Prospec(ve  Applica(ons  of  Structure-­‐Based  Drug  Design  Methods:  Comparing  to  

Intui(on  and  Other  Typical  Scoring  Methods  

Woody  Sherman  VP,  Applica5ons  Science  

Thermodynamic  Decomposi(on  of  Ligand/Protein  Binding  

Solvated Ligand Solvated Apo Protein

Desolvated Ligand

Solvated Ligand in Bioactive

Conformation

Ligand-Induced Desolvated Protein

Binding Site

Solvated Protein in Ligand-Induced

Conformation

Protein/Ligand/Water Complex

€

ΔGbind = ΔGi=1

5

∑ i( )

∆G(5) ∆H reward ∆Srot/trans penalty

∆G(1) ∆Hconf penalty ∆Sconf penalty

∆G(3) ∆H penalty ∆S reward

∆G(2) ∆Hconf penalty ∆Sconf penalty

∆G(4) ∆H ? ∆S ?

•  Key  factors:  •  Ligand  strain  •  Protein  strain  •  Interac5ons  •  Desolva5on  •  Bridging  waters  •  pKa/tautomers  •  Polariza5on  •  Other…?  

Ques(on:  How  can  we  accurately  predict  binding  energies  if  we  only  look  at  a  subset?  Answer:  By  luck  Why:  Because  we  never  know  when  a  par5cular  term  will  be  important.    

Interplay  Between  Terms  •  We  know  that  these  energy  terms  are  related  – More  polar  H-­‐bond  partner  is  oNen  offset  by  greater  desolva5on  – Opening  a  hydrophobic  pocket  in  the  protein  has  an  energe5c  cost  – Displacing  a  water  molecule  may  be  favorable  or  unfavorable  – Stronger  enthalpic  interac5ons  can  reduce  the  entropy  of  the  system  

•  Quan5fica5on  is  key  – The  balance  between  terms  can  be  subtle,  but  must  be  quan5ta5ve  to  have  a  predic5ve  method  

Some(mes  Simple  Scoring  Func(ons  Work  There  are  many  examples  (too  many  to  list);  here  is  one:  

How  many  5mes  has  a  force  field  interac5on  energy  calcula5on  worked  for  you  in  a  prospec5ve  project?  

Simple  Scoring  Func(ons  Are  Not  Robust  •  We  have  all  seen  cases  where  something  works  some5mes  •  Then  we  try  it  somewhere  else  and  it  does  not  work  •  Why  did  it  work?  – Cancela5on  of  errors  – Some  terms  might  not  be  relevant  for  a  given  target/series/system  

•  Can  it  be  useful?  – Some5mes,  but  chances  of  failure  are  high  when  moving  outside  the  valida5on  space  •  Different  interac5ons  •  Different  balance  of  terms  •  Different  cancela5on  of  errors  

What  is  Needed  

•  Claim:  Accurate  scoring  can  only  be  achieved  through:  

1.   Accurate  representa(on  of  the  energy  surface  2.   Sufficient  sampling  

Accurate  Representa(on  of  the  Energy  Surface  •  Quantum  mechanic  treatment  is  ideal,  but  computa5onally  expensive  – Must  think  about  the  tradeoff  between  more  accurate  energy  surface  versus  more  sampling  

•  An  accurate  force  field  is  likely  sufficient  to  achieve  accuracy  to  ~1  kcal/mol  – Some  subtle  QM  aspects  might  be  missed  – Could  be  added  indirectly  

•  e.g.  Using  QM  charges  

Sufficient  Sampling  •  Flexible  protein  is  needed  – Energy  surface  is  steep  and  we  must  know  when  the  protein  can  move  

•  Explicit  waters  must  be  considered  – Water  is  really  not  a  high  dielectric  con5nuum  

•  Relevant  5mescales:  – ~ns  simula5ons  should  be  sufficient  for  sampling  local  minima  – 10-­‐100  ns  for  some  side-­‐chain  transi5ons  and  loop  explora5on  – >μs  for  more  significant  movements  

•  Enhanced  sampling  methods  can  help,  but  do  not  come  without  their  own  issues  

How  Can  We  Do  This?  •  Rigorous  free  energy  approaches  – Free  energy  perturba5on  (FEP)  – Thermodynamic  integra5on  (TI)  

•  Compute  all  energe5c  terms  simultaneously  and  consistently  

•  Computa5onally  much  more  expensive  than  other  approaches,  but  tractable  for  ns  simula5on  5mes     ΔΔGbinding      =  ΔG1  –  ΔG2  

       =  ΔGA  –  ΔGB    

A  

B  

1   2  

Summary  of  FEP+  from  Schrödinger  •  We  have  an  automated  solu5on  for  predic5ng  binding  free  energies  to  an  accuracy  of  ~1  kcal/mol  

•  Valida5on  has  been  performed  on  over  1000  compounds  –  Iden5cal  protocol  – Mix  of  retrospec5ve  and  prospec5ve  applica5ons  – Many  target  classes  covered  

•  Kinases  •  Proteases  •  Bromodomains  •  GPCRs  •  PPIs  

•  Automated  system  provides  solu5on  for  non-­‐FEP  experts  

Recent  Publica(on  •  10  retrospec5ve  targets  – ~200  ligands  

•  2  prospec5ve  projects  

Accurate and Reliable Prediction of Relative Ligand Binding Potencyin Prospective Drug Discovery by Way of a Modern Free-EnergyCalculation Protocol and Force FieldLingle Wang,† Yujie Wu,† Yuqing Deng,† Byungchan Kim,† Levi Pierce,† Goran Krilov,† Dmitry Lupyan,†

Shaughnessy Robinson,† Markus K. Dahlgren,† Jeremy Greenwood,† Donna L. Romero,‡ Craig Masse,‡

Jennifer L. Knight,† Thomas Steinbrecher,† Thijs Beuming,† Wolfgang Damm,† Ed Harder,†

Woody Sherman,† Mark Brewer,† Ron Wester,‡ Mark Murcko,† Leah Frye,† Ramy Farid,† Teng Lin,†

David L. Mobley,⊥ William L. Jorgensen,∥ Bruce J. Berne,§ Richard A. Friesner,§ and Robert Abel*,†

†Schrodinger, Inc., 120 West 45th Street, New York, New York 10036, United States‡Nimbus Discovery, 25 First Street, Suite 404, Cambridge, Massachusetts 02141, United States§Department of Chemistry, Columbia University, 3000 Broadway, New York, New York 10027, United States∥Department of Chemistry, Yale University, New Haven, Connecticut 06520, United States⊥Departments of Pharmaceutical Sciences and Chemistry, University of CaliforniaIrvine, Irvine, California 92697, United States

*S Supporting Information

ABSTRACT: Designing tight-binding ligands is a primary objectiveof small-molecule drug discovery. Over the past few decades, free-energy calculations have benefited from improved force fields andsampling algorithms, as well as the advent of low-cost parallelcomputing. However, it has proven to be challenging to reliablyachieve the level of accuracy that would be needed to guide leadoptimization (∼5× in binding affinity) for a wide range of ligandsand protein targets. Not surprisingly, widespread commercialapplication of free-energy simulations has been limited due to thelack of large-scale validation coupled with the technical challengestraditionally associated with running these types of calculations.Here, we report an approach that achieves an unprecedented level of accuracy across a broad range of target classes and ligands,with retrospective results encompassing 200 ligands and a wide variety of chemical perturbations, many of which involvesignificant changes in ligand chemical structures. In addition, we have applied the method in prospective drug discovery projectsand found a significant improvement in the quality of the compounds synthesized that have been predicted to be potent.Compounds predicted to be potent by this approach have a substantial reduction in false positives relative to compoundssynthesized on the basis of other computational or medicinal chemistry approaches. Furthermore, the results are consistent withthose obtained from our retrospective studies, demonstrating the robustness and broad range of applicability of this approach,which can be used to drive decisions in lead optimization.

■ INTRODUCTION

Protein−ligand binding is central to both biological functionand pharmaceutical activity. Some ligands simply inhibit proteinfunction, while others induce protein conformational changesand hence can modulate key cell-signaling pathways. In eithercase, achieving a desired therapeutic effect is dependent uponthe magnitude of the binding affinity of ligand to targetreceptor. Designing tight-binding ligands while maintaining theother ligand properties required for safety and biologicalefficacy is a primary objective of small-molecule drug discoveryprojects.A principal goal of computational chemistry and computer-

aided drug design (CADD) is therefore the accurate predictionof protein−ligand free energies of binding (i.e., binding

affinities).1,2 The most rigorous approach to this problem isfree-energy simulation. A variety of free-energy simulationmethods, such as free-energy perturbation (FEP), thermody-namic integration (TI), and λ dynamics, employ an analysis ofatomistic molecular dynamics or Monte Carlo simulations todetermine the free-energy difference between two relatedligands via either a chemical or alchemical path.3−9 In drugdiscovery lead optimization applications, the calculation ofrelative binding affinities (i.e., the relative difference in bindingenergy between two compounds) is generally the quantity ofinterest and is thought to afford significant reduction in

Received: December 15, 2014Published: January 27, 2015

Article

pubs.acs.org/JACS

© 2015 American Chemical Society 2695 DOI: 10.1021/ja512751qJ. Am. Chem. Soc. 2015, 137, 2695−2703

FEP+   No  FEP  

JACS,  2015,  137  (7),  pp  2695–2703  

Several  New  Technologies  Needed  for  a  Robust  Solu(on  

Improved  force  field  

Enhanced  sampling    

Hardware  accelera(on  

Automated  setup  

Error  es(mates  

OPLS2  

REST  

GPU  

FEP  Mapper  

Cycle  Closure  

•  Over  500  perturba5ons  tested  w/  iden5cal  protocol    – Glide  SP,  Glide  XP,  and  Prime  MM-­‐GBSA  

 

FEP  Retrospec(ve  Comparison  

System Num. Ligand

Affinity Range

(kcal/mol)

PDB ID

FEP RMSE (kcal/mol)

FEP R-value

Expected R-value*

Glide XP R-value

Glide SP R-value

MM-GB/SA R-value

MW R-value

BACE 36 3.5 4DJW 1.12 0.61 0.62 ± 0.18 0.01 -0.28 -0.04 0.14 CDK2 16 4.2 1H1Q 1.26 0.56 0.70 ±0.22 -0.69 -0.59 -0.59 -0.48

JNK1 21 3.4 2GMX 0.76 0.88 0.62 ± 0.26 0.21 -0.14 0.64 -0.39

MCL1 42 4.2 4HW3 1.22 0.69 0.69 ± 0.14 0.61 0.43 0.42 -0.55 p38 34 3.4 3FLY 1.62 0.75 0.66 ± 0.16 0.07 0.26 -0.07 -0.46

PTP1B 26 4.1 2QBS 1.11 0.74 0.69 ± 0.18 -0.05 0.66 0.75 -0.84 Thrombin 11 1.7 2ZFF 0.72 0.68 0.34 ± 0.54 -0.57 0.28 0.54 -0.48

Tyk2 16 4.3 4GIH 0.64 0.92 0.72 ± 0.22 0.47 0.49 0.72 0.00 Wt. Av. - - - 1.14 0.72 0.65 ± 0.20 0.11 0.16 0.27 -0.38

Prospec(ve  Project  Impact  

GPCR  Collabora(on  with  Ad  Ijzerman  (Leiden  University)  

•  Retrospec5ve  study  – A2a,  β1AR,  OPRD,  and  CXCR-­‐4  

•  Prospec5ve  work  on  A2a  

A2a  

Beta-­‐1  

OPRD   CXCR4  

Legend:  A2a:    adenosine  receptor  A2a  β1AR:    β1  adrenergic  receptor  OPRD:    δ-­‐opioid  receptor  CXCR-­‐4:    chemokine  receptor  CXCR4  

Spearheaded  by  PhD  student  Bart  Lenselink  at  Leiden  University,  Netherlands;  did  summer  internship  at  Schrodinger  in  NYC  summer  2014  

GPCR  Retrospec(ve  Results  

GPCR  Result  Details  Dataset/Receptor No.  of  

Ligands Experimental  Min/Max.  ΔG  

R2 MUE  (Min/Max) Hysteresis  Avg/Max    

Mines  et  al.1    Adenosine  A2a  

9   -­‐8.61/-­‐11.56   0.61   0.68  (0/2.52)   0.66/2.11  

Piersan5  et  al.2  Adenosine  A2a  

7   -­‐9.80/-­‐11.07   0.08   0.68  (0.05/1.44)   0.43/1.15  

Thoma  et  al.3    Chemokine  CXCR4  

9   -­‐6.81*/-­‐11.03   0.60   1.12  (0.14/2.97)   1.81/4.28  

Yuan  et  al.  4  δ-­‐opioid  

11   -­‐9.57/-­‐12.85   0.72   0.69  (0.06/1.72)   0.93/2.02  

Christopher  et  al.5  β1-­‐adrenergic  

9   -­‐7.90/-­‐9.77   0.16   1.07  (0.13/2.6)   0.72/2.22  

 (1)  Mines,  P.  et  al.,  2-­‐n-­‐Butyl-­‐9-­‐methyl-­‐8-­‐[1,  2,  3]  triazol-­‐2-­‐yl-­‐9  H-­‐purin-­‐6-­‐ylamine  and  Analogues  as  A2A  Adenosine  Receptor  Antagonists.  Design,  Synthesis,  and  Pharmacological  Characteriza5on.  Journal  of  medicinal  chemistry  2005,  48,  6887-­‐6896.  

 (2)  Piersan5,  G.  et  al.,  P.  Synthesis  and  Biological  Evalua5on  of  Metabolites  of  2-­‐n-­‐Butyl-­‐9-­‐methyl-­‐8-­‐[1,  2,  3]  triazol-­‐2-­‐yl-­‐9  H-­‐purin-­‐6-­‐ylamine  (ST1535),  A  Potent  Antagonist  of  the  A2A  Adenosine  Receptor  for  the  Treatment  of  Parkinson’s  Disease.  Journal  of  medicinal  chemistry  2013,  56,  5456-­‐5463.  

 (3)  Thoma,  G.  et  al.,  Orally  bioavailable  isothioureas  block  func5on  of  the  chemokine  receptor  CXCR4  in  vitro  and  in  vivo.  Journal  of  medicinal  chemistry  2008,  51,  7915-­‐7920.    (4)  Yuan,  Y  et  al.,  Design,  Synthesis,  and  Biological  Evalua5on  of  14-­‐Heteroaroma5c-­‐Subs5tuted  Naltrexone  Deriva5ves:  Pharmacological  Profile  Switch  from  Mu  Opioid  Receptor  

Selec5vity  to  Mu/Kappa  Opioid  Receptor  Dual  Selec5vity.  Journal  of  medicinal  chemistry  2013,  56,  9156-­‐9169.    (5)  Christopher,  J.  A.  et  al.,  Biophysical  fragment  screening  of  the  β1-­‐adrenergic  receptor:  iden5fica5on  of  high  affinity  arylpiperazine  leads  using  structure-­‐based  drug  design.  Journal  

of  medicinal  chemistry  2013,  56,  3446-­‐3455.  

Gedng  a  Baseline  for  Potency  Improvements  •  17  metabolites  of  1  – Nothing  much  bezer  than  1  (7.5  nM  versus  11  nM)  

Synthesis and Biological Evaluation of Metabolites of 2‑n‑Butyl-9-methyl-8-[1,2,3]triazol-2-yl‑9H‑purin-6-ylamine (ST1535), A PotentAntagonist of the A2A Adenosine Receptor for the Treatment ofParkinson’s DiseaseGiovanni Piersanti,*,† Francesca Bartoccini,† Simone Lucarini,† Walter Cabri,‡ Maria Antonietta Stasi,‡

Teresa Riccioni,‡ Franco Borsini,‡ Giorgio Tarzia,† and Patrizia Minetti*,‡

†Dipartimento di Scienze Biomolecolari, Universita degli Studi di Urbino, Piazza Rinascimento 6, I-61029 Urbino (PU), Italy‡Sigma-Tau Industrie Farmaceutiche Riunite S.p.A., Via Pontina Km 30,400, I-00040 Pomezia, Italy

*S Supporting Information

ABSTRACT: The synthesis and preliminary in vitro evalua-tion of five metabolites of the A2A antagonist ST1535 (1) arereported. The metabolites, originating in vivo from enzymaticoxidation of the 2-butyl group of the parent compound, weresynthesized from 6-chloro-2-iodo-9-methyl-9H-purine (2) byselective C−C bond formation via halogen/magnesium exchange reaction and/or palladium-catalyzed reactions. The metabolitesbehaved in vitro as antagonist ligands of cloned human A2A receptor with affinities (Ki 7.5−53 nM) comparable to that ofcompound 1 (Ki 10.7 nM), thus showing that the long duration of action of 1 could be in part due to its metabolites. Generalbehavior after oral administration in mice was also analyzed.

■ INTRODUCTIONThe high attrition rate associated with drug discovery hasprompted increasing attention to the metabolism of the earlydrug candidates as a relevant factor in successful drugdevelopment.1 Although in most cases the metabolism ofdrugs leads to pharmacologically inactive compounds, some-times active or toxic metabolites are found. Species-dependentmetabolic transformation add further complexity to thedevelopment process, and for these reasons metaboliteidentification and examination have become an integral partof early drug discovery programs.2 Knowledge of the metabolicpathway of drug candidates is useful to improve their safety andpharmacological profile. We recently reported a series of 8-(2H-1,2,3-triazol-2-yl)purine derivatives as potent A2A adenosinereceptor antagonists (Figure 1) for their potential utility in thetreatment of Parkinson’s disease (PD).3 Compound 1(ST1535), a member of this series, has been evaluated invarious pharmacological models of Parkinson’s disease.4

Investigation of the in vitro metabolic fate of 1 by liquidchromatography/mass spectrometry (LC/MS) indicated thepresence of four metabolites with MW = MWST1535 + 16,accompanied by small amounts of two other oxidizedcompounds with MW = MWST1535 + 14 (data not shown).5

In humans, the principal metabolite is a product with MW =MWST1535 + 14.6 In all cases, MS of the isolated metabolitesindicated enzymatic oxidation of the 2-butyl group. Severalpossibilities for hydroxylation, oxidation, and allylic oxidationwere considered, and we report here the synthesis, in vitroaffinity, and intrinsic activity on adenosine receptors ofcompounds (Figure 1) that match the physicochemical

properties of five of the isolated metabolites. Moreover, generalbehavior after oral administration in mice was evaluated.

■ CHEMISTRYThe procedure that was selected for the synthesis of the 6-amino-2-(hydroxyalkyl)-9-methyl-8-(2H-1,2,3-triazol-2-yl)-9H-purines is based on the general I−Mg exchange reaction of 2,6-dihalopurines proposed by Dvorak7 to yield 3, which was thentransformed into compound 14 or on the C2-selectivepalladium-catalyzed cross-coupling of terminal alkynes with2,6-dihalopurines to give 4, 5a−c, and 6, which are thentransformed into 15, 16a−c, 17, and 18 (Scheme 1),respectively. In the case of halogen/Mg exchange, it is knownthat chlorine cannot be exchanged in the reaction of 6-chloropurine derivatives with iPrMgCl,8 whereas selectiveexchange of iodine for magnesium in mixed 6-chloro-2-iodopurine derivatives are reported.7 In the case of palladium-catalyzed cross-coupling reaction, some reports9 corroborateour hypothesis that cross-coupling reactions with 6-chloro-2-iodopurine intermediates would show selectivity for alkynyla-tion9a in the 2-carbon position. Products of such reactionswould carry a chlorine atom suitable for further synthetictransformations such as cross-coupling reactions or nucleophilicsubstitution.The synthesis of the purine derivatives 14−18 was pursued

according to the method adopted for the synthesis of 110 fromthe readily available 6-chloro-2-iodopurine (2) (Scheme 1).

Received: April 5, 2013Published: June 12, 2013

Article

pubs.acs.org/jmc

© 2013 American Chemical Society 5456 dx.doi.org/10.1021/jm400491x | J. Med. Chem. 2013, 56, 5456−5463

J.  Med.  Chem.  2013,  56,  5456−5463  

Synthesis and Biological Evaluation of Metabolites of 2‑n‑Butyl-9-methyl-8-[1,2,3]triazol-2-yl‑9H‑purin-6-ylamine (ST1535), A PotentAntagonist of the A2A Adenosine Receptor for the Treatment ofParkinson’s DiseaseGiovanni Piersanti,*,† Francesca Bartoccini,† Simone Lucarini,† Walter Cabri,‡ Maria Antonietta Stasi,‡

Teresa Riccioni,‡ Franco Borsini,‡ Giorgio Tarzia,† and Patrizia Minetti*,‡

†Dipartimento di Scienze Biomolecolari, Universita degli Studi di Urbino, Piazza Rinascimento 6, I-61029 Urbino (PU), Italy‡Sigma-Tau Industrie Farmaceutiche Riunite S.p.A., Via Pontina Km 30,400, I-00040 Pomezia, Italy

*S Supporting Information

ABSTRACT: The synthesis and preliminary in vitro evalua-tion of five metabolites of the A2A antagonist ST1535 (1) arereported. The metabolites, originating in vivo from enzymaticoxidation of the 2-butyl group of the parent compound, weresynthesized from 6-chloro-2-iodo-9-methyl-9H-purine (2) byselective C−C bond formation via halogen/magnesium exchange reaction and/or palladium-catalyzed reactions. The metabolitesbehaved in vitro as antagonist ligands of cloned human A2A receptor with affinities (Ki 7.5−53 nM) comparable to that ofcompound 1 (Ki 10.7 nM), thus showing that the long duration of action of 1 could be in part due to its metabolites. Generalbehavior after oral administration in mice was also analyzed.

■ INTRODUCTIONThe high attrition rate associated with drug discovery hasprompted increasing attention to the metabolism of the earlydrug candidates as a relevant factor in successful drugdevelopment.1 Although in most cases the metabolism ofdrugs leads to pharmacologically inactive compounds, some-times active or toxic metabolites are found. Species-dependentmetabolic transformation add further complexity to thedevelopment process, and for these reasons metaboliteidentification and examination have become an integral partof early drug discovery programs.2 Knowledge of the metabolicpathway of drug candidates is useful to improve their safety andpharmacological profile. We recently reported a series of 8-(2H-1,2,3-triazol-2-yl)purine derivatives as potent A2A adenosinereceptor antagonists (Figure 1) for their potential utility in thetreatment of Parkinson’s disease (PD).3 Compound 1(ST1535), a member of this series, has been evaluated invarious pharmacological models of Parkinson’s disease.4

Investigation of the in vitro metabolic fate of 1 by liquidchromatography/mass spectrometry (LC/MS) indicated thepresence of four metabolites with MW = MWST1535 + 16,accompanied by small amounts of two other oxidizedcompounds with MW = MWST1535 + 14 (data not shown).5

In humans, the principal metabolite is a product with MW =MWST1535 + 14.6 In all cases, MS of the isolated metabolitesindicated enzymatic oxidation of the 2-butyl group. Severalpossibilities for hydroxylation, oxidation, and allylic oxidationwere considered, and we report here the synthesis, in vitroaffinity, and intrinsic activity on adenosine receptors ofcompounds (Figure 1) that match the physicochemical

properties of five of the isolated metabolites. Moreover, generalbehavior after oral administration in mice was evaluated.

■ CHEMISTRYThe procedure that was selected for the synthesis of the 6-amino-2-(hydroxyalkyl)-9-methyl-8-(2H-1,2,3-triazol-2-yl)-9H-purines is based on the general I−Mg exchange reaction of 2,6-dihalopurines proposed by Dvorak7 to yield 3, which was thentransformed into compound 14 or on the C2-selectivepalladium-catalyzed cross-coupling of terminal alkynes with2,6-dihalopurines to give 4, 5a−c, and 6, which are thentransformed into 15, 16a−c, 17, and 18 (Scheme 1),respectively. In the case of halogen/Mg exchange, it is knownthat chlorine cannot be exchanged in the reaction of 6-chloropurine derivatives with iPrMgCl,8 whereas selectiveexchange of iodine for magnesium in mixed 6-chloro-2-iodopurine derivatives are reported.7 In the case of palladium-catalyzed cross-coupling reaction, some reports9 corroborateour hypothesis that cross-coupling reactions with 6-chloro-2-iodopurine intermediates would show selectivity for alkynyla-tion9a in the 2-carbon position. Products of such reactionswould carry a chlorine atom suitable for further synthetictransformations such as cross-coupling reactions or nucleophilicsubstitution.The synthesis of the purine derivatives 14−18 was pursued

according to the method adopted for the synthesis of 110 fromthe readily available 6-chloro-2-iodopurine (2) (Scheme 1).

Received: April 5, 2013Published: June 12, 2013

Article

pubs.acs.org/jmc

© 2013 American Chemical Society 5456 dx.doi.org/10.1021/jm400491x | J. Med. Chem. 2013, 56, 5456−5463

moderate yield from 8, as reported with the unfunctionalizedalkyl derivatives (Scheme 1).9

Purine-derived Grignard reagents react with aldehydes toproduce substituted purine derivatives bearing α-hydroxyalkylgroups in the 2-position, which are not easily accessible byother methods. We sought to extend this procedure to thesynthesis of β-hydroxyalkyl derivative, such as compound 15, bytreating the 6-chloro-2-magnesiated purines with the appro-priate epoxide (1,2-epoxybutane). Using different conditionsand catalyst systems, the epoxide-opening reaction failed to givethe desired product as did the use of purine copperintermediates prepared by oxidative addition of highly reactivezerovalent copper.11

Although a wide variety of C6 substituted purine nucleosidesbearing β-functionalized alkyl groups are known, the samecannot be said for the C2 position.12 After several unsuccessfulattempts,13 we chose to synthesize the β-hydroxyalkyl derivativevia a Sonogashira type coupling with butyne, followed byregioselective hydration of the substituted alkyne and reductionof the resulting ketone. We then focused on the preparation ofpurine derivatives having an acetylenic side chain and on theregioselective (β) hydration of unsymmetrical carbon−carbontriple bond in order to provide direct access to β ketones.Preparation of 2-alkynylpurines from 2-halogenated purine iseffectively catalyzed by palladium in the presence of copper anda base.9a Thus, when butyne was bubbled through ahomogeneous reddish solution containing 2, Pd(PPh3)2Cl2,and CuI in TEA until the reaction mixture turned black, theregioselective formation of 4 was achieved in high yield. Weinvestigated the conversion of alkynyl purine into ketone 7 byhydration. Traditionally, mercury(II) compounds or expensivetransition-metal catalysts have been employed to give thecorresponding Markovnikov product,14 whereas a generalmethod for selective anti-Markovnikov hydration reaction ofinternal alkynes has proven unsuccessful.15 Therefore, weexplored the methodology developed by Ackermann et al. forthe efficient regioselective indirect anti-Markovnikov hydrationof unsymmetrically substituted internal alkynes, consisting ofTiCl4-catalyzed hydroamination and subsequent hydrolysis ofthe generated imine/enamine to the desired ketone.15d On thatbasis, treatment of 4 with the sterically hindered t-butylamineand mesitylamine in toluene in the presence of catalytic amountof TiCl4, followed by treatment with acidic water, afforded theregioselective ketone 7 in moderate yield. Standard NaBH4reduction gave chemoselectively16 the β secondary alcohol 9,which yielded compound 15 via subsequent substitution of thechlorine atom in 6-position with an amino group, bromination,and triazolation at C8.This palladium−copper-catalyzed cross-coupling reaction can

be extended further to include other hydroxyl functionalizedterminal alkynes.17 For example, racemic propargyl alcohol (3-butyn-2-ol) and its enantiomers are commercially available.Reaction of these agents with 2 under the above cross-couplingconditions gave the alkynes 5a−c in excellent yields. No base-promoted isomerization from intermediates was detected.18

The hydroxyl-bearing terminal alkyne, 3-butyn-1-ol, was usedwith 2 under standard Sonogashira alkynylation condition toafford the primary alcohol 6 in good yield.Unfortunately, attempted hydrogenation of 5a−c, and 6

resulted in both alkyne reduction and dehalogenation. Thedehalogenation result was circumvented by selective C2amination of 5a−c and 6, followed by hydrogenation of theresulting adenines 10a−c and 11 to furnish 12a−c and 13,

respectively, in almost quantitative yields, even though highertemperature and longer times were required. Finally, thestandard procedure of C-8 bromination and subsequentreaction with triazole gave compounds 16a−c and 17. Ketone18 was obtained from 16a by oxidation with 2-iodoxybenzoicacid.

■ RESULTS AND DISCUSSIONThe synthesized compounds were evaluated for their bindingaffinity for cloned hA1 and hA2A receptors and for their intrinsicactivity on the hA2A receptor (Table 1).19

The main objective was to investigate the activities of themetabolites but in the process limited insight into thestructure−activity relationships of the adenine structural classwas obtained. Table 1 shows that all metabolites behave asadenosine receptor antagonists at the human A2A receptor withaffinities and intrinsic activities comparable to that of 1 (0.16−1.40-fold) and (0.20−1.40-fold), respectively. This indicatesthat metabolic hydroxylation of the butyl chain retains 1-likeactivity. Compared to compound 1, the hydroxylatedmetabolites 14, 15, and 17 showed slightly decreased affinitytoward the hA2A receptor, whereas hydroxylation on the butylside chain in γ position leads to metabolite 16a that is slightlymore active than the parent compound, confirming that highlyflexible substituents can be accepted in this area by the A2Areceptor, as reported.3,20

Assuming that 1 and its metabolites dock into the A2Areceptor similarly to ZM 241385,20a the alkyl and hydroxyalkylside chain would fit into the lipophilic cavity defined bytransmembrane helices VI and VII, in which one ordered watermolecule is present. Presumably, metabolite 16a, in contrast to1, will form a hydrogen-bonding water-mediated interactionwith His264 or Asn253. Although it was not the intention ofthis work, in order to further refine this hypothesis, it would be

Table 1. Binding Affinity and Inhibition Activity on Agonist-Mediated cAMP Accumulation in Human A2A ReceptorOverexpressing Cells and Binding Affinity to Human A1Receptor

compdsA2A bindingKi [nM]a,b

cAMP inhibitionIC50 [nM]a

A1 bindingKi [nM]a,b

1 11 430 100(5.8−20) (190−970) (59−180)

14 22 2300 160(8.0−57) (290−18000) (130−210)

15 64(37−110)

16a 8.4 450 34(4.5−16) (130−1600) (28−40)

16b 19 260 53(9.1−38) (55−1200) (43−66)

16c 7.5 420 34(4.2−13) (130−1300) (26−43)

17 53 2100 150(25−110) (360−12000) (120−190)

18 12 990 200(6.7−23) (160−6300) (150−260)

a95% confidence intervals. bKi values were calculated from IC50 values,obtained from competition curves by the method of Cheng andPrusoff, and are the mean of four determinations performed intriplicate.

Journal of Medicinal Chemistry Article

dx.doi.org/10.1021/jm400491x | J. Med. Chem. 2013, 56, 5456−54635458

We began by investigating the halogen-magnesium exchangereaction with 6-chloro-2-iodo-9-methyl-9H-purine (2). Reac-tion of 2 with i-PrMgCl at −78 °C, followed by addition ofbutanaldehyde at the same temperature, afforded the expected6-chloro-2-(1-hydroxybutyl)purine derivative 3 in 59% yield.However, when the I−Mg exchange reaction was performed at0 °C and addition of butanaldehyde, purine derivatives with thesecondary alcohol moiety at the 8 position was obtained in 41%yield, similarly to that found by Dvorak.7

As evidence of the regioselectivity, 13C NMR spectroscopyrevealed the disappearance of the signal of the C2−I of 2 (δ

116.0 ppm) and appearance of a new signal for C2-CHOH (δ166.3 ppm). Furthermore, the structure of compound 3 wasconfirmed by its mass spectrum (ESI+: 241−243), whichshowed the presence of the two isotopes of the chlorine atomin a 3/1 ratio. This regioselectivity is similar to that observed inSonogashira or Suzuki coupling when applied to the samedihalopurine (2). Regioselective amination of the 6 position of3 with aqueous ammonia in dioxane in a sealed tube at 60 °Cyielded the 6-amino compound (8). Selective nucleophilicbromination at 8-position using bromine in buffered solution(pH 4), followed by reaction with 1H-1,2,3-triazole, gave 14 in

Figure 1. Structure of five metabolites of 1.

Scheme 1. Synthesis of Metabolites 14−18a

aReagents and conditions: (i) (1) THF, iPrMgCl, −78 °C, (2) CH3(CH2)2CHO, −78 °C to rt, 20 h; (ii) NH3(aq)/dioxane, 70 °C, 16 h; (iii)MeOH, THF, acetate buffer pH 4, Br2, −14 °C to rt, 20 min; (iv) DMF, Cs2CO3, 1H-1,2,3-triazole, 90 °C, 16 h; (v) dioxane, alkyne, TEA,Pd(PPh3)2Cl2, CuI, rt, 1 h; (vi) (1) TiCl4, t-BuNH2, toluene, 1,3,5-Me3C6H2NH2, 120 °C, 14 h, (2) HCl 2N, rt, 1 h; (vii) MeOH, NaBH4, rt, 1 h;(viii) EtOH, H2 (4 atm), Pd/C, 50 °C, 40 h; (ix) DMSO, THF, IBX, rt, 20 h.

Journal of Medicinal Chemistry Article

dx.doi.org/10.1021/jm400491x | J. Med. Chem. 2013, 56, 5456−54635457

Expansion  of  Ideas  •  Run  FEP+  on  ~50  ideas  

•  Synthesize  and  test  top  ideas  

Modifica5ons  to  Mines  compounds    

New  ideas  

A2a  Prospec(ve  Results  •  Top  2  compounds  experimentally  >10x  bezer  than  25d  –  Improved  from  15  nM  (25d)  to  1.1  nM  (LUF7510)  

Star5ng  compound  

Bezer  

Worse  

Collabora(on  B  Case  Study  •  Goal:  To  find  new  single-­‐digit  nM  binders  – Collaborator  provided  us  idea  molecules    – We  used  FEP+  to  priori5ze  – Focused  on  maximizing  true  posi5ves,  not  minimizing  false  nega5ves  

•  Means  running  more  calcula5ons  to  find  compounds  with  high  confidence  –  Predic5ng  a  compound  to  be  5  nM  from  a  10  nM  lead  would  be  low  confidence  

•  FEP+  analysis  also  helped  to  iden5fy    – a  poten5ally  high-­‐value  backup  series  – a  new  potency-­‐driving  interac5on  mo5f  

13  TP   1  FN  12  FP   15  TN  

FEP+  Prospec(ve  Performance  in  Collabora(on  B    •  Goal:  To  find  molecules  with  pIC50-­‐Expt  >  8  •  Strategy  1:  Recommend  all  molecules  with  pIC50-­‐FEP  >  8  

Molecule   Expt.  pKi   Pred.  pKi   Abs  Err.   Molecule   Expt.  pKi   Pred.  pKi   Abs  Err.  1   8.2   10.6   2.4   22   7.6   8.2   0.5  2   7.8   10.2   2.5   23   7.9   8.1   0.2  3   8.6   10.1   1.5   24   6.5   8.1   1.7  4   8.0   9.9   1.9   25   8.0   8.0   0.0  5   8.0   9.9   1.8   26   8.0   7.8   0.2  6   8.2   9.7   1.5   27   7.9   7.7   0.2  7   8.6   9.5   1.0   28   6.9   7.6   0.7  8   8.0   9.5   1.5   29   6.6   7.2   0.6  9   7.5   9.4   1.8   30   7.0   7.1   0.1  10   8.3   9.3   1.0   31   7.1   6.8   0.2  11   7.6   9.2   1.6   32   7.2   6.7   0.5  12   6.5   9.1   2.6   33   6.8   6.6   0.2  13   8.5   8.9   0.4   34   6.6   6.3   0.3  14   8.2   8.8   0.6   35   6.2   6.2   0.0  15   8.1   8.7   0.7   36   6.3   6.1   0.3  16   8.5   8.7   0.2   37   6.1   6.0   0.1  17   7.4   8.7   1.3   38   6.7   5.6   1.1  18   7.9   8.5   0.7   39   5.6   5.6   0.0  19   7.3   8.3   1.0   40   6.4   5.2   1.3  20   7.9   8.2   0.3   41   6.3   5.1   1.2  21   7.2   8.2   1.0           MUE:   0.9    

               

%TP  =  52%  %TN  =  94%  

12  TP   2  FN  5  FP   22  TN  

FEP+  Prospec(ve  Performance  in  Collabora(on  B    •  Goal:  To  find  molecules  with  pIC50-­‐Expt  >  8  •  Strategy  2:  Recommend  only  molecules  with  pIC50-­‐FEP  >  8.7  

Molecule   Expt.  pKi   Pred.  pKi   Abs  Err.   Molecule   Expt.  pKi   Pred.  pKi   Abs  Err.  1   8.2   10.6   2.4   22   7.6   8.2   0.5  2   7.8   10.2   2.5   23   7.9   8.1   0.2  3   8.6   10.1   1.5   24   6.5   8.1   1.7  4   8.0   9.9   1.9   25   8.0   8.0   0.0  5   8.0   9.9   1.8   26   8.0   7.8   0.2  6   8.2   9.7   1.5   27   7.9   7.7   0.2  7   8.6   9.5   1.0   28   6.9   7.6   0.7  8   8.0   9.5   1.5   29   6.6   7.2   0.6  9   7.5   9.4   1.8   30   7.0   7.1   0.1  10   8.3   9.3   1.0   31   7.1   6.8   0.2  11   7.6   9.2   1.6   32   7.2   6.7   0.5  12   6.5   9.1   2.6   33   6.8   6.6   0.2  13   8.5   8.9   0.4   34   6.6   6.3   0.3  14   8.2   8.8   0.6   35   6.2   6.2   0.0  15   8.1   8.7   0.7   36   6.3   6.1   0.3  16   8.5   8.7   0.2   37   6.1   6.0   0.1  17   7.4   8.7   1.3   38   6.7   5.6   1.1  18   7.9   8.5   0.7   39   5.6   5.6   0.0  19   7.3   8.3   1.0   40   6.4   5.2   1.3  20   7.9   8.2   0.3   41   6.3   5.1   1.2  21   7.2   8.2   1.0           MUE:   0.9    

               

%TP  =  71%  %TN  =  92%  

Note:  True  posi5ve  (TP)  percent  will  con5nue  to  increase  as  FEP  cutoff  increases,  but  need  to  study  more  compounds  to  find  high-­‐confidence  hits  

Pyrazole  to  Thiazole  Core-­‐Hop  in  Collabora(on  B  •  Thiazole  core  molecule  was  not  scheduled  for  synthesis  prior  to  FEP+  calcula5ons  – FEP+  suggested  this  compound  to  be  at  least  as  good  as  the  lead  

•  Synthesis  and  experimental  tes5ng  verified  potency  – Thiazole  core  is  now  the  new  back  up  series  

N

N NH2

NO

O

F

F

title: New Entry

NN

N NH2

NO

O

F

F

title: New Entry

R1#

R2#R3#

N

N NH2

NO

O

F

F

title: New Entry

NN

N NH2

NO

O

F

F

title: New Entry

R1#

R2#R3# Expt.&pIC50&=&8.3&Expt.&pIC50&=&8.1&

•  Collaborator  became  interested  in  further  developing  a  known  weak  binding  compound:  

 

•  Amide  chemistry  made  it  easy  to  consider  alterna5ves  to  the  cyclobutyl  group  

R-­‐group  Scan  in  Collabora(on  B  

pIC50Expt.=6.9  

R-­‐group  Scan  in  Collabora(on  B  •  We  evaluate  the  following  amide  coupling  library:  – The  compounds  were  later  synthesized  

pIC50FEP+=6.1  pIC50FEP+=5.6   pIC50FEP+=6.6  

pIC50FEP+=6.1  pIC50FEP+=6.3   pIC50FEP+=5.2  

pIC50FEP+=6.3  pIC50FEP+=5.2   pIC50FEP+=6.0  

pIC50Expt.=6.5  pIC50Expt.=5.3   pIC50Expt.=6.4  

pIC50Expt.=6.5  pIC50Expt.=6.5   pIC50Expt.=6.5  

pIC50Expt.=6.6  pIC50Expt.=6.5   pIC50Expt.=6.5  

•  Good  agreement  • MUE  =  0.5  log  units  

•  But,  we  wanted  a  5ght  binders  •  Need  project  impact,  no  just  correla5on  

•  Could  we  suggest  anything  bezer?  •  Need  to  explore  more  ideas!  

R-­‐group  Scan  in  Collabora(on  B  •  Built  an  amide  coupling  library  of  commercially  available  small  carboxylic  acid  building  block  molecules  

•  Applied  chemistry  filters  to  prune  the  designs  to  248  compounds  

•  Compounds  were  docked  using  Glide  XP  •  Following  manual  inspec5on  of  poses,  22  compounds  were  selected  for  FEP+  analysis  

R-­‐group  Scan  in  Collabora(on  B  •  The  selected  22  compounds  were  diverse,  including  the  following:  

Core  Core  

Core  

Core  Core  

Core  

•  Only  a  single  compound  of  the  22  was  predicted  by  FEP+  to  bind  bezer  than  pIC50=8.0  –  Lead  crystal  structure  compound  (cyclobutyl)  had  pIC50Expt=6.9  

•  Based  on  observed  SAR  and  the  crystal  structure,  collaborators  were  skep5cal,  but,  they  did  synthesize  the  compound  

•  Results  were  promising:    –  pIC50Expt=7.9  –  pIC50FEP=8.5  

R-­‐group  Scan  in  Collabora(on  B  

Core%Cyclic  carbamate  

Lead  

R-­‐group  Scan  in  Collabora(on  B  

•  There  was  good  reason  to  be  skep5cal:  – The  crystal  structure  of  their  5ghtest  binding  member  of  the  series  filled  the  pocket  with  a  small  hydrophobic  moiety  

– Few  obvious  hydrogen  bonding  opportuni5es  for  the  cyclic-­‐carbamate  

pIC50Expt=6.9  

R-­‐group  Scan  in  Collabora(on  B  Star5ng  structure  for  FEP+  didn’t  look  great  

Star(ng    structure  

Representa(ve    structure  from    trajectory    

During  the  FEP,  the  binding  mode  substan5ally  reorganized    

FEP+  insights  into  SAR  •  More  than  simply  making  predic5ons,  we’d  also  like  to  use  FEP  to  bezer  understand  interes5ng  cases  like  this  

 

•  What  about  subtle  effects  hard  to  spot  by  eye?  – Protein  or  ligand  entropic  effects?  – Solva5on  effects?  – Rela5ve  favorability  of  different  protein-­‐ligand  contacts?  

•  One  can  speculate  about  such  effects  using  empirical  methods  and  visualiza5on,  but  free  energy  calcula5ons  can  help  to  establish  such  explana5ons  more  rigorously  

An  Example  from  Protein  Kinase  A  •  The  ester  variant  of  this  pair  of  ligands  has  been  a  consistent  outlier  in  the  development  of  our  docking  methods:    

Breitenlechner,  et  al.  J.  Med.  Chem.  2004,  47,  1375-­‐1390.  

ΔΔGbind    =  2.5  kcal/mol  

linker:    ester  →  vinyl  1sve   1svh  

Docking  rou5nely  predicts  these  compounds  to  be  equipotent  

An  Example  from  Protein  Kinase  A  •  X-­‐ray  structure  shows  no  obvious  favorable  direct  interac5ons  between  the  ester  linker  and  the  protein  nor  water:  

FEP+  analysis    

ΔΔGbind    =  +2.5  kcal/mol  

linker:    ester  →  vinyl  

1sve   1svh  

Method   ΔΔGbind  [kcal/mol]  

Experiment   +2.5  

Docking   0.0  

FEP+  (1SVE)   +1.6  

FEP+  (1SVH)   +2.3  

•  FEP+  analysis  does  correctly  iden5fy  the  favorability  of  the  ester  

•  Can  we  bezer  understand  why?  

FEP+  analysis    •  Esters  are  typically  much  more  soluble  than  vinyls:    

   •  Why  didn’t  desolva5on  of  the  ester  kill  the  compound?  – We  can  inves5gate  this  with  solva5on  free  energy  calcula5ons  

Molecule   Expt.  ΔGsolv    [kcal/mol]  

Expt.  ΔΔGsolv    [kcal/mol]  

FEP+  ΔGsolv    [kcal/mol]  

FEP+  ΔΔGsolv    [kcal/mol]  

ethyl  acetate   -­‐3.1   0   -­‐4.2   0  1-­‐butene   1.4   4.5   2.1   6.3  1-­‐hexene   1.7   4.8   2.4   6.6  

Ligand   Linker   ΔGsolv  [kcal/mol]  

FEP+  ΔΔGsolv    

ester  (1sve)   -­‐C(=O)-­‐O-­‐   -­‐43.7   0.0  vinyl  (1svh)   -­‐C=C-­‐   -­‐42.3   1.4  

The  ester  linker  here  is  much  less  favorable  in  water  than  might  be  expected.  

FEP+  analysis  •  S5ll,  the  ester  should  be  less  favorable  than  the  vinyl  based  on  desolva5on:  

•  The  ester  variant  is  some  how  gaining  3.4  kcal/mol  more  interac5on  free  energy  with  the  receptor  than  the  vinyl  variant  

•  Where  could  this  be  coming  from?    

Ligand   Expt.  ΔΔGbind    [kcal/mol]  

FEP+  ΔΔGbind    [kcal/mol]  

FEP+  ΔΔGsolv    [kcal/mol]  

FEP+  ΔΔGint  [kcal/mol]  

Ester  à  Vinyl   2.5   ~2.0   -­‐1.4   ~3.4  

FEP+  analysis  •  Analysis  of  the  trajectory  shows  no  new  visually  obvious  direct  interac5ons  from  between  the  ester  and  the  protein  

•  Further,  the  binding  modes  of  the  compounds  show  lizle  mo5on:  

SID  report  may  provide  a  hint  

Complex   Solvent  

Complex   Solvent  

Vinyl  

Ester  

Minimum  in  the  torsion  energy  

Maximum    

FEP+  analysis  •  The  torsion  angle  analysis  suggested  the  vinyl  compound  might  have  a  more  strained  bioac5ve  compound  than  the  ester  variant  

•  To  fully  validate  this  picture,  we  can  rerun  FEP  with  the  ligand  torsions  harmonically  restrained  to  the  bioac5ve  conforma5on  –  If  this  hypothesis  is  correct,  the  ligands  should  appear  equipotent  when  simulated  under  these  condi5ons    

Restrained  Torsion  Angle  FEP  

solvent  FEP:    no  restraint  

solvent  FEP:  k=1.5  kcal/mol  

Method   restraint  force    [kcal/mol]  

ΔΔGbind    [kcal/mol]  

FEP+  (1SVH)   0.0   2.25  

FEP+  (1SVH)   1.5   1.35  

FEP+  (1SVH)   15.0   -­‐0.03  

solvent  FEP:  k=15  kcal/mol  

Conclusions  •  Predic5ng  binding  energies  is  not  easy  – Many  compe5ng  factors  must  be  accounted  for  simultaneously  

•  To  predict  binding  energies  reliable,  we  need  sufficient  sampling  and  an  accurate  representa5on  of  the  energy  surface  

•  The  new  OPLS3  force  field  represents  the  best  in  ligand  and  protein  parameters  – Appears  to  be  sufficient  to  azain  ~1  kcal/mol  accuracy  in  FEP  

•  Challenging  cases  will  of  course  arise  – Large  sampling  barriers  – Electronic  effects  not  covered  by  the  force  field  – pKa,  waters,  and  more  

Acknowledgements  Development  Robert  Abel  Ed  Harder  Byungchan  Kim  Jen  Knight  Goran  Krilov  Teng  Lin  Levi  Pierce  Lingle  Wang  Yujie  Wu      

Applica(ons  Thijs  Beuming  Daniel  Cappel  Markus  Dahlgren  Michelle  Hall  Roy  Kimura  Fiona  McRobb  Daniel  Robinson  Devleena  Shivakumar  Thomas  Steinbrecher  Dora  Warshaviak  Friesner  Lab  at  Columbia  

Management  Ramy  Farid  Rich  Friesner    Scien(fic  Advisors  Bruce  Berne  John  Chodera  Bill  Jorgensen  Ron  Levy  David  Mobley  Mark  Murcko  Vijay  Pande