X‐Ray Crystallography and Free Energy Calculations Reveal the Binding Mechanism of A2A Adenosine Receptor Antagonists

Receptor Binding X-Ray Crystallography and Free Energy Calculations Reveal the Binding Mechanism of A2A Adenosine Receptor Antagonists** Willem Jespers,Gr8gory Verdon, Jhonny Azuaje,Maria Majellaro,Henrik Ker-nen, Xerardo Garc&a-Mera, Miles Congreve,Francesca Deflorian, Chris de Graaf,Andrei Zhukov, Andrew S. Dor8,Jonathan S. Mason, Johan cqvist, Robert M. Cooke,Eddy Sotelo,and Hugo Guti8rrez-de-Ter#n* Abstract: We present arobust protocol based on iterations of free energy perturbation (FEP) calculations,chemical synthesis,biophysical mapping and X-raycrystallography to reveal the binding mode of an antagonist series to the A2A adenosine receptor (AR). Eight A2AAR binding site mutations from biophysical mapping experiments were initially analyzed with sidechain FEP simulations,performed on alternate binding modes.The results distinctively supported one binding mode,which was subsequently used to design new chromone derivatives.Their affinities for the A2AAR were experimentally determined and investigated through acycle of ligand-FEP calculations,validating the binding orientation of the different chemical substituents proposed. Subsequent X-ray crystallography of the A2AAR with alow and ahigh affinity chromone derivative confirmed the predicted binding orientation. The new molecules and structures here reported were driven by free energy calculations,and provide new insights on antagonist binding to the A2AAR, an emerging target in immunooncology. Introduction Computational estimation of shifts in binding free energy, associated with ligand modifications or point mutations in the receptor macromolecule,can provide the missing link between the structure of aprotein-ligand complex and apanel of experimental binding affinities.Rigorous free energy perturbation (FEP) methods have been used for decades to understand the structure-affinity relationships (SAR) around agiven chemical scaffold, and recent advances now allow the use of this technique routinely in ligand design projects.[1] The same methodology can be used to analyze this problem from acomplementary perspective,that is to estimate the gain or loss in binding free energy from site-directed mutagenesis (SDM) data. Theidea of in silico mutagenesis,initially introduced almost three decades ago by Kollman to study the binding and catalysis of subtilisin,[2] was recently implemented in computational pipelines that pursue asystematic characterization of the effect on point mutations on for example, ligand-binding or protein stability.[3–5] Thecombination of both ligand and residue FEP simulations can provide afull energetic landscape of the molecular interactions governing protein-ligand binding,which underlies the design of two complementary protocols in our lab,namely QligFEP[6] and QresFEP,[3] integrated in the molecular dynamics (MD) software package Q.[7,8] One area where this approach is particularly promising is the design of ligands for G-protein-coupled receptors (GPCRs), asuperfamily of seven-transmembrane (7TM) cellular receptors[9] that mediate the therapeutic effects of about 30%ofall marketed drugs.[10] There is alarge amount of SAR and SDM data available for these receptors,which can be combined with the increasing growth of structural knowledge of many GPCR targets.The first integrated approach of ligand and residue FEP simulations was published by Boukharta et al. to characterize antagonist binding to the Y1neuropeptide receptor,[11] which we later expanded to other GPCR families,including the related neuropeptide receptor Y2,[12] the orphan receptor GPR139[13] and several members of the family of adenosine receptors.[14–16] Among [*] W. Jespers, Dr.H.Ker-nen, Prof. J. bqvist, Dr.H.Guti8rrez-de-Ter#n Department of Cell and Molecular Biology,Uppsala University, BMC, Biomedical Center Box 596, Uppsala (Sweden) E-mail:h[email protected] Dr.G.Verdon, Dr.M.Congreve, Dr.F.Deflorian, Dr.C.deGraaf, Dr.A.Zhukov,Dr. A. S. Dor8,Dr. J. S. Mason, Dr.R.M.Cooke Sosei Heptares Steinmetz Granta Park, Great Abington, Cambridge CB21 6DG (UK) Dr.J.Azuaje, Dr.M.Majellaro, Prof. X. Garc&a-Mera, Prof. E. Sotelo Departament of Organic Chemistry,Faculty of Farmacy,Universidade de Santiago de Compostela (Spain) Dr.J.Azuaje, Dr.M.Majellaro, Prof. E. Sotelo Centro Singular de InvestigaciknenQu&mica Biolkxica yMateriais Moleculares (CIQUS), Universidade de Santiago de Compostela (Spain) Dr.H.Ker-nen Present address:H.Lundbeck A/S Ottiliavej 9, 2500 Valby (Denmark) [**] Aprevious version of this manuscript has been deposited on apreprint server (https://doi.org/10.26434/chemrxiv.11444877.v1). Supporting information and the ORCID identification number(s) for the author(s) of this article can be found under: https://doi.org/10.1002/anie.202003788. T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA. This is an open access article under the terms of the Creative Commons AttributionLicense, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. A ngewandte Chemie Research Articles How to cite: Angew.Chem. Int. Ed. 2020,59,16536–16543 International Edition: doi.org/10.1002/anie.202003788 German Edition: doi.org/10.1002/ange.202003788 16536 T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA, Weinheim Angew.Chem.Int. Ed. 2020,59,16536 –16543 the latter, the adenosine A2A receptor (A2AAR) was one of the first GPCRs to be crystallized[17] and today stands out as one of the better characterized GPCRs from astructural perspective.Several structures of the inactive and active forms of the receptor have been solved within the last decade,and the integration of the available experimental data has strongly aided ligand design programs for this receptor.[18,19] Many A2AAR antagonists have been developed targeting anumber of pathologies,[20–22] including recent clinical candidates in immuno-oncology.Anumber of these antagonists have been co-crystallized with the A2AAR, providing unique structural information which, in combination with the extensive SDM data available,[23] allow envisaging ligand binding mechanisms and structure-based drug design (SBDD) programs of antagonist molecules.[18] However,suboptimal properties of traditional scaffolds,such as poor pharmacokinetics and low selectivity profiles,motivate the search of novel chemical entities as A2AAR antagonists,[24] frequently through high throughput screening (HTS) campaigns.Inthese cases,itisnot common to obtain acrystal structure of the receptor-ligand complex, which can hamper further hit to lead optimization. Instead, approximate binding modes are often inferred from the experimental data extracted from SAR of ligand series and SDM data,[23] which can be complemented by computational models of the protein-ligand complex.[25] Biophysical Mapping (BPM), is an integrated approach that has been used with success in antagonist design programs on the A2AAR and other GPCRs.[27–30] Here,the binding affinity of aligand series is evaluated via surface plasmon resonance (SPR) on apanel of mutant receptors,each bearing asingle-point mutation within the putative binding site.[27] Theresulting matrix of binding affinity shifts from wild type (WT) affinities combined with SAR data and mapped to areceptor-ligand model provide further insights in the determinants of binding of the scaffold. Thefirst application of this technique was based around the co-crystallized A2AAR antagonist ZM241385 in combination with 8receptor mutants (see Figure 1A).[22] These mutations involve residues in direct contact with the ligand, such as N2536.55,L853.33,M1775.38, N1815.42 and I662.64 (Ballesteros Weinstein numbering[31] in superscript), as well as residues not directly in contact with the ligand, namely S2777.42,Y2717.36 and L167EL2 (see Figure 1). This approach was extended for 1,2,4-triazines as A2AAR antagonists,[32] and the binding mode was later confirmed by X-ray crystallography (see Figure 1B).[29] In the same HTS campaign, aseries of chromones were identified as anovel family of A2AAR antagonists,[32] and consecutively optimized to yield the potent and selective Chromone 14 (see Figure 2).[30] At that point, the leadoptimization program was successful in improving the affinity of the initial HTS hit, while not focusing on pharmacokinetic optimization (i.e., the most potent compound Chromone 14 contains ametabolically unstable ester group). Interestingly, this structure-based optimization was guided by the interpretation of the BPM data and acomputational model of the complex generated by docking.This led to the proposal of two putative binding modes compatible with the BPM data. Even if the SAR of the generated series seemed to favor one of them, the question remained open due to the lack of an X-ray structure with any of these compounds in the original study.[30,32] Here,weinitially examine the SPR data available for these A2AAR antagonist families through arecently developed in silico mutagenesis tool based on free energy perturbation (FEP) simulations.[11,14,15] Theresults pointed to aunified binding mode of the chromone series,which is here used as abasis for the design, synthesis and pharmacological evaluation of an extended series of compounds aimed to further explain the underlying SAR of chromones as A2AAR antagonists.Finally,experimental structures of two chromone-A2AAR complexes were solved which confirmed the binding mode hypothesis from the computational studies. Results and Discussion Free Energy Perturbation Calculations on Existing BPM Data TheBPM data obtained for the three chemotypes of A2AAR antagonists were collected from reference [27],and relative binding free energy changes between mutant and WT receptor were calculated from Kivalues (see Table 1, DDGexp bind). Thereafter,relative binding free energies for each ligand were calculated based on 3D models of the receptorligand complexes (DDGcalc bind,Table 1). In the case of ZM241385, the model was directly extracted from the highresolution structure with the A2AAR (see Figure 1), and the curated receptor model (see Supporting Information, Methods) was used throughout this work. Theresults show excellent agreement with the experimental data, with amean absolute error (MAE) of 0.41 kcalmol@1and acorrelation coefficient of R2=0.94 (Table 1). Next, the same set of BPM mutations was analyzed for triazine 4g.Inthis case,the starting configuration of the complex was obtained by Figure 1. Binding mode and chemical structures of antagonists ZM241385 (A, crystal structure 4EIY[26])and triazine 4b (B). The experimental pose of the triazine (cyan) was superimposed on the same crystal structure of the receptorshown in panel A(ribbons). Both compounds had been characterizedbyBPM (residues labelled and depicted in gray sticks). Receptor-ligand hydrogen bonds are depicted as magenta lines. A ngewandte Chemie Research Articles 16537Angew.Chem. Int.Ed. 2020,59,16536 –16543 T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA, Weinheim www.angewandte.org aligning the triazine 4g-A2AAR complex (3UZA) to the curated model described before,retaining only the coordinates of the ligand (Figure 1B).[33] Thecalculated free energies (DDGcalc bind,Table 1) were in general good agreement with the experimental data (MAE=0.94 kcalmol@1;R 2= 0.66). In contrast to the previous case,two mutations (Met1775.38 and Asn1815.42)show qualitative discrepancy with the experimental data for this ligand. Subsequent analysis of the MD trajectories suggests that this is likely due to asuboptimal representation of the (water mediated) H-bond network upon mutation, which has previously been recognized as achallenging factor for this particular scaffold.[34,35] Once the QresFEP protocol was validated to reproduce the BPM data on experimentally known structures,wemoved on to the chromone scaffold, for which there is analogous BPM data, but no crystal structure available.Here,two reasonable binding modes were generated by molecular docking of Chromone 14.The two poses,denoted as Aand B(see Figure 2), form at least one hydrogen bond with Asn2536.55 (a highly conserved interaction in AR ligand recognition)[23] and are related by asymmetry axis along the bicycliccore of the chromone scaffold. According to the docking score function used, both poses were energetically equivalent (@9.30 vs. @9.17 for pose Aand Brespectively) making it difficult at this point to discern the correct one,in line with the previous binding hypothesis by Andrews et al.[30,32] In pose A(Figure 2A), Asn2536.55 forms an H-bond with the nitrogen in the 4-methylthiazole group,leaving the carbonyl group potentially exposed to the internal water network stabilized by residues in TM7, as observed in the A2AAR crystal structure with ZM241385[26] (see Supporting Information, Figure S1). Conversely,inpose B(Figure 2B) this carbonyl forms an additional hydrogen bond with N2536.55.Wehypothesized that in principle,the symmetry axis would allow exchanging substituents at R6and R7 between the binding poses,something that we would explore later in this work (see below). Our strategy to select the most reliable binding mode was to compute the effect on ligand binding of the 8mutations from the BPM panel[27] for each pose,and compare the results with the experimental values.The results (Figure 2C,Table 1 and Supporting Information, Table S1) highlight pose Aas the binding mode with the best correlation to experimental data. AMAE of 0.50 kcalmol@1was observed for this pose, which is comparable to the results obtained for the cocrystallized antagonist ZM241385.Inaddition, the correlation coefficient of R2=0.74 falls between those observed for ZM241385 and triazine 4g.Conversely,the corresponding values calculated on pose Bare much higher (MAE= 1.53 kcalmol@1)and the correlation is completely lost (R2= 0.03). Additionally,the computed loss of binding affinity upon the N2536.55Amutation is closer to experiment in pose Athan in pose B, indicating that the additional H-bond between Asn2536.55 and the carbonyl in pose Bwould not contribute to more favorable binding free energies.From the MD simulations,wefurther observed that the binding modes are not related by asymmetry axis as initially hypothesized, and the substituents at positions R6and R7are exploring different positions in the binding site,making them no longer readily interchangeable. Figure 2. Putative binding modes A(green) and B(magenta)of Chromone 14 to the A2AAR (H-bonds in magenta). C) Experimental and calculated changes in binding free energies for each mutation in the BPM. The error bars correspondt ot he s.e.m. of the replica calculations for the calculated values, or are adjusted to the reported value of 0.1 pKDunit in the case of experimental data.[27] Table 1: Comparison between experimental and calculated relative binding free energies (DDGbind in kcalmol@1)for A2AAR mutants. Mutant[a] ZM241385Triazine 4g DDGexp bind[b] DDGcalc bind DDGexp bind[b] DDGcalc bind I66A2.64 0.14 0.83:0.34 0.41 1.94:0.34 L85A3.33 2.45 3.30:0.41 1.09 1.65:0.37 L167A5.28 0.00 0.60:0.31 @0.14 @0.39:0.36 M177A5.38 0.14 @0.09:0.44 @0.27 1.66:0.49 N181A5.42 1.23 1.47:0.57 0.82 @0.63:0.55 N253A6.55 +5.86[c] 5.81:0.57 +4.36[c] 5.64:0.56 Y271A7.36 1.09 0.84:0.74 0.41 @0.1:0.68 [a] Data for the mutant receptor constructs reported in reference [27]. [b] Experimentalrelative binding free energies were calculated from KD values as DDGexp bind ¼RTln Kmut D= KWT D .- with experimental errors in all cases reported as approximately 0.1 pKDunit, that is, less than 0.1 kcalmol@1.[27] [c] Binding affinity of the ligand to the (mutant) receptor was lower than the experimental threshold(pKD<5inall cases). Errors are standard error of the mean (s.e.m.) over atotal of 10 replicates. A ngewandte Chemie Research Articles 16538 www.angewandte.org T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA, Weinheim Angew.Chem.Int. Ed. 2020,59,16536–16543 Design, Synthesis and Pharmacological Evaluation of Chromone Derivatives On the basis of the binding mode hypothesis assessed by the FEP calculations,wedesigned asmall collection of chromone derivatives (Scheme 1), aiming to systematically explore two different variables on A2AAR affinity:methylation at position 2(series 5)and the aforementioned effect of switching the oxygenated function (@OH or @OCOMe) from the original position 7(series 4)toposition 6(series 8), swapping the alkyl substitution (Pr/H) accordingly.Acollection of 11 molecules was prepared as depicted in synthetic Scheme 1. Structure and synthetic pathways employed to assembly chromones 4,5and 8. A ngewandte Chemie Research Articles 16539Angew.Chem. Int.Ed. 2020,59,16536 –16543 T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA, Weinheim www.angewandte.org Scheme 1. Briefly,the key 2-hydroxyphenyl ketones 3were obtained by either the Hoesch method[30] or Claisen condensation,[36] employing phenols 1or esters 6and thiazole derivatives 2a and 2b as reactive precursors,respectively. Treatment of ketones 3with orthoesters (formate or acetate) enabled the efficient chromone core formation,[37] thus defining the substituent pattern at position 2(HorMe). Finally,the required acetates were prepared by reaction of the corresponding phenols with acetyl chloride.The full synthetic methodology is provided in the Supporting Information. Theaffinity of these compounds for the WT A2AAR was then evaluated with the same SPR assay used to generate the BPM data, using the A2AAR-STAR2 construct (see Supporting Information, Methods).[27] Thedata confirmed that methylation at position 2(compounds 5a–d)isgenerally unfavorable for binding,compared to the parent series (compounds 4a–d). This is in agreement with previous work by Andrews et al.,[30] where methylation of the low affinity compound 4a (corresponding to chromone 8asreported by Andrews et al.,[30] with pKi=5.7) resulted in arelative decrease in affinity of 0.6 log units (Chromone 12, pKi= 5.1). We observe asimilar difference for this pair of compounds (Table 2, DpKD=0.85), while this effect is amplified in the case of the high affinity precursor (see 4d as compared to 5d,DpKD=2.7). More surprising was the observation of asimilar effect when the substituents between R6and R7are swapped (compounds 8a–d), which again was most pronounced for the high affinity compound 4d (see Table 2), with adrop of affinity of 2.7 log units for the corresponding methylated derivative 8d. Computational Evaluation of the Proposed Binding Modes From the pharmacological data on the expanded series of chromones and the initial MD simulations,itappears that the two binding poses differ more than just purely on the rotation axis of the chromone scaffold. Thefollowing step was then to investigate whether all compounds in the series would adopt the preferred binding mode A, or if binding mode Bisaccessible by some of the compounds,depending on the pattern of substitutions.Such ahypothesis was experimentally observed for caffeine,which in contrast to other xanthine derivatives presents adual binding mode to the A2AAR (see Figure 3A). Indeed, an isoenergetic dual binding mode for caffeine was previously hypothesized on the basis of free energy calculations,[38] before experimental observation of dual-occupancycrystal structures in complex with the A2AAR.[39,40] Conversely,the X-ray structure of the N7-demethylated analogue theophylline shows that this molecule adopts only one of these binding modes,where the acidic hydrogen at position N7makes an additional H-bond contact with N2536.55 (Figure 3). We thus evaluated whether our recently developed dual-topology protocol QligFEP was suitable to capture this different behavior, by ad irect estimation of relative binding free energies between the two poses,inanalogy to the efficiency of this protocol to compare topologically unrelated ligands (e.g., scaffold hopping).[6] Theresults for the A2A-xanthine system shown in Table 3show that this is the case:the negligible calculated free energy difference between the two poses for caffeine is in line with the equally populated dual binding mode in the crystal structure with the A2AAR (PDB code 5MZP,Figure 3A). In contrast, the single binding mode observed in the crystal structure of theophylline (blue color in Figure 3A)isenergetically favored by 1.6 [email protected] application of the same strategy in the generated chromone series showed asimilar energy gap between the two binding poses considered for the simplest chromone in our series (4a,Figure 3B), with pose Abeing 1.7 kcalmol@1more favorable than pose B. Notably,this energy gap increases significantly for the highest affinity compound 4d,suggestive Table 2: SPR affinity data for the series of Chromone derivatives synthesized in this work. Compound Substituents pKD[a] R2R5R6R7 4a[b] HH HH5.95 4b[b] HHC 3H7H6.20 4c HCH 3HH5.80 4d[b] (Chromone 14)H HC 3H7COCH38.60 5a[b] CH3HH H5.10 5b CH3HC 3H7H5.42 5c CH3CH3HH5.36 5d CH3HC 3H7COCH35.90 8a HH HH5.50 8b HH HC 3H75.70 8d HHCOCH3C3H75.90 [a] Experimental errors in all cases reported as approximately 0.1 pKD unit, that is, less than 0.1 kcalmol@1.[b] Compounds previously reported in ref.[30]. Figure 3. A) Dual binding mode of caffeine,asextracted from the A2AAR crystal structure with the A2AAR (PDB code 5MZP). Colour code is green (binding mode A) and magenta (binding mode B). B) modelled binding modes of Chromone 4a,following the same colouring Scheme as in panel A. A ngewandte Chemie Research Articles 16540 www.angewandte.org T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA, Weinheim Angew.Chem.Int. Ed. 2020,59,16536–16543 of an optimal anchoring of the R6acetate and R7n-propyl substituents in pose A(Table 3). We consequently retained the selected binding pose for each scaffold (i.e.,xanthine and chromone) and computed the affinity shifts observed upon methylation. Thecalculations reproduced the experimental decrease in affinity observed for caffeine,ascompared to the N7demethylated analogue theophylline (Table 3). Similarly,the corresponding FEP calculations capture the experimentally observed negative effect on A2A affinity due to methylation on R2in the original chromones series 4(see series 5,see Table 2).[30] Thetwo pair comparisons performed (4a !5aand 4d !5d,see Table 3) qualitatively capture this effect, which is experimentally more pronounced in the last case.Notably,acomparison of the endpoint configurations of the 4d !5dFEP transformation provide the structural reason of this effect:toavoid the induced steric hindrance of the new methyl group with Asn2536.55,the core scaffold of 5dmoved approximately 1.5 c away from the original pose Aof4d while maintaining the interaction between the nitrogen in the 4-methylthiazole group and the key asparagine residue. X-Ray Crystallography In an effort to unequivocally reveal the binding mode of the chromone scaffold, the A2AAR-STAR2 construct[27] was crystallized with compound 4d(PDB code:6ZDR) and its 2methyl analogue 5d(PDB code:6ZDV). Thestructures were obtained following the in meso soaking approach (see Supporting Information, Methods), and could be refined down to aresolution of 1.92 and 2.13 crespectively.Statistics for data collection and refinement are given in the Supporting Information, Table S2. Theoverall structure of the A2AAR receptor is highly similar to previously solved structures with other antagonists (see Figure 4A), with an RMSD of 0.46 c for the Catrace as compared to the ZM241385 structure (PDB 4EIY,calculated using PyMOL[42]). Thetwo structures show clear positive omit density at 1sfor the presence of the chromone compounds in the orthosteric binding site (Figure 4B and C), in both cases adopting binding mode A (Figure 4D and E). All residues in the binding site show aconserved rotameric state between the two structures and the A2AAR-ZM241385 complex. Theonly exception is Y2717.36,which slightly rotates outward served to accommodate the alkyl tail of the chromone at R6(see Supporting Information, Figure S2A), an induced-fit effect that was previously observed for other A2AAR ligands.[33] Thestructure with the high affinity ligand 4d shows outstanding agreement with the computational model (see Figure 5A) with an RMSD of 0.67 cbetween the docked and experimental poses,and is indeed highly similar to previously reported binding modes of the compound.[30,32] Thecore of the scaffold is anchored by an H-bond between the nitrogen in the 4-methylthiazole ring (coplanar with the chromone scaffold) and N2536.55,and complemented by a p-pstacking with F168EL2,inline with the binding pattern observed for many other A2AAR antagonists.[20] Theester at position R7is stabilizing the Glu1695.30–His2647.29 ionic pair, which closes the EL2-EL3 interface,while the propyl at R6makes contacts with Tyr2717.53 and Met2707.52.Given the good resolution of the structure,wecould observe awell-defined water network of the interface between the carbonyl moiety of the chromone core and the receptor (Figure 4D). While most water positions are conserved as compared to the A2AARZM241385 complex, ligand 4d displaces anumber of water molecules observed in the first hydration shell around that ligand (see Supporting Information, Figure 2B). Some of these waters were previously associated to ahigh-energy or “unhappy” state,[30] which could partially explain the high affinity of this particular compound. TheA 2AAR-5b complex (Figure 4E)further shows that methylation at position R2displaces the chromone core by 1.79 c,inaccordance with observations from the FEP simulations (Table 3and Figure 5B,RMSD =1.24 cbetween docked and experimental ligand configuration). The4methylthiazole ring of compound 5d remains almost coplanar with the chromone scaffold (torsional angle of 16 degrees), and the H-bond interaction between the nitrogen in this ring and Asn2536.55 is maintained. Indeed, the methylation does not disrupt the position of Asn2536.55,which remains linked to Glu1695.30 via aconserved water molecule. Similarly,the water molecule deeper in the binding crevice remains,mediating acommon anchoring point between the carbonyl group of the two ligands and His2767.43.However, the relative displacement of the chromone core on 5dsituates the substituents at R6and R7in sub-optimal pockets (Figure 4D), displacing two of the water molecules described for compound 4d,while anew water molecule in the 5dcomplex replaces the role of the acetate in 4d (see Figure 5D)inthe stabilization of the Glu1695.30–His2647.29 ionic pair. This could explain why the inclusion of an acetyl moiety in compound 5d could not recover the affinity of the parent 5b(DpKD=0.48, Table 2) as much as it was observed for the 4b/4d pair Table 3: Calculated free energy differencebetween two alternative poses for A2AAR antagonists. Pose comparison A !BLigand perturbation Ligand 1(H) !Ligand 2(CH3) Ligand 1 DDGcalc [kcalmol@1]Ligand 2 DDGexp [kcalmol@1]DDGcalc [kcalmol@1][a] Caffeine 0.47:0.49 –– – Theophylline 1.59:0.87 Caffeine 0.6[b] 0.31:0.32 4a 1.65:0.93 5a 1.16:0.1[c] 2.06:0.72 4d 8.76:0.81 5d 3.68:0.1[d] 1.66:1.09 [a] Calculations performed in the selected pose A(see text). [b] DGbind (caffeine—theophylline), extracted from ChEMBL.[41] [c] DGbind (5a–4a)and [d] DGbind (5d–4d), data from Table 2. A ngewandte Chemie Research Articles 16541Angew.Chem. Int.Ed. 2020,59,16536 –16543 T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA, Weinheim www.angewandte.org (DpKD=2.6). Finally,the crystal structures offered the opportunity to revise the 4d !5d FEP transformation, which was somewhat underestimated (see Table 3). However, the calculated value using the starting pose from the experimental coordinates did not change significantly, (DDGcalc =1.54:1.22 kcalmol@1)indicating that the docking pose was an accurate enough starting point for the FEP calculations. Conclusion We describe arobust workflow to iteratively improve receptor-ligand binding models,based on mapping of available experimental data onto structural information via free energy calculations.Weapplied this protocol to provide new insights in the binding mode of arecent series of A2AAR antagonists.Aninitial binding mode hypothesis was generated based on the exploration of BPM and SAR data of the original chromone series using FEP.This constituted the basis for the design, synthesis and evaluation of an expanded series of chromone derivatives.The experimental results were conveniently interpreted with the aid of asecond iteration of FEP calculations,which reinforced the binding hypothesis. Finally,X-ray crystallography experimentally confirmed this binding mode,supporting the rational design of these compounds.These structures,combined with the FEP calculations,provide structural and energetic insights in the determinants of high affinity binding of the chromone scaffold series.Weexpect the presented workflow to be of general applicability in structure-based drug design, particularly in the case of GPCRs where structures of receptor-ligand complexes are increasingly available. Figure 4. A) Crystal structures of the A2AAR and compound 4d (PDB code:6ZDR), ligand shown in sticks and sodium ion shown as asphere. Electron densitiesofchromones 4d (B) and 5d (C;PDB code:6ZDV). Omit maps are 2Fo@Fcat 1sigma (light blue mesh) and Fo-Fcat3sigma (green mesh). Binding mode of compound 4d (D) and 5d (E);ligands and the conserved residue N2536.55 shown as sticks, water molecules in red spheres. Figure 5. Crystal structure (orange) and modelled coordinates (cyan) of (A) the highest affinity compound 4d (PDB code:6ZDR) and (B) the methylated derivative 5d (PDB code:6ZDV) with the A2AAR. Hbond interactions are indicated in magenta. A ngewandte Chemie Research Articles 16542 www.angewandte.org T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA, Weinheim Angew.Chem.Int. Ed. 2020,59,16536–16543 Acknowledgements This work wasfinancially supportedbythe SwedishResearch Council(Grant521-2014-2118); Conseller&adeCultura,EducacilneOrdenacilnUniversitariaofthe Galician Government (Grant ED431B2017/70);CentroSingulardeInvestigacilndeGalicia accreditation2016–2019 (Grant ED431G/09), andthe European Regional DevelopmentFund(ERDF). Additional supportfromthe Swedishstrategic research program eSSENCEisacknowledged. Thecomputationswere performedonresources provided by theSwedish National Infrastructure forComputing (SNIC).Thisresearchprogram hasbeendeveloped in theframe of theEuropeanCOSTaction ERNEST (Grant CA 18133) andGLISTEN (Grant CA 1207). Conflict of interest Theauthors declare no conflict of interest. Keywords: adenosine receptors ·biophysical mapping (BPM) · free energy perturbation (FEP) · Gprotein-coupled receptor (GPCR) [1]Z.Cournia, B. Allen, W. Sherman, J. Chem. Inf. Model. 2017,57, 2911–2937. [2]S.N.Rao,U.C.Singh,P.A.Bash, P. A. Kollman, Nature 1987,328, 551–554. [3]W.Jespers,G.V.Isaksen,T.A.H.Andberg, S. Vasile,A.Van Veen, J. cqvist,B.O.Brandsdal,H.Guti8rrez-De-Ter#n, J. Chem. Theory Comput. 2019,15,5461–5473. [4] T. B. Steinbrecher,M.Dahlgren, D. Cappel,T.Lin, L. Wang,G. Krilov,R.Abel,R.Friesner, W. Sherman, J. Chem. Inf. Model. 2015,55,2411–2420. [5]V.Gapsys, S. Michielssens,D.Seeliger,B.L.deGroot, J. Comput. Chem. 2015,36,348–354. [6]W.Jespers, M. Esguerra,J.cqvist, H. Guti8rrez-de-Ter#n, J. Cheminf. 2019,11,1–16. [7] J. Marelius, K. Kolmodin, I. Feierberg,J.cqvist, J. Aqvist, J. Mol. Graphics Modell. 1998,16,213–225. [8]P.Bauer,A.Barrozo,M.Purg, B. A. Amrein,M.Esguerra,P.B. Wilson,D.T.Major,J.Aqvist, S. C. L. Kamerlin, SoftwareX 2018,7, 388–395. [9] A. L. Hopkins,C.R.Groom, Nat.Rev.DrugDiscovery 2002,1, 727–730. [10] A. S. Hauser,M.M.Attwood, M. Rask-Andersen, H. B. Schiçth, D. E. Gloriam, Nat. Rev.Drug Discovery 2017,16,829–842. [11]L.Boukharta, H. Guti8rrez-de-Ter#n, J. cqvist, PLoS Comput. Biol. 2014,10,e1003585. [12] B. Xu,S.Vasile,S.Østergaard, J. F. Paulsson, J. Pruner,J.cqvist, B. S. Wulff,H.Guti8rrez-De-Ter#n, D. Larhammar, Mol. Pharmacol. 2018,93,323–334. [13] A. C. Nøhr,W.Jespers,M.A.Shehata,L.Floryan, V. Isberg,K.B. Andersen, J. cqvist, H. Guti8rrez-de-Ter#n, H. Br-uner-Osborne, D. E. Gloriam, Sci. Rep. 2017,7,1–9. [14] H. Ker-nen, H. Guti8rrez-de-Ter#n, J. cqvist, PLoS One 2014,9, e108492. [15] H. Ker-nen, J. cqvist, H. Guti8rrez-de-Ter#n, Chem. Commun. 2015,51,3522–3525. [16]W.Jespers, A. Oliveira,R.Prieto-D&az, M. Majellaro, J. cqvist, E. Sotelo,H.Guti8rrez-De-Ter#n, Molecules 2017,22,1945. [17]V.-P. Jaakola, M. T. Griffith, M. A. Hanson, V. Cherezov,E.Y.T. Chien,J.R.Lane,A.P.Ijzerman,R.C.Stevens, Science 2008,322, 1211–1217. [18] A. Jazayeri,S.P.Andrews,F.H.Marshall, Chem. Rev. 2017,117, 21–37. [19]H.Guti8rrez-de-Ter#n, J. Sallander,E.Sotelo, Curr.Top.Med. Chem. 2017,17,40–58. [20] B. B. Fredholm, A. P. IJzerman,K.A.Jacobson,J.Linden, C. E. Mgller, Pharmacol. Rev. 2011,63,1–34. [21] C. E. Mgller,K.A.Jacobson, Biochim. Biophys. Acta Biomembr. 2011,1808,1290–1308. [22]J.-F. Chen, H. K. Eltzschig,B.B.Fredholm, Nat.Rev.Drug Discovery 2013,12,265–286. [23] W. Jespers, A. C. Schiedel, L. H. Heitman, R. M. Cooke,L.Kleene, G. J. P. van Westen, D. E. Gloriam,C.E.Mgller,E.Sotelo, H. Guti8rrez-de-Ter#n, Trends Pharmacol.Sci. 2018,39,75–89. [24]C.M.Richardson, R. J. Gillespie,D.S.Williamson,A.M.Jordan, A. Fink, A. R. Knight, D. M. Sellwood,A.Misra, Bioorg.Med. Chem. Lett. 2006,16,5993–5997. [25]V.Lounnas,T.Ritschel, J. Kelder,R.McGuire,R.P.Bywater,N. Foloppe, Comput. Struct.Biotechnol. J. 2013,5,e201302011. [26] W. Liu, E. Chun,A.A.Thompson,P.Chubukov, F. Xu,V.Katritch, G. W. Han, C. B. Roth,L.H.Heitman,A.P.IJzerman, et al., Science 2012,337,232–236. [27] A. Zhukov,S.P.Andrews,J.C.Errey,N.Robertson, B. Tehan, J. S. Mason, F. H. Marshall,M.Weir, M. Congreve, J. Med. Chem. 2011, 54,4312–4323. [28] M. Congreve,C.J.Langmead, J. S. Mason, F. H. Marshall, J. Med. Chem. 2011,54,4283–4311. [29] M. Congreve,S.P.Andrews,A.S.Dore,K.Hollenstein,E.Hurrell, C. J. Langmead, J. S. Mason, I. W. Ng,B.Tehan, A. Zhukov,etal., J. Med.Chem. 2012,55,1898–1903. [30] S. P. Andrews,J.S.Mason, E. Hurrell,M.Congreve, MedChemComm 2014,5,571. [31] J. A. Ballesteros,H.Weinstein, Methods Neurosci. 1995,25,366– 428. [32] C. J. Langmead,S.P.Andrews,M.Congreve, J. C. Errey, E. Hurrell, F. H. Marshall, J. S. Mason, C. M. Richardson, N. Robertson, A. Zhukov,etal., J. Med. Chem. 2012,55,1904–1909. [33] P. Rucktooa,R.K.Y.Cheng,E.Segala,T.Geng, J. C. Errey, G. A. Brown, R. M. Cooke,F.H.Marshall, A. S. Dor8,Sci. Rep. 2018,8, 41. [34] A. Bortolato, B. G. Tehan, M. S. Bodnarchuk, J. W. Essex, J. S. Mason, J. Chem.Inf.Model. 2013,53,1700–1713. [35] A. Bortolato,B.G.Tehan,R.T.Smith, J. S. Mason in Computational Methods for GPCR Drug Discovery. (Eds.: A. Heifetz), MethodsinMolecular Biology,Vol. 1705,Humana Press, New York, 2018,pp. 207–232. [36] K. S. Gudmundsson,B.A.Johns,S.H.Allen, Bioorg.Med. Chem. Lett. 2008,18,1157–1161. [37] M. Deodhar,K.Wood,D.S.Black, N. Kumar, Tetrahedron Lett. 2012,53,6697–6700. [38] Y. Liu, S. K. Burger,P.W.Ayers,E.Vçhringer-Martinez, J. Phys. Chem. B 2011,115,13880–13890. [39] R. K. Y. Cheng, E. Segala,N.Robertson,F.Deflorian,A.S.Dor8, J. C. Errey,C.Fiez-Vandal, F. H. Marshall, R. M. Cooke, Structure 2017,25,1275–1285.e4. [40] A. S. Dor8,N.Robertson,J.C.Errey, I. Ng,K.Hollenstein, B. Tehan, E. Hurrell,K.Bennett,M.Congreve,F.Magnani,etal., Structure 2011,19,1283–1293. [41] A. Gaulton, L. J. Bellis, P. Bento,J.Chambers,M.Davies,A. Hersey, Y. Light, S. McGlinchey, D. Michalovich, B. Al-Lazikani, et al., Nucleic AcidsRes. 2012,40,D1100-7. [42] ThePyMOL Molecular GraphicsSystem, Version 1.4Schrçdinger, LLC. Manuscript received:March 13, 2020 Revised manuscript received: May 18, 2020 Acceptedmanuscript online: June 16, 2020 Version of record online: July 22, 2020 A ngewandte Chemie Research Articles 16543Angew.Chem. Int.Ed. 2020,59,16536 –16543 T2020 The Authors. Published by Wiley-VCH Verlag GmbH &Co. KGaA, Weinheim www.angewandte.org