1

Dose-Response Modeling: Past, Present, and Future

Rory B. Conolly, Sc.D.Center for Computational Systems Biology

& Human Health AssessmentCIIT Centers for Health Research

(919) 558-1330 - voicerconolly@ciit.org - e-mail

SOT Risk Assessment Specialty Section, Wednesday, December 15, 2004

2

Outline

• Why do we care about dose response?

• Historical perspective– Brief, incomplete!

• Formaldehyde

• Future directions

3

Perspective

• This talk mostly deals with issues of cancer risk assessment, but I see no reason for any formal separation of the methodologies for cancer and non cancer dose-response assessments– PK

– Modes of action

– Tumors, reproductive failure, organ tox, etc.

4

Typical high dose rodent data – what do they tell us?

Res

pons

e

Dose

5

Not much!R

espo

nse

Dose

Interspecies

6

PossibilitiesR

espo

nse

Dose

Interspecies

7

PossibilitiesR

espo

nse

Dose

Interspecies

8

PossibilitiesR

espo

nse

Dose

Interspecies

9

PossibilitiesR

espo

nse

Dose

Interspecies

10

Benzene Decision of 1980

• U.S. Supreme Court says that exposure standards must be accompanied by a demonstration of “significant risk”– Impetus for modeling low-dose dose response

11

1984 Styrene PBPK model(TAP, 73:159-175, 1984)

A physiologically based description of the inhalation pharmacokinetics of styrene in rats and humans

John C. Ramseya and Melvin E. Andersenb

a Toxicology Research Laboratory, Dow Chemical USA, Midland, Michigan 48640, USAb Biochemical Toxicology Branch, Air Force Aerospace Medical Research Laboratory (AFAMRL/THB), Wright-Patterson Air Force Base, Ohio 45433, USA

12

Biologically motivated computational models(or)

Biologically based computational models

• Biology determines– The shape of the dose-response curve

– The qualitative and quantitative aspects of interspecies extrapolation

• Biological structure and associated behavior can be– described mathematically

– encoded in computer programs

– simulated

13

Experiments to understandmechanisms of toxicity andextrapolation issues

Computationalmodels

Riskassessment

Biologically-based computational models: Natural bridges between research and risk

assessment

14

Garbage in – garbage out

• Computational modeling and laboratory experiments must go hand-in-hand

15

Refining the description with research on pharmacokinetics and pharmacodynamics

(mode of action)R

espo

nse

Dose

Interspecies

16

Refining the description with research on pharmacokinetics and pharmacodynamics

(mode of action)R

espo

nse

Dose

Interspecies

17

Refining the description with research on pharmacokinetics and pharmacodynamics

(mode of action)R

espo

nse

Dose

Interspecies

18

Res

pons

e

Refining the description with research on pharmacokinetics and pharmacodynamics

(mode of action)

Dose

Interspecies

19

Formaldehyde nasal cancer in rats:

A good example of extrapolations across doses and species

20

21

Swenberg JA, Kerns WD, Mitchell RI, Gralla EJ, Pavkov KL

Cancer Research, 40:3398-3402 (1980)

Induction of squamous cell carcinomas of the rat nasal cavity by inhalation exposure to formaldehyde vapor.

1980 - First report of formaldehyde-induced tumors

22

Formaldehyde bioassay results

0

10

20

30

40

50

60

T

um

or R

espo

nse

(%)

0 0.7 2 6 10 15

Exposure Concentration (ppm)

Kerns et al., 1983

Monticello et al., 1990

23

Mechanistic Studies and Risk Assessments

24

What did we know in the early ’80’s?

• Formaldehyde is a carcinogen in rats and mice

• Human exposures roughly a factor of 10 of exposure levels that are carcinogenic to rodents.

25

1982 – Consumer Product Safety Commission (CPSC) voted to ban urea-formaldehyde foam insulation.

26

Casanova-Schmitz M, Heck HD

Toxicol Appl Pharmacol 70:121-32 (1983)

Effects of formaldehyde exposure on the extractability of DNA from proteins in the rat nasal mucosa.

1983 - Formaldehyde cross-links DNA with proteins - “DPX”

27

DPX

(FORMALDEHYDEIN AIR)

MUCUS

RESPIRATORYEPITHELIUM

CH2

CH2

CH2

28

Starr TB, Buck RD

Fundam Appl Toxicol 4:740-53 (1984)

The importance of delivered dose in estimating low-dose cancer risk from inhalation exposure to formaldehyde.

1984 - Risk Assessment Implications

29

1985 – No effect on blood levels

Heck, Hd’A, Casanova-Schmitz, M, Dodd, PD, Schachter, EN, Witek, TJ, and Tosun, T

Am. Ind. Hyg. Assoc. J. 46:1. (1985)

Formaldehyde (C2HO) concentrations in the blood of humans and Fisher-344 rats exposed to C2HO under controlled conditions.

30

1987 – U.S. EPA cancer risk assessment

• Linearized multistage (LMS) model

– Low dose linear

– Dose input was inhaled ppm

– U.S. EPA declined to use DPX data

31

Summary: 1980’s

• Research– DPX – delivered dose

– Breathing rate protects the mouse (Barrow)

– Blood levels unchanged

• Regulatory actions– CPSC ban

– US EPA risk assessment

32

Key events during the ’90s

• Greater regulatory acceptance of mechanistic data for

risk assessment (U.S. EPA)

• Cell replication dose-response

• Better understanding of DPX (Casanova & Heck)

• Dose-response modeling of DPX (Conolly, Schlosser)

• Sophisticated nasal dosimetry modeling (Kimbell)

• Clonal growth models for cancer risk assessment

(Moolgavkar)

33

1991 – US EPA cancer risk assessment

• Linearized multistage (LMS) model– Low dose linear

– DPX used as measure of dose

34

Monticello TM, Miller FJ, Morgan KT

Toxicol Appl Pharmacol 111:409-21 (1991)

Regional increases in rat nasal epithelial cell proliferation following acute and subchronic inhalation of formaldehyde.

1991, 1996 - regenerative cellular proliferation

35

Normal respiratory epithelium in the rat nose

36

Formaldehyde-exposed respiratory epitheliumin the rat nose (10+ ppm)

37

Dose-response for cell division rate

2.00E-04

3.00E-04

4.00E-04

5.00E-04

6.00E-04

7.00E-04

0 1 2 3 4 5 6 7

ppm formaldehyde

(Raw data)

38

DPX submodel – simulation of rhesus monkey data

1 2 3 4 5 6 710

-4

10-3

10-2

10-1

PPM

DPX (pmol/mm

3)

DPX dose-response for Rhesus monkey

Vmax: 91.02. pmol/mm3/min

Km: 6.69 pmol/mm3 kf: 1.0878 1/min Tissue thickness ALWS: 0.5401 mm MT: 0.3120 mm NP: 0.2719 mm

39

Summary: Dose-response inputs to the clonal growth model

• Cell replication– J-shaped

• DPX– Low dose linear

40

CFD Simulation of Nasal Airflow(Kimbell et. al)

41

2-Stage clonal growth model(MVK model)

Normalcells (N)

Division

N)

Death/differentiation

(N)

Mutation(N )

Initiatedcells (I)

(I)

(I)

Mutation( )

Cancercell

Tumor

(delay)

42

Dose-response for cell division rate

2.00E-04

3.00E-04

4.00E-04

5.00E-04

6.00E-04

7.00E-04

0 1 2 3 4 5 6 7

ppm formaldehyde

(Raw data)

2.0000E-04

2.5000E-04

3.0000E-04

3.5000E-04

4.0000E-04

4.5000E-04

5.0000E-04

5.5000E-04

0 1 2 3 4 5 6 7

ppm formaldehyde

(Hockey stick transformation)

43

Simulation of tumor response in rats

44

CIIT clonal growth cancer risk assessment for formaldehyde

(late ’90’s)

• Risk assessment goal– Combine effects of cytotoxicity and mutagenicity to

predict the tumor response

45

1987 U.S. EPA

Tumor response

Inhaled ppm

Cancer model(LMS)

46

1991 U.S. EPA

Inhaled ppm

Tissue dose(DPX)

Cancer model(LMS)

Tumor response

47

1999 CIIT

Cell killing Cell proliferation

Mutagenicity(DPX)

Cancer model(Clonal growth)

Tumor response

Inhaled ppm

Tissue dose

CFD modeling

48

Formaldehyde: Computational fluid dynamics models of the nasal airways

F344 RatRhesus Monkey

Human

49

Human assessment

respiratory tracttumor data(control only)

cell replicationdose-response (rat)

site-specific flux intorespiratory tract epithelium

DPX dose-responseprediction (scale-upfrom rat and monkey)

mode of actiondose-responsesubmodels

cells at risk inrespiratory tract

single-path lungdosimetry model

cell replicationin control rats

human tumorincidence

CFD nasaldosimetry model

Inhaledformaldehyde

exposure scenario

2-STAGECLONALGROWTH MODEL

maximum likelihoodestimation of baselineparameter values

2-STAGECLONALGROWTHMODEL

50

Baseline calibration against human lung cancer data

51

DPX and direct mutation

• Direct mutation is assumed to be proportional to the amount of DPX:

• Is KMU big or small?

DPXKMUmutation ⋅=

52

Grid search

53

Optimal value of KMU is zero

54

Upper bound on KMU

55

Calculation of the value of KMU

• Grid search• Optimal value of KMU was zero

– Modeling implies that direct mutation is not a significant action of formaldehyde

• 95% upper confidence limit on KMU was estimated

56

Human risk modeling

Normalcells (N)

Division

N)

Death/differentiation

(N)

Mutation(N )

Initiatedcells (I)

(I)

(I)

Mutation( )

Cancercell

Tumor

(delay)

57

Final model: Hockey stick and 95% upper confidence limit on value of KMU

2.0000E-04

2.5000E-04

3.0000E-04

3.5000E-04

4.0000E-04

4.5000E-04

5.0000E-04

5.5000E-04

0 1 2 3 4 5 6 7

1 2 3 4 5 6 710

-4

10-3

10-2

10-1

PPM

DPX (pmol/mm

3)

DPX dose-response for Rhesus monkey

Vmax: 91.02. pmol/mm3/min

Km: 6.69 pmol/mm3 kf: 1.0878 1/min Tissue thickness ALWS: 0.5401 mm MT: 0.3120 mm NP: 0.2719 mm

95% UCL on KMU

58

Predicted human cancer risks(hockey stick-shaped dose-response for cell

replication; optimal value for KMU)

2.0000E-04

2.5000E-04

3.0000E-04

3.5000E-04

4.0000E-04

4.5000E-04

5.0000E-04

5.5000E-04

0 1 2 3 4 5 6 71 2 3 4 5 6 7

10-4

10-3

10-2

10-1

PPM

DPX (pmol/mm

3)

DPX dose-response for Rhesus monkey

Vmax: 91.02. pmol/mm3/min

Km: 6.69 pmol/mm3 kf: 1.0878 1/min Tissue thickness ALWS: 0.5401 mm MT: 0.3120 mm NP: 0.2719 mm

Optimal value of KMUKMU = 0.

59

“Negative risk” using raw dose-response for cell replication

2.00E-04

3.00E-04

4.00E-04

5.00E-04

6.00E-04

7.00E-04

0 1 2 3 4 5 6 7

1 2 3 4 5 6 710

-4

10-3

10-2

10-1

PPM

DPX (pmol/mm

3)

DPX dose-response for Rhesus monkey

Vmax: 91.02. pmol/mm3/min

Km: 6.69 pmol/mm3 kf: 1.0878 1/min Tissue thickness ALWS: 0.5401 mm MT: 0.3120 mm NP: 0.2719 mm

95% UCL on KMU

60

Make conservative choices when faced with uncertainty

• Use hockey stick-shaped cell replication

• Use a 95% upper bound on the dose-response for the directly mutagenic mode of action– Statistically optimal model has 0 (zero) slope

• Risk model predicts low-dose linear risk.

• Optimal, data based model predicts negative risk at low doses

61

Summary: CIIT Clonal Growth Assessment

• Either no additional risk or a much smaller level of risk than previous assessments

• Consistent with mechanistic database– Direct mutagenicity

– Cell replication

62

Summary: CIIT Clonal Growth Assessment

• International acceptance– Health Canada– WHO– MAK Commission (Germany)– Australia– U.S. EPA (??)

• Peer-review

63

IARC 2004

• Classified 1A based on nasopharyngeal cancer

• Myeloid leukemia data suggestive but not sufficient– Concern about mechanism

– British study negative

• Reclassification driven by epidemiology

• In my opinion inadequate consideration of regional dosimetry

64

nasopharynx

Anteriornose

Wholenose

65

IARC: hazard characterization vs. dose-response assessment

66

Formaldehyde summary

• Nasal SCC in rats

• Mechanistic studies

• Risk Assessments

• Implications of the data

• IARC

67

The future

68

Outline

• Long-range goal

• Systems in biological organization

• Molecular pathways

• Data

• Example– Computational modeling

– Modularity

69

Long-range goal

• A molecular-level understanding of dose- and time-response behaviors in laboratory animals and people.– Environmental risk assessment

– Drug development

– Public health

70

Levels of biological organization

Populations

Organisms

Tissues

Cells

Organelles

Molecules(systems)

(systems)

(systems)

(systems)

(systems)

Descriptive Mechanistic

71

Levels of biological organization

Populations

Organisms

Tissues

Cells

Organelles

Molecules(systems)

Today

72

Molecular pathways

73

Segment polarity genes in Drosophila

Albert & Othmer, J. Theor Biol. 223, 1 – 18, 2003

74

ATM curatedPathway fromPathway Assist®

75

Approach

• Initial pathway identification– Static map

• Existing data

• New data

• Computational modeling– Dynamic behavior

– Iterate with data collection

76

Initial pathway identification

• Use commercial software that can integrate data from a variety of sources (Pathway Assist)– Scan Pub Med abstracts to identify “facts”

– Create pathway maps

– Incorporate other, unpublished data

• Quality control– Curate pathways

77

Computational modeling

• To study the dynamic behavior of the pathway

• Analyze data– Are model predictions consistent with existing data?

• Make predictions– Suggest new experiments

– Ability to predict data before it is collected is a good test of the model

78

DNA damage and cell cycle checkpoints

(a) G1/S Checkpoint (b) G2/M Checkpoint (a) G1/S Checkpoint (b) G2/M Checkpoint

79

p21 time-course data and simulation

Experimental data

80

IR

Mu

tatio

n F

ract

ion

Ra

te

model calculated values

(Redpath et al, 2001)

Mutations dose-response and model prediction

81

Data

82

(Fat)

Liver

Rest of Body

Air-bloodinterface

Venousblood

Tissue dosimetry is the “front end” to a molecular pathway model

83

Gain-of-function and loss-of-function screens to study network structure

• Selectively alter behavior of the network– Loss-of-function

• SiRNA

– Gain-of-function

• full-length genes

• Look for concordance between lab studies and the behavior of the computational model– Mimic gain-of-function and loss-of-function changes in

the computer

84

Example

• Skin irritation

• MAPK, IL-1, and NF-B computational “modules”

• High throughput overexpression data to characterize IL-1 – MAPK interaction with respect to NF-B

85

Skin Irritation

Epidermis

Dermis

Dead cells

Blood vessels

(keratinocytes)

Nerve Ending

s(fibroblasts)

Chemical

Tissue damage

Tissue damage

A cascade of inflammatory responses

(cytokines)

• Study on the dose response of the skin cells to inflammatory cytokines contributes to quantitative assessment of skin irritation

86

Modular Composition of IL-1 Signaling

IL-1R

IL-1

Secondary messenger

NF-B MAPK Others

Extracellular

Intracellular

IL-6, etc.Transcriptional factors

IL-1 specific top module

Constitutive downstream NF-B module

87

IL-1

IL-1R

MyD88

P

IRAK

Degraded

IRAK gene

IRAK

IRA

K

P

TRAF6

TAB2

IRA

K

P

TRAF6

TA

K1

TA

B1

P

IK PIK

Nucleus

Cytoplasm

Top IL-1 Signaling Module

NF-kB moduleSelf-limiting mechanism

88

Top Module Simulation

• IL-1 receptor number and ligand binding parameters from human keratinocytes

• Other parameters constrained by reasonable ranges of similar reactions/molecules, and tuned to fit data

Time (hrs) Time (hrs)

TA

K1

*

IRA

Kp

Increasing IRAKp

degradation

89

IK P

NFBIB

P

NFB

IB Degraded

IB geneNFB

IB

IL-6 geneNFB NFB

IB

PNFBIB

Nucleus

Cytoplasm

Constitutive NF-B Signaling Module

IK

Negativefeedback

Input signal

90

NF-B Module Simulation

• Parameters from existing NF-B model (Hoffmann et al., 2002) and refined to fit experimental data in literature

Time (hrs)

Con

cen

trat i

on

(M

) IB

NF-B+ _

Time (hrs)

Con

cen

trat i

on

(M

)

IL-6

Add constant

input signal

Longer delay

Smoothened oscillations

91

The IB–NF-B Signaling Module: Temporal Control and Selective Gene ActivationAlexander Hoffmann, Andre Levchenko, Martin L. Scott, David Baltimore

Science 298:1241 – 1245, 2002

6 hr

92

MAPK intracellular signaling cascades

http://www.weizmann.ac.il/Biology/open_day/book/rony_seger.pdf

93

Growth factor

MAPKKK

MAPKK

MAPKPLA2

AA

PKC

PLA2

AA

PKC

MKPMKP

94

MAPK time-course and bifurcation after a short pulse of PDGF

Input pulse

Growth factor

MAPKKK

MAPKK

MAPKPLA2

AA

PKC

PLA2

AA

PKC

MKPMKP

95

PIRAK

DegradedIRAK gene

TRAF6

TAB2 TAK1TAB1

I

?

K

IL-1

IL-1R

MyD88

P

IRAK

P

IRAKIRAK

I

?

KP

P

NF

?

B module

NF

?

B-dependenttranscription

MAP3K1

IL-1 MAPK crosstalk and NFkB activation

96

Gain-of-function screen

97

Model prediction

98

Future directions

• Computational modeling and data collection at higher levels of biological organization– Cells

• Intercellular communication

– Tissues

– Organisms

• NIH initiatives

• Environmental health risk, drugs ==> in vivo

99

Summary

• Biological organization and systems

• Molecular pathways– identification

– Computational modeling

• Data– Gain-of-function

– Loss-of-function

• Skin irritation example– 3 modules

– Crosstalk

– Targeted data collection

100

Acknowledgements

• Colleagues who worked on the clonal growth risk assessment– Fred Miller, Julian Preston, Paul Schlosser, Julie

Kimbell, Betsy Gross, Suresh Moolgavkar, Georg Luebeck, Derek Janszen, Mercedes Casanova, Henry Heck, John Overton, Steve Seilkop

101

Acknowledgements

• CIIT Centers for Health Research– Rusty Thomas

– Maggie Zhao

– Qiang Zhang

– Mel Andersen

• Purdue– Yanan Zheng

• Wright State University– Jim McDougal

• Funding– DOE

– ACC

102

End