Advances in fragment-based drug discovery platforms

1. Introduction

2. Methods for obtaining

structural information for

FBDD

3. Methods for obtaining binding

properties for FBDD

4. Chemoinformatics techniques

in FBDD

5. Combined use of technologies:

examples of FBDD

6. Conclusions

7. Expert opinion

Review

Advances in fragment-based drugdiscovery platformsMasaya Orita†, Masaichi Warizaya, Yasushi Amano, Kazuki Ohno &Tatsuya NiimiDrug Discovery Research, Astellas Pharma, Inc., 21 Miyukigaoka Tsukuba, Ibaraki 305-8585,

Japan

Background: Fragment-based drug discovery (FBDD) has been established as a

powerful alternative and complement to traditional high-throughput screen-

ing techniques for identifying drug leads. At present, this technique is widely

used among academic groups as well as small biotech and large pharmaceu-

tical companies. In recent years, > 10 new compounds developed with FBDD

have entered clinical development, and more and more attention in the

drug discovery field is being focused on this technique. Objective: Under the

FBDD approach, a fragment library of relatively small compounds (molecular

mass = 100 – 300 Da) is screened by various methods and the identified

fragment hits which normally weakly bind to the target are used as starting

points to generate more potent drug leads. Because FBDD is still a relatively

new drug discovery technology, further developments and optimizations in

screening platforms and fragment exploitation can be expected. This review

summarizes recent advances in FBDD platforms and discusses the factors

important for the successful application of this technique. Conclusion: Under

the FBDD approach, both identifying the starting fragment hit to be devel-

oped and generating the drug lead from that starting fragment hit are

important. Integration of various techniques, such as computational technol-

ogy, X-ray crystallography, NMR, surface plasmon resonance, isothermal

titration calorimetry, mass spectrometry and high-concentration screening,

must be applied in a situation-appropriate manner.

Keywords: biological assays at high concentrations, computational technology,

fragment-based drug discovery, isothermal titration calorimetry, mass spectrometry, NMR,

surface plasmon resonance, X-ray crystallography

Expert Opin. Drug Discov. (2009) 4(11):1125-1144

1. Introduction

Fragment-based drug discovery (FBDD) is a method of identifying potentiallyuseful new compounds that emerged in the past decade and has proven to be anovel paradigm for small molecule drug discovery [1-8]. The methodology is acomplementary and alternative approach to high-throughput screening (HTS),starting from the fragment hits (molecular mass = 100 – 300 Da) that normallybind weakly (low micromolar to millimolar) to the target and building upprogressively more potent drug leads (submicromolar to low nanomolar). Althougha theoretical basis regarding the benefits of the fragment-linking approach wasreported by Jencks in 1981 [9,10] and a fundamental computational fragment-basedconcept, the multi-copy simultaneous search method, was introduced by Karplusand co-workers [11,12], development of FBDD for use in the field of drug discoverywas triggered by the introduction of the structure–activity relationship by NMR(SAR by NMR) method by Fesik et al. at Abbott Laboratories in 1996 [13]. FBDD is,therefore, a relatively new drug discovery technology, and a list of FBDD-derivedcompounds currently in the clinical development stage may be viewed in Table 1.

10.1517/17460440903317580 © 2009 Informa UK Ltd ISSN 1746-0441 1125All rights reserved: reproduction in whole or in part not permitted

Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

Table

1.List

ofclinicalcandidatesoffragment-derivedco

mpounds.

Compound

Structure

Company

Status

Target

Therapyareas

ABT-263

S

SN

N

O

O

OO

N

N

NO

FF

F

S

O

Cl

Abbott

Gen

entech

PhaseII

Clinical

Bcl-xL

Small-celllungcancer

Chronic

lymphocytic

leuke

mia

Lymphoma

Hem

atological

neo

plasm

Can

cer

ABT-869

F N

ON

NN

H2N

Abbott

PhaseII

Clinical

VEG

Fan

dPD

GF

receptortyrosine

kinasefamily

mem

bers

Non-small-celllung

cancer

Myelodysplastic

syndrome

Acute

myelogen

ous

leuke

mia

Ren

alcellcarcinoma

Hep

atocellular

carcinoma

Breasttumor

Can

cer

Colorectal

tumor

AT-7519

N

OC

l

Cl

N

O

NN

N

O

S

O

OH

Astex

PhaseII

Clinical

CDKfamily

mem

bers

Multiple

myeloma

Can

cer

AT-9283

N

NN

NN

N

N

O

O

Astex

PhaseII

Clinical

Aurora

kinasefamily

mem

bers

Flt3

tyrosinekinase

Jak2

tyrosinekinase

Abltyrosinekinase

Hem

atological

neo

plasm

Solid

tumor

Thedataareextractedfrom

theonlinedatab

ase,

ThomsonPh

arma(http://w

ww.thomson-pharma.com/)[107].

ICAM:Intercellularad

hesionmolecule;IND:Investigational

new

drug;PD

GF:

Platelet-derived

growth

factor.

Orita, Warizaya, Amano, Ohno & Niimi

1126 Expert Opin. Drug Discov. (2009) 4(11)

Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

Table

1.List


mpounds(continued).

Compound

Structure

Company

Status

Target

Therapyareas

DG-051

N

O

O

OH

O

Cl

HC

l

deC

ODE

PhaseII

Clinical

Leuko

trieneA4

hydrolase

Myocardialinfarction

Indeg

litazar

(PLX

-204)

O

OH

NSO

OO

O

Plexxiko

nPh

aseII

Clinical

PPARalphaPP

AR

delta

PPARgam

ma

Inflam

matory

disea

seCardiovasculardisea

seNon-insulin

dep

enden

tdiabetes

LY-517717

N

O

NN

O

N

NLilly

Tularik

PhaseII

Clinical

FactorXa

Thrombosis

NVP-AUY-922

O N

ON

N

O

HO

OH

VernalisNovartis

PhaseII

Clinical

ATP

ase

Hsp

90

Can

cerSo

lidtumor

ABT-518

SO

N

O

OO

O

O

OF

F

F

OH

HH

Abbott

PhaseI

Clinical

Gelatinase

Metalloprotease-2

Metalloprotease-9

Solid

tumor

AT-13387

Nostructuraldataavailable

Astex

PhaseI

Clinical

Hsp

90

Can

cer


theonlinedatab

ase,

ThomsonPh

arma(http://w




new

drug;PD

GF:

Platelet-derived

growth

factor.

Advances in fragment-based drug discovery platforms

Expert Opin. Drug Discov. (2009) 4(11) 1127

Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

Table

1.List


mpounds(continued).

Compound

Structure

Company

Status

Target

Therapyareas

IC-776

S N+

O

O-

N

N

O

O

Lilly

ICOS

PhaseI

Clinical

CD11a

ICAM

Inflam

matory

disea

sePsoriasis

Autoim

munedisea

se

PLX-4032

O

S

O

NN

ON

Cl

F

F

Plexxiko

nRoche

PhaseI

Clinical

Raf

Bprotein

kinase

Melan

omaCan

cer

PLX-5568

Nostructuraldataavailable

Plexxiko

nRoche

PhaseI

Clinical

Raf

protein

kinase

Pain

Polycystic

kidney

disea

se

SNS-314

OS

O

OH

N

S

N

NS

NN

O

NC

l

Sunesis

PhaseI

Clinical

Aurora

protein

kinase

Can

cerSo

lidtumor

AT-13148

N

O

OH

FO

N

O

Astex

ICR

CRT

AstraZe

neca

PhaseItrials

areplanned

AKTprotein

kinase

Can

cer

SGX-393

OH

NN

NN

Lilly

(SGX)

AnIND

fora

PhaseItrial

has

bee

nfiled

Abltyrosinekinase

Can

cer


theonlinedatab

ase,

ThomsonPh

arma(http://w




new

drug;PD

GF:

Platelet-derived

growth

factor.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

Despite only a short amount of time having passed sincethe widespread application of this approach was initiated,> 10 compounds have already entered clinical developmentand many more may be expected to enter clinical developmentin the future (Table 1).

FBDD mainly consists of two steps [8]: the identificationof fragment hits to be developed, and the conversion of frag-ment hits to leads (Figure 1). However, for each step, manychoices must be made. For example, in the first step, thefragment-library, fragment-screening methods and methodsfor the prioritization of fragment hits must be selected. In thesecond step, approaches to converting fragment hits to leads(namely fragment evolution, fragment linking, fragment opti-mization or fragment self-assembly) and structural informa-tion for the molecular design must be determined. Giventhe above range of choices available in using the FBDDtechnique, this method may exist in different forms basedon the research institution, involving integration of varioustechniques which harness the strong points of each company,such as X-ray crystallography, NMR, surface plasmon reso-nance (SPR), isothermal titration calorimetry (ITC), massspectrometry (MS), high-concentration screening (HCS) andcomputational technology.

In this review, we first present protein crystallography andNMR for fragment-based projects, as ample structure data arenecessary to effectively prioritize and develop drug lead struc-tures from fragments hits. We also describe the methods forclarifying binding properties as guidelines in fragment selec-tion and development, and summarize recent advances incomputational approaches to using FBDD platforms. Finally,we discuss the factors important to the successful applicationof this technique.

2. Methods for obtaining structuralinformation for FBDD

2.1 Protein crystallography for FBDDProtein crystallography is one of the most important tech-nologies for promoting FBDD, providing two important bitsof information: evidence of specific binding of fragment hitswith target proteins, and relevant and detailed descriptions of3D interaction. Because protein crystallography analysis candetect low intermolecular affinity, it is ideal for identifyingpromising fragment hits, given that these hits usually have lowaffinity but exhibit favorable binding modes with high ligandefficiency for target proteins (see also Section 4.1) [14,15], acritical step in the FBDD process. Hit rates for fragmentscreening are generally expected to be higher than those forHTS [16], and biological and biophysical techniques usuallyapplied in fragment screenings are susceptible to artifactformation, which may induce inherent false positives. Withthese properties, high-throughput protein crystallographyplays an essential role in the identification of multiple frag-ment hits and subsequent analysis of those interaction modesfor further optimization.

In the field of drug discovery, protein crystallography isdivided into two stages. The first involves producing crystals ofprotein and the compound, while the second involves collect-ing and analyzing X-ray diffraction data obtained from thecrystals. Below, we describe the general procedures and recentadvances made with regard to these two steps.

2.1.1 Crystallization methodsTo produce the protein and compound complex crystals, twodifferent methods are generally used. The first, co-crystalliza-tion, involves precipitating complex crystals from a mixtureof protein and the compound. The second method, knownas ‘soaking,’ involves obtaining complex crystals by soakingligand-free crystals in a compound solution (Table 2).

Co-crystallization includes two advantages. First, this par-ticular method tolerates the low solubility of certain com-pounds. Compounds with low solubility in aqueous solutionscompared to organic solvents such as dimethylsulfoxide canbe dissolved in aqueous solutions containing proteins. Giventhat compounds in the early stages of FBDD are generallyhighly soluble in aqueous solution, we have not seriouslyconsidered issues regarding compound solubility during pro-tein crystallography at the present stage. However, consider-ation should be given towards maintaining solubility in laterstages of FBDD, when hydrophobic moieties are introducedinto the original compounds to obtain greater affinity to thetarget protein. The second advantage to the co-crystallizationmethod is its robustness to large, binding-induced confor-mational changes. In some cases, despite the low affinity ofthe fragments, protein conformation is drastically changed,and crystal packing, space groups and lattice constants aredifferent from ligand-free crystals. In these situations, only theco-crystallization method is able to produce complex crystals.Further, these conformational changes in the protein arehighly important for drug discovery, as the pockets generatedby conformational changes may be used to increase the affinityor novelty of well-known compounds. A disadvantage of theco-crystallization method is the large amount of proteinrequired to use it (0.1 – 0.2 mg/compound). While thisproblem often stalls high-throughput protein crystallography,it can be resolved using an ultra-small volume dispenser suchas the Mosquito from TTP LabTech [17], the Hydra fromMatrix Technologies [18] or the ScreenMaker from Innova-dyne Technologies [19]. These dispensers are able to make100 – 200 nl drops on crystallization plates, thus enablingone-fifth to one-tenth reduction in the amount of protein usedfor analysis of each compound.

In contrast, the soaking method can overcome problemsencountered using the co-crystallization method on high-throughput protein crystallography. Once crystallizing con-ditions for a ligand-free target protein are established, severalhundred or even thousands of crystals can be produced on asingle 96-well crystallization plate, which is typically preparedusing 500 – 1000 mg of protein. Further, production of asufficient number of ligand-free crystals in advance would



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

facilitate the timely determination of complex structures.This point is particularly favorable with regard to FBDD,as synthesis of new compounds, activity and affinity evalu-ation, derivation of complex structures, and designing ofnew compounds can be sped up to obtain new potentcompounds. As production of ligand-free crystals for soakingis not possible in some cases, the crystals are instead preparedfrom complex crystals by a method known as back-soaking.Back-soaking involves placing complex crystals into the solu-tion without the compounds, to clear out pockets of pro-teins [20]. Because the feasibility of back-soaking depends onseveral factors, including crystal packing, affinity of com-pound or solubility of compound, this method cannot beapplied to all complex crystals.

In certain cases, the soaking method is used to screenfragment libraries. Astex Therapeutics (Cambridge, UK)established the Pyramid system for FBDD, which involvesscreening fragment libraries by soaking fragment cock-tails [21]. This method allows HTS by protein crystallogra-phy, which detects fragments with weak affinity but specificbinding to target proteins. Obviously, data regarding thecomplex structure of multiple detected fragments areobtained simultaneously.

At Astellas Pharmaceuticals, Inc. (Tokyo, Japan), methodsfor determining complex structures for 23 proteins have beenestablished since 2006. The soaking method was applied to17 proteins, and the co-crystallization method to 3 (Figure 2).For the each of remaining proteins, some compounds were

• 3D structure information regarding fragment-protein complex (this information is also utilized for fragment hits to leads conversion in structure-based drug)

• Information regarding location of fragment- protein interaction (for protein-based NMR screening)

• Thermodynamic properties of binding

• Kinetic properties of binding

• Enzyme inhibitory activity

Crystallography 1K

NMR 1K

ITC 1 – 2K

SPR 2 – 5K

HCS 5 – 30K

Fragment to hitFragment hit identification

Fragmentlibrary

(number ofsamples which

can generally beevaluated)

Fragmentscreening

(throughput)

Fragment hitsprioritization

(type ofinformation)

Conversion offragment hits

to leads

Large size High throughput

Small size Low throughput

Figure 1. The fragment-to-lead process in FBDD. Comparison of crystallography, NMR, ITC, SPR and HCS for the size of fragmentlibrary and fragment screening; a summary of the strengths of various technologies for prioritizing fragment hits and converting fragmenthits to leads.FBDD: Fragment-based drug discovery; HCS: High-concentration screening; ITC: Isothermal titration calorimetry; SPR: Surface plasmon resonance.

Table 2. Comparison of methods for preparing complex crystals.

Method of crystal

preparation

Protein volume Time

required

Advantages

Co-crystallization 100 – 200 mg/compound

1 – 10 days Detects large conformational changes in protein;detects compounds with low solubility

Soaking 1 – 5 mg/compound

1 – 10 h Low protein consumption; fast crystal preparation;enables fragment screening



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

detected using both methods, while others were detected usingonly the co-crystallization method. Whether or not a soakingmethod can be applied, therefore, depends on both the proteinand compound. These observations suggest the importance ofdetermining an appropriate crystallographic method for use inFBDD for new target proteins by observing whether or notsoaking is possible with the compound in question, enablingtime- and cost-saving protein crystallography in FBDD.

2.1.2 Data collection and structure determinationSteady advances in synchrotron facility technology haveenabled high-speed data collection. Among these newly devel-oped technologies are optics with brighter and better colli-mated X-ray beams, high-speed shutters and diffractometers,and more sensitive charge-couple-devices (CCDs) with shorterreadout times [22]. Third generation sources such as theEuropean Synchrotron Radiation Facility, Advanced PhotonSource (APS) and Super Photon Ring-8 typically requireseveral minutes per crystal for data collection. Second gener-ation sources such as Photon Factory (PF), which has a multi-pole wiggler insertion light source, have been able to obtainhigh-flux beams, with data collection efficiency comparable tothat of third generation sources [23]. Further, the developmentof high-intensity X-ray generators coupled with improvedoptics, which are both commercially available, has dramaticallyincreased the efficiency of in-house data collection [24].

In parallel to these improvements in high-speed datacollection technology, automated sample-mounting robotsand alignment systems have been developed and integratedwith synchrotron beamlines and in-house systems [25]. Atpresent, fully automated high-throughput data collection(typically 50 – 100 complete data sets per 24 h) is availableat most synchrotron facilities and some academic or industriallaboratories. Technological improvements such as these have

facilitated the steady annual increase in the number ofX-ray crystal structures submitted to the Protein Data Bankfrom synchrotron facilities, with the number of such sub-missions accounting for > 80% of total submissions since2007 (Figure 3).

At many pharmaceutical companies, structural informa-tion has generally been used to identify hit fragments or inhit-to-lead optimization of disease-related targets. This use issupported by the fact that many pharmaceutical companiesregularly use synchrotron beamlines together with the auto-mated system. For example, Novartis and Hoffman La Rochecooperatively funded and built their own beamline (beamlineX10SA) at the Swiss Light Source. Naturally, these companieshave priority in its use, but non-reserved beam time is availablefor use by other academic or industrial users. Additionally,SGX, acquired by Eli Lilly in 2008, has a beamline for its ownuse at the APS (beamline 31-ID). Several pharmaceuticalcompanies have organized a consortium for the regular useof the synchrotron beamline (summarized in Table 3). Withits increasing opportunities to access synchrotrons, the crys-tallography group of GlaxoSmithKline determined some400 ligand–protein complex structures in 2005 [26] and 650in 2008 [27].

Given the increasing demand for data regarding ligand–protein interaction in various drug discovery projects, Astellasacquired the priority to use a high-throughput synchrotronbeamline, newly built at PF-AR beamline NE3A, in 2009 [28].The concept of the beamline is to rapidly and automaticallycollect a large amount of X-ray data. Available apparatusinclude optics that give a high-flux X-ray beam, a high-gainand fast-readout CCD detector, and a sample exchange robotwhich is able to treat cryo-cooled samples stored in a vacuumflask. Further, the robot has been equipped with the doubletong system, which was developed by a structural biologygroup at PF [29]. In a typical single tong-equipped robotsystem, a sample mounted at the X-ray beam position isinitially withdrawn into a sample-storage vacuum flask, afterwhich the next sample is picked up and mounted. A robotequipped with the double tong system can simultaneouslywithdraw a mounted sample and set the next sample, thereby,accelerating the sample exchange rate to a speed greater thanthat of other systems. Using this method, > 100 X-ray data setscan be automatically collected in a single day, potentiallycontributing to increased efficiency in lead generation basedon structural information obtained from FBDD.

In addition to technological advancements in hardware forautomated high-throughput data collection, software for rapidand automated structure determination, including data pro-cessing, phasing, ligand-fitting and refinement, has beendeveloped which integrates publicly available programs forX-ray crystallography [30,31]. However, the major bottleneck instructure determination remains the manual interpretation ofelectron density maps; fragment screening will typically pro-duce hundreds of compound for X-ray data collection, andeven experienced crystallographer may take several hours in

17

3

3

Soaking

Co-crystallization

Soaking or co-crystallization

Figure 2. Classification of target proteins in AstellasPharmaceuticals by the crystal preparing method (last3 years).



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

front of a computer screen to identify whether or not a com-pound is bound to a protein. Further, this manual procedurecan be very subjective, and the fragment structures areextremely simple, inadvertently facilitating incorrect assign-ment to an unambiguous electron density map, particularlyif data resolution is low. Although automatic ligand-fittingsoftware such as X-Ligand [32] are commercially available andcan be integrated into the automated software, the electrondensity map for each data set should be visually inspected atleast once, as ligand molecules may still be assigned to theelectron density map with a wrong conformation. Still, furtherimprovement in development of such programs is needed toresolve this issue.

Two novel next generation detectors have been developedfollowing recent technological improvements in data col-lection efficiency. One of these detectors, based on single-photon-counting technology, has already been installed inseveral synchrotron beamlines and in-house systems [33]. Theother detector is based on an amorphous selenium membraneand a matrix field emitter array (HARP-FEA) [34]. Bothdetectors have an ~ 300-fold shorter readout time (nearlyinstantaneous) and are more sensitive than widely used CCDdetectors, thus, enabling a complete data set to be obtainedin higher quality and much shorter time (equal to X-raybeam exposure time + sample exchange time). Together withcurrently available automated systems, use of these detectors

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

Nu

mb

er

0

10

20

30

40

50

60

70

80

90

100

A: Number of PDB deposition of X-ray structures

B: Number of PDB deposition of X-ray structures from synchrotron facilities

Ratio of B to A (%)

(%)

Figure 3. Annual PDB deposition of X-ray structures from synchrotron facilities according to BioSync (http://biosync.rcsb.org/BiosyncStat.html).PDB: Protein Data Bank.

Table 3. Industrial beamline in synchrotron facility.

Company or

consortium

Synchrotron Country Beamline Link

Novartis,Hoffman La Roche

SLS Switzerland X10SA http://sls.web.psi.ch/view.php/beamlines/px2/index.html

Eili Lilly (SGX) APS US 31-ID https://beam.aps.anl.gov/pls/apsweb/beamline_display_pkg.display_beamline?p_beamline_num_c=45

Takeda, GNF* ALS US 5.0.3 http://www.als.lbl.gov/als/techspecs/bl5.0.3.html

Plexxikon etc. ALS US 8.3.1 http://www.als.lbl.gov/als/techspecs/bl8.3.1.html

Astellas PF Japan NE3A http://pfweis.kek.jp/index.html

PCPROT‡ SPring-8 Japan BL32B2 http://www.pcprot.gr.jp/index_e.html

IMCA§ APS US 17-ID,17-BM http://www.imca.aps.anl.gov/

*Genomics Institute of the Novartis Research Foundation.‡Pharmaceutical Consortium for Protein Structure Analysis.§Industrial Macromolecular Crystallography Association.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

may drastically increase data collection efficiency, particularlyin synchrotron facilities.

High-throughput X-ray crystallography may indeed bepossible, as described above. For pharmaceutical companies,use of this technology would stimulate the rate of hit-fragmentselection using X-ray crystallography, thereby, contributingto more efficient hit-to-lead cycles, including compoundevaluation, synthesis and structural information in FBDD.

2.2 NMR for FBDDX-ray crystal structures provide credible evidence of com-pounds binding to a target protein. Further, X-ray crystalstructures also provide structural information critical forlead optimization. However, in some cases, electron densityfor fragments bound to protein cannot be determined forexperimental reasons, and other methods of obtaining struc-tural information regarding fragment/protein binding arestill needed.

NMR fragment screening was first introduced as ‘SARby NMR’ in 1996 by Fesik et al. at Abbott laboratories [13].With this method, the binding of small molecular fragmentsto 15N-labeled proteins was measured using 15N and 1H 2Dhetero-nuclear single quantum coherence (HSQC) NMR.NMR screening technology has developed significantly inthe decade since then, and this method is now widely usedto identify and validate novel fragment hits.

Two strategies have been developed for NMR screening.The first involves measuring the protein signals (target-based),while the other involves measuring the ligand signals (ligand-based). NMR techniques that are frequently applied toFBDD are summarized in Table 4, and a schematic represen-tation is available in Figure 4. In target-based screening,changes in the crosspeaks of 1H-13C or 1H-15N for a labeledprotein can be detected when target protein binds to acompound. This method allows for detection of interactionin the range of nanomolar to millimolar and also providesdata regarding the binding site. However, target-based screen-ing requires large quantities of labeled protein and is thuslimited to screening small proteins with high solubility [35].In ligand-based screening, changes in the ligands can bedetected when a compound binds to a target protein. Thismethod is suitable for ligands with affinities from~ 100nM – 10 mM and presents several advantages overtarget-based screening. Namely, the protein consumption isquite smaller than that required for target-based screening,and ligand-based screening does not require an isotopic label.Further, ligand aggregation can be validated. Dalvit et al.previously used 19F NMR to detect the displacement weak-binding molecule, given the several advantages in using the19F signal: good sensitivity, narrow line widths and widechemical shift range [36].

To the best of our knowledge, clinical stage compounds(ABT-263 (Abbott, Bcl-xl), NVP-AUY-922 (Novartis/Vernalis, HSP90), ABT-518 (Abbott, MMP-2,9), AT13387(Astex, HSP90), IC-776 (Lilly/ICOS, LFA-1)) were identified

using NMR as a detection method (Table 1). A number ofpapers have cited examples of NMR-based fragment screen-ing. For example, research group at Wyeth applied NMRfragment screening to identify protein–protein interactioninhibitors [37]. The researchers conducted 1H-15N HSQCexperiments on a library of 825 fragments to find the hitfragments which bound to the C-terminal region of ZipA.Seven hits were identified, and the binding mode of the bestone was revealed by X-ray crystallography. In another exper-iment, researchers at AstraZeneca identified novel inhibitorsof prostaglandin D2 synthase through FBDD [38]. A libraryof fragments (2500 compounds) was screened using 2DNMR, leading to the identification of 24 primary hits.Two iterative cycles were then carried out, including NMRscreening, molecular modeling, X-ray crystallography andin vitro biochemical assay. The IC50 value of the strongestinhibitor was found to be ~ 20 nM.

Despite the advantages of NMR, however, the throughputof NMR screening is lower than ITC, SPR and HCS. Therelationship between compound library size and initial screen-ing method is shown in Figure 1, where we can see that thefragment library size for NMR screening is small, due to lowthroughput. Throughput is typically improved by conductingfragment screening in a compound mixture [39].

3. Methods for obtaining binding propertiesfor FBDD

Thermodynamic variables such as binding enthalpy andentropy have long been avoided as guiding indices within drugdiscovery. However, due to recent improvements in technol-ogy with high sensitivity and high throughput, this conven-tion is rapidly changing. Optimization of binding enthalpyis clearly a critical step in obtaining compounds with highaffinity, and other drug properties such as selectivity andsolubility are in turn affected by a compound’s thermody-namic properties (enthalpy and entropy). Interestingly, Freireshowed that, for HIV-1 protease and HMG-CoA reductase,the first-in-class compounds were not enthalpically opti-mized, while best-in-class compounds were enthalpicallyoptimized, indicating that enthalpic optimization of drugcandidates may take years and only appears in second-generation products [40]. Because enthalpic optimization isquite difficult to attain due to enthalpy and entropycompensation, the enthalpy contribution to binding ener-gies may be a good guiding index for selecting startingleads. Additionally, the kinetic parameters of compoundsmay gain further importance in lead optimization process.For example, resarchers at Proteros showed that the evo-lution of fragments towards the fully optimized inhibitorBIRB796 included modulation of the residence time as wellas the affinity [41]. Comprehensive characterization of theinteraction between a protein and a ligand is significantwhen hundreds of potential hits must be investigated beforeoptimization of select hits.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

When a fragment binds to a protein, the active site watermolecule is expelled with bulk water. This expulsioin contri-butes both enthalpically and entropically to the binding freeenergy. The Frisner group proposed a novel method of ana-lyzing the thermodynamic properties of hydration sites inprotein pockets [42,43], demonstrating that the active sites ofproteins provide diverse environments for solvating water. Suchinformation on the thermodynamic properties of hydrationsites may be useful in the selection of starting fragments.In the field of FBDD, starting fragments are selected based

on several binding properties, including thermodynamic andkinetic properties, and these methods are widely used in pri-mary fragment screening. Below, we provide a quick intro-duction to the experimental methods currently used to obtainbinding properties.

3.1 Surface plasma resonanceSPR biosensors detect changes in the surface’s refractive indexresulting from the binding and subsequent separation oftwo molecules, one of which is bound to the sensor surface.

In this manner, kinetic properties for binding and separationcan be easily obtained, and thermodynamic properties canbe deduced by analyzing temperature dependence. As themolecules in the solution flowing over the surface binds tothe fixed molecules, the refractive index near the sensor sur-face increases, leading to a shift in the SPR angle. We canthen calculate thermodynamic equilibrium binding data andkinetic constants (Kd, kon, koff) from the shift in the resonanceposition. SPR has recently come into wide use as an initialscreening method. Ekstrom et al. applied SPR to identify low-affinity ligands (low micromolar to millimolar) for allostericbinding sites on human liver glycogen phosphorylase [44].Several technology-based companies have developed excellenttechnologies for use in this field, such as Graffinity’s uniqueSPR-based process (rapid array informed structure evolution;RAISE) [45].

3.2 Isothermal titration calorimetryITC is one of the most promising methods available formeasuring biomolecular interactions, enabling simultaneousdetermination of all binding parameters in a single experi-ment, including binding constants, reaction stoichiometry,enthalpy and entropy. When a ligand binds to a protein, heatis either generated or absorbed, and the ITC thermodynamictechnique is able to directly measure the heat released orabsorbed during this biomolecular binding event, enablingaccurate determination of binding constants, reaction stoichi-ometry, enthalpy and entropy. In this way, ITC provides acomplete thermodynamic profile of the molecular interactionin a single experiment, explaining its recently achieved statusas a popular primary screening technique [46].

3.3 Mass spectrometryThe basics of MS are twofold: compounds are ionized togenerate charged molecules and measurement is conductedbased on mass:charge ratios. The MS assay accurately quan-tifies binding affinity, stoichiometry and specificity over a widerange of ligand Kd values. Seth et al. developed an MS-basedfragment discovery method, calling their concept ‘structure–activity relation by mass spectrometry (SAR by MS)’. Theyinitially identified a weak hit to the ribosomome IIA sub-domain of hepatitis C, and then optimized it, obtaining asubmicromolar inhibitor [47]. Researchers at Sunesis developeda method for identifying fragments that bind to specific siteson a protein, known as ‘tethering’ [48,49], and MS spectrometryis often used to identify hit fragments in this method.

4. Chemoinformatics techniques in FBDD

Thanks to the recent rapid advances in structural biology,as mentioned above, a large amount of data on complexstructures is now available. Often, > 100 structures may bedetermined for a single project in which a wide range of com-pounds of varying molecular size and structure are complexedto a target protein. These structural data guide medicinal and

Table 4. Comparison of NMR techniques frequently

applied to FBDD.

Detection Parameter Characteristics

Protein Chemical shift(1H,15N,13C)

Detects binding epitopeon protein; generallyrestricted to small proteins;isotopic labeling usuallyneeded; relatively largeamount of protein needed

Ligand Relaxation (1H) Detects binding epitopeon ligand; increase inperformance by increasingprotein size

Ligand Relaxation (19F) Detects binding epitopeon ligand; restricted tocompounds with F atoms;relatively small amount ofprotein needed; increase inperformance by increasingprotein size

Ligand Cross-relaxationin the protein-fragmentcomplex (transferred-NOE or STD-NMR)

Detects binding epitopeon ligand; increase inperformance by increasingprotein size; isotopiclabeling not needed

Ligand Cross-relaxationbetween the fragmentand protein-boundwater molecules(water-LOGSY)

Detects binding epitopeon ligand; suitable forhydrophilic targets andligands; isotopic labelingnot needed; relativelylarge amount of proteinneeded

FBDD: Fragment-based drug discovery; NOE: Nuclear Overhauser effect;

STD: Saturation transfer difference; WaterLOGSY, Water-ligand observed via

gradient spectroscopy.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

computational chemists in prioritizing the many fragmenthits as well as in developing structurally-validated designsto convert fragment hits into leads. However, methods ofselecting promising fragment hits and converting them toleads remain too indirect, even given the large volume ofaccumulated 3D-structure data. To address these issues, newbiophysical technologies such as SPR and ITC as well asseveral computational approaches have been developedrecently. In the following section, we discuss recent advancesmade in the field of FBDD with regard to the computationaltechniques (Figure 5).

4.1 Ligand efficiencyLigand efficiency (LE), defined as the binding energy pernon-hydrogen heavy atom (HA), was originally proposed as auseful parameter in selecting and optimizing a lead com-pound [50]. More recently, however, LE is now being appliedin FBDD, in the prioritization of fragment hits and conver-sion of fragment hits to leads. The Abbott group carried outa retrospective analysis of 18 highly-optimized inhibitors,demonstrating the remarkably linear relationship betweenpotency and molecular mass during ideal optimization,thereby, indicating that LE values stay almost constant duringthe ideal fragment-to-lead process [51]. A search through theliterature turns up 30 unique examples in the field of FBDDin which similar tendencies were observed [8].

While LE provides an important metric for assessing thepotential of fragment hits, its value is intrinsically related to

molecular size, and smaller molecules have inherently greaterLE values than larger ones. The Johnson & Johnson groupfirst proposed an empirical ‘fit quality’ metric to compensatefor this dependency by calculating the maximum LE for thenumber of HAs with affinity data derived from the bindingdatabase [52,53]. Our group also introduced a different index(% LE) [8] derived using Kuntz’s data [54], and the AstraZenecagroup proposed the measurement of size-independent ligandefficiency [55]. Recently, the Merck group published a reporton using the unempirical factor LELP (logP/LE) to reflectthe lipophilic component of activity [56]. These indices aresummrized in Table 5.

Although these modified LE values are admittedly conve-nient for assessing the quality of fragment hits, they do notreflect the mode of interaction of fragment hits with the targetprotein, and are mainly intended for primary filtering offragment hits.

4.2 Structural interaction fingerprints and hydrationanalysisIn FBDD projects, a wide variety of binding modes are oftenobserved with regard to complex structures of fragment hitsand target proteins. For example, fragments may interact withvarious regions of a pocket or even with a different pocketentirely. On observation of binding, fragment hits must beanalyzed and prioritized based on the mode of interaction.

To this end, we must determine how to best represent andunderstand the intermolecular interactions between multiple

Irradiation

H

(3) ligand,relaxation (19F)

(2) ligand,relaxation (1H)

(4) ligand,STD-NMR

(5) ligand,water-LOGSY

Ligand

Protein Protein

H H

15N 13C

HLigand

HLigandF

H2O

IrradiationMagnetization transfer

Observation Observation Observation

Magnetizationtransfer

Observation

A. B. C.

(1) protein,chemical shift

Protein

H2O

Figure 4. Schematic representation of NMR techniques frequently applied to FBDD: A.) 1) chemical shifts in protein, 2) relaxation(1H), 3) relaxation (19F); B) 4) STD-NMR; c) 5) WaterLOGSY.FBDD: Fragment-based drug discovery; STD: Saturation transfer difference; WaterLOGSY, Water-ligand observed via gradient spectroscopy.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

fragment hits and a target protein, a critical step in theutilization of complex structure data. Structural interactionfingerprints (SIFts) are particularly useful in addressingthis issue; SIFts translate 3D structural binding data into a1D binary string [57]. From this data string, we can derivethe residue in contact with compounds as well as the typeof interaction occurring (main-chain/side-chain and polar/non-polar/hydrogen bonding). Further, the string can beused to visualize the binding mode of a given compound,classify compounds from the viewpoint of the bindingmode and mine new structures containing key interactions.This residue-based interaction fingerprint has recently beenextended to atom-based interaction fingerprints, allowing forbetter descriptions of ligand–protein interactions, includingmeasurements of interaction strength [58,59].

After determining the best method of representing theinteraction, we must then determine what the interactionmode of the promising fragment is and to which regionthe promising fragment binds. Clarifying and understandingthe character of the ideal fragment hits is essential for thisissue. The ideal fragment hit becomes the scaffold of the finallead compound [60], indicating that good SAR can be observedfor additional portions outside the ideal fragment moiety andbinding mode usually remains during fragment-to-lead con-version. It is, therefore, important to identify which region canbe the molecular recognition motif that recognizes the scaf-fold. One typical, recently identified motif is characterized bya hydrophobic enclosure in which the sides of the cavity areformed by hydrophobic protein side chains, and the ligandforms correlated hydrogen bonds with the target proteins [42].

Librarydesign

Fragmentscreening

Fragment hitsprioritization

Fragment hit tolead

To be considered Approaches referencedin this paper

Chemical group

Predicted activity

Ligand efficiency

SIFtHydration analysis

Hydration analysis

HybridizationEnergy calculation

DiversitySynthetic feasibility

Property

Docking mode

Synthetic feasibilityActivityInteraction mode

Property

Property

Figure 5. Computational approach in the processes of FBDD.FBDD: Fragment-based drug discovery.

Table 5. Ligand efficiency indices.

Name Definition Comment Ref.

LE -RTln(Kd)/(HA) ~ -RTln(pKi)/(HA) Original definition about ligand efficiency [50]

BEI pKi (or pKd)/MW Similar to LE [108]

SEI pKi (or pKd)/PSA Simultaneous use with BEI [108]

LLE pKi - cLogP (or LogD) Metric of acceptable lipophilicity per unit of in vitro potency [109]

FQ LE/(-0.064 + 0.873 *e(-0.026*HA)) Empirical score for correction of size-dependency [52]

FQ LE/(0.0715 + 7.5328/(HA)+25.7079/(HA)2+ –361.4722/(HA)3

Empirical score for correction of size-dependency [53]

%LE LE/(1.614log2(10/HA))*100 Empirical score for correction of size-dependency [8]

LELP logP/LE Metric of contribution of lipophilic component to LE [56]

SILE -RTln(pKi)/(HA)03 Empirical score for correction of size-dependency [55]

FQ: Fit quality; HA: Number of non-hydrogen heavy atoms; LE: Ligand efficiency; LLE: Ligand-lipophilicity efficiency; LELP: Ligand-efficiency-dependent lipophilicity;

SEI: Surface-binding efficiency index; SILE: Size-independent ligand efficiency.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

This particular motif’s contribution to overall binding affinityis often found to be greater than expected, owing in part to thefact that the compound’s release of a water molecule into thebulk fluid from the suboptimal environment in the cavityresults in an increase in binding affinity.

Hydration analysis of the region shows that several watermolecules form a rigid, stable cluster with the hydrogen-bonding network within the cluster as well as with thetarget protein, exhibiting unusual entropic instability andenthalpic stability [43]. The displacement of this cluster bythe compound that can mimic the favorable interactionof water molecules with target proteins and recapture theprotein–water interaction energy leads to the large entropiccontribution to the binding affinity [43,61,62]. In other words,identification of the water cluster in the hydrophobic enclo-sure via hydration analysis allows for identification of themost favorable binding sites and interaction modes. Indeed,several reports have found that Schrodinger’s docking pro-gram Glide XP, with its new scoring function implementingthe concept of hydrophobic enclosure, was able to predict thebinding mode of fragment hits for several protein targets withhigh accuracy [63]. Further, our in-house FBDD programsfound that the promising fragment hits expels entrolpic,unstable water clusters and appropriately interacted withprotein as observed in the protein–water complex.

Hydration analysis also enables identification of hydra-tion sites with anomalous unfavorable energy near the knownactive compounds, proving its efficacy not only in prioritiz-ing fragment hits but also in suggesting regions suitable foradding additional chemical groups to improve activity andselectivity during the fragment-to-lead process [61].

4.3 Hybridization methodsAlthough SIFt and hydration analysis are admittedly pow-erful tools for classifying and prioritizing fragment hitswith respect to 3D complex structures, these methods arenot intended for use in de novo design. Thus, the next goal ofresearchers is to design new compounds while takingfull advantage of the structural information obtained from> 100 complex structures. To accomplish this, the novelde novo design technique BREED has been proposed [64].Using the BREED technique, two complex structures aresuperposed and hybridization is performed by merging orswapping substructures between the compounds along spa-tially overlapped bonds. This method is relatively straightfor-ward to medicinal chemists and can be carried out manuallywith a few complex structures. For experiments involving largenumbers of complex structures, however, this technique isbetter performed automatically. N complex structures willresult in (N � (N-1))/2 pairs, and the generated compoundscan be used as input compounds in subsequent procedures; inthis manner, a large set of novel compounds can be createdfrom a small number of starting compounds.

In contrast to other de novo programs, the BREED methodis based entirely on experimental data and is not affected by

protein flexibility, because the compounds are superposed intheir active conformation. Further, the method is applicableto complex structures from not only the same protein or samefamily, but also to structures from unrelated proteins withtopologically similar binding pockets. To take advantage ofthis ability and thereby further improve drug design, severalapproaches can be utilized for detecting topological similarityand alignment of two protein pockets [65-67].

The BREED methodology is now being implemented oncommercial platforms [68,69], and several approaches relatedto the classical BREED methodology have recently beenreported. The MEDIT SA group has developed a sequentialcomputational drug design protocol for FBDD consisting ofthe following: ligand detection for complex structures, frag-mentation of ligands, alignment of complex structures bytopological similarity of binding pockets and combinationof substructures into new hybrid compounds [70]. The Bayergroup has presented the idea of fragment shuffling, whichinvolves a scoring scheme for the incremental construction ofnovel ligands [71].

Of particular note is the fact that all of these methods aremore effective with fragment hits than with larger compounds,such as HTS hits. Because a larger compound will sometimesinteract in the suboptimized binding mode, due to interfer-ence by inappropriate side chains, its substructures will intrin-sically generate inappropriate compounds on hybridization.The exponential increase in the number of complex structureshas clearly stimulated the investigation into efficient proto-cols using these experimental data, such as the hybridizationmethods mentioned above, and it can be expected that thesemethods will also be widely used in FBDD.

4.4 Binding energy calculationOne of the most vexing problems in the field of computa-tional chemistry is finding an efficient method of analyzingof protein–ligand binding energies accurately [72,73]. In theFBDD field, an understanding of protein–ligand bindingenergy is especially useful in converting a fragment hit toa lead.

Thus far, many methods have been presented regardinginvestigating the contribution of protein–ligand complexesto binding free energies (Table 1). For example, molecularmechanics-Possion-Boltzmann surface area and molecularmechanics-generalized Born surface area, both commonlyused methodologies in obtaining binding energies and ana-lyzing the energy contribution of protein–ligand binding, areavailable for use in library design for fragment-to-lead con-version [74,75]. However, accuracy with these two methods islimited, as many important contributions are not well con-sidered. In contrast, free energy perturbation and thermody-namic integration may be the most rigorous to perform andis the strictest of techniques [76,77]. These methods have beenused frequently to calculate the free energy difference betweensimilar molecules. However, obtaining absolute bindingenergies is extremely inefficient because these methods require



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

long time simulation to maintain the reversibility about thework associated with the decoupling process. Recently, thePande group developed the Bennett acceptance ratio method,a novel binding free energy method based on Jarzynski’stheory, and used this method to calculate the absolute bind-ing free energies for FKBP ligand complexes in which theFKBP binding modes have been well investigated [72]. Further,Tanida et al. demonstrated that the absolute binding freeenergies have an effective linear relationship between com-puted and experimental values when binding theophyllineand its analogue to an RNA aptamer [73]. While no reporthas been made regarding using these types of accurate yetrigorous calculations in the fragment-to-lead process, thiscalculation method may be one of the most promising foraccurately predicting binding free energies for protein–ligandcomplexes and investigating thermodynamic properties.

4.5 Protein flexibilityProtein flexibility plays an important role in protein–ligandinteraction [78] and should be taken into account and analyzedfor efficient drug design [79]. A wide range of conformationalchanges have been noted from large interdomain movementsto small side-chain rearrangements in binding pocket residues,even in the absence of a ligand [80,81]. Ligand binding-inducedconformational changes in proteins are more common thanin apo-form proteins [81,82]. Ligands with diverse chemicalscaffolds are known to induce a range of changes in proteinconformation [83].Further, some proteins can adopt different conformations

to accommodate similar ligands. Analysis of 206 bindingsite pairs of structurally similar ligands binding to the sameproteins, which were listed in the Protein Data Bank, revealedside-chain movements in 50% of the pairs; changes weredetermined to have occurred if the RMSD for all side-chainatoms at one residue within 5 A

�of the ligand exceeded

1.0 A�

[84]. Conformational changes due to loop or domainmovements can occur, albeit rarely, upon binding of simi-lar analogues [84,85]. It is also possible to induce conforma-tion changes by the small size of compounds like fragmenthits. The Abbott group found that one fragment hit iden-tified by second-site NMR screening for HSP90 adoptedmultiple binding modes in two distinct protein conforma-tions, open and closed forms [86]. Other groups have alsoobserved fragment-induced conformational changes due toloop movements [87,88].Protein flexibility must be considered in virtual screening

of fragment libraries, prioritization of fragment hits and inthe fragment-to-lead process. There are several approaches toincorporate protein flexibility in docking. Soft docking [89],which allows some overlap between the protein and the ligandduring docking simulation, is the most computational effi-cient method, but does not work for large conformationalchanges. Ensemble docking [90,91] and multiple docking [92]

torelate large conformational change, but generation ofmuliple reliable conformations requires great computational

costs (molecular dynamic simulation, Monte Carlo samplingor normal mode analysis) and/or experimental structural data.Induced-fit docking approach has been recently proposed foruse in prediction of ligand binding-induced conformationalchange for a given compound [93]. However, prediction ofconformational changes for a target protein remains challeng-ing, and knowledge-based approaches using a large volume ofcomplex structure data will become more effective [94].

5. Combined use of technologies: examples ofFBDD

Many researchers have pointed out the importance of tak-ing multidimensional proprieties into account when select-ing starting compounds, and two papers in particular havereported on these properties. In their study, Ciulli et al. usedWaterLOGSY (Water-ligand observed via gradient spectros-copy) NMR spectroscopy, ITC under low c value conditions,inhibition studies and site-directed mutagenesis to probe‘hot spots’ at cofactor-binding sites of a model dehydrogenase,Escherichia coli ketopantoate reductase [95]. They found thatthe 2¢-phosphate and the reduced nicotinamide groups con-tributed significantly to the compound’s binding energy.These authors’ approach can similarly be applied to determine‘hot spots’ in fragment hits. In their study, Muzammil et al.used ITC and crystallography to analyze the binding prop-erties of several medium-to-low picomolar protease inhibitorsof wild-type protease, thermodynamically demonstrating thatthe good response of tipranavir arises from its unique behav-ior: tipranavir compensates for entropic losses either by actualenthalpic gains or by sustaining minimal enthalpic losseswhen binding the mutants [96]. As mentioned above, severalmethods are now available for selecting starting fragments,each with its own advantages and disadvantages. We must,therefore, consider several important points (affinity, complexstructure, thermodynamics properties, computational bindingenergy) when selecting starting fragments.

Our example of FBDD, including the design and syn-thesis of non-peptidic inhibitors for the Syk C-terminal Srchomology 2 (SH2) domain, is described in Figure 6 [97]. Sykbelongs to a family of hematopoietic cell-specific proteintyrosine kinases that play a critical role in mediating cellularresponses activated by the interaction between antigens andantibody receptors. Given these properties, this protein has,therefore, emerged as a potentially useful therapeutic targetfor immune suppression. In the present study, we were ableto successfully determine the structure for the Syk C-terminalSH2 domain via a series of triple-resonance experiments usingNMR. Results showed that the Syk C-terminal SH2 domaincontains the same three major pockets (pY, pY+1 and pY+3) asseveral other SH2 domains. Novel hit fragments for the pY andpY+1 pockets were identified using in silico screening, NMRand SPR (Compound 1 for pY pocket, IC50 = 5900 µM;Compound 2 for pY+1 pocket, IC50 = 8000 µM).Compound 3 (IC50 = 350 µM) was designed and synthesized



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

via the fragment-linking approach using the distance infor-mation obtained from NOE experiments with NMR andCompound 4 (IC50 = 38 µM), a non-peptidic inhibitor ofthe Syk C-terminal SH2 domain, was obtained with our frag-ment evolution approach. Results showed that Compound 4,which was made by combining the fragment linking andfragment evolution approaches, exhibited activity comparableto that of the monophosphorylated natural peptide ligandVpYTGLS (IC50 = 17 µM).

6. Conclusions

FBDD is a lead discovery process whereby new, high-affinitylead compounds are generated starting from low molecularmass fragments. Recent advancements in the technology usedin this technique have increased its speed of result generationand thereby made broad application possible. Here, we reviewrecent advances in experimental and computational technol-ogy in the field of FBDD which are key to its implementa-tion. The continued advancement of these technologies mayfurther increase the efficiency of the application of FBDD.

7. Expert opinion

The Abbott group used NMR in the first fragment-screeningby the SAR by NMR approach [13]. However, more recently,several non-NMR methods for fragment-screening have beendeveloped, including X-ray crystallography, SPR, ITC andHCS. Many reviews have pointed out both the advantagesand disadvantages in terms of sensitivity of detection, through-put, required instrumentation and the level of informationgenerated [6,15,98]. Fragment screening, therefore, requiresselection and integration of various techniques based on thetechnology available to each company at the time.

Further, the importance of the method used to integratevarious technologies for identification and validation of truehit fragments after primary screening cannot be overempha-sized. Although the size of the fragment library is importantin the successful application of FBDD, the number of thecompounds which can be assayed during screening of pri-mary fragments depends on the screening method [98]. NMRand X-ray crystallography can both provide useful structuralinformation clarifying fragment binding and informing

pYpocket

pY+1pocket

pY+3pocket

Fragmentlinking

Fragmentevolution

O

OOH

OH

NH

OS

N

OH

O

OO

Compound 1IC50 = 5900 µM


OHO

OOHN

H

OS

N

OH

O

OO

OHO

OOH

SN

OH

O

OO



Linking Evolution

Linking

Evolution

pYpocket

pY+1pocket

pY+3pocket

pYpocket

pY+1pocket

pY+3pocket

A. B. C.

Figure 6. Structural models of compounds 1 – 4 and the Syk C-terminal SH2 domain complex. The surface of the Syk C-terminalSH2 domain is colored according to the pockets on the protein surface. pY, pY+1 and pY+3 pockets are colored magenta, cyan and orange,respectively. A) Compounds 1 and 2 are initial hit fragments. B) Compound 3 was obtained via the fragment-linking approach.C) Compound 4 was further optimized via the fragment evolution approach.SH2: Src homology 2.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

researchers regarding introduction of functional groups toimprove fragment activity. However, screening throughputfor these methods is low, and only small fragment librariescan be examined due to the large amount of target proteinrequired. In contrast, throughput for SPR, ITC, MS andHCS is high, and large fragment libraries can be evaluated,although these methods can only provide data regardingbinding properties, with no structural information. In par-ticular, because it only functions as a simple biochemical assayat concentrations above 100 mM, HCS is cost effective, fastand can access any biological target. However, HCS as well asSPR, ITC and MS, often provide false positives, which mustthen be removed using another validation method such asX-ray crystallography. In general, the richer the informationobtained by a screening method, the smaller the size of theevaluable fragment library. Therefore, the size and contents of

the fragment-library should be considered based on the pri-mary screening method used (Figure 1). Some groups, includ-ing bio ventures, have developed specific technologies, suchas Graffinity Pharmaceuticals’ RAISE [45], Evotec’s single-molecule Fluorescence Correlation Spectroscopy detectiontechnique [99], ZoBio’s target immobilized NMR screeningmethod [100], Sunesis’s tethering [48] and substrate activityscreening of University of California, Berkeley [101,102]. Totake advantage of this technology, other pharmaceutical com-panies may opt to collaborate on research projects withbioventures such as these [103].

When designing a fragment library, the solubility of thefragments in aqueous solution is extremely important, asscreening will be conducted at a high concentration (above100 µM). Computational prediction of aqueous solubilityis attractive when simple judgment is possible, and this

0

0.2

0.4

0.6

0.8

1

1.2

1.4

0 5 10 15 20 25 30 35

HA

LE

0123456789

0 5 10 15 20 25 30 35

HA

pK

iHypothetical criteriaof drug leads

(pKi = 8)

Hypothetical criteria of drug leads (pKi= 8)

Hitfragment

Fragmentoptimization

Fragmentevolution

Fragment linking/self-assembly

Hit fragment B

Hit fragment A

Fragment optimization • Only one hit fragment required• Minor structural modification, but difficult to improve activity of the hit fragment while keeping molecule size almost constant

Hitfragment

Hit fragment B

Hit fragment A

Fragment evolution• Only one hit fragment required• Standard hit-to-lead approach which adds the functional groups that bind to additional parts of the target protein• Structural information is useful• Synergy with combinatorial chemistry

Fragment linking • More than two hit fragments required• Structural information is essential• Simple and rational approach to joining hit fragments, but difficult to design suitable linkers

Fragment self-assembly • More than two hit fragments required• Structural information is essential• Rational approach to making new scaffold combining partial structures of hit fragments, but difficult to design the molecule

+

A. B.

Figure 7. Concept of four different approaches to converting fragment hits to leads: fragment optimization, fragmentevolution, fragment linking and fragment self-assembly. A. Plot of HA versus LE and a hypothetical criteria line of drug leads (pKi = 8).Under the fragment optimization approach, hit fragments are optimized to drug leads while maintaining an almost constant molecule size,contributing to a perpendicular tendency in the plot. Under the fragment evolution approach, hit fragments are generally optimized to drugleads while maintaining an almost constant LE value, thereby contributing to a horizontal tendency in the plot. B. Plot of HA versus activityvalue (pKi) and a hypothetical criteria line of drug leads (pKi = 8). Under both the fragment linking and fragment self-assembly approaches,the pKi of the drug lead is expected to equal the sum of the pKi of hit fragments A and B. Therefore, in this plot, they serve as approacheswhich add the vector of fragments A and B. The four approaches are also described and compared in the lower part of this figure.HA: Heavy atom; LE: Ligand efficiency.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

technique was actually used by Astellas when designing afragment library. However, accurate prediction using thismethod remains difficult, because both the crystal andsolution states of the compound must be considered. Exper-imental determination of the aqueous solubility of a selectedfragment is, therefore, required. Recently, a group withAstraZeneca developed and applied a high-throughput aque-ous solubility assay and approach to managing the risk ofselecting poorly soluble fragments [104]. With regard to physi-cochemical parameters of the compounds suitable for a frag-ment library, the ‘rule of 3’ proposed by Astex Therapeuticsis well known [4]. This rule states that fragments shouldhave a molecular mass £ 300 g/mol, £ 3 hydrogen-bonddonors, £ 3 hydrogen-bond acceptors and a ClogP of £ 3,with optimal additional criteria of £ 3 rotatable bonds anda polar surface area £ 60 A

� 2[105]. Many research groups have

suggested different physicochemical filters for use in assemblingfragment libraries [105]. However, most of these approaches canbe related with the rule of 3. When designing a fragmentlibrary, we must also consider molecular diversity, synthetictractability, structural novelty, chemical stability, drug-likenessand privileged medicinal chemistry scaffolds, among otherimportant factors.

The many different approaches available for convertingfragment hits to leads are categorized into the followingfour types: fragment optimization, fragment linking, fragmentself-assembly and fragment evolution (Figure 7)[106]. Selectingthe appropriate approach here is important in convertingfragments to drug leads. Fragment optimization is the tradi-tional fragment-to-lead approach, which maintains an almostconstant molecule size. Further, fragment activity is improvedby minor structural change to the fragment, an effect whichis similar to that observed in common optimization researchof lead compounds by medicinal chemists.

Fragment linking and fragment self-assembly require twoor more fragments which bind to different parts of thebinding pocket. The fragment linking approach involves usinga linker, while the fragment self-assembly approach makesa completely new scaffold combining partial structures ofhit fragments. The Abbott group used fragment linking intheir SAR by NMR method [13], and if the linking or self-assembly is carried out well, further great improvements inactivity can be expected. However, in general, the fragment

linking and fragment self-assembly approaches are not easy tocarry out, due to difficulties in finding more than two frag-ments which bind to different parts in the binding pocket andin designing compounds without disrupting the bindingmodes of hit fragments.

Fragment evolution is an approach to fragment growingwhich adds functional groups that bind to additional partsof the target protein. Of the four methods for convertingfragment hits to leads, this approach has been the mostapplicable and successful. Unlike the fragment linking andfragment self-assembly, fragment evolution does not need asecond fragment which binds to different parts of the pocket,and structure-based drug design is able to accelerate efficientgrowing of fragments to drug-leads. Further, practical use ofcombinatorial chemistry in the growing step is also possible,and success using this simple fragment evolution techniquehas been reported more and more frequently.

Successful application of the FBDD technique will requireaccurate measurement of the weak affinities of fragmentsand enhancement of fragments to drug leads. Recentadvances in hard- and software, as well as computationaltechniques, have increased the throughput of structuralmethods such as X-ray crystallography and NMR, and theuse of such biophysical techniques such as SPR, ITC, HCSand MS is becoming increasingly efficient. In the past10 years, FBDD has gained popularity among pharmaceu-tical companies as a simple, quick, productive and cheapapproach to indentifying leads. More than 10 compoundshave already entered clinical development thanks to theFBDD technique, and this approach will probably continueto attract further attention and lead to development of evenmore compounds.

Acknowledgements

The authors thank the many Astellas scientists, K Suzumura,M Sekiguchi, S Ogino, H Sakashita, T Hondo, N Katayama,A Moritomo, M Oku, K Mori, H Fuji and Y Matsumotofor helpful discussions.

Declaration of interest

The authors are employees of Astellas Pharma, Inc.



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

Bibliography1. Albert JS, Blomberg N, Breeze AL,

et al. An Integrated approach to

fragment-based lead generation:

philosophy, strategy and case studies

from astrazenecas drug discovery

programmes. Curr Top Med Chem

2007;7:1600-29

2. Alex AA, Flocco MM. Fragment-based

drug discovery: what has it achieved so far?

Curr Top Med Chem 2007;7:1544-67

3. Chessaria G, Woodhead AJ. From

fragment to clinical candidate–a historical

perspective. Drug Discov Today

2009;14(13-14):668-75

4. Congreve M, Carr R, Murray C, et al. A

’Rule of Three’ for fragment-based lead

discovery? Drug Discov Today

2003;8(19):876-80

5. de Kloe GE, Bailey D, Leurs R,

de Esch IJP. Transforming fragments into

candidates: small becomes big in medicinal

chemistry. Drug Discov Today

2009;14(13-14):630-46

6. Erlanson DA, McDowell RS, O’Brien T.

Fragment-based drug discovery.

J Med Chem 2004;47(14):3463-82

7. Hajduk PJ, Greer J. A decade of

fragment-based drug design: strategic

advances and lessons learned. Nat Rev

Drug Discov 2007;6(3):211-9

8. Orita M, Ohno K, Niimi T. Two ‘golden

ratio’ indices in fragment-based drug

discovery. Drug Discov Today

2009;14(5-6):321-8

9. Jencks WP. On the attribution and

additivity of binding energies. Proc Natl

Acad Sci USA 1981;78(7):6-4050

10. Leach AR, Hann MM, Burrows JN,

Griffen EJ. Fragment screening: an

introduction. Strucutre-Based Drug

Design. In: Hubbard RE, editor, RSC

Publishing; 2007. p. 142-72

11. Miranker A, Karplus M. Functionality

maps of binding-sites - a multiple copy

simultaneous search method.

Proteins Struct Funct Genet

1991;11(1):29-34

12. Caflisch A, Miranker A, Karplus M.

Multiple copy simultaneous search

and construction of ligands in

binding-sites - application to inhibitors of

hiv-1 aspartic proteinase. J Med Chem

1993;36(15):2142-67

13. Shuker SB, Hajduk PJ, Meadows RP,

Fesik SW. Discovering high-affinity

ligands for proteins: SAR by NMR.

Science 1996;274(5292):1531-4

14. Bembenek SD, Tounge BA, Reynolds CH.

Ligand efficiency and fragment-based drug


2009;14(5-6):278-83

15. Carr RA, Congreve M, Murray CW,

Rees DC. Fragment-based lead discovery:

leads by design. Drug Discov Today

2005;10(14):987-92

16. Hann MM, Leach AR, Harper G.

Molecular complexity and its impact on

the probability of finding leads for drug

discovery. J Chem Inf Comput Sci

2001;41(3):856-64

17. Available from: http://www.ttplabtech.

com/ [cited]

18. Available from: http://www.

matrixtechcorp.com/ [cited]

19. Available from: http://www.innovadyne.

com/ [cited]

20. Davies TG, VerdonkML, Graham B, et al.

A structural comparison of inhibitor

binding to PKB, PKA and PKA-PKB

chimera. J Mol Biol 2007;367(3):882-94

21. Hartshorn MJ, Murray CW,

Cleasby A, et al. Fragment-based lead

discovery using x-ray crystallography.

J Med Chem 2005;48(2):403-13

22. Tickle I, Sharff A, Vinkovic M, et al.

High-throughput protein crystallography

and drug discovery. Chem Soc Rev

2004;33(8):558-65

23. Available from: http://pfwww.kek.jp/

index.html [cited]

24. Available from: http://www.rigaku.com/

index_world.html [cited]

25. Available from: http://smb.slac.stanford.

edu/robosync/index.html [cited]

26. Skarzynski T, Thorpe J. Industrial

perspective on X-ray data collection and

analysis. Acta Crystallogr D

Biol Crystallogr 2006;62:102-7

27. Campobasso N. Harnessing fragments in

big pharma: examples and processes

at GlaxoSmithKline. AstraZeneca,

Alderley Park, UK: RSC BMCS

Fragments; 2009

28. Yamada Y, Hiraki M, Sasajima K, et al.

AR-NE3A, a new macromolecular

crystallography beamline for

pharmaceutical applications at the photon

factory. AIP Conference Proceedings

2009. In press

29. Hiraki M, Watanabe S, pHonda N, et al.

High-throughput operation of

sample-exchange robots with double tongs

at the photon factory beamlines.

J Synchrotron Radiat 2008;15:300-3

30. Available from: http://accelrys.com/ [cited]

31. Available from: http://www.ccp4.ac.uk/

[cited]

32. Oldfield TJ. X-LIGAND: an application

for the automated addition of flexible

ligands into electron density.

Acta Crystallogr D Biol Crystallogr

2001;57:696-705

33. Available from: http://www.dectris.com/

sites/dectris.html [cited

34. Miyoshi T, Igarashi N, Matsugaki N,

Yamada Y, Hirano K, Hyodo K, et al.

Development of an X-ray HARP-FEA

detector system for high-throughput

protein crystallography.

J Synchrotron Radiat 2008;15:281-4

35. Klages J, Coles M, Kessler H. NMR-based

screening: a powerful tool in

fragment-based drug discovery.

Mol Biosyst 2006;2(6-7):319-31

36. Dalvit C, Fagerness PE,

Hadden DTA, et al. Fluorine-NMR

experiments for high-throughput

screening: theoretical aspects,

practical considerations, and range of

applicability. J Am Chem Soc

2003;125(25):7696-703

37. Tsao DHH, Sutherland AG, Jennings LD,

et al. Discovery of novel inhibitors of the

ZipA/FtsZ complex by NMR fragment

screening coupled with structure-based

design. Bioorg Med Chem

2006;14(23):7953-61

38. Hohwy M, Spadola L, Lundquist B, et al.

Novel prostaglandin D synthase inhibitors

generated by fragment-based drug design.

J Med Chem 2008;51(7):2178-86

39. Lepre CA. Library design for NMR-based

screening. Drug Discov Today

2001;6(3):133-40

40. Freire E. Do enthalpy and entropy

distinguish first in class from best in class?

Drug Discov Today

2008;13(19-20):869-74

41. Neumann L, Ritscher A, Muller G,

Hafenbradl D. Fragment-based lead

generation: identification of seed fragments

by a highly efficient fragment screening

technology. J Comput-Aided Mol Des

2009;23(8):501-11



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

42. Friesner RA, Murphy RB, Repasky MP,

et al. Extra precision glide: docking and

scoring incorporating a model of

hydrophobic enclosure for protein-ligand

complexes. J Med Chem

2006;49(21):6177-96

43. Young T, Abel R, Kim B, et al. Motifs for

molecular recognition exploiting

hydrophobic enclosure in protein-ligand

binding. Proc Natl Acad Sci USA

2007;104(3):808-13

44. Ekstrom JL, Pauly TA, Carty MD, et al.

Structure-activity analysis of the purine

binding site of human liver glycogen

phosphorylase. Chem Biol

2002;9(8):915-24

45. Neumann T, Junker H-D, Schmidt K,

Sekul R. SPR-based fragment screening:

advantages and applications. Curr Top

Med Chem 2007;7(16):1630-42

46. Williams G. Integrating biophysical

methods into fragment-baed screening.

San Diego, CA: Fragment-based Lead

Discovery Conference; 2008

47. Seth PP, Miyaji A, Jefferson EA, et al. SAR

by MS: discovery of a new class of

RNA-binding small molecules for the

hepatitis C virus: internal ribosome entry

site IIA subdomain. J Med Chem

2005;48(23):7099-102

48. Erlanson DA, Braisted AC, Raphael DR,

et al. Site-directed ligand discovery.

Proc Natl Acad Sci USA

2000;97(17):9367-72

49. Oslob JD, Romanowski MJ, Allen DA,

et al. Discovery of a potent and selective

aurora kinase inhibitor. Bioorg Med

Chem Lett 2008;18(17):4880-4

50. Hopkins AL, Groom CR, Alex A. Ligand

efficiency: a useful metric for lead selection.

Drug Discov Today 2004;9(10):430-1

51. Hajduk PJ. Fragment-based drug design:

how big is too big? J Med Chem

2006;49(24):6972-6

52. Reynolds CH, Bembenek SD, Tounge BA.

The role of molecular size in ligand

efficiency. Bioorg Med Chem Lett

2007;17(15):4258-61

53. Reynolds CH, Tounge BA, Bembenek SD.

Ligand binding efficiency: trends, physical

basis, and implications. J Med Chem

2008;51(8):2432-8

54. Kuntz ID, Chen K, Sharp KA,

Kollman PA. The maximal affinity of

ligands. Proc Natl Acad Sci USA

1999;96(18):9997-10002

55. Nissink JWM. Simple size-independent

measure of ligand efficiency. J Chem

Inf Model 2009;49(6):1617-22

56. Keseru GM,Makara GM. The influence of

lead discovery strategies on the properties

of drug candidates. Nat Rev Drug Discov

2009;8(3):203-12

57. Deng Z, Chuaqui C, Singh J.

Structural Interaction Fingerprint (SIFt):

a novel method for analyzing

three-dimensional protein-ligand binding

interactions. J Med Chem

2004;47(2):337-44

58. Kelly MD, Mancera RL. Expanded

interaction fingerprint method for

analyzing ligand binding modes in

docking and structure-based drug design.

J Chem Inf Comput Sci

2004;44(6):1942-51

59. Mpamhanga CP, Chen B, McLay IM,

Willett P. Knowledge-based interaction

fingerprint scoring: a simple method for

improving the effectiveness of fast scoring

functions. J Chem Inf Model

2006;46(2):686-98

60. Murray CW, Verdonk ML. The

consequences of translational and

rotational entropy lost by small molecules

on binding to proteins. J Comput Aided

Mol Des 2002;16(10):741-53

61. Abel R, Young T, Farid R, Berne BJ, et al.

Role of the active-site solvent in the

thermodynamics of factor xa ligand

binding. J Am Chem Soc

2008;130(9):2817-31

62. Thijs B, Ramy F, Woody S.

High-energy water sites determine

peptide binding affinity and specificty of

PDZ domains. Protein Sci

2009;18(8):1609-19

63. Loving K, Salam N, ShermanW. Energetic

analysis of fragment docking and

application to structure-based

pharmacophore hypothesis generation.

J Comput Aided Mol Des

2009;23(8):541-54

64. Pierce AC, Rao G, Bemis GW. BREED:

generating novel inhibitors through

hybridization of known ligands.

Application to CDK2, P38, and HIV

protease. J Med Chem

2004;47(11):2768-75

65. Jambon M, Imberty A, Delege G,

Geourjon C. A new bioinformatic

approach to detect common 3D sites in

protein structures. Proteins Struct

Funct Bioinform 2003;52(2):137-45

66. Ramensky V, Sobol A, Zaitseva N, et al.

A novel approach to local similarity of

protein binding sites substantially

improves computational drug design

results. Proteins Struct Funct Bioinform

2007;69:349

67. Schmitt S, Kuhn D, Klebe G. A new

method to detect related function among

proteins independent of sequence and fold

homology. J Mol Biol

2002;323(2):387-406

68. MOETM (Molecular Operating

Environment), Chemical Computing

Group Inc, Montreal, Canada

69. Schrodinger package, Schrdinger LLC,

New York, NY

70. Moriaud F, Doppelt-Azeroual O,

Martin L, et al. Computational

fragment-based approach at pdb scale by

protein local similarity. J Chem

Inform Model 2009;49(2):280-94

71. Nisius B, Rester U. Fragment shuffling: an

automated workflow for three-dimensional

fragment-based ligand design. J Chem

Inf Model 2009;49(5):1211-22

72. Fujitani H, Tanida Y, Ito M, et al. Direct

calculation of the binding free energies of

FKBP ligands. J Chem Phys

2005;123(8):084108.1-084108.5

73. Tanida Y, Ito MS, Fujitani H. Calculation

of absolute free energy of binding for

theophylline and its analogs to RNA

aptamer using nonequilibrium work

values. Chem Phys 2007;337(1-3):135-43

74. Artis DR, Lin JJ, Zhang C, et al.

Scaffold-based discovery of indeglitazar, a

PPAR pan-active anti-diabetic agent.

Proc Natl Acad Sci USA

2009;106(1):262-7

75. Card GL, Blasdel L, England BP, et al.

A family of phosphodiesterase inhibitors

discovered by cocrystallography and

scaffold-based drug design. Nat Biotechnol

2005;23(2):201-7

76. Beveridge DL, DiCapua FM. Free energy

via molecular simulation: applications to

chemical and biomolecular systems.

Annu Rev Biophys Biophys Chem

1989;18:431-92

77. Kollman P. Free-energy

calculations - applications to chemical

and biochemical phenomena. Chem Rev

1993;93(7):2395-417

78. Mobley DL, Dill KA. Binding of

small-molecule ligands to proteins: “what



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

you see” is not always “what you get”.

Structure 2009;17(4):489-98

79. Davis AM, St-Gallay SA, Kleywegt GJ.

Limitations and lessons in the use of

X-ray structural information in drug

design. Drug Discov Today

2008;13(19-20):831-41

80. Ma BY, Shatsky M, Wolfson HJ,

Nussinov R. Multiple diverse ligands

binding at a single protein site: A matter of

pre-existing populations. Protein Sci

2002;11(2):184-97

81. Najmanovich R, Kuttner J, Sobolev V,

Edelman M. Side-chain flexibility in

proteins upon ligand binding.

Proteins Struct Funct Genet

2000;39(3):261-8

82. Gutteridge A, Thornton J.

Conformational changes observed in

enzyme crystal structures upon substrate

binding. J Mol Biol 2005;346(1):21-8

83. Teague SJ. Implications of protein

flexibility for drug discovery. Nat Rev

Drug Discov 2003;2(7):527-41

84. Bostrom J, Hogner A, Schmitt S. Do

structurally similar ligands bind in a similar

fashion? J Med Chem

2006;49(23):6716-25

85. Kim K. Outliers in SAR and QSAR: 2. Is a

flexible binding site a possible source of

outliers? J Comput Aided Mol Des

2007;21(8):421-35

86. Huth JR, Park C, Petros AM, et al.

Discovery and design of novel HSP90

inhibitors using multiple fragment-based

design strategies. Chem Biol Drug Des

2007;70:1-12

87. Hubbard R. Recent lessons in

fragment-based discovery. San Diego, CA:

Chembridge Healthcare Institute’s

Fourth Annual: Fragment-Based Drug

Discovery; 2009

88. Kuglstatter A. Impact of fragment

screening on the rapid discovery of novel

and selective protein kinase inhibitors.

San Diego, CA: Chembridge Healthcare

Institute’s Fourth Annual: Fragment-Based

Drug Discovery; 2009

89. Jiang F, Kim SH. Soft docking - matching

of molecular-surface cubes. J Mol Biol

1991;219(1):79-102

90. Knegtel RMA, Kuntz ID, Oshiro CM.

Molecular docking to ensembles of

protein structures. J Mol Biol

1997;266(2):424-40

91. Huang SY, Zou XQ. Ensemble docking of

multiple protein structures: considering

protein structural variations in molecular

docking. Proteins 2007;66(2):399-421

92. Sutherland JJ, Nandigam RK, Erickson JA,

ViethM. Lessons in molecular recognition.

2. Assessing and improving cross-docking

accuracy. J Chem Inf Model

2007;47(6):2293-302

93. Sherman W, Day T, Jacobson MP,

Friesner RA, Farid R. Novel procedure for

modeling ligand/receptor induced fit

effects. J Med Chem 2006;49(2):534-53

94. Subramanian J, Sharma S, B-Rao C. A

novel computational analysis of

ligand-induced conformational changes in

the ATP binding sites of cyclin dependent

kinases. J Med Chem

2006;49(18):5434-41

95. Ciulli A, Williams G, Smith AG, et al.

Probing hot spots at protein-gand binding

sites: a fragment-based approach using

biophysical methods. J Med Chem

2006;49(16):4992-5000

96. Muzammil S, Armstrong AA, Kang LW,

et al. Unique thermodynamic response of

tipranavir to human immunodeficiency

virus type 1 protease drug resistance

mutations. J Virol 2007;81(10):5144-54

97. Niimi T, Orita M, Okazawa-Igarashi M,

et al. Design and synthesis of non-peptidic

inhibitors for the Syk C-terminal SH2

domain based on structure-based in-silico

screening. J Med Chem

2001;44(26):4737-40

98. Siegal G, Ab E, Schultz J. Integration of

fragment screening and library design.

Drug Discov Today

2007;12(23-24):1032-9

99. Hesterkamp T, Barker J, Davenport A,

Whittaker M. Fragment based drug

discovery using fluorescence correlation

spectroscopy techniques: challenges and

solutions. Curr Top Med Chem

2007;7:1582-91

100. Vanwetswinkel S, Heetebrij RJ,

van Duynhoven J, et al. TINS, target

immobilized NMR screening: an efficient

and sensitive method for ligand discovery.

Chem Biol 2005;12(2):207-16

101. Wood WJL, Patterson AW, Tsuruoka H,

et al. Substrate activity screening: a

fragment-based method for the rapid

identification of nonpeptidic protease

inhibitors. J Am Chem Soc

2005;127(44):15521-7

102. Patterson AW, Wood WJL, Ellman JA.

Substrate activity screening (SAS): a

general procedure for the preparation and

screening of a fragment-based

non-peptidic protease substrate library for

inhibitor discovery. Nat Protocols

2007;2(2):424-33

103. Warr WA. Fragment-based drug discovery.

J Comput Aided Mol Des

2009;23(8):453-8

104. Colclough N, Hunter A, Kenny PW, et al.

High throughput solubility determination

with application to selection of compounds

for fragment screening. Bioorg Med Chem

2008;16(13):6611-16

105. Brewer M, Ichihara O, Kirchhoff C, et al.

Fragment-Based Drug Discovery: A

Practical Approach. In: Zartler ER, editor,

Wiley Publishing; 2008. p. 39-62

106. Rees DC, Congreve M, Murray CW,

Carr R. Fragment-based lead discovery.

Nat Rev Drug Discov 2004;3(8):660-72

107. Thomson Pharma (Copyright � 2009

Thomson Reuters).Available from: http://

www.thomson-pharma.com/ [cited]

108. Abad-Zapatero C, Metz JT. Ligand

efficiency indices as guideposts for drug


2005;10(7):464-9

109. Leeson PD, Springthorpe B. The influence

of drug-like concepts on decision-making

in medicinal chemistry. Nat Rev

Drug Discov 2007;6(11):881-90

AffiliationMasaya Orita†, Masaichi Warizaya,

Yasushi Amano, Kazuki Ohno & Tatsuya Niimi†Author for correspondence

Drug Discovery Research,

Astellas Pharma,

Inc., 21 Miyukigaoka Tsukuba,

Ibaraki 305-8585, Japan

Tel: +81 29 863 6768; Fax: +81 29 856 2558;

E-mail: [email protected]



Exp

ert O

pin.

Dru

g D

isco

v. D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y U

nive

rsity

of

Was

hing

ton

on 1

1/05

/14

For

pers

onal

use

onl

y.

Documents

Advances in fragment-based drug discovery platforms