computational predictive admet: guiding...

30
Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University of Liverpool 9 th February 2016

Upload: duongque

Post on 25-Apr-2018

225 views

Category:

Documents


5 download

TRANSCRIPT

Page 1: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Computational predictive ADMET:Guiding Medicinal Chemistry

Neil Berry

University of Liverpool

9th February 2016

Page 2: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Outline• Why do we need predictive ADMET?

• ADMET compound quality – physicochemical and structural

• Quantitative Structure Activity Relationship (QSAR) – qualifying ADMET predictions

• Metabolism

• Case studies from University of Liverpool and Liverpool School of Tropical Medicine

Malaria

Chronic pain

Filariasis

• Summary

• Future work

• Acknowledgements

2

Page 3: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Introduction - NeedWhy do we need predictive ADMET?

• Chances of drug candidate reaching Phase II ~37%

• Probability of success in Phase II ~34%

• Physicochemical properties have key role in success

• ADMET characteristics - huge impact on

EfficacySelectivity

⇒ ADMET key factor in drug discovery

⇒ Robust ADMET prediction expedites success

Nature Reviews Drug Discovery, 2013, 12, 948Nature Reviews Drug Discovery, 2003, 2, 192 3

Page 4: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

• ADMET Compound quality - no universal definition

Physicochemical Quality

• Permeability/solubility

Pfizer analysis of drugs and candidates. 90% oral drugs: MW<500, LogP<5, #(OH+NH)<5, #(O+N)<10

• Receptor promiscuity

AZ analysis of >2000 compounds in Cerep assayscLogP<3 – decrease riskcLogP>4 – increase riskLLE > 5 decrease risk (Lipophilic ligand efficiency = pIC50 – cLogP)

• ADMET

GSK analysis of ~30k compoundsMW<400 and cLogP<4 reduces ADMET risk

Introduction – Compound Quality4

Advanced Drug Delivery Reviews, 1997, 23, 3Nature Reviews Drug Discovery, 2007, 6, 881Journal of Medicinal Chemistry, 2008, 51, 817

Page 5: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Introduction – Compound QualitySubstructural Quality - Filters and flagging

• Identify compounds with undesirable chemical features at early stage

ToxicophoresMetabolically labileHTS false positives (e.g. pan-assay interference compounds)Interfere with biochemical assays (e.g. fluorescent, coloured, aggregators)

• Usually exclusion – removal/flagging of unwanted compounds

• Can be inclusion – at least one polar atom, at least one rotable bond etc.

• Usually hard cutoffs – more recent overall desirability measures

QED – Quantitative estimate of drug likenessQED outperforms

Lipinski4/400Veber (≤10 rotable bonds & PSA<140Å2)etc.

5

Drug Discovery Today 2003, 8, 86Nature Reviews Drug Discovery, 2013, 12, 948

Nature, 2014, 513, 481Journal of Medicinal Chemistry, 2015, 58, 7076

Page 6: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Quantitative Structure Activity Relationships - QSAR

6

Quantification of ADMET predictions

• QSAR - empirical models giving quantitative prediction via algorithm

• ∆ (Measured end point) = f (∆Structure)

End point – e.g. potency, solubility, ADMET readout etc.

• Relate molecular properties with measured end point

• Several thousand properties can be calculated

Accounts of Chemical Research, 1993, 26, 147Expert Opinion in Drug Discovery, 2010 5: 633

Molecular Informatics, 2010, 29, 476

Molecular property Interaction Example descriptors

Lipophilicity Hydrophobic logP

Polarisability van-der-Waals Molar refractivity

Electron density Ionic, dipole-dipole, hydrogen bonds, charge transfer

HOMO, LUMO

Topology Steric hindrance, geometrical fit Distances, volumes

• Machine learning methods used to relate measurements with calculated properties

Page 7: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Challenging

• Fraught with difficulties, including:

Errors in dataErrors in compound structureErrors in calculated molecular propertiesMolecular properties not capturing key molecular informaitonErrors in machine learningExtrapolation of models beyond “domain of applicability”

• Domain of applicability

No one universal definition

QSAR

7

Nature Reviews Drug Discovery, 2013, 12, 948http://www.oecd.org/chemicalsafety/testing/oecdquantitativestru

cture-activityrelationshipsprojectqsars.htm

Page 8: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Examples

• Multiple examples of successful QSAR

Approved drugs and lead optimisation programmes

Automated QSAR

• Use of information technology to build and update QSAR with little intervention

• Example of successful automated QSARs in drug discovery

AZ – manual logD model not as good in predictive power over time

QSAR

8Reviews in Computational Chemistry, 1990, 1, 335

Nature Reviews Drug Discovery, 2013, 12, 948

Page 9: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Introduction

• Metabolism is one of major clearance pathways for ~75% of drugs

• Biotransformation can give metabolites with substantially altered compound profile

PhysicochemicalPhysiologicalPharmacologicalToxicological

• Metabolism main factor in mediating (de)activation, (de)toxification

• Metabolic systems highly complex

• Expression and substrate specificity vary greatly

• Inter- and intra-individual factors

Gender, Genetic polymorphisms, intestinal flora, lifestyle, medication etc.

Metabolism

9Nature Reviews Drug Discovery, 2015, 14, 387

Page 10: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Towards computational prediction

• Knowing metabolic properties of molecule – help optimise compound stability

⇒ In vivo half life

⇒ Risk/benefit ratio as a therapy

• Experimental approaches demanding

EquipmentExpertiseCostTime

• Computational methods – especially very early stage discovery

Higher throughputLower cost

Metabolism

10Nature Reviews Drug Discovery, 2015, 14, 387

Page 11: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Computational approaches

• Scope includes prediction – for a small molecule

Sites of metabolismMetabolitesMetabolic ratesInteraction with metabolising enzymesToxicological effects of metabolites

• Utilise a variety of underlying technologies

QSARDockingData miningMachine learningetc.

• Vast number of software available

Free, commercial, download, online etc.

Metabolism

11Nature Reviews Drug Discovery, 2015, 14, 387

Page 12: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Strategy → Replace Me with CN

Reduce chain length

Replace CN with

Sulfonamide

Example

• Bayer - Non-steroidal mineralocorticoid receptor antagonist

• Metabolite identification – limited throughput

• Optimisation of metabolic properties in vitro/vivo with predictions sites of metabolism

Fmax - % compound remaining CL – blood clearance in rats

Metabolism12

Nature Reviews Drug Discovery, 2015, 14, 387

Page 13: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 1 - Malaria

13

Why do we need a drug against malaria?

• Global health problem

• Best estimates put the number of clinical episodes of malaria at 0.5 billion

• ~650,000 deaths and >400 million drug treatments/year

• Huge human and socioeconomic burden

• Current therapies are failing

• Parasite resistance remains a major threat

• Very few attractive drug targets

PfNDH2

World Malaria Report 2012, World Health Organisation, Geneva, Switzerland

Page 14: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 1 - Malaria

14

PfNDH2

• The mitochondria is proven drug target - e.g. Atovaquone in MalaroneTM

• Provides intermediates for pyrimidine biosynthesis

• Inhibition leads to mitochondiral dysfunction and parasite death

Methods in Enzymology; Allison, W. S., Scheffler, I. E., Eds.; AcademicPress: New York, 2009; Vol. 456, Chapter 17 , pp 303−320.

• Enzyme “choke point” in electron transport chain

• Pf type II NADH:quinone oxidoreductase (PfNDH2) outstanding therapeutic target

• Only one selective inhibitor known (HDQ)

• HDQ – poor pharmacokinetics and non drug-like

Page 15: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 1 - Malaria

15

Hit Identification

• HTS has been developed, validated and scaled up

Journal of Medicinal Chemistry, 2012, 55, 3144

Goal

• Identify several novel molecular scaffolds which inhibit PfNDH2

• Identification achieved via HTS with compounds selected using chemoinformatics

1) Explore the chemical space around the active hits (similarity searching)

2) Identify new active chemotypes (scaffold-hopping)

3) Favour drug/lead like compounds

⇒ Select 16000 compounds for HTS from 750000 library

• Scaffolds identified

⇒ Medicinal chemistry programme

Page 16: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 1 - Malaria

16

Chemoinformatics Methods

Similarity searches

Diversity selection

Scoring compounds - Lead and Drug likeness bias

Lipinski and Veber filters

⇒ 16000 "leadlike“, diverse compounds selected via chemoinformatic approach for HTS

Compound Score = 4 * (Similarity) + 1*f(LogS) + 1*f(LogP) + 2*f(MW)

Journal of Medicinal Chemistry, 2012, 55, 3144

Page 17: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 1 - Malaria

17

PNAS, 2012, 109 8298Journal of Medicinal Chemistry, 2012, 55, 1844Journal of Medicinal Chemistry, 2012, 55, 1831Journal of Medicinal Chemistry, 2012, 55, 3144

Page 18: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 1 - Summary

18

Summary

• Potent inhibitors active against blood stages of malaria

• nM inhibition against enzyme and parasite (sensitive and resistant strains)

• Rapid selective depolarisation of mitachondria => parsite death

• Potent oral acitivty in mouse model, favourable PK aligned with single-dose treatment

• Ease of synthesis low cost of goods

• Fulfil target product profile for potent, safe, inexpensive drug for clinical evaulation

PNAS, 2012, 109 8298Journal of Medicinal Chemistry, 2012, 55, 1844Journal of Medicinal Chemistry, 2012, 55, 1831Journal of Medicinal Chemistry, 2012, 55, 3144

Page 19: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 2 – Chronic Pain

19

Why do we need a drug for chronic pain?

• Longstanding unmet need

• Chronic pain affects ~20% adults in USA and Europe

• Wrecks lives

• Huge economic impact – lost working days

• Current medication only effective on ~40% sufferers

• Pain market >$6 billion by 2017

O

OH

H2N

Pregabalin

O OH

NH2

Gabapentin

European Journal of Pain, 2006, 10, 287The Oncologist, 2010, 15, 24

Nature Medicine, 2010, 16, 1241

Page 20: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Strategy → Modulate pKa

EC50 α1 Gly (µM) 4.8 ± 1.2 0.0067 ± 0.003

Case Study 2 – Chronic Pain

20

Rationale

4-Chloropropofol Enhances Glycinergic Chloride Curre nts

• Breakthrough in vitro proof-of-concept

• Strychnine-sensitive glycine receptors

• >1000 fold more potent than propofol0.0%

100.0%

200.0%

300.0%

400.0%

0.1 1.0 10.0 100.0 1000.0 10000.0

4-chloropropofol [nM]

% Potentiation of the glycine [10 µM] response

EC50 = 6.7 ± 0.3 nM

Page 21: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 2 – Chronic Pain

21

• CNS penetration required

• Guidelines suggested - different to Lipinski’s

• MPO score - probability of CNS penetration

• 74% CNS drugs have MPO score > 4

• Six key physicochemical parameters

• Traffic light system

• Design and drive medicinal chemistry

Property Raw Trans .

ClogP 3.9 0.550

ClogD 3.7 0.150

TPSA 20.23 0.012

MW 178 1.000

HBD 1 0.833

pKa 11 0.000

MPO 2.5

MPO – Multiparameter Optimisation

ACS Chemical Neuroscience, 2010, 1, 435

Page 22: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 2 – Chronic Pain

22

OH

N

O

O

OH

N

O

OH

Cl

F

Strategy → Metabolic blockade

Reduce cLogP Increase water solubility

Increase metabolic stability

EC50 α1 Gly (µM) 0.0007 ± 0.003 0.001 >100 0.00035 ±0.00002

IC50 GABA A (µM) 5% @ at 100 µM Modulation @ 0.12 Modulation @ 30 >30

MPO 1.8 2.9 3.4 4.4

Progression

Page 23: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 2 – Summary

23

PharmacokineticsOral bioavailability (rat) 86% ����

Stability in rat liver microsomes 740 mins ����

Stability in human hepatocytes 56 mins ����

Brain:CSF levels >10 times EC50 at 2h (oral dose 3mg/kg) ����

Plasma half life >180 mins ����

In vitro plasma protein binding (rat) 62% ����

Cytochrome P450 inhibition >10 µM (CYP3A4, 2D6, 2C9, 2C19,1A2) ����

Cytochrome P450 induction >10 µM (CYP3A4, 2D6, 2C9, 2C19,1A2) ����

SafetyFunctional hERG assay >30 µM ����

Cytotoxicity in HepG2 cells >100 µM ����

Genotoxicity: Ames Negative (at 250 µg/mL) ����

Absence of metabolic alerts None present ����

7-day preliminary tox study in rodents No organ toxicity observed (x100 therapeutic dose) ����

• Proceed towards optimised candidate selection

• Supported by optimised chemically distinct backups

Page 24: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 3 - Filariasis

24

Introduction

• Filariasis - parasitic disease caused by an infection with roundworms

• Serious health problem

• Leading cause of global morbidity

Lymphatic filariasis Onchoceriasis

Elephantiasis River blindness

120 million 37 million

Symbiosis, 2010, 51, 55

Page 25: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 3 - Filariasis

25

Wolbachia

• Bacteria lives in gut of filarial worms

• Wolbachia essential for filarial worms

• Drugs which target wolbachia should be effective

• Improved clinical case management - pathology and inflammatory

Current Therapy

• 4-6 week – antibiotic doxycycline

Issues

• Relatively long treatment – compliance

• Contraindicated in pregnancy and children

Symbiosis, 2010, 51, 55

Page 26: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Case Study 3 - Filariasis

26

Astra-Zeneca Collaboration

• Provides

Measured and predicted PK paramaters

logD7.4

Aqueous solubility

Human PPB

Human microsome clearance

Rat heptocytes clearance

Hit Identification

• HTS has been developed, validated and scaled up

• Screening of multiple libraries (>2M compounds) – informed by chemoinformatics

• Several hit series identified

⇒ Medicinal chemistry optimisation

Journal of Biomolecular Screening, 2014, 1

Page 27: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Correlation coefficient = 0.80 Correlation coefficient = 0.87

Case Study 3 - Filariasis

Measured vs. Predicted Aqueous Solubility Measured vs. Predicted LogD7.4

How do we use the predicted data?

• Help prioritise proposed synthetic targets in each series in the lab

• Used alongside QSAR models for potency

27

Page 28: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Added value

• Predicted PK is

Extremely good and incredibly useful

• Enables us to place our effort in the molecules with the best chance of success

• Warns us away from misleading literature

• Two current lead compounds – from distinct chemotypes

• Very good ADMET profile

• Further optimisation ongoing towards candidate selection

Case Study 3 - Summary

Series 1 Series 2

28

Page 29: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Summary

29

• Physicochemical and structural properties have key role in success

• ADMET characteristics - huge impact on

EfficacySelectivity

• Compound quality – physicochemical and structural

• Computational predictive

QSAR

ADMET

in harness with in vitro/vivo data

⇒ Enables rational, rapid, realisation of quality compounds for development

Page 30: Computational predictive ADMET: Guiding …maggichurchouseevents.co.uk/bmcs/Downloads/Archive/DMPK...Computational predictive ADMET: Guiding Medicinal Chemistry Neil Berry University

Liverpool School of Tropical Medicine

Steve Ward

Giancarlo Biagini

Mark Taylor

30

Acknowledgements*

University of Liverpool

Paul O’Neill

Martin Leuwer

Funding

CROs

AWOL work was supported by a grant from the Bill and Melinda Gates Foundation awarded to the Liverpool School of Tropical Medicine as part of the Anti-Wolbachia consortium

AstraZenecaPeter Webborn

Mark Wenlock

* PIs