Proprietary and Confidential © AstraZeneca 2008 FOR INTERNAL USE ONLY David Cook Global Safety Assessment AstraZeneca The Application of Systems Biology.

Proprietary and Confidential © AstraZeneca 2008FOR INTERNAL USE ONLY

David CookGlobal Safety AssessmentAstraZeneca

The Application of Systems Biology to Safety Assessment

• The assessment of the safety of medicines is taken very seriously by the industry and regulatory authorities

• Getting the toxicological risk assessment wrong can have significant impacts on patient health

• The perception of a risk can reduce the benefit of a potential medicine

It benefits no-one to produce a medicine with an unacceptable safety profile

Beyond the risk to the patientCost of toxicological failure

Kola and Landis Nature Reviews Drug Discovery 3, 711-716 (August 2004)

• >20% of candidate drugs fail due to unpredicted toxicology• Additionally, some drugs fail to reach efficacy due to dose-limiting toxicology• Each compound failure in the clinic costs between $10M and >$100M depending on

when it fails• Better prediction of potential risk early

• Avoid the problem• Better understanding of potential risks in patients (or subsets of patients)

• Manage the risk• Only small changes = huge benefits

Influencing choice in drug discovery

• Successful drug discovery and development is about making the right decision at the right time

• The “big” decision points (milestones, tollgates etc.) are not the important ones

• The right decision requires access to the right information• The right time is dictated by the phase of the drug-discovery process

• Scale approaches to deliver to the decision-making cycle• data delivered late, might as well have not been generated at all!

Influence design here Understand and mitigate issues here

Influencing choice in drug discoveryNeeds: Scaling approaches to the volume and rate of analysis

In Silico

In Vitro Screen

In Vitro Functional Assay

In Vivo Confirmatory Assay

GLP studies

Volume of analysis Approach

MS

1M

S2

MS

3M

S4

MS

5M

S1

MS

2M

S3

MS

4M

S5

Milestone Rate of analysis

Minutes

Hours

Days

Weeks

Months

• Cannot simply move the “traditional” testing paradigm to earlier phases in drug discovery

• Unethical and incompatible with 3Rs and animal usage

• Cannot handle the volume of analysis

• Cannot handle the rate of data delivery

• Need to adopt more in vitro and in silico approaches• Computational Biology

Toxicologists are Systems Biologists

Efficacy consideration• One disease

• One mechanism in one disease

• One target in one mechanism in one disease

• One therapy against one target in one mechanism in one disease

Toxicological consideration•One therapy (perturbation)

•Multiple mechanisms• Primary effects

• Predicted secondary effects

•Effect(s) in healthy volunteers• Effects on normal biochemistry

•Effect(s) in the patient• Effects on potentially abnormal

biochemistry

• Interaction with other therapies

•Effect(s) in a population of patients• Idiosyncrasy

“Reductionist Drive”

“Systems Drive”

Has the drive produced here limited our understanding here?

• The “single protein” model of cause and effect

Excitation initiated in the sino-atrial node spreads through the heart

Action potential morphology varies according to cardiac region

The wave of excitation can be detected on the body surface: the electrocardiogram (ECG)

Cardiac Ion channel liabilitiesBackground biology: Origin of the ECG

Background biologyInformation derived from the ECG: PR, QRS & QT intervals

PR(PQ)

QRS

QT

PR(PQ): an index of conduction through the atrio-ventricular node

QRS: an index of conduction through the ventricles

QT: an index of action potential duration in the ventricles

P T

Q

S

R

From a pre-clinical perspective, this molecular understanding is fundamental to being able to prevent or minimise ECG risk

Kv4.3

Nav1.5Cav1.2

Kv7.1

K+K+

Na+Na+

Ca2+Ca2+

K+K+

Na+

K+

Na+

K+

HCN

Ca2+Ca2+

Cav3.2

Kv1.5

K+K+

Ventricular myocyte action potential

Atrial myocyte action potential

Kv11.1 (hERG)

Bers (2001). Excitation-Contraction Coupling and Contractile Force. Kluwer Academic Publishers, Netherlands. ISBN 0-7923-7157-7.

Background biologyKey ion channels underlying action potentials*

inside

outside

* Only a sub-units shown

Na+Na+

NaV1.5

(INa)

Ca2+Ca2+

K+K+

hKv11.1

(hERG)(IKr)

CaV1.2

(ICa,L)

AV nodal VentricularVentricular

Increase PR interval Increase QRS duration Increase QT duration

AV block Ventricular tachycardia Torsades de Pointes

What’s the problem?Effect of channel block on action potentials & ECG

What’s the problem?Strong evidence that inhibition of cardiac ion channels can lead to life-threatening arrhythmias

Channel Congenital “loss of function” mutations can lead to:

Pharmacological inhibition can lead to:

Example drugs

Nav1.5 Atrial fibrillation; Ventricular fibrillation; Sick Sinus Syndrome

Ventricular Tachycardia

Encainide; Flecainide1

Cav1.2 ST segment elevation AV block Verapamil2; Diltiazem

Kv11.1(hERG)

Torsades de Pointes Torsades de Pointes Astemizole; Cisapride; Droperidol;Terfenadine; Thioridazine;Terodiline3

1 Echt et al., N Engl J Med. (1991); 324, 781-8. 2 Cohen et al. Neurology (2007); 69, 668-75. 3 see Redfern et al. Cardiovasc Res (2003) 58, 32-45.

In silico

In Silico Cardiac Ion Channel strategy

Predicted Activity at:hERGNav1.5

Prediction of activity at individual channels

Prediction of effect on ventricular action potential duration based on measured

activity at individual channels

Channel DataNav1.5 inactiveKv4.3 inactiveCav1.2 IC50 10 mMKv7.1 inactiveKv11.1 IC50 5 mM

?

Test compound

NR 2

R 3

X R 1

L O G D 6 5L O G D 7 4A R O ML O G PH M O _ R E S O N _ _ E N E R G YN U M _ R I N G SN O N P O L A R _ C O U N T

D o c k in g S c o r e

T r a d i t io n a lD e s c r ip to r s

P h a rm a c o p h o r eF e a tu re s

H idd enInp ut O u tp utw i j

f ( s iw i j)

N e u ra l N e tw o rk s

T e r m in a lN o d e

I N A C T IV E

T e r m in a lN o d e

I N A C T IV E

T e r m in a lN o d e

I N A C T IV E

T e r m in a lN o d e

A C T I V E

L e a f N o d e

N = 1 5 4

L e a f N o d e

N = 2 7 4

L e a f N o d e

N = 8 1 0

T e r m in a lN o d e

A C T I V E

R o o t N o d e

N = 1 2 0 3

D e c is io n T re e s

-10

-5

0

5

-20 -10 0 10

t[3]

t[1]

P L S

C o n se n su sh E R G

P r e d ic t io n

C o n se n su sh E R G

P r e d ic t io n

A stra Ze n e c a h ERG Q SA R :D ive rse M o le c u la r D e sc rip to rs a n d Sta tistic a l M e th o d s to G e n e ra t e a ’ C o n se n su s’ P re d ic tio n

P O L _ S U R F _ A R E AN E G C H A R G E _ G A S TP O S C H A R G E _ G A S TC H A R G E _ G A S TD I P O L E _ M O M E N TM O L _ V O L U M EE tc … … … …

hERG QSAR in AstraZeneca

Impact : Less hERG related cardiac arrhythmia liability over time

0%

20%

40%

60%

80%

100%

2004 2005 2006 2007 2008

Year Measured in Ionworks

Mea

sure

d h

ER

G IC

50

> 10 µM

3 to 10 µM

< 3µM

In silico

In Silico Cardiac Ion Channel strategy

Predicted Activity at:hERGNav1.5

Prediction of activity at individual channels

Prediction of effect on ventricular action potential duration based on measured

activity at individual channels

Channel DataNav1.5 inactiveKv4.3 inactiveCav1.2 IC50 10 mMKv7.1 inactiveKv11.1 IC50 5 mM

?

Test compound

Modelling of Action Potential ’System’

Multi-Scale Modelling: Assessing Cardiac Safety

Modelling of Interactions on the Protein Level

R

S

T

Q

Prediction of Effects on Q-T Interval

Prediction of Effects on Action Potential Duration

Systems Model of Cardiac Ion Channels

Modelling of Action Potential ’System’

Potent, selective hERG blocker

Systems Model of Cardiac Ion Channels

Modelling of Action Potential ’System’

Potent, relatively non-selective hERG blocker

Systems Model of Cardiac Ion Channels

Modelling of Action Potential ’System’

Low potency, non-selective blocker

Systems Model of Cardiac Ion Channels

Modelling of Action Potential ’System’

Compound that activates some channel types and blocks others

↑QT (4%

)

"other"CV tox (73%)

Haemodynamic

Remodelling

Myopathy

Contractility

11%

MI

CV Tox

Arrhythmia (23%)

e.g. Other toxicities, Efficacy, Portfolio etc.

Moving beyond arrhythmiasCardio-Vascular toxicity and Drug Withdrawals post Phase I

Lipophilicity

Charge, Hydrogen bonding

Size

Moving beyond arrhythmias

• QSAR modelling for compounds with CV toxicity• Molecules with similar properties are plotted close together• Plot of withdrawn compounds overlaid on all compounds in DrugBank

• No clear structural bias of compounds with CV toxicity beyond a tendency towards lipophilic molecules (shared with most withdrawn compounds)

• Cannot predict CV liability solely based on molecular structure

• Despite data complexity, too much “biology” for this approach to work

• Biological understanding is lacking: what are the molecular mechanisms?

• Need to improve the basic science before we can develop further models

Withdrawn CV (Arrhythmia)Drugbank

Withdrawn CV (Long QT syndrome)

Withdrawn (other CV tox)

Withdrawn (other)

Dynamic modelling: Focus on idiosyncratic DILI

• Drug-induced liver injury (DILI)

• Intrinsic: predictable, dose dependent e.g. acetaminophen

• Idiosyncratic: unpredictable, dose independent (?)

• For pharmaceuticals, idiosyncratic DILI accounts for a significant number of patient deaths annually

• These occur in a minority (by definition) of patients

• Occurs late in the clinical development phase or even post-marketing• Cost the industry $$$$$

• Regulators are demanding larger and larger trials, beyond that required to establish efficacy, in attempts to detect idiosyncratic drug reactions

• Cost $$$$

• Delays getting new medicines to patient

• Need new approaches to the early prediction of idiosyncratic DILI• Preclinical screens (in vitro, in vivo)

• Early clinical trials (biomarkers)

• People are not even a good model of people!

• Can dynamic modelling render the unpredictable, predictable?

Idiosyncratic DILI is multi-factorial due to a “perfect storm” of factors

Idiosyncratic DILI is…well…complicated!

Idiosyncratic DILI has a spacial componentIdiosyncratic D

ILI has a temporal com

ponent

Goals of DILI-sim

Aim is to provide tools that can help integrate and interpret structural, in vitro and in vivo data to predict likely hepatic responses in preclincal species and ultimately man

The DILI-sim Modeling Approach: Multi-Scale“Middle-out” approach

Kuepfer 2010, Molecular Systems Biology

DILIsym™ Model v1.0 Sub-model Interactions: Drug Metabolism, GSH, and Mito. Dysfunction

RM

GSH depletion

&recovery

Drug distribution

& metabolism

Mitochondrial dysfunction

Form to Function Approach Links Dynamic Changes in Hepatocytes to Liver Function

RM

GSH depletion&

recovery

Drug distribution&

metabolism

Mitochondrial dysfunction

Hepatocyte life-cycle

Biomarkers

Immune mediators

‘Form tofunction’

Good Agreement Between Simulations and Measured Data in Rats Following APAP Overdose

Inter Quartile Range & 95% Confidence Interval shown

RATS

Preclinical data and simulation results

Population Sample Generation – Humans

Schiodt 2001

39g mean APAP dose34 hr mean NAC delay

n = 37

* Red lines indicate simulated humans

HUMANS

Clinical data and simulation results

CONCLUSIONS

• Toxicology is intrinsically a problem in systems biology• “Pathology with numbers”

• Lots of data and information but often little knowledge

• Understanding of key drives such as hERG and cardiac ion channels are not always known

• Mutlifactoral, temporal responses involving environmental and genetic factors

• Understanding and prediction demands a quantitative approach

• First generation models are coming on line

• Summarising and organizing information – knowledge repositories

• May fail, but in organizing the data will help us understand gaps

• Investments in systems models for safety are easier to justify

• Models have both longevity and breadth of application

• Used for many projects over many years

• Investments in large-scale approaches can be justified because of the nature of the problem, when it occurs and returns if successful

• Huge scope for pre-competitive working in this space

Has Systems Biology finally found a true home in pharmaceutical R&D?

Acknowledgements

Cardiac Modelling

• Scott Boyer (AZ)

• Mark Davies (AZ)

• Claire Gavaghan (Umetrics)

• Najah Abi-Gerges (AZ)

• Leyla Hussein (AZ)

• Sherri Matis-Mitchell (AZ)

• Hitesh Mistry (AZ)

• Chris Pollard (AZ)

• Stephaine Roberts (AZ)

• Jonathan Swinton (AZ)

• Jean-Pierre Valentin (AZ)

DILI Modelling

• Gerry Kenna (AZ)

• Brett A. Howell (Research Scientist, IDSS)

• Scott Q. Siler (Siler Consulting)

• Jeffrey L. Woodhead (Postdoctoral Fellow, IDSS)

• Paul B. Watkins (Director, IDSS)

• Entelos, Inc. (no longer affiliated, but previously contributed)