systems pharmacology; an industrial perspective … · , systems pharmacology of the ngf pathway;...

SYSTEMS PHARMACOLOGY; AN INDUSTRIAL

PERSPECTIVE

(Warwick school of Engineering vacation school)

Dr Neil Benson. Director, Xenologiq Ltd.

[email protected]

www.xenologiq.com Quantitative drug discovery solutions

Interested in PKPD modelling, dose prediction, PBPK and systems pharmacology

mailto:[email protected]

Why should drug discovery be interested in systems pharmacology? Attrition


Attrition – a small proportion of the ideas worked on in drug discovery become drugs -> significant $ spent on ideas that never become products

10 years of evaluation & proposals and arguably nothing has changed


Kola & Landis, Nature Rev DD (2004); PHII attrition due to lack of efficacy or safety key issue

Paul et al, Nature reviews, DDT, (2010) – PHII attrition by far largest contributor to productivity

Hay et al (2014); Nature biotech; The attrition statistics showed a decrease in probability of success compared to previous evaluations

2004 2010 2014

Why drug discovery needs mathematical models •Kola & Landis, Nature Rev DD (2004); Paul et al, Nature reviews, DDT, (2010), Hay et al (2014) – PHII attrition by far largest contributor to productivity

•Mainly due to lack of efficacy or safety (ie we didn’t understand the biology -> consequences perturbing a complex system)

Mathematical modelling & simulation proven approach for tackling complexity in many areas of science and engineering

What is systems pharmacology ?

Time

Effect

Concen

tration

PD

PK

PKPD

Understanding of behaviour of a drug in a disease www.xenologiq.com Quantitative drug discovery solutions

Requires mathematical modelling approach

Example 1 – PKPD & systems pharmacology approach to Toll-like 7 receptor agonists for Hepatitis C virus

•HCV unmet medical need

•Hypothesis - TLR7 agonism -> interferon response -> improved anti HCV effect


Project questions at preclinical stage


1. Can we achieve a dose that will deliver efficacy ?

2. What is the minimum anticipated biological effect level ?

Benson et al Antimicrobial Agents and Chemotherapy, March 2010, p. 1179-1185, Vol. 54, No. 3

Toll-like 7 receptor agonists for Hepatitis C virus; aims

Challenge:

No animal model for HCV Link between TLR7 agonism and antiviral effect not established Dose prediction for interferon inducers for hepatitis C ?

Opportunity:

IFNα available as translatable biomarker Clinical PK and PKPD data for recombinant IFNα Published math model for antiviral efficacy of IFNα

Solution:

Integrated mechanism-based PKPD approach linking exposurebiomarkeroutcome


www.xenologiq.com 9

Very complex – mathematical modelling solution required

Predicting dose/ antiviral response in man

TLR7 – projecting human interferon induction

• Data generated in the mouse

• Rapid and robust induction of IFNa observed from zero baseline

5 15 25

5 15 25

TIME

100

300

100

300

100

300

DV

AMT: 100 AMT:300

AMT: 500 AMT1000

AMT: 5000 AMT:1000.00

5 15 25

5 15 25

TIME

100

300

100

300

100

300

DV

AMT: 100 AMT:300

AMT: 500 AMT1000

AMT: 5000 AMT:1000.00

Plasma IFNa (IU/ml)


How do we use this data in our model ?

TLR7 – indirect PKPD model of IFNa production

.RCSC

.CS

dt

dR

p50

pmax

outk

Data described using a modified indirect effect eqn


Where; dR/dt = rate of production IFNa (IUml-1h-1), Smax = max rate of production IFNa ( IUml

-1h-1) SC50 (ng/ml) = conc drug for ½ max production IFNa kout = elimination rate constant IFNa (h

-1), Cp = predicted plasma concentration (ng/ml) R = plasma IFNa conc (IUml-1)

5 15 25

5 15 25

TIME

100

300

100

300

100

300

DV

AMT: 100 AMT:300

AMT: 500 AMT1000

AMT: 5000 AMT:1000.00

5 15 25

5 15 25

TIME

100

300

100

300

100

300

DV

AMT: 100 AMT:300

AMT: 500 AMT1000

AMT: 5000 AMT:1000.00

Fit to PKPD data

Mathematical model of time & dose dependence

PK

tktkea

a ae ee)kV(k

Dose.kC

+

PD

TLR7 agonist PKPD fit results

Parameter Estimate CV(%)

kout (h-1) 0.958 0.1

SC50 (ng/ml) 135 24

Smax (IU/ml/h) 294 8

IIV Smax (%) 70 26

Residual error

(IU/ml)

65 19

Individual predicted IFNa in plasma (IU/ml)

0 200 400 600O

bse

rve

d I

FN

a in p

lasm

a (

IU/m

l)

0

200

400

600


0 10 20 30 40 50 60 70 80

101

102

103

104

105

Time (hours)

Co

nce

ntr

atio

n (

dru

g (

ng

/ml IF

Na

pg

/ml)

PK

PD

Example individual

TLR7 – linking IFNa to antiviral effect

VTT ).1(dsdt

dT IVT ).1(

dt

dI

Parameter estimates from clinical data (Neumann, A.U., et al, Science, 1998. 282(5386): p. 103-7)

HCV = hepatitis C viral load

c the viral decay rate constant,

δ the infected hepatocyte decay

rate constant.

ε = inhibition of viral release (p)

= inhibition of viral infection s&d = Target cell

production/degradation rates

II TT

Infection Rate

c

Clearance

p

Virions/d

Target Cell

Infected

Cell

Loss

cVpI ).1(dt

dV

IFNa blocks viral release

www.xenologiq.com

Can be represented as a set of equations

IFNIC

IIFN

50

max

Modelling enabled an integrated mechanism-based PKPD approach for predicting antiviral efficacy of TRL7 agonists

Rat PK

Mouse IFNα induction

PKPD

Human IFNα PKPD Human PK

Human IFNα induction

PKPD

Human antiviral PKPD

PBPK

Clinical

Literature

Viral load

PKPD model

Experimental

Literature

In silico

0 20 40 60

time (days)

101102103104105106107

101102103104105106107

101102103104105106107

vir

al lo

ad

(R

NA

co

pie

s/m

L)

dose: 1 dose: 3

dose: 10 dose: 30

dose: 100

Trial simulation

Human dose

prediction

www.xenologiq.com

Conclusions

• Collected & integrated all available data for decision making

• Allowed conversion of information into knowledge for; 1. Dose prediction – likely to achieve efficacy at a practical dose

2. Minimum anticipated biological effect level (MABEL) ID

3. Sensitivity analysis – what is important? (compound potency <

system properties)


Example 2; systems pharmacology & technology feasibility

Drug molecule – binds the target but has poor pharmacokinetics (the ‘war head’)

Molecule coupled to the drug; improves the PK whilst retaining the drug effect

General outline of the approach;

Acknowledgement;Tomomi Matsuura & Piet van der Graaf, Pfizer

Gives improved PK

‘War head’ binds the target


+

Pharmacophore + Linker Antibody CovX- body

Technology feasibility

Project question for target Z

• The coupled drug exhibits pharmacological action, but...

• The war head strongly binds human serum albumin (>>90% bound in plasma)

• Given dose of the new drug is limited by cost & what can be delivered...

• Can the required receptor occupancy be achieved via this approach in this case?


Model of the system

Serum albumin

Drug

R= receptor

Parameters measured (eg surface plasmon resonance) or derived from pharmacokinetic pharmacodynamic analysis of in vivo experiments


Summary

•FR: free receptor concentration/initial receptor concentration

•Green is good, red is bad

•At higher ppb require v low target Kd –project terminated 19

Perspective

• A plausible scenario is that this molecule would have exhibited;

Good in vitro efficacy

Clean pre-clinical safety profile (no pharmacology)

Good PK prediction

Good safety & PK in Phase 1 trials (no pharmacology)

Phase II negative (no pharmacology)


• Model based approach - terminating this early saved $$’s

Example 3; NFKB pathway

Ihekwaba AE, Broomhead DS, Grimley RL, Benson N, Kell DB.(2004). Systems Biology 1(1): 93-103.

& Science, 306, Oct 2004, 704 - 708.

• Project question – what are the optimal targets in the pathway and what % effect/inhibition is required?

• Use published ODE model; 64 parameters & 26 variables

0

0.4

0.8

1.2

1.6

2

0 100 200 300 400 500 600

Time (min) after TNFa

Am

plitu

de

B

215 350 440 (min)

0 50 170

A


NFKBn

Acknowledgement; L Cucurull-Sanchez, Karen Spink

Further detail


Cucurull-Sanchez, K.S Spink & S. Moschos, Drug Discovery Today, 17, 13-14, 665-, 2012

Project approach – Small interfering RNA (siRNA)

Challenge: IL-1

Target: IKK-2

Biomarker: IL-8


Max attenuation [IKK] – 80% ?

Sensitivity analysis identified IKK as a potential target

IKK would appear to be a good potential target

dk

tdx )( Change in the species (eg [NFKBn])

Change in variable (eg parameter or reactant conc)


But high % inhibition required to fully dampen NFKB oscillations..

0 1 2 3 4 5 6 7 8 9 10

x 104

0

0.01

0.02

0.03

0.04

0.05

0.06

Time

Sta

tes

States versus Time

IkBa

NF_kB

IkBa_NF_kB

IkBb

IkBb_NF_kB

IkBe

IkBe_NF_kB

IKKIkBa

IKKIkBa_NF_kB

IKK

IKKIkBb

IKKIkBb_NF_kB

IKKIkBe

IKKIkBe_NF_kB

NF_kBn

IkBan

IkBan_NF_kBn

IkBbn

IkBbn_NF_kBn

IkBen

IkBen_NF_kBn

source

IkBa_t

sink

IkBb_t

IkBe_t

Basal 80% knock-down of IKK

0 1 2 3 4 5 6 7 8 9 10

x 104

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

Time

Sta

tes

States versus Time

IkBa

NF_kB

IkBa_NF_kB

IkBb

IkBb_NF_kB

IkBe

IkBe_NF_kB

IKKIkBa

IKKIkBa_NF_kB

IKK

IKKIkBb

IKKIkBb_NF_kB

IKKIkBe

IKKIkBe_NF_kB

NF_kBn

IkBan

IkBan_NF_kBn

IkBbn

IkBbn_NF_kBn

IkBen

IkBen_NF_kBn

source

IkBa_t

sink

IkBb_t

IkBe_t

Time (s) Time (s)

[NF-

B

(nu

c)]

(M

)

[NF-

B

(nu

c)]

(M

) •Assuming ~remove oscillation to suppress IL8

•and given technology typically gives maximum 80% decrease in target

•Risk that strong attenuation of biomarker may not be achieved


Maximum achievable inhibition via siRNA

Validation of model prediction

Thanks; Sterghios Moschos Karen Spink for data

•80% inhibition of IKK enzyme

•~30% attenuation of IL8 production

•Project terminated


Perspective

• Laboratory costs to test siRNA hypothesis considerable

• Failure of outcome could have been predicted before any work was done

• Model based approach - $$ savings and focus on a less risky target/approach


Example 4; systems pharmacology of the nerve growth factor pathway

•Extracellular Regulated Kinase (ERK) activation controls pain response (eg trka levels, ion channel activity etc)

•We assume dppERKnuc is proportional to pain

•Team question – what are the best pain targets ?

0 20 40 60 80 1000

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Time (mins)

dppE

RK

nuc(µ

M)

Acknowledgement; Cesar Pichardo, Pinky Dua & Piet van der Graaf (Pfizer Neusentis) and at ISB Oleg Demin)


Nat Cell Biol

Models in literature Fujioka et al, J Biol Chem, 2006

0 2 4 6 8 10 12 14 16 18 200

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Pfizer model Sasagawa et al

ppnuc

ERK

Time (mins)

Sasagawa et al, Nat Cell Biol, 2005

Benson & Dua et al, Systems pharmacology of the NGF pathway; Interface Focus, 6 April 2013 vol. 3 no. 2, 20120071.


• One compartment integrated ‘systems biology’ model constructed (59 molecular species and 233 parameters)

0 10 20 30 40 50 60

10-2

100

102

104

106

Species

Tim

e in

teg

ral d

[dp

pE

RK

nu

c]/d

[X]

Finding optimal targets; sensitivity analysis of the model


dk

tdx )(Change in the species (eg [dppERKnuc])

Change in variable (eg parameter or reactant conc)

Rank

(dppERKnuc)

Species Initial

value

Unit Description

1 NGFext 30 pM NGF concentration in extracellular matrix

2 Grb2_SOS_pShc_pTrkA 0 µM Downstream multi-protein complex

3 pShc_pTrkA 0 µM Downstream multi-protein complex

4 Shc_pTrkA 0 µM Downstream multi-protein complex

5 pTrkA 0 µM Phosphorylated TrkA concentration

6 L_NGFR 0 µM NGF NGFR complex

7 FRS2_pTrkA 0 µM Downstream multi-protein complex

8 pFRS2_pTrkA 0 µM Downstream multi-protein complex

9 Crk_C3G_pFRS2_pTrkA 0 µM Downstream multi-protein complex

10 Grb2_SOS_pShc_pTrkA_endo 0 µM Downstream multi-protein complex

NGF

By calculating integrals -> rank of importance

Finding optimal targets; sensitivity analysis (SA) of the model


• SA provides a way to rank targets & further triage by druggability criteria -> interesting targets;

1. NGF (Consistent with NGF mAb efficacy in pain; eg Lane et al, 2010 )

2. RAS activity (Mutations in NF1 gene associated with a chronic pain phenotype (J Neurophysiol 94: 3659-3660, 2005)

3. Trka kinase activity (eg Expert Opin Ther Pat. 2009 Mar;19(3):305-19. Trk kinase inhibitors as new treatments for cancer and pain).

Systems biology -> systems pharmacology


Systems biology model; One compartment No physiological data No easy way to compare eg

mAb’s and small molecules Gives large dose over-

prediction

Systems pharmacology model; Two (n) compartments Physiological information (eg

compartment volume) Separate compartments enable

comparison of small molecules and mAb’s

Can be used for dose prediction

But what about dose? The NGF Systems pharmacology model


2 compartments

1. Extracellular body water (~15L)

2. Neuronal compartment (~0.001 L)

•Signal transduction elements in neuronal compartment

•Enables dose prediction for mAbs and small molecules

•Model predicted max efficacious dose hypothetical mAb at ~ 1000xKD (a non-intuitively high ratio)

Benson & Dua et al, Systems pharmacology of the

NGF pathway; Interface focus, 2013.

Model predictions quantitatively concordant with data for prototype NGF mAbs eg tanezumab


http://www.fda.gov/downloads/AdvisoryCommi

ttees/CommitteesMeetingMaterials/Drugs/Arthr

itisAdvisoryCommittee/UCM301305.pdf

• Cmax ~ 30 nM (=3000 x KD) • -> Model prediction consistent with clinical pain data • -> efficacy & dose could have been predicted from ‘05 model

http://www.page-

meeting.org/pdf_asset

s/7203-

PageAthensCleton.pdf

• Model predicts higher than expected drug concentration (>1000 x KD) for maximum efficacy

Perspective

• ‘05 published model predicted efficacy & (non-intuitive) dose for NGF mAb well ahead of PHII data

• Model provides mechanistic insight into why this is so;

ie due to a negative feedback loop in the NGF pathway (Interface focus 6 April 2013 vol. 3 no. 2, 20120071.)


Example 5; application of systems pharmacology to the clinical development of Fatty acid amide

hydrolase (FAAH) inhibitors for pain.


‘A systems pharmacology perspective on the clinical development of Fatty acid

amide hydrolase (FAAH) inhibitors for pain’. CPT Pharmacometrics Syst. Pharmacol. (2014) 3, e91; doi:10.1038/psp.2013.72

Fatty acid amide hydrolase (FAAH) inhibitors for pain.

• Cannabinoid receptors CB1&2 target for THC – active ingredient of cannabis

• Endogenous Anadamide (AEA) is an agonist of CB1&2

• Wealth of interesting biological data eg increased AEA attenuates nerve firing

(eg Bacci et al, Nature 431,312–6).

• Some evidence of efficacy of cannabis preparations in pain

(eg Nurmikko et al., 2007,Pain, 133, 210-220)


Acknowledgements; van der Graaf, PV D. Nichols & G. L. Li (Pfizer) & Demin,O (Institute for SB, Moscow).

FAAH • AEA degraded by FAAH

• Project hypothesis;

• FAAHi PF-04457845 animal model data positive

• PF-04457845; good pharmacology & PK

• Appears to be very promising drug discovery target


Aim for systems pharmacology approach ‘09

• Project questions;

1. will we test the pharmacology fully?

2. can we predict receptor occupancy and level required for efficacy?

• Systems pharmacology approach with available internal & external data prior to PII data

• Collaboration with Demin & Metelkin, ISB in 2009 as PF-4457845 –> phase 1


Literature review; main conclusions 1. Some useful data (eg AEA disposition, enzyme kinetics to

enable model construction)

2. Endocannabinoid biology complex & not elucidated eg with

apparent contradictory impacts for eg AEA on CB1 & TRPV1 ( Piscitelli and Di MarzoChem. Neurosci. 2012, 3, 356−363)

? Steady state impact

3. Link between occupancy & nerve firing attenuation – no quantitative data

Can’t specify what success would be

in receptor occupancy vs time in man www.xenologiq.com Quantitative drug discovery solutions

Presynaptic neuron

Reaction

or transport

Activation

Inhibition

Reaction

or transport

Activation

Inhibition

signal

CB1RCa2+

Ca2+

med

iato

rs

Ca2+

AEA

NA-PE

PECa2+

?

AEA

?

AEACOX-2, LOX FAAH

NAT

NAPE-PLD

G proteins

G proteins

signal

2-AG

DAG

PI

?

2-AG

PLC

DAG lipase

degradation

?

rece

pto

rs

COX-2, LOX MGL

degradation

2-AG

Summary of model construction

1. XEA disposition; inter-

compartmental equilibration of

XEA’s, location of enzymes

involved, tissue binding

• Large ODE model variables – 39, reactions – 75. 5 main components…

The model

FAAH + AEA FAAH_AEA FAAH + P

+

XEA

FAAH_XEA

FAAH + P

.XEAK

KAEAK

Vmax.Sv

xeaFAAH,m,

aeaFAAH,m,

FAAHm,

2. Multi-substrate binding of FAAH competitive

3. 2 step bio-synthesis of XEA’s

FAAH + I FAAH_I*

kinh

4. Irreversible inhibition of FAAH

& PK;clinical data parameters

5.Ligand partitioning &

receptor occupancy

theoretical model KD,obs=Kp

-1*KD

An additional clearance mechanism?

400 350 300 250 200 150 100 50 0

10

9

8

7

6

5

4

3

2

1

400 350 300 250 200 150 100 50 0

260

240

220

200

180

160

140

120

100

80

60

40

20

0

With unknown enzyme Without unknown enzyme

Plateau

No plateau

Pla

sm

a A

EA

, n

M

Pla

sm

a A

EA

, nM

Time, h Time, h

Modelling results Modelling results

Enzymes

• FAAH-2

• NAAA N-acylethanolamine

hydrolysing acid amidase

• COX-2 cycloxygenase-2

• 12-LOX 12-lipoxygenase

• 15-LOX 15-lipoxygenase

• CYP2D6 P450

• CYP4F2 P450

• CYP3A4 P450

Consume

all XEA’s Incorporating quantitative perspective -> hypothesis NAAA most likely

Dosage: 10 mg PF

OEA

PEA

LEA

AEA

TIME, hours

FA

As

co

nc

en

tra

tio

n,n

M

350300250200150100500

50

45

40

35

30

25

20

15

10

5

0

The model simulated observed clinical biomarker

data

0 50 100 150 200 250 300 350

Time after dose (h)

0

5

10

15

20

25

CB

1 r

ecepto

r o

ccupancy (

%)

0.1

0.3

1

3

10

20

40

PF-04457845 (mg)

Simulation of CB1 RO

• FAAH inhibition >> 96%

•Receptor occupancy limited at ~ 25% (on basis of model hypothesise this is; N-acylethanolamine hydrolysing acid amidase (NAAA))

•Is 25% enough to give pharmacological effect?

Pre phase II clinical feedback 2010

• Model suggested at doses planned we will test pharmacology as far as possible (FAAH inhibition >96%).

• But alternative clearance of AEA exists that limits effect – hypothesis NAAA

• No quantitative data to link receptor occupancy and pain outcome; success criteria open question

• Model suggests some receptor occupancy, but data to confirm or rule this out desirable; ‘Pillar 2’ (Morgan et al, Drug Discovery Today Volume 17, Issues 9–10, May 2012, Pages 419–424)

• High risk project & will we learn from the experiment?

Epilogue


•2012 PF-04457845 failed in OA

•? Confidence in animal model translation

•? Did we express the pharmacology

•Systems pharmacology perspective –> build knowledge & technology before any further phase II

Perspective

• Model contains numerous assumptions –> ‘wrong’

• Value not ‘description’ but as a tool to ask better questions

• This process highlighted risks that were not apparent via other means –> ‘useful’



A vision for the future of drug discovery; systems models central in all projects as a tool to support decision making

Conclusions

• PKPD & Systems pharmacology models can help identify optimal targets & approaches and estimate dose at very early stage

• Growing body of evidence that systems models can add value in drug discovery (eg Benson et al, Proceedings of the 11th International Conference on Systems Biology Advances in Experimental Medicine and Biology. 2012, Volume 736, Part 5)

• Can be applied to efficacy and safety toxicity/questions -> make better decisions in drug discovery & tackle attrition


Questions/ comments?


Backups


Initial clinical data; how could AEA exhibit pharmacology?

Dosage: 10 mg PF

OEA

PEA

LEA

AEA

TIME, hours

FA

As

co

ncen

tra

tio

n,n

M

350300250200150100500

50

45

40

35

30

25

20

15

10

5

0

•AEA binds strongly to HSA (fu ~ 0.0001)

•Primary pharmacology ~300 nM •Peak total AEA ~ 10nM

•Peak free = 1 pM

How could AEA exhibit pharmacology?


Some evidence AEA binds within membrane (THE JOURNAL OF BIOLOGICAL CHEMISTRY, Vol. 280, 2005).

ie ~ order of magnitude for the hypothesis that we can drive CBR pharmacology

Log D7.4 anandamide ~5.67^ KD = 2.1e-6*300nM = ~1pM

^brain:plasma rat Kp= 7.2 Thanks Katie Critchell PDM Pfizer.

systems pharmacology; an industrial perspective … · , systems pharmacology of the ngf pathway;...

Documents