Systems toxicology

1
Systems toxicology Predicting Drug Induced
Liver Injury Kalyanasundaram Subramanian,
Ph.D. Strand Life Sciences
2
Overview

• The Hepatotoxicity problem
• Modeling Approach
• Validation
• Summary

3
Liver is highly susceptible to toxicity

• 60 liver failures are due to toxicity.
• 2-4 jaundice is associated with drugs.
• Main Problems
• Loss of Functional Liver Cells via Cell Death
• Impaired Bile Flow
• Faulty Fat Processing

4
Modes of cell Death
5
Bile flow is an important function of the liver
Systemic Circulation
Bile Duct
Portal Systemic Shunt
Liver
Intestine
6
Impaired Beta-Oxidation can lead to Fatty Liver
Diet
Beta oxidation
Free fatty acids
triglycerides
Assembled lipoprotein
7

Why is hepatotoxicity prediction hard?

• A drug may not cause toxicity but its metabolites
  might
• People may respond differently to the same drug
• Physiological status (e.g., obesity) may modulate
  toxic response

8
Biotransformation can cause toxic metabolites to
be formed
Phase II (transferases)
Phase I (CYP450)
Glutathione, Glucuronic acid, Sulphate, Glycine Gl
utamine, acetylation
Oxidation, dealkylation
RH
excreted
excreted
9
People respond differently to the same drug
10
Physiological/disease factors
estrogen
11
Hepatotoxicity is the result of complex
interactions
drug/metabolites
patient
12
Liver toxicity is inferred from blood parameters
Toxin/ Virus
Biomarkers AST, ALT, bilirubin
Detected via blood analysis Non-specific,
non-unique
13
Specific Problems to be Addressed

• Given an NCE, can I predict the concentration
  range in which the drug is safe?
• Can I predict a toxic dose range?
• Can I predict the mechanism by which the drug
  will injure the liver?
• Can I identify specific biomarkers associated
  with each injury mechanism?

14
Building a Top-Down Systems Model
increasing detail
cascade of biological pathways linking target
to clinical endpoints
target
clinical endpoint
Liver Lobule
Proteins
Liver cells
15
Top-down Model Development Leads to Novel Insights
Clinical data
Explicit hypotheses reverse engineered to
fill knowledge gaps
High throughput data In vitro data Animal model
data
16
Our strategy for building a predictive model

• Strategy
• Build a comprehensive model of liver homeostasis
  (normal or steady state)
• Treat toxicity as a case of drug-induced
  perturbations
• Computationally mine the network to identify key
  pathway (combinations)
• Create assays that measure effect of
  drug/metabolite on the pathways
• Predictive platform is a combination of assays
  and model
• Generate mechanism specific biomarkers
• Alternatives
• QSAR
• genomics

17
Identifying Mechanisms

• Identify drugs reported to be liver toxic in
  literature
• Identify the molecular mechanism of toxicity for
  each such drug (e.g., cell death, impaired bile
  flow etc)
• Identify root causes for these mechanisms (e.g.,
  oxidative stress, transporter inhibition)
• Model these root causes (identify pathways for
  each, and kinetics for each pathway)

18
Clues from biomarkers on the injury mechanisms
Drugs can cause cell death (necrosis) to the same
extent as Ischemia ATP depletion is one of the
root causes of Ischemia So ATP depletion could be
a root cause of drug induced cell death
19
ATP and Glutathione depletion can lead to necrosis
ATP,GSH,Ca...
20
Transporter Inhibition can lead to impaired Bile
Flow
Systemic Circulation
Bile Duct
Portal Systemic Shunt
Liver
Intestine
21
Fat import and export is an important function of
the liver
22
Handling Metabolism

• Modeling ATP depletion also addresses metabolism
  effects implicitly
• How?

23
Drug Metabolism linked to Injury
24
The liver is functionally asymmetric
Zone 1 High Oxygen High gluconeogenesis
Zone 3 Low oxygen High CYPs, glycolysis
Toxicity could be linked to metabolism
25
Handling Physiology Effects

• Physiological Factors which exacerbate toxicity
• Obesity
• Diabetes

26
Handling Patient Variations

• Models can handle genetic variations in key
  proteins involved
• Key enzymes can also point to source of
  variability

27
Equation Types Roadblocks
28
Simplification rules for complex pathways

• Rapid equilibrium approximation is used whenever
  applicable
• Single substrate Michealis-Menten can be analysed
  via standard approximation when applicable
  (linear or zero-order)
• Multiple flux processes can be clubbed together
  when they dont change over the time-scales
  considered
• Multi-substrate enzyme kinetics can be reduced to
  single substrate when other substrates are near
  saturation
• In a cascade
• Rate of an enzyme (Rate of the overall cascade)
  Steady state

29
Cellular ATP The Fluxes involved

• Rate of mitochondrial ATP production by
  fof1ATPase Vfof1atpase
• Rate of transport of ATP-ADP across
  mitochondriaVant
• Rate of transport of Pi across mitochondriaVpicar
  rier
• Rate of adenylate kinase (interconversion of
  adenine nucleotides)Vadk
• Rate of cytosolic ATP production through
  glycolysis by pyruvate
• kinase and phosphoglycerate kinase Vpk Vpgk
• Rate of mitochondrial ATP utilisation
  Vutilisation(mitochondria)
• Rate of cytosolic ATP utilisation
  Vutilisation(cytosol)

30
Differential Equations
31
Conservation Laws
PHOSPHATE POOL IN THE CELL
Psum3ATPe2ADPeAMPePie(3ATPm2ADPmPim)/Rcmc
onstant
ADENINE NUCLEOTIDE POOL IN THE CELL
Asum ATPeADPeAMPe (ATPmADPmAMPm)/Rcmcons
tant

• AMP DOES NOT TRAVERSE THROUGH THE MITOCHONDRIAL
  MEMBRANE.
• 
  
  (Dransield Aprille Arch. Biochem. Biophys.
  313156-165)
• 2) ADP ATP ARE EXCHANGED BETWEEN THE
  CYTOSOL AND MITOCHONDRIA
• VIA ANT ANTIPORT

TWO CONSERVATION LAWS FOR ADENINE NUCLEOTIDE
POOLS IN THE CELL
ADENINE NUCLEOTIDE POOL IN THE CYTOSOL
1. Asum,e Total adenine pool in the cytosol
ATPeADPeAMPeconstant
ADENINE NUCLEOTIDE POOL IN THE MITOCHONDRIA
2. Asum,mTotal adenine pool in the
mitochondriaATPmADPmconstant
e- cytosol m- mitochondria
32
Enzyme Kinetics
Flux V freactants
Reactants considered either variables or as
constant parameters.
e.g. The kinetic expression for fof1atpase
33
Modeling Fluxes Typical Roadblocks solutions
Problems Solutions
Hepatocyte data unavailable Use non-dimensional parameter values (e.g. S/Km) from other sources.
Km value needs to be estimated for single substrate MM kinetics For a cascade of reactions the homeostatic flux value of the cascade can be equated to the flux value of any enzyme in the cascade at steady state
Km values for the substrates that take part in more than one important metabolic network e.g., ATP, NADPH Important metabolites operate near saturation, hence two substrate enzyme kinetics can be modified to single substrate kinetics
For mitochondrial enzymes and transporters, experiments are usually done in isolated mitochondria Protein content ratio of cell protein to mitochondrial protein is used to express the flux value with respect to whole cell
In vitro experiment does not mimic the in vivo combinations of effects of cellular regulators Simulate the in vivo condition with the help of experimental information
34
How to estimate in vivo value of a parameter from
in vitro data - I
VPFK Rate determining step for glycolytic
cascade Very highly regulated
allosteric enzyme by various cellular
metabolites e.g. ATP, AMP, phosphate, ammonium
ion, Fructose-1,6-bisphosphate,
Fructose-2,6-bisphosphate, pH, citrate
Vmax Obtained from literature. (JBC
1797 254 5584-5587, Annu. Rev. Biochem. 1968 37
377-390) In vitro Km for PFK w.r.t. F6P is 6 mM
while In vivo
concentration of F6P is 0.06 mM (Biochemistry
1980,191477-1484)
35
How to estimate in vivo value of a parameter from
in vitro data - II

• Flux of PFK, Vpfk f(ATP, AMP, Pi, FBP)
• For an allosteric enzyme
• nf1(ATP, AMP, Pi, FBP)
• Kmf2(ATP, AMP, Pi, FBP)
• Objectives-
• Calculate f1 and f2.
• Combine them to obtain proper physiological
  allosteric behaviour of PFK .

JBC 1974 249 7824-7831. Biochemistry 1980 19
1477-1484. Biochemistry 1980 19 1484-1490. Annu.
Rev. Physiol 1992 54 885-909. Annu. Rev. Biochem
1983 52 617-653. Annu. Rev. Biochem 1968 37
249-302. JBC 1979 254 5584-5587. PNAS 1980 77
5861-5864.
36
Parameter Fitting for ATP regulation
Km,ATP(1.63ATP1.36)
nATP1.26(12.88ATP/(0.162ATP))
37
Combining all the Regulators
Regulator Factor Effect on Km
ATP 6.25 Increase
AMP 0.41 Decrease
Pi 0.21 Decrease
FBP 0.51 Decrease
(ATP, Pi, AMP, FBP) correction factor
ATP only
Km6 mM
Km0.06 mM
38
Steady State Flux Balance Analysis Simple
but Useful
This is not standard FBA, but an understanding
that parts of the system is coupled via fluxes
Vant Vpicarrier Vadk0 Vutilisation(cytoso
l) VantVpkVpgk
Helps to calculate unknown parameter values for
Vpicarrier
Validated by experimental evidence.
Used to calculate the parameters for clubbed
utilisation term.
Numerical solution of the equations provides the
steady state metabolite concentrations and flux
values
39
The Results of All of this

• 2.5 years of about 10 people working on this
  problem
• A non-linear highly ODE system that consists of
  109 states and
• continues to grow in size and complexity
• Can describe certain aspects of the liver
  reasonably well
• Consists of several modules

40
Modules
41
Definition of homeostasis for a minimal model
cell viability
Cytotoxicity cell death
bile acids Bilirubin Actin skeleton
Cholestasis/ impaired bile flow
Hepatotoxicity in the clinic
predictive model in silico
fatty acids
Steatosis/ fatty liver
42
Glutathione-ROS-Lipid Peroxidation

• Scope
• To capture intracellular GSH and ROS metabolism,
  the lipid peroxidation process
• the interdependence among the three modules in
  homeostasis to predict drug metabolism induced
  changes in GSH and intracellular effects of
  increased ROS.
• Major pathways
• Intracellular antioxidant interactions
• Basic scheme of lipid peroxidation
• GSH synthesis, efflux and the redox cycle
• Upon completion, the model will predict
• GSH depletion caused by increased ROS (due to
  drug metabolism) or the conjugation of the drug
  with GSH (eg. EA, Acetaminophen)
• The increase in lipid peroxidation caused by
  increased ROS and imbalance of antioxidant levels
  (including GSH).

43
ATP Conservation

• Scope
• Metabolic network for ATP synthesis
• Understanding the regulation and connections
  among the different pathways involved
• Major pathways
• Glycolysis, malate-aspartate shuttle,
  Tri-carboxylic acid (TCA) cycle, Oxidative
  phosphorylation
• Upon completion, this module will predict
• target (or targets) which when perturbed can
  cause drug induced necrotic death of cell due to
  ATP depletion.
• the distribution among different pathways (e.g.
  Glycolysis and Oxidative phosphorylation) for the
  total ATP pool in the cell, under normal and
  perturbed state
• Time scale of cell survival under toxic exposure.

44
Fatty Acid Metabolism

• Scope
• To understand the partitioning of free fatty acid
  flux in the hepatocyte to identify the key
  event(s) and/or metabolite(s) concentrations that
  could lead to the development of fatty liver
  (steatosis)
• Major pathways
• mitochondrial beta oxidation, triglyceride
  synthesis and storage, ketone body formation,
  fatty acid synthesis
• Upon completion, the module can explain
• the development of steatosis from the inhibition
  of any of the above processes. For e.g.,
  tetracycline, amiodarone, inhibit ?-oxidation
  leading to steatosis.
• Alcohol-induced steatosis
• Hormonal control of VLDL secretion from the
  triglyceride stores.

45
Actin Cytoskeleton

• Scope
• Quantity and rate of actin polymerization, the
  number and length of filaments and degree of
  branching
• The impact of the cytoskeletal function on
  bile-flow related processes
• Major pathways
• the actin polymerization pathway with the role of
  six actin binding proteins, pH, electrolytes
• second messengers that modulate the pathway (e.g.
  PIP2)
• Given quantitative data, the module can explain
• Effects of drugs that alter the above mentioned
  modulators and hence, actin architecture
  function
• The degree of impact on canalicular
  contractility, microvilli integrity and bile-
  transporter function

46
Bile Salt, Bilirubin, Bicarbonate

• Scope
• To understand the metabolism and transport of
  bile-salts, bilirubin and bicarbonate ions in the
  hepatocyte
• To understand the bile-salt dependant and
  independent flow of bile in the body
• Major pathways
• Bile salt and bilirubin metabolism
• Upon completion this module will explain
• Cholestasis and necrosis due to dysregulation of
  the pathways modeled
• The impact of drugs on these pathways (given in
  vitro data)

47
Validation Studies

• Validate homeostasis
• Module level and whole system level
• Validate effect of drugs and toxins
• Validate known genetic diseases
• Look for insights

48
Validation HomeostasisATP module
Simulations Experimental Results
Cytosolic ATP 2950 mM 2760 mM
Cytosolic ADP 200 mM 315 mM
Mitochondrial ATP 9000 mM 10380 mM
Mitochondrial ADP 7000 mM 5380 mM
Cytosolic Pi 3375 mM 3340 mM
Mitochondrial Pi 14000 mM 16800 mM
Cytosolic AMP 60 mM 130 mM
ATP generated by Glycolysis 33 38
ATP generated by Oxidative- phosphorylation 66 57
Eur J Biochem 1978, 84413-420 Eur J
Biochem 1999 263671-685
49
Validation HomeostasisActin Cytoskeleton module

• Rate of filament growth is linear and constant
  both at the pointed and the barbed end, Pollard,
  J. Cell Biol. 1986 (103) 2747-54

50
Validation Homeostasis Steatosis module

• We examined how the fatty acid flux was
  distributed between esterification and
  ß-oxidation in differing nutritional states and
  compared against known values in the literature

of flux entering mitochondrial oxidation of flux entering mitochondrial oxidation
Nutritional State Simulations Experimental Value
Fed 74.6 70
Fasted 35 30
Ontko, J A. JBC,1972,vol247,1788-1800
51
Effect of toxins BSO
Validation Effect of toxinGSH module

• Model reproduces the extent and time-course of
  GSH depletion due to BSO

Toxin Buthionine Sulfoximine (BSO) Target
?-glutamylcysteine synthetase
52
Validation Effect of DrugGSH module
Drug Ethacrynic Acid (EA) Target
Glutathione-S-transferase
Experiment
2
3
1

• Mitochondria depletion of GSH is also reproduced

53
Validation Effect of DrugBile-salt module
Drug Fusidate Target Bile Salt Export Pump
(BSEP)

• ATP Dependent transport of taurocholate inhibited
  by fusidate with a Ki of 2.2 ?M 1
• Simulate effect of 100 mg/Kg dose given
  intravenously, use PK data from literature 2
• Simulations show that the rate of transport of
  taurocholate inhibited by 85
• Compares well with experimental value of 80 1

1 Bode KA, et. al., Biochem Pharmacol. 2002 Jul
164(1)151-8 2 Taburet AM et. al., J Antimicrob
Chemother. 1990 Feb25 Suppl B23-31
54
Validation Genetic DiseaseBilirubin module
Literature Simulations
UGT activity 10-33 wild type 20 activity
UCB in serum lt70 µM 50 µM
55
Validation Genetic DiseaseBilirubin module
Literature Simulations
UGT activity 0-10 wild type 110 activity
UCB in serum 100-430 µM 106-219 µM
56
Novel InsightsATP module
Our simulations show a bell shaped dependence of
glycolysis-rate on ATP1
Further analysis shows that there is a critical
concentration for ADP in the system
1 Ataullakhanov Vitvitsky Bioscience Reports.
2002 22501-511
57
Biological Insights
Novel InsightsGSH module

• The capacity of the liver to recover from
  reactive hydrogen shock

58
Enhancing the model using NLPDrugs involved in
Steatosis
59
Enhancing the model using NLPDrugs involved in
Cholestasis
60
The Overall Hepatotoxicity Platform..
drug candidate
toxic pathways toxic concentrations biomarkers etc
assay results
Assay Panel
Liver Model
61
Extensions

• Acute to Chronic
• Idiosyncrasy
• Organ architecture
• Other toxicity endpoints

62
Summary

• A liver model that can.
• Reproduce homeostasis (behave like a normal
  liver)
• Quantify the evolution homeostasis to disease and
  simulate toxicity
• Identify important pathways involved in disease
• Evaluate toxicity potential of drug candidates
• Generate mechanism specific biomarkers

63
Team

• Anupama Rajan Bhat
• R. Rajesh, Ph.D.
• Dr. Nalini, R.
• Dr. Narasimha, M.K., Ph.D.
• Rajeev Kumar
• Sai Jagan Mohan, Ph.D.
• Sonali Das, Ph.D.
• Sowmya Raghavan, Ph.D.
• Raghunathan Srivatsan, Ph.D.
• Kas Subramanian, Ph.D.