using dilisym to predict hepatotoxicity risk during pre

38
Using DILIsym to Predict Hepatotoxicity Risk During Pre-clinical Development DVDMDG Symposium March 2019 Paul Michalski

Upload: others

Post on 13-Mar-2022

4 views

Category:

Documents


0 download

TRANSCRIPT

Using DILIsym to Predict

Hepatotoxicity Risk During

Pre-clinical Development

DVDMDG Symposium

March 2019

Paul Michalski

DILIsym Overview

2

Paul Michalski. DVDMDG Symposium. March 19, 2019.

PK Model Dose-Exposure

Predictions

Seahorse Mitochondrial

Stress Test

Transporter/Enzyme inhibitionCholestasis: BSEP, MRP3/4, NTCP

Bilirubin: MRP2, OATP1B1, OATP1B3, UGT1A1

Cell HealthCell count, Nuclear Size, DNA structure,

Mitochondrial mass and membrane

potential, oxidative stress, GSH content,

and cellular ATP

Mass Spec Intracellular

Concentrations

DILIsym (Drug Induced Liver Injury QST Model)DILIsym is a Quantitative Systems Toxicology (QST) model providing dose-specific predictions of drug induced liver injury

• Model simulates liver homeostasis of pathways known to be involved with drug induced liver injury

• Dose-exposure relationship of compound in liver determined using standard physiologically based pharmacokinetic (PBPK) modelling

• Using the DILIsym input panel, the model simulates concentration-specific interactions of the drug on liver homeostasis, predicting time and

compound dependent toxicities

Inputs

DILIsym Model

Outputs

Translational Toxic & Lethal Dose PredictionConfidence in Dose Safety Margins

Translational Scaling of Tox Studies (Rat-Dog-Human)

Reduction in attrition from false positive /negatives in

current heptotox strategy

Clinical DILI BiomarkersBetter Clinical Monitoring,

Lower risk to patients

Primary Mechanism of ToxicityBetter Defined Lead Screening,

Reduced attrition

Species: Human, Dog, Rat, Mouse

Reported Accuracy: 79%1

DILIsym reports accuracy in 52 out of 66 human

simulation scenarios.

(59 unique compounds)

Paul Michalski. DVDMDG Symposium. March 19, 2019.

DILIsym – A quick look under the hoodFull model consists of over 500 state variables

Feeding Effects Module

Meals modify the concentrations of

plasma glucose and triglycerides,

affecting liver fatty acids, glycogen,

pyruvate, etc.

Downstream effects on mitochondrial

function by modifying ratio of pyruvate

and FA oxidation, ultimately affecting

ATP levels and RNS/ROS generation.

Similarly complex networks describe

pharmacokinetics, hepatocyte life

cycle, innate immune response, etc.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

5

DILIsym currently models 14 DILI mechanismsThere are known DILI mechanisms which are not currently included in DILIsym

Liver Specific Mechanisms

Increase RNS/ROS

Increase ATP Utilization

Direct Necrosis

Direct Apoptosis

Inhibit Bile Acid Transport

Inhibit Bilirubin Transport

Reactive Metabolite Formation

Mitochondrial Toxicity Mechanisms

Electron transport chain inhibition

MT Uncoupler

Pyruvate Oxidation Inhibition

Fatty Acid Oxidation Inhibition

ATP Synthesis Inhibition

MT Permeability Transition Pore Initiator

Glycolysis Inhibition

Paul Michalski. DVDMDG Symposium. March 19, 2019.

DILIsym Example Outputs

6

Paul Michalski. DVDMDG Symposium. March 19, 2019.

7

DILIsym simulates time course of all relevant variables

DILIsym simulations provide time

courses of all relevant PK variables,

hepatotox markers, and other

biological variables of interest.

In the following slides with plots of

response vs dose, the value at the

final time point is plotted, either the

end of dosing or at the time of death.

Simulation Setup

Dosing for 1 week.

Human dosed twice daily.

Baseline parameters for average human.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

8

Concentration dependence of DILI biomarkers

By integrating PK with a QST model of DILI,

DILIsym can predict:

1) The dose at which toxicity first appears

2) The most sensitive biomarker indicating

toxicity

Simulation Setup

Dosing for 1 week.

Dog dosed once daily, human dosed twice daily.

Baseline parameters for average dog and human.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

9

Time dependence of hepatotoxicitySome doses only exhibit toxicity after long-term repeat dosing

One or two weeks dosing is not sufficient to capture the toxicity which might be observed in a long-

term clinical study.

In this example, the 50 mg/kg/day dose appeared non-toxic during a 1 week study but becomes

lethal when dosed for longer periods.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

10

Time dependence of hepatotoxicity – species differences

For dog, compound is slightly more toxic under a longer dosing regimen.

For human, extremely toxic under a longer dosing regimen.

Dog Human

Paul Michalski. DVDMDG Symposium. March 19, 2019.

11

Population variability of DILI DILIsym provides populations with drug susceptibility and biological variability

DILIsym provides the time course of all relevant variables for the entire simulated population.

Can answer questions like, “At dose X, which fraction of the population will have ALT elevated 3-fold?”

Paul Michalski. DVDMDG Symposium. March 19, 2019.

12

Population variability only manifests in a narrow dose range

At very low and very high doses, all subjects respond similarly

Paul Michalski. DVDMDG Symposium. March 19, 2019.

DILIsym PoC

13

Paul Michalski. DVDMDG Symposium. March 19, 2019.

Lead Optimization Pre-candidatePre-clinical

Development

Clinical

Development

Facts

• High throughput –

(compounds/day)

• Need to predict from structure

• High throughput assay panels and

QSAR provide flags for DILI

• Very little PK info

• Rough dose estimate

• GSK: Hepatotox strategy

• Acceptable throughput

• 7-day tox rat study typically performed

• Limited PK data

• Uncertainty in dose prediction

• Acceptable throughput

• PK/TK data available

• More accurate dose

prediction

• Understanding on DILI

phenotype in

preclinical species

• 14 - 28 day toxicology

• Available clinical PK

and liver biomarker

data

Op

po

rtu

nit

y

• Guide LO - identify the right

balance of safety / potency / PK

• Identify cases where the 7 day rodent/non-

rodent tox is likely to provide false +/-

• Failure in lead informs selection of backup

• Improved hepatotoxicity prediction

• Determine max dose / dosing regimen with no

DILI in human

• Replace 7-day tox ? Unlikely (other toxicities)

• Better DILI tox species? Unlikely (other

drivers)

• As per pre-candidate

selection opportunities

extended to 14-28 day

timeframe

• Quantify the risk of

DILI in human

• NOAEL estimate in

human

• Biomarker strategy

• Better biomarkers

• Evaluate liver damage

based on biomarkers

• Estimate patients at

risk of DILI

• Estimate probability of

DILI

• Population variability

Ch

allen

ges

• Inputs prediction from structure

• … or measured high throughput

• Simulation need to fully automated

• Predictions need to be accurate

enough to steer the chemistry in

the right direction

• Need better dose predictions

• Better that the hepatotox panel

• Requires high confidence in model

• Evaluation of non hepatotoxins

• Case studies to increase trust in the model

• Reduced cost of input assays

• Focus on translational aspects of the model

• Dedicated modeling resource

• Requires high

confidence in model

• Focus on translational

aspects of the model

• Case studies where

the model was right

and predicted

something different

compared to standard

approaches

• Accurate DILI model

required

• Emphasis on

population variability

• High performance

computing required

DILISym: Opportunities and challenges

Paul Michalski. DVDMDG Symposium. March 19, 2019.

Pre-candidate

Facts

• Acceptable throughput

• 7-day tox rat study typically performed

• Limited PK data

• Uncertainty in dose prediction

Op

po

rtu

nit

y

• Improved hepatotoxicity prediction

• Identify cases where rodent/non-rodent tox is likely to provide false +/-

• Failure in lead informs selection of backup

• Determine max dose / dosing regimen with no DILI in human

Ch

allen

ges

• Dedicated modeling resource

• Requires high confidence in model

• Evaluation of non hepatotoxins

• Case studies to increase trust in the model

• Focus on translational aspects of the model

• Reduced cost of input assays

DILISym: Opportunities and challengesFocus on pre-candidate stage

Some learnings

Paul Michalski. DVDMDG Symposium. March 19, 2019.

16

GSK DILIsym Proof of Concept Compounds8 compounds selected to cover various categories of interest

GO Low dose, assay neg. GO

STOP STOP

Low dose, assay pos.

High dose, assay neg. High dose, assay pos.

The current GSK strategy stratifies compounds based upon dose and in vitro assay flags.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

17

GSK DILIsym Proof of Concept Compounds8 compounds selected to cover various categories of interest

GO Low dose, assay neg. GO

STOP STOP

Low dose, assay pos.

High dose, assay neg. High dose, assay pos.

GSK-A – Hepatotoxic

GSK-B – Hepatotoxic

GSK-D – Hepatotoxic

Phenoxybenzamine – Non-hepatotoxic.

Zafirlukast - Hepatotoxic

GSK-C – Hepatotoxic

Fluconazole - Hepatotoxic

No compounds in this category

The current GSK strategy stratifies compounds based upon dose and in vitro assay flags.

Raloxifene – Non-hepatotoxic

Paul Michalski. DVDMDG Symposium. March 19, 2019.

DILIsym Assay Panel

18

Paul Michalski. DVDMDG Symposium. March 19, 2019.

19

Overview of DILIsym assay panel and other input data

In vivo PK Data

MitoTox Assay

High Content

Screening Assay

Transporter Assays

PBPK Models

Mechanism of mitotox and

quantitative parameters

IC50 values for BSEP,

NTCP, MRP3, and MRP4

DILIsym

Model

Experimental Data Data Analysis Model Building Simulation Results

Measures of reactive

metabolites (GSH), ROS,

mito mem. pot., along with

quantitative parameters

PK data time course

Hepatocyte

apoptosis/necrosis

DILI clinical markers:

ALT, SDH, Arg1, etc.

Time of death

Time courses of all

relevant biological

variables

Paul Michalski. DVDMDG Symposium. March 19, 2019.

20

DILIsym Assay Panel

Seahorse Assay (Mitotox)

Three measures of mitotox (OCR, ECAR, RRC)

Two time points (1 and 24 hours)

Seven concentrations (span in vivo Cmax, two orders of magnitude above and below)

Two technical replicates for each measurement

Cell Health Assay

• Eight measures of cell health

• Performed in both HepG2 (parent tox) and primary hepatocytes (metabolite tox)

• Two time points (6 and 24 hours)

• Eight concentrations (span in vivo Cmax, two orders of magnitude above and below)

• Three technical replicates for each measurement

Transporter Inhibition Assays

• Determines IC50 for four transporters (BSEP, NTCP, MRP3, MRP4)

Seven concentrations (span in vivo Cmax, two orders of magnitude above and below)

Three technical replicates for each measurement

Paul Michalski. DVDMDG Symposium. March 19, 2019.

21

PK Analysis Pipeline

Pre-clinical PK

Data

PhysChem

PropertiesGastroPlus PBPK

Model

DILIsym PBPK

Model

Originally planned to develop and validate PBPK models in GastroPlus, then use identical parameters in DILIsym.

GastroPlus is used as an intermediate because it enables rapid model development and cross-species translation.

But, parameter translation proved impossible due to differences between DILIsym and GastroPlus PBPK models.

Different number of compartments, but also differences in total flow rate, tissue flow rates, and tissue volume.

GastroPlus PBPK DILIsym PBPK

Note: Although DILIsym has an option to prescribe drug PK, it is not

recommended because it eliminates potentially important feedback

mechanisms.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

22

PK Analysis Pipeline

Pre-clinical PK Data

PhysChem Properties

GastroPlus PBPK

ModelDILIsym PBPK Model

Important Insight

We only require the DILIsym PBPK model to reproduce the predicted plasma and liver concentration time course.

Current Pipeline

(1) Develop GastroPlus PBPK models and generated predicted plasma and liver concentration profiles.

(2) Fit the predicted profiles with DILIsym. To do so, treat the DILIsym ‘muscle’ and ‘other’ tissues as general

compartments, and fit their flow rates and Kps in order to match predictions.

• Liver Kp is fixed based on the GastroPlus model.

Predicted plasma and

liver concentrations

Paul Michalski. DVDMDG Symposium. March 19, 2019.

23

Important to NOT fix the PKAs liver cells die, drug exposure can increase substantially

Dosing for 1 week Dosing for 1 year

50 mg/kg dose is not toxic for 1 week, fatal at 39 weeks. Exposure increases nearly 10-fold in final weeks.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

GSK-C

Paul Michalski. DVDMDG Symposium. March 19, 2019.

25

GSK-C Background

Low dose, assay positive, hepatotoxic.

GSK hepatotox strategy incorrectly predicts non-hepatotoxic.

In one study, several patients receiving 7.5 mg BID exhibited increased ALT levels

approximately 10 days into the study.

Another study had two patients with elevated ALT, but these were trauma patients so the

connection to drug is less clear.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

26

GSK-C Seahorse ResultsSeahorse assay indicates compounds is an ATP synthesis inhibitor

OCR UP OCR

DOWN

OCR

BIPHASIC

PROTON

GRADIENT

UP

ATP

Synthesis

Inhibition

PROTON

GRADIENT

DOWN

MT

Uncoupler

ETC

Inhibition

.

MT Perm

Transition

Paul Michalski. DVDMDG Symposium. March 19, 2019.

27

GSK-C DILIsym ResultsBaseline individual – Dosing BID for two weeks

Red line indicates dose at which

hepatotoxicity was observed in the clinic.

DILIsym predicts toxicity appears at

doses about 3-fold higher than the

clinical toxic dose.

Importantly, DILIsym predicts toxicity well

below the 100 mg threshold used in the

GSK strategy.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

28

GSK-C DILIsym ResultsPopulation Results – Dosing BID for two weeks

Red line indicates dose at which

hepatotoxicity was observed in the clinic.

As expected, population simulations

show variability in the region where

toxicity is turning on, but does not

change the qualitative conclusion of

toxicity at doses below 100 mg.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

29

GSK-C Modeling Conclusion

DILIsym correctly predicts GSK-C to be hepatoxic, in contrast to the current GSK hepatotox

strategy.

DILIsym predicts toxicity at doses 3-5 fold higher than the toxic doses observed in the clinic.

Importantly, toxicity was predicted well below the GSK threshold of 100 mg.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

30

PoC Results OverviewDILIsym performs well on four out of seven compounds.

GO Low dose, assay neg. GO

STOP STOP

Low dose, assay pos.

High dose, assay neg. High dose, assay pos.

GSK-A – Only DILIsym flag is BSEP inhibition, which was not strong enough to recapitulate observed toxicity. (GSK assay hits for BSEP and covalent binding.)

GSK-D – DILIsym flags mitotoxicity and non-specific metabolite-dependent ATP reduction. GSK assays flag Cyp MDI and covalent binding.

GSK-B – DILIsym predicts toxicity at doses 2-3 fold larger than observed in the clinic, and predicts no therapeutic window. (Mechs GSH, BSEP, cell health.)

Phenoxybenzamine – Non-hepatotoxic. DILIsym predicts hepatotoxicity at doses 5-fold higher than the largest recommended dose after 4 weeks of dosing, but it seems most patients take much lower largest dose for short periods of time.

Zafirlukast - Toxicology reports say mechanism is unknown but “clearly idiosyncratic” and likely the result of a metabolite. Probably a poor choice for DILIsym.

GSK-C – DILIsym correctly predicts hepatotoxicity at doses well below 100 mg/day. (GSK mechs BSEP, Cyp MDI.)

Fluconazole – All DILIsym assays were clean. Unable to build DILIsym model.

(Predicted Toxic Dose)/(Observed Toxic Dose)

< 3

3-10

10-100

Unsupported Mechanism

No compounds in this category

Paul Michalski. DVDMDG Symposium. March 19, 2019.

31

Comparison of GSK and DILIsym Analysis

Compound Actual Status GSK Prediction DILIsym Services

Prediction

Fluconazole Hepatotoxic Clean Hepatotoxic

Phenoxybenzamine Clean Clean Clean

Zafirlukast Hepatotoxic Clean Clean

GSK-A Hepatotoxic Clean Hepatotoxic

GSK-B Hepatotoxic Hepatotoxic Hepatotoxic

GSK-C Hepatotoxic Hepatotoxic Hepatotoxic

GSK-D Hepatotoxic Hepatotoxic Hepatotoxic

GSK: 4/7 correctly predicted

DS: 6/7 correctly predicted

Paul Michalski. DVDMDG Symposium. March 19, 2019.

32

Reason for Different Fluconazole Predictions Different interpretation of in vitro mitotox data

GSK: All assays are within the limits of normal, so there is no effect.

DILIsym: The 20% decline in OCR is a real effect indicating weak ETC inhibition. The subsequent rise back

to baseline indicates uncoupling at higher concentrations.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

33

DILIsym PoC Conclusions

1) DILIsym performs well for compounds with hepatoxicity mechanism which are

(a) included in DILIsym

(b) detectable in the DILIsym in vitro assay panel

2) PoC highlighted the importance of mitotox as a hepatotox mechanism. We are now in the

process of including mitotox screens in the GSK hepatotox strategy.

3) Assay negative compounds cannot be modeled in DILIsym (seems obvious, but now forms

an important flag in screening strategy).

4) The DILIsym assay panel does a poor job of capturing the effects of reactive metabolites.

The assay panel is being re-configured to include spheroids instead of primary hepatocytes.

5) Most importantly, DILIsym does provide added value over our current GSK hepatotox

strategy.

Paul Michalski. DVDMDG Symposium. March 19, 2019.

General Pre-clinical Modeling Considerations

How do we use DILIsym to inform go/no-go decisions?

Paul Michalski. DVDMDG Symposium. March 19, 2019.

35

Population variability only manifests in a narrow dose range

At very low and very high doses, all subjects respond similarly

Paul Michalski. DVDMDG Symposium. March 19, 2019.

36

Pre-clinical view of DILIsym

Estimated

therapeutic dose

In general we will have an estimated human therapeutic dose and the DILIsym model

results with their associated uncertainty.

Question: How do we use this information to make a go/no-go decision?

Paul Michalski. DVDMDG Symposium. March 19, 2019.

37

Pre-clinical view of DILIsym

DILIsym cannot be used in isolation. Must take into account the category of the drug, pre-

clinical tox observations, and the DILIsym predictions for human.

Example: Drug in the “Low dose, assay positive” category

Current GSK strategy says “GO”.

Exp. Pre-clinical Coverage

No data < 10 > 10

DS Pred Preclin Cov <10 >10 <10 >10 <10 >10

(DS Tox. Dose) / (Est. Eff. Dose)

< 1 STOP STOP STOP STOP GO? STOP

1 – 10 GO GO GO GO? GO GO

> 10 GO GO GO GO GO GO

Paul Michalski. DVDMDG Symposium. March 19, 2019.

38

Pre-clinical view of DILIsym

DILIsym cannot be used in isolation. Must take into account the category of the drug, pre-

clinical tox observations, and the DILIsym predictions for human.

GO GO

STOP STOP

Low dose, assay neg. Low dose, assay pos.

High dose, assay neg. High dose, assay pos.

Exp. Pre-clinical Coverage

No data < 10 > 10

DS Pred Preclin Cov <10 >10 <10 >10 <10 >10

(DS Tox. Dose) / (Est. Eff.

Dose)

< 1 STOP STOP STOP STOP GO? STOP

1 – 10 GO GO GO STOP? GO GO

> 10 GO GO GO GO? GO GO

Exp. Pre-clinical Coverage

No data < 10 > 10

DS Pred Preclin Cov <10 >10 <10 >10 <10 >10

(DS Tox. Dose) / (Est. Eff.

Dose)

< 1 STOP STOP STOP STOP GO? STOP

1 – 10 GO GO GO GO? GO GO

> 10 GO GO GO GO GO GO

Exp. Pre-clinical Coverage

No data < 10 > 10

DS Pred Preclin Cov <10 >10 <10 >10 <10 >10

(DS Tox. Dose) / (Est. Eff.

Dose)

< 1 STOP STOP STOP STOP STOP STOP

1 – 10 GO GO GO? STOP GO? GO

> 10 GO GO GO GO? GO GO

Exp. Pre-clinical Coverage

No data < 10 > 10

DS Pred Preclin Cov <10 >10 <10 >10 <10 >10

(DS Tox. Dose) / (Est. Eff.

Dose)

< 1 STOP STOP STOP STOP STOP STOP

1 – 10 STOP STOP STOP? STOP STOP STOP

> 10 Mech? Mech? Mech? STOP Mech? GO

Paul Michalski. DVDMDG Symposium. March 19, 2019.