Lecture 4

De Novo Design

Invited Guest Professorship

Université Louis Pasteur, StrasbourgUniversité Louis Pasteur, Strasbourg

Prof. Dr. Gisbert Schneider

Goethe-University, Frankfurt

9 Decembre 2008, (c) G. Schneider1

De novo design concepts

Requirements

• Structure sampling method

• Structure assessment method

• Grow• Link• Lattice• Stochastic

• Primary constraints

Implementations

• Structure assessment method

• Search method & Stop criterion

• Primary constraints(receptor, ligand)

• Secondary constraints

• Depth-first search (DFS)• Breadth-first search (BFS)• Random search• Evolutionary Algorithm• Monte Carlo / Metropolis• Exhaustive enumeration

2

Name Publication Building BlocksFragments Receptor Ligand DFSA BFSBRandom MCC EAD

Grow Link Lattice MDEStochastic

HSITE/ 2D Skeletons10,29,85 1989 X X X3D Skeletons30 1990 X X X XDiamond Lattice31 1990 X X X XBUILDER v126 1992 X X X X XLEGEND18 1991 X X X XLUDI11,12,86-88 1992 X X X X XNEWLEAD28 1993 X X X X XSPLICE58 1993 X X X XGenStar32 1993 X X X XGroupBuild16 1993 X X X XCONCEPTS37 1993 X X X XSPROUT15,55-57 1993 X X X X X XMCSS & HOOK23,25 1994 X X X XGrowMol19 1994 X X X X XMCDNLG59 1995 X X X XChemical Genesis20 1995 X X X X XDLD24,89 1995 X X X X

Fitting and clipping of planar skeletons

Building Blocks Primary target constraints Combinatorial Search Strategy Structure Sampling

PRO_LIGAND13,42,90-93 1995 X X X X X XSMoG39,40,94 1996 X X XBUILDER v227 1995 X X X XCONCERTS33 1996 X X X XRASSE21 1996 X X X XPRO_SELECT14,38 1997 X X X XSkelGen61,62 1997 X X X X XNachbar43,95 1998 X X X XGlobus47 1999 X X X XDycoBlock34,35 1999 X X X XLEA45 2000 X X X XLigBuilder22 2000 X X X X XTOPAS46 2000 X X X XF-DycoBlock36 2001 X X X XADAPT65 2001 X X X XPellegrini & Field44 2003 X X X X XSYNOPSIS53 2003 X X X XCoG48 2004 X X X X XBREED60 2004 X X X Exhaustive recombination3

Fragment Placing and Linking

Place fragments Find a suitable linker Link fragments

III

4

O

O

N

H

O

OH

O

OH

Olink

Ile56

N

H

placefragments

Link/Grow Strategy

Babine et al. (1995)Bioorg. Med. Chem. Lett. 5:1719

Ki = 16 µM

O O

OH

OO

grow

Asp37

O

O

Definebinding pocket

(FKBP-12)

Determineinteraction sites

place firstfragment

Phe46

O

O

N

H

5

N

H

N

H

O

OH

Lattice Strategy

O

O

O

O

Fill pocketwith lattice points

Find and connectinteraction points

OH

Assign molecularframework

Buildmolecule

6

Assemble, Dock & Score:Beware of “Unwanted Fragments”

Idea:• generate ligands outside the pocket• dock complete molecules• score

NH NH2

NH

O NH

Asp189

K. Bleicher, M. Stahl, G. Schneider(unpublished)

Insoluble!

S1 pocket

Lipophilicpocket

Thrombin

7

Too fine-grained (atomistic) optimization

“Wrong“ placement of start fragment(s)

Issues of Conventional Design Approaches

“Wrong“ placement of start fragment(s)

Unsuitable linker fragments

8

1. Generate a molecular skeleton based on molecular graph2. Assign real 3D substructure elements (e.g. SPROUT)

1. Link 2D molecular building blocks (SMILES, mol)

Molecule Assembly Strategies

A

B1. Link 2D molecular building blocks (SMILES, mol)2. Calculate 3D conformation (e.g. TOPAS)

• Directly link 3D molecular fragments (e.g. LUDI)

B

C

9

x

HNNH

BindingPocketInitial State

Level 1

Tree model of search space exploration by an automated structure generation method

• Grow strategy

• Depth-first search

xx

NH

O

NH

O

End State

Level 2

Designed Molecule

...

NHHO

O

NH

NN

• Structure-based

10

O

O

ON

N

OHN

N+

OH

O

O

OHO

O

OHN

NH

O

N

HO

O

OH

0.47

0.57

O

NH

NO

N

N

NN

N

N

HN

NH

O

NN0.66

Initial state

inde

x (s

imila

rity

to th

e te

mpl

ate

stru

ctur

e)

Search space exploration by an evolutionary algorithm

• Mutation / Selection

• Depth-first search

• Ligand-based (Reference: Gleevec®) Br NH

HNO

OOH

HN

NN

FN 0.81

NN

O

NH

NH

N

N

FN

0.92

NN

O

NH

NH

N

N

N

1.0

0.70

Br

NH

O

N

NHN

NN

N

N

HN

End state

Tan

imot

oin

dex

(sim

ilarit

y to

the

tem

plat

e st

ruct

ure)

(Reference: Gleevec®)

11

Generation of favourable ligand-binding positions

• CAVEAT (Lauri & Bartlett, 1984)

• GRID (Goodford, 1985 )

• LUDI (Böhm, 1992)

• MCSS (Miranker & Karplus, 1991) /CHARMM (Brooks, 1983)

12

CAVEAT (Lauri & Bartlett, JCAMD 1984, 8:51)

a) b) c)

HNO

NH

O NHO OH

3.17

IC50 = 2.4 µM

Zn2+

MMP-2

S1'

HNO

NH

O

Pro238

Leu181

de novodesign

Zn2+

S1' Gly219

13

LUDI (Böhm)

• Find interaction “hot-spots”HD: blueHA: redLipo: green

• Place fragments

Ligand H-bond donor

N

Receptor

H-bond acceptor

H

• Place fragments

• Link fragments

Böhm et al. (1996)

14

De novo Design with LUDI: Trypsin Inhibitors

Böhm et al. (1996)

15

OHO

NR

O

HO

HN

NH

O

OO

OH

HN

NH

O

OO

OH

NAD+ NADH + H+

DHODH

Dihydroorotate Orotate

De novo Design with SPROUT: DHODH Inhibitors

N

O

3.18

O

NR

X

Inhibitor template Selective inhibitor

3.19

Ala59

Structure-based design with SPROUT (Johnson et al.)

Dihydroorotate dehydrogenase (DHODH)of Plasmodium falciparum

16

De novo Design with BOMB: HIV-1 RT Inhibitors

OX

RPhenyl

N

NPyrazinyl

R

∆∆G in kcal/mol

0.0

-4.2 ± 0.4O

ClN

• Free Energy Perturbation (FEP) approach (Jorgensen et al. 2006 )

• Target: HIV-1 reverse transcriptase

X

NH

Het

Core structure

N

N

2-Pyrimidinyl

R

Pyrazinyl

S

N2-Thiazolyl

R

-11.4 ± 0.5

-9.6 ± 0.4

Cl

NH

N

NO

3.20

EC50 = 10 nM

à place core structure in receptor structureà optimize side-chains by FEP O

O

HO OHO O

O

O

HO O

Initial state

λ = 0

Intermediate state 1 Final state

λ = 1

Intermediate state 2

Step 1 Step 2 Step 3

17

CHARMM

18

19

HIV-Protease

• benzene minima

DHFR site points

• acceptors• donors• ring centroids• neutrals

20

Mol 1Mol 2

3D QSAR: Molecular Field Analysis

Probe 1 Probe 2 Probe 3

Mol 3

Mol N

21

HO O

NOH1

HOH

De novo design examples

• New antifungal agent• Candida / Mycobacterium lanosterol 14α-demethylase (CYP51)• MCSS fragement identification• LUDI fragment linking• no heme coordination (no CYP-P450 interaction)Ji et al. (2003) JMC 46:474

• New HIV-1 protease inhibitor (Ki = 42 nM)• BREED „preferred fragment“ approach

O

OHN

O

OHN

SO

O

NH

2

NH

ON

SH2N

3

• BREED „preferred fragment“ approach• First step: 4 reference molecules recombined• Second step: hybrid fusing with 100 reference structures

• New HIV-1 reverse transcriptase inhibitor (IC50 = 4.4 µM)• SYNOPSIS structure-based approach• First step: 28 designs with predicted low IC50

• Second step: expert inspection & selection, 18 synthesized• 10/18 with IC50 < 100 µM

Pierce et al. (2004) JMC 47:2768

Vinkers et al. (2003) JMC 46:2765

22

NHN

OHN

O 4

Recent combinatorial de novo design examples

• New Cdk-4 inhibitor (IC50 < 1 µM)• LEGEND with homology model• First step: candidate designs• Second step: combinatorial optimization of preferred scaffolds

(MW < 350 Da)• Third step: LUDI & LeapFrog for selectivity optimization

(side-chain optimization)

Honma et al. (2001) JMC 44:4628

CF3

F3CO

5

• New CB-1 ligand (IC50 = 0.3 µM)• TOPAS ligand-based approach• First step: designs assembled from GPCR-fragments• Second step: expert inspection & scaffold selection• 6-10% hit rate with IC50 < 10 µM

(side-chain optimization)

Rogers-Evans et al. (2004) QCS 23:426

23

Ant Colony OptimizationA Simple Combinatorial Case: Peptide Design

T-cell receptor

• Design of novel antigens presentedby MHC I (H-2Kb)

• Length: 8 residues (208 possibilities)

à Neural network ensemble + jury

Artificial ant system

(PDB: 1HSA, Madden et al. 1993)

à Artificial ant system

Assay confirms the predictions:

89% correct for binding peptides

95% correct for non-binding peptides

Hiss et al. (2007) PEDS 20, 99.Schneider et al. (1998) PNAS 95:12179.

24

Ant System for Combinatorial Design

25

De novo molecular design

DB

Known drugs and leadsKnown drugs and leads

Set of reactionsSet of reactions

Generate µ initialparent molecules

Generate λ new moleculesfrom µ parent(s)

Start

R10

OHO

DB

Stock of building blockswith reaction labels

from µ parent(s)

Determine fitness

Select µ best molecules

End

No

Yes

Terminate?

R10

NR7

OR4

R3OHO

R1

• Template compound(s)• Prediction tools• Biochemical assay

• Template compound(s)• Prediction tools• Biochemical assay

Gen

era

tio

n l

oo

p

26

([c;R1:1][10*]).([10*][c;R1:2])>>[c;R1:1]-[c;R1:2]([c;R1:1][10*]).([10*][c;R1:2])>>[c;R1:1]-[c;R1:2]

Aromatic-C + Aromatic-C

SMIRKS/ ReactionSMILES for Virtual Synthesis

AromaticReaction type & site index

+[10*] [10*]

Aromaticcarbon Member of

exactly one ring

Atom mapping index

Reaction type & site index

27

HN

F

F

O

HN

F

HN

O

HN

Rx1

O

F

F HN

F

Rx1NH ORx1

Rx1

N

O

Rx1O

NN Rx1

O

N

N

HNHN

O

F

F

Breeding new molecules

Virtual reactionVirtual reaction

MutationMutation

HN

F

F

O

HN

F

HN

O

N

NHN

CO

NHN

CO

HN

Rx1

O

F

F

HN

F

Rx1NH ORx1

Rx1

CO

Rx1NN

N

NHRx1

NH

Rx1 CO

Rx1

CO

N HN OHN

F

HN

O

F

F

HNCO N

NHN

CrossoverCrossover

28

A Universal Virtual Molecule Generator

HN

F

F

O

HN

F

HN

O

O

Parent Structure

NHN

F

F

O

HN

O

Step 1 Step 3

Child Structure

O

HN

F

F

O

OH

H2NO OH

H2N

F

NRandomly selecta reaction andretro-synthesize

Fragmented Parent Structure

Select one fragment andsubstitute by a new fragmentof the same type

HN

F

F

O

OH

H2NO OH

Fragmented Child Structure

Synthesize withreaction chosenin Step 1

Step 1

Step 2

Step 3

N

29

Compound Growth

B) Synthesis - Growth

HNON

ORx1

HN

F

F

O

HN

F

HN

O

F

Retro-SynthesisHN

F

HN

O

CO

NHRx1

HN

F

Rx1

HNO

Rx1Rx1

HNRx1Rx1HN

F

O

Rx1

HN

O

HN

Rx1

O

F

F

N

ORx1

HN

F

Rx1

NH ORx1

Rx1 NH ORx1

Rx1CO

Rx1NH

Rx1

NH ORx1

CO

NHRx1

30

Compound Shrinking

N

ORx1

HN

F

F

O

HN

F

HN

O

F

HNO

Rx1Rx1

F

A) Retro-Synthesis Synthesis - Shrinkage

N

HN

F

F

O

Rx1

HN

O

HN

Rx1

O

F

F

N

ORx1

HN

F

Rx1

HNO

Rx1Rx1

HN

F

Rx1

NH ORx1

Rx1

NH

NH2

Rx1

HN

HN

O

F

F

NH2

HN

Rx1

O

F

F

31

70

80

90

100

Random Search

How many iterations? How many compounds?

Limited resources: 300 compoundsà There is an optimum!

Trypsininhibition

Population size × Generations

1 x

300

2 x

150

3 x

100

5 x

60

10 x

30

15 x

20

20 x

15

30 x

10

37 x

8

50 x

6

60 x

5

75 x

4

100

x 3

150

x 2

IC50 [µM]

0

10

20

30

40

50

60

70

Evolution Strategy

preferred combinations

inhibition

32

Design Examples – Dopamine D3 Ligand

BP 897

• ‘recapped’ drugs

• Daylight Fingerprints

• Tanimoto Coefficient

NH

ON

NO

D3: 0.92 nMD2: 61 nMPilla et al. (1999) Nature 400:371-373

T = 0.88

ON

N

NH

O

NH

O

ON

N

HN

O

Cl

Cl

T = 0.80

N

O

NNH

T = 0.75

O

NNN

O

O

T = 0.7833

NH

O

N

N

O

Reference: BP 897

4-bond spacer

Design Examples – Dopamine D3 Ligand

3-bond spacer

ON

NHN

O

I

DesignD3: 396 nMD2: 117 nMHackling et al. (2003) J. Med. Chem. 46:3883-3899

ON

NHN

O

34

Reference compound:

Acetylpromazine

Lind et al. (2002)

lipophilic polar

De Novo Design of TAR RNA Ligands (1)

S

NO

N

Lind et al. (2002)

LIQUID pharmacophore model

of Acetylpromazine

Designed molecule in the same

LIQUID pharmacophore model

De novo Design:

F N

N

O F

FF

F

35

NN

N SF

Generation 20, Distance=1.12

O

N

Cl

ClN

N

FF

FO

N N

O

HOGeneration 1, Distance=5.16Start

De Novo Design of TAR RNA Ligands (2)

S

N

NO

Reference molecule

F

NO

FF

F F

N

Generation 60, Distance=0.88

Generation 40, Distance=1.02

F

F

NS

Generation 10, Distance=1.54

FFF

NS

N

O

Cl

Cl

Generation 5, Distance=1.75

O O

F

NO

N

End

Manually refined design

36

U40

U25

C39

G26

F

N NO

De Novo Design of TAR RNA Ligands (3)

Modified bioactive

de novo design

Superposition with

AcetylpromazineC24

FH2NNO

FNH

N

OMeO

OH

OOMe

O

a)

b) c) d)-e)

F

N NO

O

OH

Docking

37

De Novo Design of Bioactive Compounds:

Selected Success Stories

Kv1.5 potassium channel blocker Angew. Chem. Int. Ed. (2000) 39:4130

SO

O

NH

HO

HN

O

O

SO

ONH HN

O

hCB-1 cannabinoid receptor inverse agonists QSAR Comb. Sci. (2004) 23:426

NS

O

O

O

NO

CF3

F3CO

N

O

ON

O

O

F

ClCl

F

J. Med. Chem. (2008) 51:2115

Potent benzodioxole series

38

New Cannabinoid Receptor 1 (CB-1) Ligands

NS

O

O

O N

O

O

O

de novodesign

N

O

O R2

O

R1

Scaffold A

NF4.36, K i = 110 nM

Template structure

NR1

R2

Scaffold B

N

CF3

F3CO

Ki = 300 nM

• Virtual combinatorialchemistry hit

Roger-Evans et al. (2004) QCS 23:426. 39

The Kv1.5 Story

K+

SO

O

NH

HO

HN

O

OO

SO

ONH HN

O

Icagen template, IC50 < 1 µM design, IC50 = 7 µM

R1

de novodesign

SO

ONH HN

O

optimized, IC50 = 1 µM

R1

Pharmacophoremodel

SO

ONH HN

O

lead structure, IC50 < 1 µM

SO

ONH HN

O

Side-chain Linker Scaffold

variable constant

virtualcombichem

The target

Schneider et al. (2000)Angew. Chem. Int. Ed. 39:4130.

40

Design of a druglike hKv1.5 channel blocker

WDIRECAP Building

BlocksTOPAS

3bcombinatorialoptimization

24,563~ 46,000Reference molecule

aromaticnucleophilicsubstitution

aromaticnucleophilicsubstitution

1-fluoro-2-nitrobenzene o-anisidine reduction(Pd-catalytic hydrogenation)

reduction(Pd-catalytic hydrogenation)

condensationcondensation

41

W215

N98

G216

D189

W60DY60A

G219

A190

H57

L99S195

NH2NH

NH

O

6

1. Placement of fragments2. Combinatorial optimization• Preferred reaction (red. amination)• „needle“ approach

1DWB, 3.16 Å

Thrombin Inhibitors:Automatic vs. Manual Design

Ki = 10 nMBöhm et al. (1999) JCAMD 13:51

W215

N98

G216

D189

W60DY60A

G219

A190

3.1

W215

N98

G216

D189

W60DY60A

G219

A190

3.1

NN

NH2HN

O

OF

7

1. Placement of central scaffold2. Modelling3. Fluorine scan

1OYT, 1.67Å

1DWB, 3.16 Å

Ki = 6 nM

Böhm et al. (1999) JCAMD 13:51

Obst et al. (1997) Chem. Biol. 4:287Olsen et al. (2003) Angew. Chem. Int. Ed. 42:2507

42

COLIBREE®Combinatorial Library Breeding

COLIBREE®Combinatorial Library BreedingCombinatorial Library BreedingCombinatorial Library Breeding

43

Disassembly Rules

Ar Ar ArRetain building blocks withMW < 200 Da

à 7,184 Building Blocksavg MW = 141 ± 35 Da

Retain building blocks withMW < 200 Da

à 7,184 Building Blocksavg MW = 141 ± 35 Da

44

Linker Library

45

Molecule Dissection

46

Positive Design

47

Positive and Negative Design

PPARα(negative weight)

PPARγ(positive weight)

48

Thrombin inhibitors PPARα modulators

Land ho! Exploration of chemical space

SO H

N

O

O

NH

O

N

NH2H2N

NAPAP

HOO

OHN

O

S

N

CF3

GW590735

49

Let’s be positive!De novo design of Thrombin inhibitors

CATS descriptor no. 10: PP0 =

Top-ranking designed compounds O O

NH

OHN

SO

O

HN NH2

O

NH

O N

NHH2N

HN

SO

OSCl

O O

NH

OHN

SO

O

S

Cl

HN NH2

O O

NH

OHN

SO

O

HN NH2

O O

NH

OHN

SO

OF S

NHH2N

50

Climbing “Acid hill”:De novo design of PPARαααα agonists

CATS descriptor no. 19:

DN1 =

HO

O

Top-ranking designed compounds

OHO O

HN

O

O

Cl

OHO O

HN O

CF3

O

HOO

HNO

NS

O

HOO

HNO

NS

Cl

CF3

OCF3

OHO O

HN

O

O

OHO O

HN

O

N

OHO O

HN

O

O

I

HO

O

HOO

HNO

NS

CF3

O

HOO

HNO

NS

S

O

HOO

HNO

NS

Cl

Cl

51

Voyages to the unknown: Bioisosters

NO

HO

NH

O

Design 1

inactive!

PPARα

referenceHO

O

O

NH

OS

N

CF3

NO

HO

NH

ODesign 2

EC50 = 0.5 µM

52

Summary Statements

De novo design generatesnew scaffolds & ideas.

Scoring is the most critical part

Human intervention is essential

Accept moderate bioactivity of the designed compounds

53

Selected Reading

Books:

1. G. Schneider, K.-H. Baringhaus (2008) Molecular Design, Wiley-VCH:Weinheim.

2. H.-D. Höltje, W. Sippl, D. Rognan, G. Folkers (2008) Molecular Modeling, Wiley-

VCH:Weinheim.

3. R. O. Duda, P. E. Hart, D. E. Stork (2000) Pattern Recognition, Wiley:New York.

Review articles:

Y. Tanrikulu Y, G. Schneider (2008) Pseudoreceptor models in drug design: bridging ligand- and

receptor-based virtual screening. Nat. Rev. Drug Discov. 7:667-77.

M.J. Stoermer (2006) Current status of virtual screening as analysed by target class. Med.

Chem. 2:89-112.

P. Willett (2006) Similarity-based virtual screening using 2D fingerprints. Drug Discov. Today

11:1046-53.

M. Congreve, C.W. Murray, T.L. Blundell (2005) Structural biology and drug discovery. Drug

Discov. Today 10:895-907.

G. Schneider, U. Fechner (2005) Computer-based de novo design of drug-like molecules. Nat.

Rev. Drug Discov 4:649-63.

54