency and ca. 50-fold higher selectivity to hA₃AR than other subtypes (Table S1). For additional consideration of rotameric states of the key residues for ligand binding, the side chains of ten residues (i.e., Leu90, Leu91, Thr94, Phe168, Met177, Trp243, Leu246, Asn250, Ser271, and His272) were set to be flexible during the docking process. Finally, the best models that produced reasonable binding modes for all of the representative agonists and antagonists were selected (Table S2) and further optimized through the MD simulation.

Using the best agonist- and antagonist-based hA₃AR homology models, we performed the docking study in two steps to allow full protein flexibility. First, the docking was carried out with side-chain flexibility permitted for the 10 residues important for ligand binding. Then, the resulting docked conformations were used for the following induced-fit docking, which fully allows flexibility of the entire binding site. The results showed reasonable binding modes when the agonists and antagonists were docked to the agonist-based and antagonist-based models, respectively.

The docked structures were further optimized through MD simulation. The Membrane Builder module⁴² in CHARMM-GUI (www.charmm-gui.org)⁴³ was used to insert the receptor structure into the lipid bilayer system. The MD simulation is performed using Gromacs v.2020.2⁴⁴ with CHARMM36 force field.⁴⁵ The force field for the ligand molecule was generated by CHARMM general force field (CGenFF) program.⁴⁶ The preequilibration process was conducted under NVT (constant particle number, volume, and temperature) for the 10 ns simulation time, followed by 40 ns simulation under NPT (constant particle number, pressure, and temperature) ensemble for system equilibration. Then, the production run was performed under NPT for 300 ns. A time step of 2 fs was used in all simulations, and the data were sampled every 0.1 ns for the analysis.

For more detailed theoretical explanation, we adopted network analysis based on graph theory to the protein structure of hA₃AR bound with each ligand. The residue interaction network was constructed by representing each amino acid residue as a single node, and the betweenness centrality (C_B) values were calculated for each node. The residues with the increased betweenness centrality (|ΔC_B| > 0.02) are mapped to the structure as shown in Figure 6. All computations except MD simulation were carried out on Intel XEON Gold 6254 Processor (3.1 GHz, 18-core) with Linux Cent OS release 7.7. MD simulation was performed using the Korea Institute of Science and Technology Information (KISTI) supercomputer. More detailed procedures of the computational studies are described in the Supporting Information.

ABBREVIATIONS

AR

adenosine receptor

EN-NMA

elastic network-normal mode analysis

hA₃AR

human A₃ adenosine receptor

MD

molecular dynamics

MRC

multiple receptor conformation

TM

transmembrane

[bmim]⁺BF₄⁻

1-butyl-3-methylimidazolium tetrafluoroborate