Dr. Sander B. Nabuurs

Computational Drug Discovery groupCenter for Molecular and Biomolecular InformaticsRadboud University Medical Centre

The road to new drugs.

How to find new hits?

High Throughput Screening (HTS)

Virtual Screening (VS)

Integration HTS and VS

Molecular docking.

Considering protein flexibility.

Structure-based drug design in practice: Influenza case study.

WHAT DO WE WANT?

The goal of a drug is tomodulate the function ofits target receptor whichresult in a pharmacologicaleffect in the human body.

HOW DO WE GET THERE?

Developing a new drug: is extremely difficult.

takes a lot of time (> 10 years).

is very expensive (~ 1 billion $).

target compound pharmacologicaleffect

Marketing &Sales

RegistrationClinical

DevelopmentPre-Clinical

DevelopmentLead

OptimizationLead

DiscoveryTarget

Discovery

RESEARCH DEVELOPMENT

O

NH

S

F

CN

Lead Discovery

Hit / Lead Candidate

Lead Optimization

Development Candidate

Computer Aided Drug Design

Target Discovery

TargetIdentification

TargetValidation

AssayDevelopment

O

NH

S

F

CN

Bioinformatics

In Vitro

In Vivo

Computer-Aided Drug Design (CADD) refers to theapplication of informatics methods within rationaldrug design, to discover, design and optimizebiologically active compounds.

Ligands unknown

Target protein structure unknown

high-throughputscreening

Target protein structure known

TWO SCENARIOS

Target geneTarget DiscoveryLigands unknown

HTS is the most importantsource of new hits.

Pharmaceutical companieshave screening libraries upto a few million compounds.

Chemical space of drug-likemolecules is > 1080

.

Building a good screeningcollection is crucial!

Active compounds

Compound collection

In-housecollection

Availableto buy

Everything Diverseselection

Focusedselection

Screening set?

DIVERSE SELECTION FOCUSED SELECTION

Diverse selection:identify

dissimilarcompounds

Focused selection:Identifysimilar

compounds

DIVERSE SELECTION

Chemical Descriptor 2

Ch

em

ical

Des

crip

tor

1

HOW CAN WE MEASURE SIMILARITY?

In the selection ofscreening compoundsthe ‘Rule of Five’ isoften used.

It summarizes typicalproperties of knowndrugs.

These rules are oftenused as a first filteringstep.

In 1997 Chris Lipinski observedfor many drugs:

molecular weight < 500

lipophilicity (LogP) < 5

H-bond donors < 5

H-bond acceptors < 10

rotatable bonds < 10

DIVERSE SELECTION FOCUSED SELECTION

Chemical Descriptor 2

Ch

em

ical

Des

crip

tor

1

Chemical Descriptor 2

Ch

em

ical

Des

crip

tor

1

Sampling around known active sub-structures or structuralfragments can improve the quality of the library.

dopamine derivative

The use of chemical (ormolecular) descriptors isbased on the similarproperty principle.

Molecules with similarstructures and similarproperties should alsoexhibit similar activity.

Chemical Descriptor 2

Ch

emic

al D

esc

rip

tor

1

Fingerprints consistof various descriptorsencoded into bitstrings.

These descriptorscan be fragments orthe presence orabsence of otherproperties.

dopamine derivative

HOW CAN WE MEASURE SIMILARITY?

xA = 8xAB = 5

xB = 6

ABBA

ABAB

xxx

xS

Tanimoto coefficient

56.09

5

568

5

ABS

Note: this is just one of many different similarity measures!

1IN HIV Protease inhibitorTanimoto Similarity 0.47Tanimoto Similarity 0.47

VAC HIV Protease inhibitor

XN1 HIV Protease inhibitorTanimoto Similarity 0.63

MK1 HIV Protease inhibitorTanimoto Similarity 0.63

BEB HIV Protease inhibitorTanimoto Similarity 0.60

BEH HIV Protease inhibitorTanimoto Similarity 0.60

PZQ NOT HIV Prot. inhibitor Tanimoto Similarity 0.49

TI3 NOT HIV. inhibitor TS Tanimoto Similarity 0.48

A high Tanimoto Similarity can beuseful for prioritization.

However, no guarantees!

flutamide retro-flutamide

progesterone receptor 4 nM 6 nM

glucocorticoid receptor 25 nM 38 nM

androgen receptor 0.5 nM 55 nM

flutamide retro-flutamide

NH

O OHCF3

BrO2N

NH

OHCF3

BrO2N

O

Despite being the majorsource of new hits, HTS hasits drawbacks:

It’s expensive. In practice only accessible to

industry. Logistical errors. e.g. ‘frequent hitters’

Measurement errors. e.g. suboptimal readout

Strategic errors. e.g. assay variability

Active compounds

Compound collection

Ligand(s) unknown

Target protein structure unknown

high-throughputscreening

Target protein structure known

virtual screening

TWO SCENARIOS

Target geneTarget DiscoveryLigand(s) unknown

In Virtual Screening (VS)compounds are selectedusing computer programsto predict receptor binding.

VS is much cheaper and isable to process much morecompounds in less time.

Experimental validation ishowever always required! Active compounds

Compound database

A few success stories from virtual screening

STRUCTURE-BASED VS

Predict the orientation(and affinity) of a smallmolecule binding to aprotein target.

Requires the availability ofa 3D target structure!

Structure-based virtual screening

Active compounds

• Compounds to purchase.• Compounds from in-house library.• Virtual compounds.

Compound database

Docking

program

Target

protein

Compounddatabase

Docking

program

Target-Compoundcomplexes

Activecompounds

Despite its advantages VS alsohas its drawbacks:

Experimental validation isalways required.

Protein structure errors. e.g. ‘induced fit’

Sampling errors. e.g. faulty poses due to solvent

Scoring errors. e.g. false positives / negatives Active compounds

Compound database

Screening library

Focused library Hits

High ThroughputScreening

Virtual Screening

Hypothesis generation

Focused and sequential screening

VS hits

Hits

Analysis

Virtual Screening

Parallel and independent screening

HTS hits

High ThroughputScreening

Screening library

The road to new drugs.

How to find new hits?

High Throughput Screening (HTS)

Virtual Screening (VS)

Integration HTS and VS

Molecular docking .

Considering protein flexibility.

Structure-based drug design in practice: Influenza case study.

The docking problem involvesmany degrees freedom:

Translational.

Rotational.

Configurational

(Ligand + Receptor!)

Since the early eightiesseveral docking algorithmshave been devised.

These can be characterizedby the number of degreesfreedom that they ignore.

Target

protein

Compound

Docking

program

Target-Compoundcomplex

Fully flexible docking

Induced fit docking

Flexible ligand docking

Rigid body docking

Receptor

flexibility

Ligand

flexibility

Ligand

translations

Ligand

rotations

A number of flexible ligand dockingprograms:

Dock[Kuntz et al, J Mol Biol, 161:269-288, 1982]

Autodock[Morris et al, J Comput Chem, 19:1639-1662, 1994]

FlexX[Rarey et al, J Mol Biol, 261:470-489, 1996]

Gold[Jones et al, J Mol Biol, 267:727-748, 1997]

Glide[Friesner et al, J Med Chem, 47:1739-1749, 2004]

Molecular docking typically consists of two separate stages:

1. Exploration of conformational and configurational space.

2. Evaluation of the strength of the receptor-ligand interaction.

Targetprotein

Dockingprogram

Compound

Target-Compoundcomplex

Sampling

Scoring

Prior to ligand placement, mostdocking programs will create asimplified description of thetarget binding site.

This is typically done usingsimple geometry descriptors,like spheres or triangles .

These geometrical descriptorsare usually combined withchemical and electrostaticdescriptors to guide ligandplacement.

Receptor

Ligand

Example 1

Example 2

Docking programs generate alarge number of differentdocking poses.

In general one can distinguishtwo different scenarios:

1. Many different poses of thesame ligand need to be rankedfor accuracy.

2. Different poses of differentligands need to be rankedbased on their receptoraffinity.

The ideal scoring function workswell in both cases...

1 …52 3 4

1 …52 3 4

• First principles scoring functionsgenerally use a Molecular Mechanicsforce field.

• Such force fields typically containintra-molecular terms:

– Bond lengths

– Bond angles

– Dihedral terms

• And inter-molecular terms:

– Van der Waals contacts (non-polar)

– Electrostatic interactions (polar)Ebind = Eintra + Enonpolar + Epolar

• Empirical scoring functions have beendeveloped to score ligands very rapidly.

ΔGbind= ΔG0 +

ΔGpolar · Σ f(Complex) +

ΔGnon-polar · Σ f(Complex) +

ΔGrot · Nrotatable-bonds

• ΔG0, ΔGpolar, ΔGnon-polar, and ΔGrot areempirically parameterized weights.

• f(Complex) is a penalty function aimedat penalizing any unfavorableinteraction geometries.

In practice moleculardocking is generally used toanswer two different typesof questions:

1. Which compounds in mycompound collection couldbe active on receptor A?

2. How does the complex lookthat is formed by receptorA and compound B?

Compound BReceptor A

+

Complex?

docking

Compoundcollection

Receptor A

+

Actives?

docking

The road to new drugs.

How to find new hits?

High Throughput Screening (HTS)

Virtual Screening (VS)

Integration HTS and VS

Molecular docking .

Considering protein flexibility.

Structure-based drug design in practice: Influenza case study.

Drug targets are flexiblebiomolecules and theirdynamics play an important rolein ligand binding.

Insight in receptor flexibility canbe valuable when interpretingstructure activity relationships(SAR) and optimizing leadcompounds.

Predicting ligand binding inflexible binding sites is howeverproblematic !

1 ligand1 receptor

conformation

10 ligands10 receptor

conformations

LOCK AND KEY INDUCED FIT

+Receptor A

X

Complex A-X

+Receptor A

Y

Complex A’-Y

Introduceflexibility

Target Induced fitcomplex

Optimizecomplexes

Generatecomplexes

Compound

Gln

Asn

His

Singlereceptorstructure

flexibleensemble

flexibleresidue

selection

binding siteHis/Gln/Asn

sampling

binding siterotamer

sampling

Selection is based on structural analyses of:

• apo structures• other holo structures• temperature factors

The Fleksy approach docks intoan ensemble of receptorstructures.

The approach is based on aunited protein descriptiongenerated from an ensemble ofprotein structures.

In our case the ensemblecontains the generated set ofside chain rotamers andsampled Asn/Gln/His sidechains.

2 Asnstates 15 side chain

rotamers

8 Hisstates

crystal structureapo form

crystal structureholo form

identifyflexible

residues

sampleAsn/His

?

RMSD

0.6 Å

Ligand(s) unknownLigand(s) known!

Ligand(s) unknown

Target protein structure unknown

high-throughputscreening

Target protein structure known

virtual screening by docking

TWO SCENARIOS

Target geneTarget Discovery

SBDD

Structure based drugdesign relies on structuralknowledge of the targetprotein to design andoptimize lead compounds.

This knowledge is obtainedfrom either experimentalstructures or compu-tational predictions.

A requirement is the availability of receptor structures: NMR spectroscopy

X-ray crystallography

In practice protein X-ray crystallography is the major source of structural information.

Protein Expression and Purification

Data Collection

Analysis / Design

Structure Building

Refinement

Crystallisation

A receptor structure canoften explain:

Binding

Specificity

Inhibition

Flexibility

Reaction mechanism

And it allows predictions tobe made!

1918 Influenza Epidemic Influenza Virus

NEURAMIDASE POCKET SIALIC ACID

RELENZA SIALIC ACID

RELENZA TAMIFLU

5.000.000 doses in NL

Trouw – 3 maart 2009

RELENZA TAMIFLU

WT Ki = 1.0H274Y Ki= 1.9

WT Ki = 1.0H274Y Ki =265

SIALIC ACID

H274YH274Y

UMC St. Radboud

Jacob de Vlieg

Gijs Schaftenaar

Schering-Plough

Scott Lusher

Markus Wagener

Ross McGuire

Hans Raaijmakers

Software

BiosolveIT

Accelrys

Molecular Networks

Funding

Dutch Organization for Scientific Research (NWO)