DRAFT - DO NOT CITE OR QUOTE
EPA/635/R-10/002C
www.epa.gov/iris
vvEPA
TOXICOLOGICAL REVIEW OF
FORMALDEHYDE
INHALATION TOXICITY
(CAS No. 50-00-0)
In Support of Summary Information on the
Integrated Risk Information System (IRIS)
VOLUME IV of IV
Appendices
March 17, 2010
NOTICE
This document is an Inter-Agency Science Consultation draft. This information is distributed
solely for the purpose of pre-dissemination peer review under applicable information quality
guidelines. It has not been formally disseminated by EPA. It does not represent and should not
be construed to represent any Agency determination or policy. It is being circulated for review
of its technical accuracy and science policy implications.
U.S. Environmental Protection Agency
Washington, DC
-------
DISCLAIMER
This document is a preliminary draft for review purposes only. This information is
distributed solely for the purpose of pre-dissemination peer review under applicable information
quality guidelines. It has not been formally disseminated by EPA. It does not represent and
should not be construed to represent any Agency determination or policy. Mention of trade
names or commercial products does not constitute endorsement or recommendation for use.
This document is a draft for review purposes only and does not constitute Agency policy.
IV-ii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS—TOXICOLOGICAL REVIEW OF FORMALDEHYDE
(CAS No. 50-00-0)
LIST OF TABLES xi
LIST OF FIGURES xx
LIST 01 ABBREVIATIONS AM) ACRONYMS xxv
FOREWORD xxxii
AUTHORS, CONTRIBUTORS, AND REVIEWERS xxxiii
VOLUME I
1. INTRODUCTION 1-1
2. BACKGROUND 2-1
2.1. PI IYSICOCIIHYIICAI. PROPERTIES OF FORMALDEHYDE 2-1
2.2. PRODUCTION, USES, AND SOURCES OF FORMALDEHYDE 2-1
2.3. ENVIRONMENTAL LEVELS AND HUMAN EXPOSURE 2-4
2.3.1. Inhalation 2-5
2.3.2. Ingestion 2-10
2.3.3. Dermal Contact 2-11
3. TOXICOKINETICS 3-1
3.1. CHEMICAL PROPERTIES AND REACTIVITY 3-1
3.1.1. Binding of Formaldehyde to Proteins 3-1
3.1.2. Endogenous Sources of Formaldehyde 3-3
3.1.2.1. Normal Cellular Metabolism (Enzymatic) 3-3
3.1.2.2. Normal Metabolism (Non-Enzymatic) 3-5
3.1.2.3. Exogenous Sources of Formaldehyde Production 3-5
3.1.2.4. FA-GSH Conjugate as a Method of Systemic Distribution 3-6
3.1.2.5. Metabolic Products of FA Metabolism (e.g., Formic Acid) 3-6
3.1.2.6. Levels of Endogenous Formaldehyde in Animal and Human
Tissues 3-6
3.2. ABSORPTION 3-9
3.2.1. Oral 3-9
3.2.2. Dermal 3-9
3.2.3. Inhalation 3-9
3.2.3.1. Formaldehyde Uptake Can be Affected by Effects at the
Portal of Entry 3-10
3.2.3.2. Variability in the Nasal Dosimetry of Formaldehyde in
Adults and Children 3-12
3.3. DISTRIBUTION 3-13
3.3.1. Levels in Blood 3-13
3.3.2. Levels in Various Tissues 3-15
3.3.2.1. Disposition of Formaldehyde: Differentiating Covalent
Binding and Metabolic Incorporation 3-16
This document is a draft for review purposes only and does not constitute Agency policy.
IV-iii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
3.4. METABOLISM 3-20
3.4.1. In Vitro and In Vivo Characterization of Formaldehyde Metabolism 3-20
3.4.2. Formaldehyde Exposure and Perturbation of Metabolic Pathways 3-23
3.4.3. Evidence for Susceptibility in Formaldehyde Metabolism 3-24
3.5. EXCRETION 3-25
3.5.1. Formaldehyde Excretion in Rodents 3-26
3.5.2. Formaldehyde Excretion in Exhaled Human Breath 3-27
3.5.3. Formaldehyde Excretion in Human Urine 3-31
3.6. MODELING THE TOXICOKINETICS OF FORMALDEHYDE AND DPX 3-32
3.6.1. Motivation 3-32
3.6.2. Species Differences in Anatomy: Consequences for Gas Transport and
Risk 3-34
3.6.3. Modeling Formaldehyde Uptake in Nasal Passages 3-40
3.6.3.1. Flux Bins 3-41
3.6.3.2. Flux Estimates 3-41
3.6.3.3. Mass Balance Errors 3-42
3.6.4. Modeling Formaldehyde Uptake in the Lower Respiratory Tract 3-42
3.6.5. Uncertainties in Formaldehyde Dosimetry Modeling 3-44
3.6.5.1. Verification of Predicted Flow Profiles 3-44
3.6.5.2. Level of Confidence in Formaldehyde Uptake Simulations 3-45
3.6.6. PBPK Modeling of DNA Protein Cross-Links (DPXs) Formed by
Formaldehyde 3-48
3.6.6.1. PBPK Models for DPXs 3-48
3.6.6.2. A PBPK Model for DPXs in the F344 Rat and Rhesus
Monkey that uses Local Tissue Dose of Formaldehyde 3-50
3.6.6.3. Uncertainties in Modeling the Rat and Rhesus DPX Data 3-51
3.6.7. Uncertainty in Prediction of Human DPX Concentrations 3-53
VOLUME II
4. HAZARD CHARACTERIZATION 4-1
4.1. HUMAN STUDIES 4-1
4.1.1. Noncancer Health Effects 4-1
4.1.1.1. Sensory Irritation (Eye, Nose, Throat Irritation) 4-1
4.1.1.2. Pulmonary Function 4-11
4.1.1.3. Asthma 4-19
4.1.1.4. Respiratory Tract Pathology 4-26
4.1.1.5. Immunologic Effects 4-30
4.1.1.6. Neurological/Behavioral 4-42
4.1.1.7. Developmental and Reproductive Toxicity 4-45
4.1.1.8. Oral Exposure Effects on the Gastrointestinal Tract 4-56
4.1.1.9. Summary: Noncarcinogenic Hazard in Humans 4-56
This document is a draft for review purposes only and does not constitute Agency policy.
IV-iv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
4.1.2. Cancer Health Effects 4-57
4.1.2.1. Respiratory Tract Cancer 4-57
4.1.2.2. Non-Respiratory Tract Cancer 4-84
4.1.2.3. Summary: Carcinogenic Hazard in Humans 4-107
4.2. ANIMAL STUDIES 4-109
4.2.1. Noncancer Health Effects 4-110
4.2.1.1. Reflex Bradypnea 4-110
4.2.1.2. Respiratory Tract Pathology 4-120
4.2.1.3. Gastrointestinal Tract and Systemic Toxicity 4-201
4.2.1.4. Immune Function 4-216
4.2.1.5. Hypersensitivity and Atopic Reactions 4-225
4.2.1.6. Neurological and Neurobehavioral Function 4-250
4.2.1.7. Reproductive and Developmental Toxicity 4-285
4.2.2. Carcinogenic Potential: Animal Bioassays 4-324
4.2.2.1. Respiratory Tract 4-324
4.2.2.2. Gastrointestinal Tract 4-326
4.2.2.3. Lymphohematopoietic Cancer 4-328
4.2.2.4. Summary 4-335
4.3. GENOTOXICITY 4-335
4.3.1. Formaldehyde-DNAReactions 4-335
4.3.1.1. DNA-Protein Cross-Links (DPXs) 4-336
4.3.1.2. DNA Adducts 4-341
4.3.1.3. DNA-DNA Cross-Links (DDXs) 4-343
4.3.1.4. Single Strand Breaks 4-344
4.3.1.5. Other Genetic Effects of Formaldehyde in Mammalian Cells 4-345
4.3.2. In Vitro Clastogenicity 4-345
4.3.3. In Vitro Mutagenicity 4-347
4.3.3.1. Mutagenicity in Bacterial Systems 4-347
4.3.3.2. Mutagenicity in Non-mammalian Cell Systems 4-353
4.3.3.3. Mutagenicity in Mammalian Cell Systems 4-353
4.3.4. In Vivo Mammalian Genotoxicity 4-360
4.3.4.1. Genotoxicity in Laboratory Animals 4-360
4.3.4.2. Genotoxicity in Humans 4-362
4.3.5. Summary of Genotoxicity 4-370
4.4. SYNTHESIS AND MAJOR EVALUATION OF NONCARCINOGENIC
EFFECTS 4-371
4.4.1. Sensory Irritation 4-376
4.4.2. Pulmonary Function 4-379
4.4.3. Hypersensitivity and Atopic Reactions 4-382
4.4.4. Upper Respiratory Tract Histopathology 4-383
4.4.5. Toxicogenomic and Molecular Data that May Inform MOAs 4-385
4.4.6. Noncancer Modes of Actions 4-387
4.4.7. Immunotoxicity 4-389
This document is a draft for review purposes only and does not constitute Agency policy.
IV-v DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
4.4.8. Effects on the Nervous System 4-390
4.4.8.1. Irritant Threshold Detection 4-390
4.4.8.2. Behavioral Effects 4-391
4.4.8.3. Neurochemistry, Neuropathology, and Mechanistic Studies 4-392
4.4.8.4. Summary 4-392
4.4.8.5. Data Gaps 4-393
4.4.9. Reproductive and Developmental Toxicity 4-393
4.4.9.1. Spontaneous Abortion and Fetal Death 4-393
4.4.9.2. Congenital Malformations 4-396
4.4.9.3. Low Birth Weight and Growth Retardation 4-396
4.4.9.4. Functional Development Outcomes (Developmental
Neurotoxicity) 4-397
4.4.9.5. Male Reproductive Toxicity 4-398
4.4.9.6. Female Reproductive Toxicity 4-399
4.4.9.7. Mode of Action 4-400
4.4.9.8. Data Gaps 4-402
4.5. SYNTHESIS AND EVALUATION OF CARCINOGENICITY 4-402
4.5.1. Cancers of the Respiratory Tract 4-402
4.5.2. Lymphohematopoietic Malignancies 4-408
4.5.2.1. Background 4-408
4.5.2.2. All LHP Malignancies 4-410
4.5.2.3. All Leukemia 4-414
4.5.2.4. Subtype Analysis 4-418
4.5.2.5. Myeloid Leukemia 4-419
4.5.2.6. Solid Tumors of Lymphoid Origin 4-421
4.5.2.7. Supporting Evidence from Animal Bio-Assays for
Formaldehyde-Induced Lymphohematopoietic Malignancies 4-423
4.5.3. Carcinogenic Mode(s) of Action 4-427
4.5.3.1. Mechanistic Data for Formaldehyde 4-428
4.5.3.2. Mode of Action Evaluation for Upper Respiratory Tract
Cancer (Nasopharyngeal Cancer, Sino-Nasal) 4-439
4.5.3.3. Mode(s) of Action for Lymphohematpoietic Malignancies 4-446
4.5.4. Hazard Characterization for Formaldlehyde Carcinogenicity 4-453
4.6. SUSCEPTIBLE POPULATIONS 4-454
4.6.1. Life Stages 4-454
4.6.1.1. Early Life Stages 4-455
4.6.1.2. Later Life Stages 4-459
4.6.1.3. Conclusions on Life-Stage Susceptibility 4-459
4.6.2. Health/Disease Status 4-460
4.6.3. Nutritional Status 4-461
4.6.4. Gender Differences 4-462
4.6.5. Genetic Differences 4-462
This document is a draft for review purposes only and does not constitute Agency policy.
IV-vi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
4.6.6. Co-Exposures 4-464
4.6.6.1. Cumulative Risk 4-464
4.6.6.2. Aggregate Exposure 4-465
4.6.7. Uncertainties of Database 4-465
4.6.7.1. Uncertainties of Exposure 4-465
4.6.7.2. Uncertainties of Effect 4-466
4.6.8. Summary of Potential Susceptibility 4-467
VOLUME III
5. QUANTITATIVE ASSESSMENT: INHALATION EXPOSURE 5-1
5.1. INHALATION REFERENCE CONCENTRATION (RfC) 5-2
5.1.1. Candidate Critical Effects by Health Effect Category 5-3
5.1.1.1. Sensory Irritation of the Eyes, Nose, and Throat 5-3
5.1.1.2. Upper Respiratory Tract Pathology 5-5
5.1.1.3. Pulmonary Function Effects 5-6
5.1.1.4. Asthma and Allergic Sensitization (Atopy) 5-10
5.1.1.5. Immune Function 5-16
5.1.1.6. Neurological and Behavioral Toxicity 5-17
5.1.1.7. Developmental and Reproductive Toxicity 5-25
5.1.2. Summary of Critical Effects and Candidate RfCs 5-33
5.1.2.1. Selection of Studies for Candidate RfC Derivation 5-33
5.1.2.2. Derivation of Candidate RfCs from Key Studies 5-40
5.1.2.3. Evaluation of the Study-Specific Candidate RfC 5-66
5.1.3. Database Uncertainties in the RfC Derivation 5-69
5.1.4. Uncertainties in the RfC Derivation 5-72
5.1.5. Previous Inhalation Assessment 5-74
5.2. QUANTITATIVE CANCER ASSESSMENT BASED ON THE NATIONAL
CANCER INSTITUTE COHORT STUDY 5-74
5.2.1. Choice of Epidemiology Study 5-75
5.2.2. Nasopharyngeal Cancer 5-76
5.2.2.1. Exposure-Response Modeling of the National Cancer
Institute Cohort 5-76
5.2.2.2. Prediction of Lifetime Extra Risk of Nasopharyngeal Cancer
Mortality 5-79
5.2.2.3. Prediction of Lifetime Extra Risk of Nasopharyngeal Cancer
Incidence 5-81
5.2.2.4. Sources of Uncertainty 5-83
5.2.3. Lymphohematopoietic Cancer 5-88
5.2.3.1. Exposure-Response Modeling of the National Cancer
Institute Cohort 5-88
This document is a draft for review purposes only and does not constitute Agency policy.
IV-vii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
5.2.3.2. Prediction of Lifetime Extra Risks for Hodgkin Lymphoma
and Leukemia Mortality 5-91
5.2.3.3. Prediction of Lifetime Extra Risks for Hodgkin Lymphoma
and Leukemia Incidence 5-93
5.2.3.4. Sources of Uncertainty 5-95
5.2.4. Conclusions on Cancer Unit Risk Estimates Based on Human Data 5-99
5.3. DOSE-RESPONSE MODELING OF RISK OF SQUAMOUS CELL
CARCINOMA IN THE RESPIRATORY TRACT USING ANIMAL DATA 5-102
5.3.1. Long-Term Bioassays in Laboratory Animals 5-104
5.3.1.1. Nasal Tumor Incidence Data 5-104
5.3.1.2. Mechanistic Data 5-105
5.3.2. The CUT Biologically Based Dose-Response Modeling 5-106
5.3.2.1. Maj or Results of the CUT Modeling Effort 5-111
5.3.3. This Assessment's Conclusions from Evaluation of Dose-Response
Models of DPX Cell-Replication and Genomics Data, and of BBDR
Models for Risk Estimation 5-111
5.3.4. Benchmark Dose Approaches to Rat Nasal Tumor Data 5-118
5.3.4.1. Benchmark Dose Derived from BBDR Rat Model and Flux
as Dosimeter 5-118
5.3.4.2. Comparison with Other Benchmark Dose Modeling Efforts 5-125
5.3.4.3. Kaplan-Meier Adjustment 5-128
5.3.4.4. EPA Time-to-Tumor Statistical Modeling 5-129
5.4. CONCLUSIONS FROM THE QUANTITATIVE ASSESSMENT OF
CANCER RISK FROM FORMALDEHYDE EXPOSURE BY INHALATION .. 5-133
5.4.1. Inhalation Unit Risk Estimates Based on Human Data 5-133
5.4.2. Inhalation Unit Risk Estimates Based on Rodent Data 5-133
5.4.3. Summary of Inhalation Unit Risk Estimates 5-135
5.4.4. Application of Age-Dependent Adjustment Factors (ADAFs) 5-136
5.4.5. Conclusions: Cancer Inhalation Unit Risk Estimates 5-137
6. MAJOR CONCLUSIONS IN THE CHARACTERIZATION OF HAZARD AND
DOSE-RESPONSE 6-1
6.1. SUMMARY OF HUMAN HAZARD POTENTIAL 6-1
6.1.1. Exposure 6-1
6.1.2. Absorption, Distribution, Metabolism, and Excretion 6-1
6.1.3. Noncancer Health Effects in Humans and Laboratory Animals 6-4
6.1.3.1. Sensory Irritation 6-4
6.1.3.2. Respiratory Tract Pathology 6-5
6.1.3.3. Effects on Pulmonary Function 6-8
6.1.3.4. Asthmatic Responses and Increased Atopic Symptoms 6-9
6.1.3.5. Effects on the Immune System 6-10
6.1.3.6. Neurological Effects 6-11
6.1.3.7. Reproductive and Developmental Effects 6-12
This document is a draft for review purposes only and does not constitute Agency policy.
IV-viii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
6.1.3.8. Effects on General Systemic Toxicity 6-13
6.1.3.9. Summary 6-14
6.1.4. Carcinogenicity in Human and Laboratory Animals 6-14
6.1.4.1. Carcinogenicity in Humans 6-14
6.1.4.2. Carcinogenicity in Laboratory Animals 6-20
6.1.4.3. Carcinogenic Mode(s) of Action 6-21
6.1.5. Cancer Hazard Characterization 6-24
6.2. DOSE-RESPONSE CHARACTERIZATION 6-25
6.2.1. Noncancer Toxicity: Reference Concentration (RfC) 6-25
6.2.1.1. Assessment Approach Employed 6-25
6.2.1.2. Derivation of Candidate Reference Concentrations 6-25
6.2.1.3. Adequacy of Overall Data Base for RfC Derivation 6-26
6.2.1.4. Uncertainties in the Reference Concentration (RfC) 6-29
6.2.1.5. Conclusions 6-32
6.2.2. Cancer Risk Estimates 6-32
6.2.2.1. Choice of Data 6-32
6.2.2.2. Analysis of Epidemiologic Data 6-33
6.2.2.3. Analysis of Laboratory Animal Data 6-36
6.2.2.4. Extrapolation Aporoaches 6-37
6.2.2.5. Inhalation Unit Risk Estimates for Cancer 6-41
6.2.2.6. Early-Life Susceptibility 6-41
6.2.2.7. Uncertainties in the Quantitative Risk Estimates 6-42
6.2.2.8. Conclusions 6-45
6.3. SUMMARY AND CONCLUSIONS 6-45
REFERENCES R-l
VOLUME IV
APPENDIX A: SUMMARY OF EXTERNAL PEER REVIEW AND PUBLIC
COMMENTS AND DISPOSITIONS A-l
APPENDIX B: SIMULATIONS OF INTERINDIVIDUAL AND ADULT-TO-CHILD
VARIABILITY IN REACTIVE GAS UPTAKE IN A SMALL SAMPLE
OF PEOPLE (Garcia et aL 2009) B-l
APPENDIX C: LIFETABLE ANALYSIS C-l
APPENDIX D: MODEL STRUCTURE & CALIBRATION IN CONOLLY ET AL.
(2003, 2004) D-l
APPENDIX E: EVALUATION OF BBDR MODELING OF NASAL CANCER IN THE
F344 RAT: CONOLLY ET AL. (2003) AND ALTERNATIVE
IMPLEMENTATIONS E-1
APPENDIX F: SENSITIVITY ANALYSIS OF BBDR MODEL FOR FORMALDEHYDE
INDUCED RESPIRATORY CANCER IN HUMANS F-l
This document is a draft for review purposes only and does not constitute Agency policy.
IV-ix DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
APPENDIX G: EVALUATION OF THE CANCER DOSE-RESPONSE MODELING
OF GENOMIC DATA FOR FORMALDEHYDE RISK ASSESSMENT G-l
APPENDIX H: EXPERT PANEL CONSULTATION ON QUANTITATIVE
EVALUATION OF ANIMAL TOXICOLOGY DATA FOR
ANALYZING CANCER RISK DUE TO INHALED FORMALDEHYDE .. H-l
This document is a draft for review purposes only and does not constitute Agency policy.
IV-x DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES
Table 2-1. Physicochemical properties of formaldehyde 2-2
Table 2-2. Ambient air levels by land use category 2-6
Table 2-3. Studies on residential indoor air levels of formaldehyde (non-occupational) 2-8
Table 3-1. Endogenous formaldehyde levels in animal and human tissues and body fluids 3-8
Table 3-2. Formaldehyde kinetics in human and rat tissue samples 3-21
Table 3-3. Allelic frequencies of ADH3 in human populations 3-25
Table 3-4. Percent distribution of airborne [14C]-formaldehyde in F344 rats 3-26
Table 3-5. Apparent formaldehyde levels in exhaled breath of individuals attending a
health fair, adjusted for methanol and ethanol levels which contribute to the
detection of the protonated species with a mass to charge ratio of 31 reported
as formaldehyde (m/z = 31) 3-29
Table 3-6. Measurements of exhaled formaldehyde concentrations in the mouth and nose,
and in the oral cavity after breath holding in three healthy male laboratory
workers 3-30
Table 3-7. Extrapolation of parameters for enzymatic metabolism to the human 3-53
Table 4-1. Cohort and case-control studies of formaldehyde cancer and NPC 4-59
Table 4-2. Case-control studies of formaldehyde and nasal and paranasal cancer 4-71
Table 4-3. Epidemiologic studies of formaldehyde and pharyngeal cancer (includes
nasopharyngeal cancer) 4-78
Table 4-4. Epidemiologic studies of formaldehyde and lymphohematopoietic cancers 4-98
Table 4-5. Respiratory effects of formaldehyde-induced reflex bradypnea in various
strains of mice 4-112
Table 4-6. Respiratory effects of formaldehyde-induced reflex bradypnea in various
strains of rats 4-113
Table 4-7. Inhaled dose of formaldehyde to nasal mucosa of F344 rats and B6C3F1
mice exposed to 15 ppm 4-116
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xi DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
Table 4-8. Exposure regimen for cross-tolerance study 4-117
Table 4-9. Summary of formaldehyde effects on mucociliary function in the upper
respiratory tract 4-127
Table 4-10. Concentration regimens for ultrastructural evaluation of male CDF rat
nasoturbinates 4-129
Table 4-11. Enzymatic activities in nasal respiratory epithelium of male Wistar rats
exposed to formaldehyde, ozone, or both 4-130
Table 4-12. Lipid analysis of lung tissue and lung gavage from male F344 rats exposed
to 0, 15, or 145.6 ppm formaldehyde for 6 hours 4-138
Table 4-13. Formaldehyde effects on biochemical parameters in nasal mucosa and lung
tissue homogenates from male F344 rats exposed to 0, 15, or 145.6 ppm
formaldehyde for 6 hours 4-139
Table 4-14. Mast cell degranulation and neutrophil infiltration in the lung of rats
exposed to formaldehyde via inhalation 4-140
Table 4-15. Summary of respiratory tract pathology from inhalation exposures to
formaldehyde—short term studies 4-143
Table 4-16. Location and incidence of respiratory tract lesions in B6C3F1 mice
exposed to formaldehyde 4-146
Table 4-17. Formaldehyde effects (incidence and severity) on histopathologic changes in
the noses and larynxes of male and female albino SPF Wistar rats exposed to
formaldehyde 6 hours/day for 13 weeks 4-148
Table 4-18. Formaldehyde-induced nonneoplastic histopathologic changes in male
albino SPF Wistar rats exposed to 0, 10, or 20 ppm formaldehyde
and examined at the end of 130 weeks inclusive of exposure 4-149
Table 4-19. Formaldehyde-induced nasal tumors in male albino SPF Wistar rats
exposed to formaldehyde (6 hours/day, 5 days/week for 13 weeks) and
examined at the end of 130 weeks inclusive of exposure 4-150
Table 4-20. Formaldehyde effects on nasal epithelium for various concentration-by-
time products in male albino Wistar rats 4-153
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
Table 4-21. Rhinitis observed in formaldehyde-treated animals; data pooled for male
and female animals 4-154
Table 4-22. Epithelial lesions found in the middle region of nasoturbinates of
formaldehyde-treated and control animals; data pooled for males and
females 4-155
Table 4-23. Cellular and molecular changes in nasal tissues of F344 rats exposed to
formaldehyde 4-156
Table 4-24. Percent body weight gain and concentrations of iron, zinc, and copper in
cerebral cortex of male Wistar rats exposed to formaldehyde via inhalation
for 4 and 13 weeks 4-158
Table 4-25. Zinc, copper, and iron content of lung tissue from formaldehyde-treated
male Wistar rats 4-158
Table 4-26. Total lung cytochrome P450 measurements of control and formaldehyde-
treated male Sprague-Dawley rats 4-159
Table 4-27. Cytochrome P450 levels in formaldehyde-treated rats 4-160
Table 4-28. Summary of respiratory tract pathology from inhalation exposures to
formaldehyde, subchronic studies 4-162
Table 4-29. Histopathologic findings and severity scores in the naso- and
maxilloturbinates of female Sprague-Dawley rats exposed to inhaled
formaldehyde and wood dust for 104 weeks 4-166
Table 4-30. Histopathologic changes (including tumors) in nasal cavities of male
Sprague-Dawley rats exposed to inhaled formaldehyde or HC1 alone and
in combination for a lifetime 4-170
Table 4-31. Summary of neoplastic lesions in the nasal cavity of f344 rats exposed to
inhaled formaldehyde for 2 years 4-173
Table 4-32. Apparent sites of origin for the SCCs in the nasal cavity of F344 rats
exposed to 14.3 ppm of formaldehyde gas in the Kerns et al. (1983)
bioassay 4-174
Table 4-33. Incidence and location of nasal squamous cell carcinoma in male F344
rats exposed to inhaled formaldehyde for 2 years 4-175
Table 4-34. Summary of respiratory tract pathology from chronic inhalation exposures
to formaldehyde 4-183
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xiii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
Table 4-35. Cell proliferation in nasal mucosa, trachea, and free lung cells isolated
from male Wistar rats after inhalation exposures to formaldehyde 4-194
Table 4-36. The effect of repeated formaldehyde inhalation exposures for 3 months on
cell count, basal membrane length, proliferation cells, and two measures of
cell proliferation, LI and ULLI, in male F344 rats 4-196
Table 4-37. Formaldehyde-induced changes in cell proliferation and (ULLI) in the nasal
passages of male F344 rats exposed 6 hours/day 4-198
Table 4-38. Cell population and surface area estimates in untreated male F344 rats and
regional site location of squamous cell carcinomas in formaldehyde
exposed rats for correlation to cell proliferation rates 4-199
Table 4-39. Summary of formaldehyde effects on cell proliferation in the upper
respiratory tract 4-202
Table 4-40. Summary of lesions observed in the gastrointestinal tracts of Wistar rats
after drinking-water exposure to formaldehyde for 4 weeks 4-206
Table 4-41. Incidence of lesions observed in the gastrointestinal tracts of Wistar rats
after drinking-water exposure to formaldehyde for 2 years 4-209
Table 4-42. Effect of formaldehyde on gastroduodenal carcinogenesis initiated by
MNNG and NaCl in male Wistar rats exposed to formaldehyde (0.5%
formalin) in drinking water for 8 weeks 4-212
Table 4-43. Summary of benign and malignant gastrointestinal tract neoplasia
reported in male and female Sprague-Dawley rats exposed to
formaldehyde in drinking water at different ages 4-214
Table 4-44. Incidence of hemolymphoreticular neoplasia reported in male and
female Sprague-Dawley rats exposed to formaldehyde in drinking water
from 7 weeks old for up to 2 years (experiment BT 7001) 4-215
Table 4-45. Battery of immune parameters and functional tests assessed in female
B6C3F1 mice after a 3 week, 15-ppm formaldehyde exposure 4-218
Table 4-46. Summary of the effects of formaldehyde inhalation on the mononuclear
phagocyte system (MPS) in female B6C3F1 mice after a 3-week, 15 ppm
formaldehyde exposure ( 6 hours/day, 5 days/week) 4-219
Table 4-47. Formaldehyde exposure regimens for determining the effects of
formaldehyde exposure on pulmonary S. aureus infection 4-221
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xiv DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
Table 4-48. Summary of immune function changes due to inhaled formaldehyde
exposure in experimental animals 4-226
Table 4-49. Study design for guinea pigs exposed to formaldehyde through different
routes of exposure: inhalation, dermal, and injection 4-232
Table 4-50. Sensitization response of guinea pigs exposed to formaldehyde through
inhalation, topical application, or footpad injection 4-233
Table 4-51. Cytokine and chemokine levels in lung tissue homogenate supernatants in
formaldehyde-exposed male ICR mice with and without Der f sensitization 4-240
Table 4-52. Correlation coefficients among ear swelling responses and skin mRNA
levels in contact hypersensitivity to formaldehyde in mice 4-249
Table 4-53. Summary of sensitization and atopy studies by inhalation or dermal
sensitization due to formaldehyde in experimental animals 4-251
Table 4-54. Fluctuation of behavioral responses when male AB mice inhaled
formaldehyde in a single 2-hour exposure: effects after 2 hours 4-259
Table 4-55. Fluctuation of behavioral responses when male AB mice inhaled
formaldehyde in a single 2-hour exposure: effects after 24 hours 4-259
Table 4-56. Effects of formaldehyde exposure on completion of the labyrinth test by
male and female LEW. IK rats 4-263
Table 4-57. Summary of neurological and neurobehavioral studies in inhaled
formaldehyde in experimental animals 4-279
Table 4-58. Effects of formaldehyde on body and organ weights in rat pups from
dams exposed via inhalation from mating through gestation 4-289
Table 4-59. Formaldehyde effects on Leydig cell quantity and nuclear damage in adult
male Wistar rats 4-298
Table 4-60. Formaldehyde effects on adult male albino Wistar rats 4-299
Table 4-61. Formaldehyde effects on testosterone levels and seminiferous tubule
diameters in Wistar rats following 91 days of exposure 4-300
Table 4-62. Effects of formaldehyde exposure on seminiferous tubule diameter and
epithelial height in Wistar rats following 18 weeks of exposure 4-302
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xv DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
Table 4-63. Incidence of sperm abnormalities and dominant lethal effects in
formaldehyde-treated mice 4-302
Table 4-64. Body weights of pups born to beagles exposed to formaldehyde during
gestation 4-303
Table 4-65. Testicular weights, sperm head counts, and percentage incidence of
abnormal sperm after oral administration of formaldehyde to male
Wistar rats 4-305
Table 4-66. Effect of formaldehyde on spermatogenic parameters in male Wistar rats
exposed intraperitoneally 4-306
Table 4-67. Incidence of sperm head abnormalities in formaldehyde-treated rats 4-307
Table 4-68. Dominant lethal mutations after exposure of male rats to formaldehyde 4-308
Table 4-69. Summary of reported developmental effects in formaldehyde inhalation
exposure studies 4-311
Table 4-70. Summary of reported developmental effects in formaldehyde oral exposure
studies 4-317
Table 4-71. Summary of reported developmental effects in formaldehyde dermal
exposure studies 4-318
Table 4-72. Summary of reported reproductive effects in formaldehyde inhalation
studies 4-319
Table 4-73. Summary of reported reproductive effects in formaldehyde oral studies 4-322
Table 4-74. Summary of reported reproductive effects in formaldehyde intraperitoneal
studies 4-323
Table 4-75. Summary of chronic bioassays which address rodent leukemia and
lymphoma 4-329
Table 4-76. Formaldehyde-DNA reactions (DPX formation) 4-340
Table 4-77. Formaldehyde-DNA reactions (DNA adduct formation) 4-343
Table 4-78. Formaldehyde-DNA interactions (single stranded breaks) 4-344
Table 4-79. Other genetic effects of formaldehyde in mammalian cells 4-346
Table 4-80. In vitro clastogenicity of formaldehyde 4-348
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xvi DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
Table 4-81 Summary of mutagenicity of formaldehyde in bacterial systems 4-350
Table 4-82. Mutagenicity in mammalian cell systems 4-355
Table 4-83. Genotoxicity in laboratory animals 4-361
Table 4-84. MN frequencies in buccal mucosa cells of volunteers exposed to
formaldehyde 4-364
Table 4-85. MN and SCE formation in mortuary science students exposed to
formaldehyde for 85 days 4-364
Table 4-86. Incidence of MN formation in mortuary students exposed to formaldehyde
for 90 days 4-365
Table 4-87. Multivariate regression models linking genomic instability/occupational
exposures to selected endpoints from the MN assay 4-369
Table 4-88. Genotoxicity measures in pathological anatomy laboratory workers and
controls 4-370
Table 4-89. Summary of human cytogenetic studies 4-372
Table 4-90. Summary of cohort and case-control studies which evaluated the
incidence of all LHP cancers in formaldehyde-exposed populations
(ICD-8 Codes: 200-209) and all leukemias (ICD-8 Codes: 204-207) 4-412
Table 4-91. Secondary analysis of published mortality statistics to explore alternative
disease groupings within the broad category of all lymphohematopoetic
malignancies 4-419
Table 4-92. Summary of studies which provide mortality statistics for myeloid
leukemia sub-types 4-420
Table 4-93. Summary of mortality statistics for Hodgkin's lymphoma, lymphoma and
multiple myeloma from cohort analyses of formaldehyde exposed workers 4-422
Table 4-94. Summary of chronic bioassays which address rodent leukemia and
lymphoma 4-424
Table 4-95. Incidence of lymphoblastic leukemia and lymphosarcoma orally dosed in
Sprague-Dawley rats 4-425
Table 4-96. Available evidence for susceptibility factors of concern for formaldehyde
exposure 4-469
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xvii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
Table 5-1. Points of departure (POD) for nervous system toxicity in key human and
animal studies 5-19
Table 5-2. Effects of formaldehyde exposure on completion of the labyrinth test by
male and female LEW. IK rats 5-23
Table 5-3. Developmental and reproductive toxicity PODs including duration
adjustments - key human study 5-31
Table 5-4. Summary of candidate studies for formaldehyde RfC development by
health endpoint category 5-36
Table 5-5. Adjustment for nonoccupational exposures to formaldehyde 5-64
Table 5-6. Summary of reference concentration (RfC) derivation from critical study and
supporting studies 5-68
Table 5-7. Relative risk estimates for mortality from nasopharyngeal malignancies
(ICD-8 code 147) by level of formaldehyde exposure for different
exposure metrics 5-78
Table 5-8. Regression coefficients from NCI log-linear trend test models for NPC
mortality from cumulative exposure to formaldehyde 5-79
Table 5-9. Extra risk estimates for NPC mortality from various levels of continuous
exposure to formaldehyde 5-80
Table 5-10. ECooos, LECooos, and inhalation unit risk estimates for NPC mortality from
formaldehyde exposure based on the Hauptmann et al. (2004) log-linear
trend analyses for cumulative exposure 5-81
Table 5-11. ECooos, LECooos, and inhalation unit risk estimates for NPC incidence from
formaldehyde exposure based on the Hauptmann et al. (2004) trend
analyses for cumulative exposure 5-82
Table 5-12. Relative risk estimates for mortality from Hodgkin lymphoma
(ICD-8 code 201) and leukemia (ICD-8 codes 204-207) by level of
formaldehyde exposure for different exposure metrics 5-90
Table 5-13. Regression coefficients for Hodgkin lymphoma and leukemia mortality
from NCI trend test models 5-90
Table 5-14. Extra risk estimates for Hodgkin lymphoma mortality from various levels
of continuous exposure to formaldehyde 5-91
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xviii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
Table 5-15. Extra risk estimates for leukemia mortality from various levels of
continuous exposure to formaldehyde 5-91
Table 5-16. ECooos, LECooos, and inhalation unit risk estimates for Hodgkin lymphoma
mortality from formaldehyde exposure based on Beane Freeman et al.
(2009) log-linear trend analyses for cumulative exposure 5-93
Table 5-17. ECoos, LECoos, and inhalation unit risk estimates for leukemia mortality
from formaldehyde exposure based on Beane Freeman et al. (2009)
log-linear trend analyses for cumulative exposure 5-93
Table 5-18. ECooos, LECooos, and inhalation unit risk estimates for Hodgkin
lymphoma incidence from formaldehyde exposure, based on Beane
Freeman et al. (2009) log-linear trend analyses for cumulative exposure 5-94
Table 5-19. ECoos, LECoos, and inhalation unit risk estimates for leukemia incidence
from formaldehyde exposure based on Beane Freeman et al. (2009)
log-linear trend analyses for cumulative exposure 5-94
Table 5-20. Calculation of combined cancer mortality unit risk estimate at 0.1 ppm 5-100
Table 5-21. Calculation of combined cancer incidence unit risk estimate at 0.1 ppm 5-100
Table 5-22. Summary of tumor incidence in long-term bioassays on F344 rats 5-105
Table 5-23. BMD modeling of unit risk of SCC in the human respiratory tract 5-125
Table 5-24. Formaldehyde-induced rat tumor incidences 5-128
Table 5-25. Human benchmark extrapolations of nasal tumors in rats using
formaldehyde flux and DPX 5-135
Table 5-26. Summary of inhalation unit risk estimates 5-136
Table 5-27. Total cancer risk from exposure to a constant formaldehyde exposure
level of 1 |ig/m3 from ages 0-70 years 5-137
Table 6-1. Summary of candidate Reference Concentrations (RfC) for co-critical studies.... 6-27
Table 6-2. Effective concentrations (lifetime continuous exposure levels) predicted
for specified extra cancer risk levels for selected formaldehyde-related
cancers 6-36
Table 6-3. Inhalation unit risk estimates based on epidemiological and experimental
animal data 6-42
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xix DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES
Figure 2-1. Chemical structure of formaldehyde 2-1
Figure 2-2. Locations of hazardous air pollutant monitors 2-5
Figure 2-3. Modeled ambient air concentrations based on 1999 emissions 2-7
Figure 2-4. Range of formaldehyde air concentrations (ppb) in different environments 2-9
Figure 3-1. Formaldehyde-mediated protein modifications 3-2
Figure 3-2. 3H/14C ratios in macromolecular extracts from rat tissues following exposure
to 14C and 3H-labeled formaldehyde (0.3, 2, 6, 10, 15 ppm) 3-18
Figure 3-3. Formaldehyde clearance by ALDH2 (GSH-independent) and ADH3
(GSH-dependent) 3-20
Figure 3-4. Metabolism of formate 3-22
Figure 3-5. Scatter plot of formaldehyde concentrations measured in ppb in direct breath
exhalations (x axis) and exhaled breath condensate headspace (y axis) 3-31
Figure 3-6. Reconstructed nasal passages of F344 rat, rhesus monkey, and human 3-36
Figure 3-7. Illustration of interspecies differences in airflow and verification of CFD
simulations with water-dye studies 3-37
Figure 3-8. Lateral view of nasal wall mass flux of inhaled formaldehyde simulated in
the F344 rat, rhesus monkey, and human 3-38
Figure 3-9. CFD simulations of formaldehyde flux to human nasal lining at different
inspiratory flow rates 3-39
Figure 3-10. Single-path model simulations of surface flux per ppm of formaldehyde
exposure concentration in an adult male human 3-43
Figure 3-11. Pressure drop vs. volumetric airflow rate predicted by the CUT CFD
model compared with pressure drop measurements made in two hollow
molds (CI and C2) of the rat nasal passage (Cheng et al., 1990) or in rats
in vivo 3-45
Figure 3-12. Formaldehyde-DPX dosimetry in the F344 rat 3-47
Figure 4-1. Delayed asthmatic reaction following the inhalation of formaldehyde after
"painting" 100% formalin for 20 minutes 4-20
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xx DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES (continued)
Figure 4-2. Formaldehyde effects on minute volume in naive and formaldehyde-
pretreated male B6C3F1 mice and F344 rats 4-115
Figure 4-3. Sagittal view of the rat nose (nares oriented to the left) 4-121
Figure 4-4. Main components of the nasal respiratory epithelium 4-122
Figure 4-5. Decreased mucus clearance and ciliary beat in isolated frog palates
exposed to formaldehyde after 3 days in culture 4-126
Figure 4-6. Diagram of nasal passages showing section levels chosen for morphometry
and autoradiography in male rhesus monkeys exposed to formaldehyde 4-135
Figure 4-7. Formaldehyde-induced cell proliferation in male rhesus monkeys exposed to
formaldehyde 4-136
Figure 4-8. Formaldehyde-induced lesions in male rhesus monkeys exposed to formaldehyde
4-137
Figure 4-9. Frequency and location by cross-section level of squamous metaplasia in
the nasal cavity of F344 rats exposed to formaldehyde via inhalation 4-172
Figure 4-10. Effect of formaldehyde exposure on cell proliferation of the respiratory
mucosa of rats and mice 4-190
Figure 4-11. Alveolar MP Fc-mediated phagocytosis from mice exposed to 5 ppm
formaldehyde, 10 mg/m3 carbon black, or both 4-223
Figure 4-12. Compressed air in milliliters as parameter for airway obstruction
following formaldehyde exposure in guinea pigs after OVA sensitization and
OVA challenge 4-235
Figure 4-13. OVA-specific IgGl (IB) in formaldehyde-treated sensitized guinea pigs
prior to OVA challenge 4-235
Figure 4-14. Anti-OVA titers in female Balb/C mice exposed to 6.63 ppm
formaldehyde for 10 consecutive days, or once a week for 7 weeks 4-236
Figure 4-15. Vascular permeability in the tracheae and bronchi of male Wistar rats
after 10 minutes of formaldehyde inhalation 4-238
Figure 4-16. Effect of select receptor antagonists on formaldehyde-induced vascular
permeability in the trachea and bronchi of male Wistar rats 4-239
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxi DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES (continued)
Figure 4-17. The effects of formaldehyde inhalation exposures on eosinophil
infiltration (Panel A) and goblet cell proliferation (Panel B) after Der f
challenge in the nasal mucosa of male ICR mice after sensitization and
challenge 4-241
Figure 4-18. NGF in BAL fluid from formaldehyde-exposed female C3H/He mice
with and without OA sensitization 4-243
Figure 4-19. Plasma Substance P levels in formaldehyde-exposed female C3H/He
mice with and without OVA sensitization 4-244
Figure 4-20. Motor activity in male and female rats 2 hours after exposure to
formaldehyde expressed as mean number of crossed quadrants ± SEM 4-256
Figure 4-21. Habituation of motor activity was observed in control rats during the
second observation period (day 2, 24 hours after formaldehyde exposure) 4-257
Figure 4-22. Motor activity was reduced in male and female LEW. IK rats 2 hours
after termination of 10-minute formaldehyde exposure 4-258
Figure 4-23. The effects of the acute formaldehyde (FA) exposures on the
ambulatory and vertical components of SLMA 4-260
Figure 4-24. Effects of formaldehyde exposure on the error rate of female LEW. IK
rats performing the water labyrinth learning test 4-264
Figure 4-25. Basal and stress-induced trunk blood corticosterone levels in male
LEW. IK rats after formaldehyde inhalation exposures 4-269
Figure 4-26. NGF production in the brains of formaldehyde-exposed mice 4-274
Figure 4-27. Mortality corrected cumulative incidences of nasal carcinomas in the
indicated exposure groups 4-325
Figure 4-28. Leukemia incidence in Sprague-Dawley rats exposed to formaldehyde
in drinking water for 2 years 4-330
Figure 4-29. Unscheduled deaths in female F344 rats exposed to formaldehyde for
24 months 4-332
Figure 4-30. Cumulative leukemia incidence in female F344 rats exposed to
formaldehyde for 24 months 4-333
Figure 4-31. Cumulative incidence or tumor bearing animals for lymphoma in
female mice exposed to formaldehyde for 24 months 4-334
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES (continued)
Figure 4-32. DNA-protein cross-links (DPX) and thymidine kinase (tk) mutants in
TK6 human lymphoblasts exposed to formaldehyde for 2 hours 4-357
Figure 4-33. Developmental origins for cancers of the lymphohematopoietic system 4-409
Figure 4-34A. Association between peak formaldehyde exposure and the risk of
lymphohematopoietic malignancy 4-415
Figure 4-34B. Association between average intensity of formaldehyde exposure and
the risk of lymphohematopoietic malignancy 4-416
Figure 4-35. Effect of various doses of formaldehyde on cell number in (A) HT-29
human colon carcinoma cells and in (B) human umbilical vein epithelial cells
(HUVEC) 4-433
Figure 4-36. Integrated MO A scheme for respiratory tract tumors 4-446
Figure 4-37. Location of intra-epithelial lymphocytes along side epithelial cells in
the human adenoid 4-450
Figure 5-1. Change in number of additions made in 10 minutes following formaldehyde
exposure at 32, 170, 390, or 890 ppb 5-21
Figure 5-2. Effects of formaldehyde exposure on the error rate of female LEW. IK rats
performing the water labyrinth learning test 5-24
Figure 5-3. Fecundity density ratio among women exposed to formaldehyde in the high
exposure index category with 8-hour time weighted average formaldehyde
exposure concentration of 219 ppb 5-27
Figure 5-4. Estimated reduction in peak expiratory flow rate (PEFR) in children in
relation to indoor residential formaldehyde concentrations 5-41
Figure 5.5. Odds ratios for physician-diagnosed asthma in children associated with in-
home formaldehyde levels in air 5-45
Figure 5-6. Prevalence of asthma and respiratory symptom scores in children
associated with in-home formaldehyde levels 5-48
Figure 5-7. Prevalence and severity of allergic sensitization in children associated
with in-home formaldehyde levels 5-49
Figure 5-8. Positive exposure-response relationships reported for in-home
formaldehyde exposures and sensory irritation (eye irritation) 5-53
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxiii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES (continued)
Figure 5-9. Positive exposure-response relationships reported for in-home
formaldehyde exposures and sensory irritation (burning eyes) 5-54
Figure 5-10. Age-specific mortality and incidence rates for myeloid, lymphoid, and
all leukemia 5-98
Figure 5-11. Schematic of integration of pharmacokinetic and pharmacodynamic
components in the CUT model 5-110
Figure 5-12. Fit to the rat tumor incidence data using the model and assumptions in
Conolly etal. (2003) 5-112
Figure 5-13. Spatial distribution of formaldehyde over the nasal lining, as
characterized by partitioning the nasal surface by formaldehyde flux to
the tissue per ppm of exposure concentration, resulting in 20 flux bins 5-120
Figure 5-14. Distribution of cells at risk across flux bins in the F344 rat nasal lining 5-120
Figure 5-15. MLE and upper bound (UB) added risk of SCC in the human nose for
two BBDR models 5-124
Figure 5-16. Replot of log-probit fit of the combined Kerns et al. (1983) and
Monticello et al. (1996) data on tumor incidence showing BMCio and
BMCLio 5-127
Figure 5-17. EPA multistate Weibull modeling: nasal tumor dose response 5-131
Figure 5-18. Multistage Weibull model fit 5-132
Figure 5-19. Multistage Weibull model fit of tumor incidence data compared with
KM estimates of spontaneous tumor incidence 5-132
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxiv DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS
ACGIH
American Conference of Governmental Industrial Hygienists
ADAF
age-dependent adjustment factors
ADH
alcohol dehydrogenase
ADS
anterior dorsal septum
AIC
Akaike Information Criterion
AIE
average intensity of exposure
AIHA
American Industrial Hygiene Association
ALB
albumin
ALDH
aldehyde dehydrogenase
ALL
acute lymphocytic leukemia
ALM
anterior lateral meatus
ALP
alkaline phosphatase
ALS
amyotrophic lateral sclerosis
ALT
alanine aminotransferase
AML
acute myelogenous leukemia
AMM
anterior medial maxilloturbinate
AMPase
adenosine monophosphatase
AMS
anterior medial septum
ANAE
alpha-naphthylacetate esterase
ANOVA
analysis of variance
APA
American Psychiatric Association
ARB
Air Resources Board
AST
aspartate aminotransferase
ATCM
airborne toxic control measure
ATP
adenosine triphosphate
ATPase
adenosine triphosphatase
ATS
American Thoracic Society
AT SDR
Agency for Toxic Substances and Disease Registry
AUC
area under the curve
BAL
bronchoalveolar lavage
BALT
bronchus associated lymphoid tissue
BBDR
biologically based dose response
BC
bronchial construction
BCME
bis(chloromethyl)ether
BDNF
brain-derived neurotrophic factor
BEIR
biologic effects of ionizing radiation
BfR
German Federal Institute for Risk Assessment
BHR
bronchial hyperresponsiveness
BMC
benchmark concentration
BMCL
95% lower bound on the benchmark concentration
BMCR
binuclated micronucleated cell ratefluoresce
BMD
benchmark dose
BMDL
95% lower bound on the benchmark dose
This document is a draft for review purposes only and does not constitute Agency policy.
IV -xxv DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
BMR benchmark response
BN Brown-Norway
BrdU bromodeoxyuridine
BUN blood urea nitrogen
BW body weight
CA chromosomal aberrations
CalEPA California Environmental Protection Agency
CAP College of American Pathologists
CASRN Chemical Abstracts Service Registry Number
CAT catalase
CBMA cytokinesis-blocked micronucleus assay
CBMN cytokinesis-blocked micronucleus
CDC U.S. Centers for Disease Control and Prevention
CDHS California Department of Health Services
CFD computational fluid dynamics
CGM clonal growth model
CHO Chinese hamster ovary
CI confidence interval
CUT Chemical Industry Institute of Toxicology
CLL chronic lymphocytic leukemia
CML chronic myelogenous leukemia
CNS central nervous system
CO2 carbon dioxide
COEHHA California Office of Environmental Health Hazard Assessment
CREB cyclic AMP responsive element binding proteins
CS conditioned stimulus
C x t concentration times time
DA Daltons
DAF dosimetric adjustment factor
DDX DNA-DNA cross-links
DEI daily exposure index
DEN diethylnitrosamine
Der f common dust mite allergen
DMG dimethylglycine
DMGDH dimethylglycine dehydrogenase
DNA deoxyribonucleic acid
DOPAC 3,4-dihydroxyphenylacetic acid
DPC / DPX DNA-protein cross-links
EBV Epstein-Barr virus
EC effective concentration
ED effective dose
EHC Environmental Health Committee
ELISA enzyme-linked immunosorbent assay
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxvi DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
EPA
U.S. Environmental Protection Agency
ERPG
emergency response planning guideline
ET
ethmoid turbinates
FALDH
formaldehyde dehydrogenase
FDA
U.S. Food and Drug Administration
FDR
fecundability density ratio
FEF
forced expiratory flow
FEMA
Federal Emergency Management Agency
FEV1
forced expiratory volume in 1 second
FISH
fluorescent in situ hybridization
FSH
follicle-stimulating hormone
FVC
forced vital capacity
GALT
gut-associated lymphoid tissue
GC-MS
gas chromatography-mass spectrometry
GD
gestation day
GI
gastrointestinal
GO
gene ontology
G6PDH
glucose-6-phosphate dehydrogenase
GPX
glutathione peroxidase
GR
glutathione reductase
GM-CSF
granulocyte macrophage-colony-stimulating factor
GSH
reduced glutathione
GSNO
S-nitrosoglutathione
GST
glutathione S-transferase
HAP
hazardous air pollutant
Hb
hemoglobin
HC1
hydrochloric acid
HCT
hematocrit
HEC
human equivalent concentration
5-HI A A
5-hydroxyindoleacetic acid
hm
hydroxymethyl
HMGSH
S-hydroxymethylglutathione
HPA
hypothalamic-pituitary adrenal
HPG
hypothalamo-pituitary-gonadal
HPLC
high-performance liquid chromatography
HPRT
hypoxanthine-guanine phosphoribosyl transferase
HR
high responders
HSA
human serum albumin
HSDB
Hazardous Substances Data Bank
Hsp
heat shock protein
HWE
healthy worker effect
I cell
initiated cell
IARC
International Agency for Research on Cancer
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxvii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
ICD
International Classification of Diseases
IF
interfacial
IFN
interferon
Ig
immunoglobulin
IL
interleukin
LP.
intraperitoneal
IPCS
International Programme on Chemical Safety
IRIS
Integrated Risk Information System
Km
Michaels-Menton constant
KM
Kaplan-Meier
LD50
median lethal dose
LDH
lactate dehydrogenase
LEC
95% lower bound on the effective concentration
LED
95% lower bound on the effective dose
LHP
ly mphohematopoi eti c
LI
labeling index
LM
Listeria monocytogenes
LMS
linearized multistage
LLNA
local lymph node assay
LOAEL
lowest-observed-adverse-effect level
LPS
lipopolysaccharide
LR
low responders
LRT
lower respiratory tract
MA
methylamine
MALT
mucus-associated lymph tissues
MCH
mean corpuscular hemoglobin
MCHC
mean corpuscular hemoglobin concentration
MCS
multiple chemical sensitivity
MCV
mean corpuscular volume
MDA
malondialdehyde
MEF
maximal expiratory flow
ML
myeloid leukemia
MLE
maximum likelihood estimate
MMS
methyl methane sulfonate
MMT
medial maxilloturbinate
MN
micronucleus, micronuclei
MNNG
N-methyl -N-nitro-N-nitrosoguani dine
MOA
mode of action
MoDC
monocyte-derived dendritic cell
MP
macrophage
MPD
multistage polynomial degree
MPS
mononuclear phagocyte system
MRL
minimum risk level
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxviii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
mRNA messenger ribonucleic acid
MVE-2 Murray Valley encephalitis virus
MVK Moolgavkar, Venzon, and Knudson
N cell normal cell
NaCl sodium choride
NAD+ nicotinamide adenine dinucleotide
NADH reduced nicotinamide adenine dinucleotide
NALT nasally associated lymphoid tissue
NATA National-Scale Air Toxics Assessment
NCEA National Center for Environmental Assessment
NCHS National Center for Health Statistics
NCI National Cancer Institute
NEG Nordic Expert Group
NER nucleotide excision repair
NGF nerve growth factor
NHL non-Hodgkin's lymphoma
NHMRC/ARMCANZ National Health and Medical Research Council/Agriculture and Resource
Management Council of Australia and New Zealand
NNK nitrosamine nitrosamine 4-(methylnitrosamino)-l-(3-pyridyl)-butanone
N6-hmdA N6-hydroxymethyldeoxyadenosine
N4-hmdC N4-hydroxymethylcytidine
N2-hmdG N2-hydroxymethyldeoxyguanosine
NICNAS National Industrial Chemicals Notification and Assessment Scheme
NIOSH National Institute for Occupational Safety and Health
NLM National Library of Medicine
NMDA N-methyl-D-aspartate
NO nitric oxide
NOAEL no-ob served-adverse-effect level
NPC nasopharyngeal cancer
NRBA neutrophil respiratory burst activity
NRC National Research Council
NTP National Toxicology Program
OR odds ratio
OSHA Occupational Safety and Health Administration
OTS Office of Toxic Substances
OVA ovalbumin
PBPK physiologically based pharmacokinetic
PC Philadelphia chromosome
PCA passive cutaneous anaphylaxis
PCMR proportionate cancer mortality ratio
PCNA proliferating cell nuclear antigen
PCR polymerase chain reaction
PCV packed cell volume
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxix DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
PEC AM
platelet endothelial cell adhesion molecule
PEF
peak expiratory flow
PEFR
peak expiratory flow rates
PEL
permissible exposure limit
PFC
plaque-forming cell
PG
peri glomerular
PHA
phytohemagglutinin
PLA2
phospholipase A2
PI
phagocytic index
PLM
posterior lateral meatus
PMA
phorbol 12-myristate 13-acetate
PMR
proportionate mortality ratio
PMS
posterior medial septum
PND
postnatal day
POD
point of departure
POE
portal of entry
PTZ
pentilenetetrazole
PUFA
polyunsaturated fatty acids
PWULLI
population weighted unit length labeling index
RA
reflex apnea
RANTES
regulated upon activation, normal T-cell expressed and secreted
RB
reflex bradypnea
RBC
red blood cells
RD50
exposure concentration that results in a 50% reduction in respiratory rate
REL
recommended exposure limit
RfC
reference concentration
RfD
reference dose
RGD
regional gas dose
RGDR
regional gas dose ratio
RR
relative risk
RT
reverse transcriptase
SAB
Science Advisory Board
see
squamous cell carcinoma
SCE
sister chromatid exchange
SCG
sodium cromoglycate
SD
standard deviation
SDH
succinate dehydrogenase; sarcosine dehydrogenase
SEER
Surveillance, Epidemiology, and End Results
SEM
standard error of the mean
SEN
sensitizer
SH
sulfhydryl
SHE
Syrian hamster embryo
SLMA
spontaneous locomotor activity
This document is a draft for review purposes only and does not constitute Agency policy.
IV -xxx DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
SMR
standardized mortality ratio
SNP
single nucleotide polymorphism
SOD
superoxide dismutase
SOMedA
N6-sulfomethyldeoxy adenosine
Spl
specificity protein
SPIR
standardized proportionate incidence ratio
SSAO
semicarbozole-sensitive amine oxidase
SSB
single strand breaks
STEL
short-term exposure limit
TBA
tumor bearing animal
TH
T-lymphocyte helper
THF
tetrahydrofolate
TK
toxicokinetics
TL
tail length
TLV
threshold limit value
TNF
tumor necrosis factor
TP
total protein
TRI
Toxic Release Inventory
TRPV
transient receptor potential vanilloid
TWA
time-weighted average
TZCA
thiazolidine-4-carboxylate
UCL
upper confidence limit
UDS
unscheduled DNA synthesis
UF
uncertainty factor
UFFI
urea formaldehyde foam insulation
ULLI
unit length labeling index
URT
upper respiratory tract
USD A
U.S. Department of Agriculture
VC
vital capacity
VOC
volatile organic compound
WBC
white blood cell
WDS
wet dog shake
WHO
World Health Organization
WHOROE
World Health Organization Regional Office for Europe
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxxi DRAFT—DO NOT CITE OR QUOTE
-------
This page intentionally left blank.
This document is a draft for review purposes only and does not constitute Agency policy.
IV-xxxii DRAFT—DO NOT CITE OR QUOTE
-------
Appendix A
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
APPENDIX A
SUMMARY OF EXTERNAL PEER REVIEW
AND PUBLIC COMMENTS AND DISPOSITIONS
[NOTE: This is a placeholder for Appendix A which will be drafted
following External Peer review and receipt of public comments.]
This document is a draft for review purposes only and does not constitute Agency policy.
A-2 DRAFT—DO NOT CITE OR QUOTE
-------
Appendix B
i
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
APPENDIX B
SIMULATIONS OF INTERINDIVIDUAL AND ADULT-TO-CHILD VARIABILITY IN
REACTIVE GAS UPTAKE IN A SMALL SAMPLE OF PEOPLE (Garcia et al., 2009)
Garcia et al. (2009) used computational fluid dynamics to study human variability in the
nasal dosimetry of reactive, water-soluble gases in 5 adults and 2 children, aged 7 and 8 years
old. They considered two model categories of gases, corresponding to maximal and moderate
absorption at the nasal lining. The nasal airway (including the nasopharynx) geometries of these
individuals were mapped out using magnetic resonance imaging or computed tomography scans.
The scans chosen for the analysis were from individuals who had normal nasal anatomies with
no pathology (as per a review carried out by a ear-nose-throat surgeon). The minute volumes of
these individuals were estimated to range from 6.8 to 9.0 L/min (adults) and 5.5 to 5.8 L/min
(children). The sample size in this study is too small to consider the results representative of the
population as a whole (as also recognized by the authors). Nonetheless, various comparisons
with the characteristics of other study populations add to the strength of this study. The range of
adult minute volumes in this study is reported by the authors to be in good agreement with that
obtained in many other studies in the literature; minute volumes for the children in the study
were found to be similar to the average minute volume of 6.1 ± 1.7 L/min obtained by Bennett
and Zeman (2004) in a study of 36 children aged 6 to 13 years; the range of nasal surface area
values for the adults agreed well with that obtained by Guilmette et al. (1997) for 45 adults; and
the range of values for the surface area to volume ratio is in good agreement with that obtained
for 40 adult Caucasians studied by Yokley (2006). The surface area to volume ratio is useful for
comparing the rate of diffusional transport of a gas out of different cavities; however in the case
of the highly non-homogenously shaped nasal lumen, this might at best be considered a gross
indicator.
We focus here only on the "maximum uptake" simulations in Garcia et al. (2009). In this
case, the gas was considered so highly reactive and soluble that it was reasonable to assume an
infinitely fast reaction of the absorbed gas with compounds in the airway lining. Although such a
gas could be reasonably considered to represent formaldehyde, these results cannot be fully
utilized to inform quantitative estimates of formaldehyde dosimetry (and that does not appear to
have been the intent of the authors either). This is because the same boundary condition
corresponding to maximal uptake was applied on the vestibular section as well as on the
transitional and transitional epithelial lining of the nasal cavity. This is not appropriate for
formaldehyde as the lining on the nasal vestibule is made of keratinized epithelium which is
considerably less absorbing than the transitional or respiratory epithelium (Kimbell et al. 2001).
This document is a draft for review purposes only and does not constitute Agency policy.
B-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
Table B-l provides results obtained by Garcia et al. (2009) for five adults and two
children for uptake for the maximal uptake scenario. Although the nasal cavities of the children
were smaller in surface area, volume and length, the surface-area-to-volume ratios were similar
in the two age groups. Overall uptake efficiency, average flux (rate of gas absorbed per unit
surface area of the nasal lining) and maximum flux levels over the entire nasal lining did not
vary substantially between adults (1.6-fold difference in average flux and much less in maximum
flux), and the mean values of these quantities were comparable between adults and children.
These results are in agreement with conclusions reached by Ginsberg et al. (2005) that overall
extrathoracic absorption of highly and moderately reactive and soluble gases (corresponding to
category 1 and 2 reactive gases as per the scheme in USEPA [1994]) is similar in adults and
children. However, none of these three quantities should be considered as reasonable indicators
of variability in the interaction of the gas with the nasal lining. For a very reactive and soluble
gas, regional absorption of the gas is highly nonhomogenously distributed; therefore
interindividual variability in this distribution will be washed out when averaged over the whole
nose. Estimates of maximum gas flux, on the other hand, correspond to extremely small
localized regions of hot spots (see chapter 3), and thus may not be a proper measure of inter-
individual variability in flux distribution patterns over the whole nose. Furthermore, numerical
error in the calculation (such as mass balance and irregularly shaped elements of the finite-
element mesh) is likely to be most pronounced when estimates are considered over extremely
small regions.
Table B-l: Variations in overall nasal uptake, whole nose flux, and key parameters
% nasal uptake MV SA/V Avg flux Maximum flux
(L) (1/mm) 10"8 kg/(s.m2) 10"8 kg/(s.m2)
left cavity right cavity left cavity right cavity
adult 1
93.5
9
1.12
1.8
1.5
10.8
10.0
adult 1
92.4
6.8
1.09
1.5
1.5
10.8
10.4
adult 1
93.1
9
0.88
1.6
1.3
11
10.6
adult 1
89.2
7.1
0.87
1.2
1.2
10.6
10.2
adult 1
91.5
6.9
0.95
1.4
1.5
10.8
10.0
child 1
92
5.5
1.13
1.9
1.5
11.8
11.0
child2
88.2
5.8
0.95
1.6
1.5
12.3
11.6
MV = minute volume, SA=nasal surface area, V=nasal volume
Source: Garcia et al. (2009).
This document is a draft for review purposes only and does not constitute Agency policy.
B-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
On the other hand, Figure 6A of Garcia et al. (2009), reproduced here as Figure B-l,
shows significant interhuman variability in flux values at specific points on the nasal walls. The
local flux of formaldehyde varies among individuals by a factor of 3 to 5 at various distances
along the septal axis of the nose. However, interpretation of the values in this plot is problematic
for reasons explained in the paper 1:
The greater variability among individuals seen for wall fluxes at specific sites of the nasal
passages (Figure 6) in comparison to the minimal variability in total uptake
(Table 2) and whole-nose dose (Tables 3 and Tables 4) indicates that fluxes of equal
magnitude do not exactly overlay the same anatomical regions of the nasal cavity in each
individual. This implies that specific anatomical regions subtended by maximum flux
could be offset from one individual to another.
Notwithstanding this difficulty in interpretion, we believe the quantities plotted in Figure B-l
provide a better perspective of the inter-individual (adult) variability in local flux than the
variation in whole nose average or in maximum flux presented in Table B-l.
A 1.0E-07
Adults - maximum uptake
Children - maximum uptake
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
the adverse response, then variation in average whole nose flux or overall nasal uptake efficiency
would be most useful since one is then interested in inter-individual variability in the overall
dose delivered to the lung. It is possible to conceive of allergic or irritation responses being
triggered by some threshold value of local flux. In such a case it may be preferable to calculate
the variability associated with the net surface area receiving flux values greater than that
threshold. On the other hand, the probability of developing a tumor at a nasal site may be non-
linearly related to the flux at that site and linearly related to the number of cells at that site. In
this case, the appropriate metric may be the nasal surface area associated with some intermediate
levels of local flux (see Appendix in Subramaniam et al. 2008).
Various caveats presented by the authors as limitations of their study should be noted:
Possible nonuniform distribution of epithelial types, enzymes, glands and other cellular
metabolic or clearance machinery were not considered in the model; only effects pertaning to
resting breathing were considered; the study sample size was small; children younger than 7
years old were not studied; and, the model assumed a rigid nasal geometry.
Garcia et al. (2009) conclude their paper as follows:
our simulations predicted no differences in the nasal dosimetry of reactive, water-
soluble gases between children and adults, suggesting that the risk factor of 10 typically
used to accommodate inter-human variability is adequate."
In addition to the obvious caveat already recognized by the authors in regards extrapolating from
a study involving just two children, this conclusion needs further qualification. Firstly, the safety
factor of 10 that is typically applied for interhuman variability does not specifically include
children. Instead, EPA practice is that unless there is reasonable evidence that childhood forms a
more susceptible lifestage, no additional factors are applied for this population. Second, it is
important not to confuse dosimetric differences across lifestages-which actually contribute to
intra-individual differences-with inter-individual differences that may contribute towards the 10-
fold uncertainty factor. This interhuman variability in susceptibility-for a given life-stage-may
be considered to arise from both pharmacokinetic and pharmacodynamic factors, and the practice
in the US has been to split these into factors of 3.3 each. Then, the roughly 3 to 5 -fold variation
estimated for adults (and also between the two children) in Figure B-l suggest that a factor of 10
may not be adequate to accommodate inter-human variability for those formaldehyde-induced
adverse responses for which the localized nature of formaldehyde flux, and therefore the inter-
individual differences in regional dosimetry, play a role.
This document is a draft for review purposes only and does not constitute Agency policy.
B-5 DRAFT—DO NOT CITE OR QUOTE
-------
This page intentionally left blank.
This document is a draft for review purposes only and does not constitute Agency policy.
B-6 DRAFT—DO NOT CITE OR QUOTE
-------
Appendix C
i
-------
1 APPENDIX C
2
3 LIFETABLE ANALYSIS
4
5 A spreadsheet illustrating the extra-risk calculation for the derivation of the lower 95%
6 bound on the effective concentration associated with a 0.05% extra risk (LECooos) for
7 nasopharyngeal carcinoma (NPC) incidence is presented in Table C-l.
8
This document is a draft for review purposes only and does not constitute Agency policy.
C-2 DRAFT—DO NOT CITE OR QUOTE
-------
Table C-l. Extra-risk calculation3 for environmental exposure to 0.0461 ppm formaldehyde (the LECooos for
NPC incidence)b using a log-linear exposure-response model based on the cumulative exposure trend results of
Hauptmann et al. (2004), as described in Section 5.2.2.
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Interval
number
(i)
Age
interval
All cause
mortality
(xl05/yr)
NPC
incidence
(xl05/yr)
All
cause
hazard
rate (h*)
Prob of
surviving
interval
(q)
Prob of
surviving
up to
interval
(S)
NPC
cancer
hazard
rate (h)
Cond
prob of
NPC
incidence
in
interval
(Ro)
Exp
duration
mid
interval
(xtime)
Cum
exp mid
interval
(xdose)
Exposed
NPC
hazard
rate (hx)
Exposed
all cause
hazard
rate
(h*x)
Exposed
prob of
surviving
interval
(qx)
Exposed
prob of
surviving
up to
interval
(Sx)
Exposed
cond
prob of
NPC in
interval
(Rx)
1
<1
728.7
0
0.0073
0.9927
1.0000
0.00000
0.000000
0
0.0000
0 0000
0.0073
0.9927
1.0000
0 00000
2
1-4
32.9
0.05
0.0013
0.9987
0.9927
0.00000
0.000002
0
0.0000 i 0.0000
0.0013
0.9987
ow?7 0 00000
3
5-9
16.4
0.03
0.0008
0.9992
0.9914
0.00000
0.000001
0
0.0000 1 0.0000
0.0008
0.9992
0.9914 0 00000
4
10-14
20.9
0.09
0.0010
0.9990
0.9906
0.00000
0.000004
0
0.0000 0 0000
0.0010
0.9990
5
15-19
68.2
0.12
0.0034
0.9966
0.9896
0.00001
0.000006
2.5
O
0.0034
0.9966
o<>X96 1 0.00001
(> 20-24 96 o.ir, 0.0048
0.9952
0.9862
0.00001 ; 0.000008
7.5
|05l" OO000
0.0048
0.9952
o'jxi.: o (loot) i
0.9815 0.00001
7
25-29
0.23
0.0050
0.9951
0.9815
0.00001 10.000011
12.5 1.7528
0.0000
0.0050
0.9951
8
30-34
116.3
0.48
0.0058
0.9942
0.9766
0.00002
0.000023
17.5
2.4539
0.0000
0.0058
0.9942
0.9766
9
35-39
162.2
0.55
0.0081
0.9919
0.9710
0.00003
0000027
22 5 3 1550
0 0000
0 0081
0.9919
0 9710
0 00003
10 I 40-44
237.3
1.14 0.0119 0.9882 0.9631
0.00006
0 000055 27 5 3 8561 0 0001 0 0119
0.9882
0 9631 0 00008
11
45-49
356
1.3
0.0178
0.9824
0.9518
0.00007
0.000061
32.5
4.5572
0.0001
0 0178
0.9823
0.9517
0.00009
12
50-54
518.6
1.72
0.0259
0.9744
0.9350
0.00009
0.000079
37.5
5 2583
0.0001
0.0260
0.9744
0.9349
0.00012
13 1 55-59
801.8
1 (.'> ()()4(i| 0.960" O'JIII " 42^ 1 ()()4(i| o
-------
a 2
^ ^ Column A: interval index number (i).
° to ST Column B: 5-year age interval (except <1 and 1-4) up to age 85.
| ^ a. Column C: all-cause mortality rate for interval i (x 105/year) (2000 data from NCHS).
i 5 § Column D: NPC incidence rate for interval i (x 105/year) (1996-2000 SEER data).
5 g I Column E: all-cause hazard rate for interval i (h*,) (= all-cause mortality rate x number of years in age interval).0
V to ^
>!
^ ^ Column F: probability of surviving interval i without being diagnosed with NPC (qO (= exp(-h*i)).
I. a Column G: probability of surviving up to interval i without having been diagnosed with NPC (SO (S| = 1: S, = S, i x q for i>l).
¦2 Column H: NPC incidence hazard rate for interval i (hL) (= NPC incidence rate x number of years in interval).
"o a, Column I: conditional probability of being diagnosed with NPC in interval i (= (h/h*,) x S, x (1-q,)). i.e., conditional upon surviving up to interval i without
i?" S- V. having been diagnosed with NPC [Ro, the background lifetime probability of being diagnosed with NPC = the sum of the conditional probabilities
~ across the intervals].
Column J: exposure duration (in years) at mid-interval (xtime).
Column K: cumulative exposure mid-interval (xdose) (= exposure level (i.e., 0.0461 ppm) x 365/240 x 20/10 x xtime) [365/240 x 20/10 converts continuous
environmental exposures to corresponding occupational exposures].
ColumnL: NPC incidence hazard rate in exposed people for interval i (hx,) (= h; x (1 + p x xdose), where (3 = 0.05183 + (1.645 x 0.01915) = 0.08333) [0.05183
per ppm x year is the regression coefficient obtained, along with its SE of 0.01915, from Dr. Hauptmann (see Section 5.2.2.1). To estimate the
LECooos, i e., the 95% lower bound on the continuous exposure giving an extra risk of 0.05%, the 95% upper bound on the regression coefficient is
used, i.e., MLE + 1.645 x SE],
Column M: all-cause hazard rate in exposed people for interval i (h*x;) (= h* + (hx - h,)).
Column N: probability of surviving interval i without being diagnosed with NPC for exposed people (qx,) (= cxpM^x,)).
Column O: probability of surviving up to interval i without having been diagnosed with NPC for exposed people (Sx,) (Sxi = 1: Sx = Sx , x qx _i. for i>l).
Column P: conditional probability of being diagnosed with NPC in interval i for exposed people (= (lix/l^x,) x Sx x (1-qx,)) [Rx, the lifetime probability of
being diagnosed with NPC for exposed people = the sum of the conditional probabilities across the intervals].
a Using the methodology of BEIRIV (1988).
b The estimated 95% lower bound on the continuous exposure level of TCE that gives a 0.05% extra lifetime risk of NPC.
0 For the cancer incidence calculation, the all-cause hazard rate for interval i should technically be the rate of either dying of any cause or being diagnosed with
the specific cancer during the interval, i.e., (the all-cause mortality rate for the interval + the cancer-specific incidence rate for the interval—the cancer-specific
mortality rate for the interval [so that a cancer case isn't counted twice, i.e., upon diagnosis and upon death]) x number of years in interval. This adjustment
was ignored here because the NPC incidence rates are small compared with the all-cause mortality rates.
^ MLE = maximum likelihood estimate, SE = standard error
O
H
b
O
-------
Appendix D
i
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
APPENDIX D
MODEL STRUCTURE & CALIBRATION IN CONOLLY ET AL. (2003, 2004)
The various studies indicated in Section 5.4.1 were followed by the development of a
biologically motivated dose-response model for formaldehyde-induced cancer as represented in a
series of papers and in a health assessment report (CUT model) (Conolly et al., 2004, 2003,
2000; Conolly, 2002; Kimbell et al., 2001a, b; Overton et al., 2001; CUT, 1999). EPA's cancer
guidelines (U.S. EPA, 2005a) suggest using a BBDR model for extrapolation when data permits
since it facilitates the incorporation of MO A in risk assessment. The CUT modeling and
available data were evaluated in a series of peer-reviewed papers (Klein et al., 2009; Crump et
al., 2008; Subramaniam et al., 2008, 2007) and debated further in the literature (Conolly et al.,
2009; Crump et al., 2009). In addition, alternatives to the CUT biological modeling (but based
on that original model) are further explored and evaluated here.
In Conolly et al. (2003), tumor incidence data in the above long-term bioassays were
modeled by using an approximation of the two-stage clonal growth model (Moolgavkar et al.,
1988) and allowing formaldehyde to have directly mutagenic action. Conolly et al. (2003)
combined these data with historical control data on 7,684 animals obtained from National
Toxicology Program (NTP) bioassays. These models are based on the Moolgavkar, Venzon, and
Knudson (MVK) stochastic two-stage model of cancer (Moolgavkar et al., 1988; Moolgavkar
and Knudson, 1981; Moolgavkar and Venzon, 1979), which accounts for growth of a pool of
normal cells, mutation of normal cells to initiated cells, clonal expansion and death of initiated
cells, and mutation of initiated cells to fully malignant cells.
The MVK model for formaldehyde accounted for two MO As that may be relevant to
formaldehyde carcinogenicity:
o An indirect MOA in which the regenerative cell proliferation in response to formaldehyde
cytotoxicity increased the probability of errors in DNA replication. This MOA was modeled
by using labeling data on normal cells in nasal mucosa of rats exposed to formaldehyde,
o A possible direct mutagenic MOA, based on information indicating that formaldehyde is
mutagenic (Speit and Merk, 2002; Heck et al., 1990; Grafstrom et al., 1985), was modeled by
using rat data on formaldehyde production of DPXs (Monticello et al., 1996, 1991).
The human model for formaldehyde carcinogenicity (Conolly et al., 2004) is
conceptually very similar to the rat model. The model uses, as input, results from a dosimetry
model for an anatomically realistic representation of the human upper airways and an idealized
This document is a draft for review purposes only and does not constitute Agency policy.
D-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
representation of the lower airways. However, the model does not incorporate any data on
human responses to formaldehyde exposure. The rat and human formaldehyde models are
detailed further below.
The following notations are used in the rest of this chapter:
N cell, normal cell
I cell, initiated cell
LI, labeling index (number of labeled cells/(number labeled + unlabeled cells)
ULLI, unit length labeling index (number labeled cells/length of basement membrane)
N, number of normal cells that are eligible for progression to malignancy
ocn, division rate of normal cells (hours-1)
Hn, rate at which an initiated cell is formed by mutation of a normal cell (per cell division
of normal cells)
ai. division rate of an initiated cell (hours-1)
Pi. death rate of an initiated cell (hours-1)
lii, rate at which a malignant cell is formed by mutation of an initiated cell (per cell
division of initiated cells)
A novel contribution of the CUT model is that cell replication rates and DPX
concentrations are driven by local dose, which is formaldehyde flux to each region of nasal
tissue expressed as pmol/mm2-hour. This dosimetry is predicted by computational fluid
dynamics (CFD) modeling using anatomically accurate representations of the nasal passages (see
Chapter 3). Such a feature is important to incorporating site-specific toxicity in the case of a
highly reactive gas like formaldehyde, for which uptake patterns are spatially localized and
significantly different across species (see Chapter 3). In the CUT model, each of these
parameters is characterized by local flux. The inputs to the two-stage cancer modeling consisted
of results from other model predictions as well as empirical data as follows:
• Regional uptake of formaldehyde in the respiratory tract predicted by using CFD
modeling in the F344 rat and human (Kimbell et al., 2001a, b; Overton et al., 2001;
Subramaniam et al., 1998)
• Concentrations of DPXs predicted by a PBPK model (Conolly et al., 2000) calibrated to
fit the DPX data in F344 rat and rhesus monkey (Casanova et al., 1994, 1991) and
subsequently scaled up to humans
This document is a draft for review purposes only and does not constitute Agency policy.
D-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
• oin inferred from LI data on rats exposed to formaldehyde (Monticello et al., 1996, 1991,
1990)
D.l. DPX AND MUTATIONAL ACTION
Formaldehyde interacts with DNA to form DPXs. These cross-links are considered to
induce mutagenic as well as clastogenic effects. Casanova et al. (1994, 1989) carried out two
studies of DPX measurements in F344 rats. In the first study, rats were exposed to
concentrations of 0.3, 0.7, 2, 6, and 10 ppm for 6 hours and DPX measurements were made over
the whole respiratory mucosa of the rat, while, in the second study, the exposure was to 0.7, 2, 6,
or 15 ppm formaldehyde for 3 hours and measurements were made at "high" and "low" tumor
sites. Overall, these studies showed statistically significantly elevated levels of DPXs at
concentrations >2 ppm, with the trend also indicating elevated DPXs at 0.7 ppm. In Conolly et
al. (2003), DPX formation is considered proportional to the intracellular dose that induces
mutations. Conolly et al. (2000) used data from the second study to develop a PBPK model that
predicted the time course of DPX concentrations as a function of regional formaldehyde flux
(estimated in the CFD modeling and expressed as pmol/mm2-hour). In Conolly et al. (2003), this
PBPK model was then used to predict regional DPX concentrations (that is, as a function of
regional formaldehyde flux) (Figure 5-11, Chapter 5). These data were incorporated into the
two-stage clonal expansion model by defining the mutation rate of normal and initiated cells as
the same linear function of DPX concentration as follows:
(J-N = (j-l = (J-Nbasal + KMU X DPX (1)
The unknown constants (J,Nbasai and KMU were estimated by fitting model predictions to the
tumor bioassay data.
D.2. CALIBRATION OF MODEL
The rat model in Conolly et al. (2003) involved six unknown statistical parameters that
were estimated by fitting the model to the rat formaldehyde bioassay data shown in Table 5-24 in
Chapter 5 (Monticello et al., 1996; Kerns et al., 1983) plus data from several thousand control
animals from all the rat bioassays conducted by the NTP. These NTP bioassays were conducted
from 1976 through 1999 and included 7,684 animals with an incidence of 13 SCCs (i.e., 0.17%
incidence). The resulting model predicts the probability of a nasal SCC in the F344 rat as a
function of age and exposure to formaldehyde. The fit to the tumor incidence data is shown in
Figure 5-12, Chapter 5. (Note: This figure also shows the fit to the data obtained by the
This document is a draft for review purposes only and does not constitute Agency policy.
D-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
implementation of this model in Subramaniam et al. [2007], which is discussed later in this
chapter.)
D.3. FLUX BINS
The spatial distribution of formaldehyde over the nasal lining was characterized by
partitioning the nasal surface by formaldehyde flux to the tissue, resulting in 20 "flux bins"
(Figure 5-13, Chapter 5). Each bin is comprised of elements (not necessarily contiguous) of the
nasal surface that receive a particular interval of formaldehyde flux per ppm of exposure
concentration (Kimbell et al., 2001a). The spatial coordinates of elements comprising a
particular flux bin are fixed for all exposure concentrations, with formaldehyde flux in a bin
scaling linearly with exposure concentration (ppm). The number of cells at risk varies across the
bins, as shown in Figure 5-14, Chapter 5.
D.4. USE OF LABELING DATA
Cell replication rates in Conolly et al. (2003) were obtained by pooling labeling data
from two phases of a labeling study in which male F344 rats were exposed to formaldehyde gas
at similar concentrations (0, 0.7, 2.0, 6.0, 10.0, or 15.0 ppm). The first phase employed injection
labeling with a 2-hour pulse labeling time, and animals were exposed to formaldehyde for early
exposure periods of 1, 4, and 9 days and 6 weeks (Monticello et al., 1991). The second phase
used osmotic mini pumps for labeling with a 120-hour labeling time to quantify labeling in
animals exposed for 13, 26, 52, and 78 weeks (Monticello et al., 1996). The combined pulse and
continuous labeling data were expressed as one exposure TWA over all sites for each exposure
concentration. aN was calculated from these labeling data by using an approximation from
Moolgavkar and Luebeck (1992). A dose-response curve for normal cell replication rates (i.e.,
aN as a function of formaldehyde flux) was then calculated as shown in Figure D-l. These steps
are carefully detailed and evaluated in Subramaniam et al. (2008), and discussion of the data will
continue in the section on uncertainties in characterizing cell replication rates.
D.5. UPWARD EXTRAPOLATION OF NORMAL CELL DIVISION RATE
The extensive labeling data collected by Monticello et al. (1996, 1991) present an
opportunity to use precursor data in assessing cancer risk. However, these empirical data could
be used to determine afflux) only for the lower flux range, 0-9,340 pmol/mm2-hour (see
Subramaniam et al. [2008] for the reasons), as shown by the solid line in Figure D-l, whereas the
highest computed flux at 15.0 ppm exposure was 39,300 pmol/mm2-hour. Therefore Conolly et
al. (2003) introduced an adjustable parameter, amax, that represented the value of aN(flux) at the
This document is a draft for review purposes only and does not constitute Agency policy.
D-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
9
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
maximum flux of 39,300 pmol/mm -hour. amax was estimated by maximizing the likelihood of
the two-stage model fit to the tumor incidence data. For 9,340 < flux < 39,300 pmol/mm2-
hour, afflux) was determined by linear interpolation from ocn(9,340) to amax, as shown by the
dashed line in Figure D-l.
0
03
Dd
0.05
0.04
0.03
0.00
0.002
0.001
0.000
0.002
0.001
0.000
CD 0.01
J-shape
Hockey
Empirical aN (from ULLI data)
Estimated aN
Estimated otj ^
/
max
/
0 2000 4000 6000
/
i
10000
20000
30000
40000
Flux (pmole/mrrr/h)
Figure D-l. Dose response of normal (ocn) and initiated (ai) cell division rate
in Conolly et al. (2003).
Note: Empirically derived values of aN (TWA over six sites) from Table 1 in
Conolly et al. (2003) and optimized parameter values from their Table 4 were
used. The main panel is for the J-shape dose response. Insets show J-shape and
hockey-stick shape representations at the low end of the flux range. The long
arrow denotes the upper end of the flux range for which the empirical unit-length
labeling data are available for use in the clonal growth model. amax is the value of
aN at the maximum formaldehyde flux delivered at 15 ppm exposure and
estimated by optimizing against the tumor incidence data, ai < aN for flux greater
than the value indicated by the small vertical arrow. Conolly et al. (2004, 2003)
assumed J3i = aN at all flux values.
Source: Subramaniam et al. (2008).
This document is a draft for review purposes only and does not constitute Agency policy.
D-6 DRAFT—DO NOT CITE OR QUOTE
-------
1 D.6. INITIATED CELL DIVISION AND DEATH RATES
2 The pool of cells used for obtaining the LI data in Monticello et al. (1996, 1991) consists
3 of largely normal cells with perhaps increasing numbers of initiated cells at higher exposure
4 concentrations. Since the division rates of initiated cells in the nasal epithelium, either
5 background or formaldehyde exposed, could not be inferred from the available empirical data,
6 Conolly et al. (2003) made what they perceived to be a biologically reasonable assumption for
7 ai, assuming ai to be linked to ocn via a two-parameter function:
8
9 ai = aN x {multb - multc x max[aN - aN(basai), 0]} (2)
10
11 where aN = afflux), a,N(basai) is the estimated average cell division rate in unexposed normal
12 cells, and multb and multc are unknown parameters estimated by likelihood optimization against
13 the tumor data.2 The value of aN(basai) was equal to 3.39 x 10^ hours-1 as determined by Conolly
14 et al. (2003) from the raw averaged LI data. The ratio ai/aN is plotted against flux in Figure D-2,
15 and ai(flux) is shown by the dotted line in Figure D-1.
1.08
1.04
z
8
8
0.96
0
10000
20000
30000
4000C
0 10000 20000 30000 4000C
16 Flux (pmole/mm^/h)
17
18 Figure D-2. Flux dependence of ratio of initiated and normal cell replication
19 rates (oci/ocn) in CUT model.
20 Note: Cell replication rate of initiated cells is less than normal cell replication rate
21 at flux exceeding the value denoted by the arrow. By assumption, the y-axis also
22 represents (ai/J3i).
23 Source: Subramaniam et al. (2008).
24
2 multb and multc were equal to 1.072 and 2.583, respectively (J-shaped aN),, and 1.070 and 2.515, respectively
(hockey-stick shaped aN)
This document is a draft for review purposes only and does not constitute Agency policy.
D-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Death rates of initiated cells (Pi) are assumed to equal the division rates of normal cells
for all formaldehyde flux values, that is
Pi(flux) = aN(flux) (3)
Conolly et al. (2003) stated that this formulation for ai and Pi provided the best fit of the
model to the tumor data.
D.7. STRUCTURE OF I II I CUT HUMAN MODEL
Subsequent to the BBDR model for modeling rat cancer, Conolly et al. (2004) developed
a corresponding model for humans for the purpose of extrapolating the risk to humans estimated
by the rat model. Also, rather than considering only nasal tumors, it is used to predict the risk of
all human respiratory tumors. DPXs observed at proximal portions of the rhesus monkey LRT
(Casanova et al., 1991) suggested that the LRT may be at risk in addition to the URT. In
addition, some epidemiologic studies (Gardner et al., 1993; Blair et al., 1990, 1986) reported an
increase in lung cancer associated with formaldehyde exposure, while others reported no such
increases (Collins et al., 1997; Stayner et al., 1988). The human model for formaldehyde
carcinogenicity (Conolly et al., 2004) is conceptually very similar to the rat model and follows
the schematic in Figure 5-11, Chapter 5. The following points need to be noted:
• The model does not incorporate any data on human responses to formaldehyde exposure.
• The model is based on an anatomically realistic representation of the human nasal
passages (in a single individual) and an idealized representation of the LRT. Local
formaldehyde flux to respiratory tissue is estimated by a CFD model for humans
(Subramaniam et al., 1998; Kimbell et al., 2001a; Overton et al., 2001).
• Rates of cell division and cell death are, with a minor modification, assumed to be the
same in humans as in rats.
• The concentration of formaldehyde-induced DPXs in humans is estimated by scaling up
from values obtained from experiments in the F344 rat and rhesus monkey. This scaling
up was discussed in chapter 3.
• The statistical parameters for the human model are either estimated by fitting the model
to the human background data, assumed to have the same value as obtained in the rat
model, or, in one case, fixed at a value suggested by the epidemiologic literature. The
delay, D, is fixed at 3.5 years, based on a fit to the incidence of lung cancer in a cohort of
British doctors (Doll and Peto, 1978). The two other parameters in the rat model that
This document is a draft for review purposes only and does not constitute Agency policy.
D-8 DRAFT—DO NOT CITE OR QUOTE
-------
1 affect the background rate of cancer (multb and Hbasai) are estimated by fitting to U.S.
2 cancer incidence or mortality data. These parameters affect the baseline values for the
3 human oti, |i\, and |ii. Since amax, multfc, and KMU do not affect the background cancer
4 rate, they cannot be estimated from the (baseline) U.S. cancer incidence rates. Therefore,
5 in Conolly et al. (2004, 2003), amax and multfc are assumed to have the same values in
6 humans as in rats, and the human value for KMU is obtained by assuming that the ratio
7 KMU/|ibasai is invariant across species. Thus,
8
9 KMU[m) = KMUla) x ) (4)
t^Nbasa!{rat)
This document is a draft for review purposes only and does not constitute Agency policy.
D-9 DRAFT—DO NOT CITE OR QUOTE
-------
This page intentionally left blank.
This document is a draft for review purposes only and does not constitute Agency policy.
D-10 DRAFT—DO NOT CITE OR QUOTE
-------
Appendix E
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
APPENDIX E
EVALUATION OF BBDR MODELING OF NASAL CANCER IN THE F344 RAT:
CONOLLY ET AL. (2003) AND ALTERNATIVE IMPLEMENTATIONS
A biologically based dose-response model for formaldehyde-induced cancer was in a
series of papers and in a health assessment report (CUT model) (Conolly et al., 2004, 2003,
2000; Conolly, 2002; Kimbell et al., 2001a, b; Overton et al., 2001; CUT, 1999). The model
structure, notations, and calibration have been described in Appendix D. In Chapter 5, an
evaluation of the uncertainties of this model and alternative approaches based on its conceptual
framework was presented in a summary form. This Appendix provides the relevant details of that
evaluation and presents a range of dose-response curves for tumor risk in the rat. It is divided
into the following major sections. First, an overview of all the issues evaluated is provided in
tabular form. The rest of the Appendix then presents only those issues which have a significant
impact on model predictions. These issues pertain to the use of history controls, the uncertainty
and variability in the dose-response for normal cell-replication rates, and sensitivity of model
results to uncertainty in the kinetics of initiated cells. The issues have significant impact on mode
of action inferences, and this is discussed in some detail.
E.l. TABULATION OF ALL ISSUES EVALUATED IN THE RAT MODELS
Table E-l summarizes model uncertainties and their impact as evaluated by EPA. The
key uncertainties are discussed in considerably more detail in additional sections in this
Appendix and in published manuscripts as denoted in the tables.
E.2. Statistical Methods Used in Evaluation
Parameters of the alternate models shown here were estimated by maximizing the
likelihood function defined by the data (Cox and Hinkley, 1974). Such estimates are referred to
as maximum likelihood estimates (MLEs). Statistical confidence bounds were computed by
using the profile likelihood method (Crump, 2002; Cox and Oakes, 1984; Cox and Hinkley,
1974). In this approach, an asymptotic 100(1 - a)% upper (lower) statistical confidence bound
for a parameter, /?, in the animal cancer model is calculated as the largest (smallest) value of fi
that satisfies
2[Lmax -L*(P)] = XI_2a (5)
This document is a draft for review purposes only and does not constitute Agency policy.
E-2 DRAFT—DO NOT CITE OR QUOTE
-------
a 2
o S.
a ^3
Grj St
li
>!
5^
W
§•
rs
sj
3
5
>!
6
Si
Table E-l. Evaluation of assumptions and uncertainties in the CUT model for nasal tumors in the F344 rat
Assumptions, approach,
and characterization of
input data
Rationale
EPA evaluation
Further
elaboration3
Hoogenveen et al. (1999)
solution method, which is
valid only for time-
independent parameters, is
accurate enough.
Errors due to this
assumption thought to
be significant only at
high concentration and
not at human
exposures.
EPA implemented a solution method valid for time-dependent
parameters. Results did not differ significantly from those
obtained assuming Hoogenveen et al.(1999) solutions.
Caveat: Impact not evaluated for the case where cell
replication rates vary in time.
Crump et al.
(2006);
Subramaniam
et al. (2007)
All SCC tumors are rapidly
fatal; no incidental tumors.
Death is expected to
occur typically within
1-2 weeks of observed
tumor (personal
communication with
R. Conolly).
1) Overall, assumption does not impact model calibration or
prediction.
2) However, since 57 animals were observed to have tumors
at interim sacrifice times, EPA implementation distinguished
between incidental and fatal tumors. Time lag between
observable tumor and time of death was significant compared
to time lag between first malignant cell and observable tumor.
Subramaniam
et al. (2007)
Historical controls from
entire NTP database were
lumped with concurrent
controls in studies.
Data is on control
animals, and number is
large (7,684).
Therefore, intercurrent
mortality was not
expected to be
substantial.
1) Tumor incidence in "all NTP" 10-fold higher than in "all
inhalation NTP" controls. Including all NTP controls is
considered inappropriate.
2) Low-dose response curve sensitive to use of historical
controls.
3) Model inference on relevance of formaldehyde's
mutagenic potential to its carcinogenicity varies from
"insignificant" to "highly significant," depending on controls
used. (See Appendix F for impact on human risk.)
Table E-2;
Subramaniam
et al. (2007);
Sec E.3.1
LI was derived from
experimentally measured
ULLI.
Derived from
correlating ULLI to LI
measured in same
experiment.
Significant variation in number of cells per unit length of
basement membrane. Spread in ULLI/LI -25%. Impact on
risk not evaluated.
Subramaniam
et al. (2008);
Crq o
§ 4
O
>!
w
oo
o
H
L
O
-------
a ^3
Grj St
!"§
Crq o
^ a
.cj ^
^ a
^ a
5^
W
§•
rs
sj
3
5
>!
6
Si
Table E-l. Evaluation of assumptions and uncertainties in the CUT model for nasal tumors in the F344 rat
Assumptions, approach,
and characterization of
input data
Rationale
EPA evaluation
Further
elaboration3
>!
Pulse and continuous
labeling data combined in
deriving aN from LI.
Continuous LI
normalized by ratio of
pulse to continuous
values for control data.
Formula used for deriving aN from LI is not applicable for
pulse labeling data. Pulse labeling is measure of number of
cells in S-phase, not of their recruitment rate into S-phase; not
enough information to derive aN from pulse data. Impact on
risk predictions could not be evaluated.
Subramaniam
et al. (2008);
Sec E.3.2.2
W
To construct dose response
for aN, labeling data were
weighted by exposure time
(t) and averaged over all
nasal sites (TWA). Flux at
an exposure concentration
was averaged over all nasal
sites.
Site-to-site variation in
LI large and did not
vary consistently with
flux. No reasonable
approach available for
incorporating time
variation in labeling in
interspecies
extrapolation.
1) TWA assigns low weight to early time LI, but aN for early t
is very important to the cancer process. Since pulse ULLI
was used for t < 13 weeks, impact of these ULLIs on risk
could not be evaluated.
2) Time dependence in aN derived from continuous ULLI
does not significantly impact model predictions.
3) Site-to-site variation of aN at least 10-fold and has major
impact on model calibration. 10-fold variation in tumor
incidence data across sites.
4) Large differences in number of cells across nasal sites
(Table E-3), so averaging over sites is problematic.
5) Histologic changes, thickening of epithelium and
metaplasia occur at later times for the higher dose and would
affect replication rate.
Figures E-l, E-
2, E-3;
Subramaniam
et al. (2008);
Sec E.3.2.3
O
-------
a 2
o S.
a ^3
Grj St
li
>!
5^
W
§•
rs
sj
3
5
>!
6
Si
Table E-l. Evaluation of assumptions and uncertainties in the CUT model for nasal tumors in the F344 rat
Assumptions, approach,
and characterization of
input data
Rationale
EPA evaluation
Further
elaboration3
Steady-state flux estimates
not affected by airway and
tissue reconfiguration due
to long-term dosing.
Histopathologic
changes not likely to be
rate-limiting factors in
dosimetry.
1) Thickening of epithelium and squamous metaplasia
occurring at later times for the higher dose will reduce tissue
flux. Not incorporated in model.
2) These effects will push regions of higher flux to more
posterior regions of respiratory tract. Likely to affect
calibration of rat model. Uncertainty not evaluated
quantitatively.
3) Calibration of PBPK model for DPXs seen to be highly
sensitive to tissue thickness.
Kimbell et al
(1997);
Subramaniam
et al. (2008)
TWA afflux) rises above
baseline levels only at
cytolethal dose. Above
such dose, afflux) rises
sharply due to regenerative
proliferation.
Variability in afflux)
rate is represented by
also considering
hockey-stick (threshold
in dose) when TWA
indicates J-shape
(inhibition of cell
division) description of
afflux).
1) Uncertainty and variability in o,n quantitatively evaluated
to be large. In addition, several qualitative uncertainties in
characterization of aN(flux) from LI.
2) Several dose-response shapes, including a monotonic
increasing curve without a threshold, were considered in order
to adequately describe highly dispersed cell replication data.
Substantial impact on low dose risk, including negative added
risk.
Figures E-l, E-
2, E-3, E-4, E-
5;
Subramaniam
et al. (2008);
Sec E.3.2
Crq o
§ 4
O
>!
w
o
H
o
o
-------
a 2
o S.
a ^3
Grj St
!"§
Crq o
^ a
.cj ^
^ a
^ a
5^
W
§•
cs
sj
3
5
>!
6
5^
S^
>!
Table E-l. Evaluation of assumptions and uncertainties in the CUT model for nasal tumors in the F344 rat
Assumptions, approach,
and characterization of
input data
Rationale
EPA evaluation
Further
elaboration3
Dose response for ai was
obtained from oin,
assuming ratio (ai /aN) to
be a two-parameter
function of flux (see
Figures 5-7, 5-9).
Parameters were estimated
by optimizing model
predictions against tumor
incidence data.
(ai /a^ was >1.0 in line
with the notion of I
cells possessing a
growth advantage over
N cells.
Satisfies Occam's razor
principle (Conolly et
al., 2009).
1) Estimated (ai /a\) in CUT modeling is <1.0 (growth
disadvantage) for higher flux values and is >1.0 only at lower
end of flux range in model (see Figure 5-9).
2) Since there are no data to inform ai, sensitivity of risk
estimates to various functional forms was evaluated. Risk
estimates for the rat were extremely sensitive to alternate
biologically plausible assumptions for ai(flux) and varied by
many orders of magnitude at <1 ppm, including negative risk.
All these models described tumor incidence data and cell
replication and DPX data equally well.
Figures D-2, E-
5, E-6;
Subramaniam
et al. (2008);
Crump et al.
(2009, 2008);
Sec E.3.3
Death rate of I cells
Pi(flux) assumed = division
rate of N cells aN(flux).
Based on homeostasis
(aN = |3n) and
assumption that
formaldehyde is equally
cytotoxic to N cells and
I cells. Satisfies
Occam's Razor
principle (Conolly et
al., 2009).
1) In general, data indicate I cells are more resistant to
cytolethality and that ADH3 clearance capacity is greater in
transformed cells. Therefore, plausibility of assumption (J3i =
a^ is tenuous.
2) Alternate assumption, J3i proportional to ai, was examined.
Risk estimates extremely sensitive to assumptions on J3i (see
Figure 5-12).
Subramaniam
et al. (2008);
Crump et al.
(2009, 2008);
Sec E.3.3
DPX is dose surrogate for
formaldehyde mutagenic
potential. DPX clearance
is rapid and complete in 18
hours.
Casanova et al. (1994).
Half-life for DPX clearance in in vitro experiments on
transformed cell lines was sevenfold longer than estimated by
Conolly et al. (2004, 2003) and perhaps 14-fold longer with
normal (non-transformed) human cells. Some DPX
accumulation therefore likely. However, model calibration
and dose response in rat is insensitive to this uncertainty.
Quievryn and
Zhitkovich,
(2000);
Subramaniam
et al. (2007);
Chap 3
W
On
o
-------
a 2
o S.
a ^3
Grj St
li
>!
5^
W
§•
rs
sj
3
5
>!
6
Si
Table E-l. Evaluation of assumptions and uncertainties in the CUT model for nasal tumors in the F344 rat
Assumptions, approach,
and characterization of
input data
Rationale
EPA evaluation
Further
elaboration3
Formaldehyde mutagenic
action takes place only
while DPXs are in place.
DNA lesions remain after DPX removal. DPX induces
further DNA and protein damage. Potential for
formaldehyde-induced mutation after DPX clearance. Thus,
formaldehyde mutagenicity may be underrepresented. Could
not quantitatively evaluate uncertainty (no data on clearance
of secondary lesions).
Barker et al.
(2005); Speit
and Schmid
(2006);
Subramaniam
et al. (2008);
Chap 4
Crq o
§ 4
O
>!
""References stated here are in addition to Conolly et al. (2004, 2003).
Note: Risk estimates discussed in this table are for the F344 rat.
W
O
H
b
O
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
where L indicates the likelihood of the rat bioassay data, Lmax is its maximum value, is, for
a fixed value of fi, the maximum value of the log-likelihood with respect to all of the remaining
parameters, and X]_2a is the 100(l-2a) percentage point of the chi-square distribution with one
degree of freedom. The required bound for a parameter, /?, was determined via a numerical
search for a value of fi that satisfies this equation.
The additional risk is defined as the probability of an animal dying from an SCC by the
age of 790 days, in the absence of other competing risks of death, while exposed throughout life
to a prescribed constant air concentration of formaldehyde, minus the corresponding probability
in an animal not exposed to formaldehyde. The MLE of additional risk is the additional risk
computed using MLEs of the model parameters.
The method described above for computing profile likelihood confidence bounds cannot
be used with additional risk because additional risk is not a parameter in the cancer model.
Instead, an asymptotic 100(1 - a)% upper (lower) statistical confidence bound for additional risk
was computed by finding the parameter values that presented the largest (smallest) value of
additional risk, subject to the inequality
^[Lmax L] ^X]-2a (6)
being satisfied, with the resulting value of additional risk being the required bound. This
procedure was implemented through use of penalty functions (Smith and Coit, 1995). For
example, the profile upper bound on additional risk was computed by maximizing the "penalized
added risk," defined as (
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
selected runs. For select cases, the bootstrap method was also used to calculate confidence
bounds in order to confirm their accuracy. Values so calculated were found to be in agreement
with those calculated by using the likelihood method.
E.3. PRIMARY UNCERTAINTIES IN BBDR MODELING OF THE F344 RAT DATA
The evaluation of the CUT model and alternatives to this model that were implemented
for the F344 rat will be discussed first. Of the issues tabulated above, the following uncertainties
in the modeling of the F344 rat data will be discussed in considerably more detail: use of
historical controls, uncertainty and variability in characterizing cell replication rates from the
labeling data, and uncertainty in model specification of initiated cell kinetics.
In their evaluation, Subramaniam et al. (2007) first attempted to reproduce the Conolly et
al. (2003) results under similar conditions and assumptions as employed in their paper, which
included the assumption that tumors were rapidly fatal. Figure 5-12 in Chapter 5 shows the
results for this case. The predicted probabilities shown in this figure were obtained by
Subramaniam et al. (2007) by using the source code made available by Dr. Conolly. These are
compared with the best-fitting model and plotted against the Kaplan-Meier (KM) probabilities.
Although the results are largely similar, there are some differences (Subramaniam et al., 2007).
Given the scope of issues to examine for the uncertainty analyses, the evaluation
proceeded in stages. First, the Hoogenveen et al. (1999) solution was replaced by one that is
valid for a model with time varying parameters (first entry in Table E-l), and tumors found at
scheduled sacrifices were assumed to be incidental rather than fatal (second entry in Table E-l ).
Second, weekly averaged solutions for DPX concentration levels were used instead of hourly
varying solutions (predicted by a PBPK model). The log-likelihood values and tumor
probabilities remained essentially unchanged. Upon quantitative evaluation, these factors,
although important from a methodological point of view, were not found to be major
determinants of either calibration or prediction of the model for the F344 rat data (Subramaniam
et al., 2007).
Subramaniam et al. (2007), as in Georgieva et al. (2003), used the DPX clearance rate
constant obtained from in vitro data instead of the assumption in Conolly et al. (2003) that all
DPXs cleared within 18 hours (Subramaniam et al., 2007). With this revision, weekly average
DPX concentrations were larger than those in Conolly et al. (2003) by essentially a constant ratio
equal to 4.21 (range of 4.12-4.36) when averaged over flux bin and exposure concentrations.
Accordingly, cancer model fits to the rat tumor incidence data using the two sets of DPX
concentrations (everything else remaining the same) provided very similar parameter estimates,
except that the parameter KMUrat in equation 1 (and equation 4) was 4.23 times larger with the
This document is a draft for review purposes only and does not constitute Agency policy.
E-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Conolly et al. (2003) DPX concentrations. In other words, the product KMU x DPX remained
substantially unchanged. However, the different clearance rate does significantly impact the
scale-up of the two-stage clonal growth model to the human.
After making the above modifications, the impact of the following uncertainties, which
had a large impact on modeling, was examined sequentially.
E.3.1. Sensitivity to Use of Historical Controls
E.3.1.1. Use of Historical Controls
Conolly et al. (2003) combined the historical controls arising from the entire NTP
database of bioassays. Tumor and survival rates in control groups from different NTP studies
are known to vary due to genetic drift in animals over time and differences in laboratory
procedures, such as diet, housing, and pathological procedures (Haseman, 1995; Rao et al.,
1987). In order to minimize extra variability when historical control data are used, the current
NTP practice is to limit the historical control data, as far as possible, to studies involving the
same route of exposure and to use historical control data from the most recent studies (Peddada
et al., 2007).
Bickis and Krewski (1989) analyzed 49 NTP long-term rodent cancer bioassays and
found a large difference in determinations of carcinogenicity, depending on the use of historical
controls with concurrent control animals. The historical controls used in the CUT modeling
controls came from different rat colonies and from experiments conducted in different
laboratories over a wide span of years, so it is clearly problematic to assume that background
rates in these historical control animals are the same as those in the concurrent control group.
There are considerable differences among the background tumor rates of SCCs in all NTP
controls (13/7,684 = 0.0017), NTP inhalation controls (1/4,551 = 0.0002), and concurrent
controls (0/341 = 0.0). The rate in all NTP controls is significantly higher than that in NTP
inhalation controls (p = 0.01, Fisher's exact test). Given these differences, the inclusion of any
type of historical controls is problematic and is thought to have limited value if these factors are
not controlled for (Haseman, 1995).
E.3.1.2. Influence on Model Calibration and on Human Model
To investigate the effect of including historical controls in the CUT model, the analyses
in Subramaniam et al. (2007) were conducted by using the following sets of data for controls (the
fraction of animals with SCCs is denoted in parentheses): only concurrent controls (0/341),
concurrent controls plus all the NTP historical control data used by Conolly et al.(2003)
This document is a draft for review purposes only and does not constitute Agency policy.
E-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
(13/8,031), and concurrent controls plus data from historical controls obtained from NTP
inhalation studies (1/4,949) (NTP, 2005).3
The results of the evaluation are shown in Table E-2. For these analyses, the same
normal cell replication rates and the same relationship (see eq 2) between initiated cell and
normal cell replication rates as used in Conolly et al. (2003) were used. In all cases, weekly
averaged values of DPX concentrations were used. Model fits to the tumor incidence data were
similar in all cases (see Figure 5-12 in Chapter 5 and Subramaniam et al. [2007] for a more
complete discussion). The biggest influence of the control data was seen to be on the estimated
basal mutation rate in rats, jUNbasai(rat), which, in turn, influences the estimated mutation effect in
humans through eq 4 (Appendix D). amax was also seen to be a sensitive parameter and is
discussed later. See Subramaniam et al. (2007) for other parameters in the calibration.
The ratio KMU/|iNbasai is of particular interest because extrapolation to human in Conolly
et al. (2004) assumed its invariance as given by eq 4 (Appendix D). Now, (J,Nbasai in the human is
estimated independently by fitting a scaled-up version of the two-stage model to human baseline
rates of tumor incidence. Thus, a decrease in the value of |i\basai estimated in the rat modeling
increases the formaldehyde-induced mutational effect in the human.
While the MLE of KMU/|iNbasai is zero in the CUT animal model (Conolly et al., 2003), it
takes a range of values from 0 to 0.9 mm3/pmol and undefined (or infinite, when [j,Nbasai = 0) in
the various cases examined in this paper. The 95% upper confidence bound on this ratio ranges
from 0.25-6.2 (these values would be four times larger had the Conolly et al. [2003] DPX
concentrations been used) to infinite. Thus, the extrapolation to human risk by using the
approach in Conolly et al. (2004) becomes particularly problematic when only concurrent
controls are used, because then the mutational contribution to formaldehyde-induced risk in
humans becomes unbounded. This issue will be discussed again toward the end of the
discussion on historical controls.
3 Three animals in the inhalation historical controls were diagnosed with nasal SCC. Of these, two of the tumors
were determined to have originated in tissues other than the nasal cavity upon further review (Dr. Kevin Morgan and
Ms. Betsy Gross Bermudez, personal communication). These two tumors were therefore not included on the advice
of Dr. Morgan. See Subramaniam et al. (2007) for more details.
This document is a draft for review purposes only and does not constitute Agency policy.
E-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
a 2
o S.
a ^3
Grj St
li
>!
5^
W
§•
rs
sj
3
5
>!
6
Si
Table E-2. Influence of control data in modeling formaldehyde-induced cancer in the F344 rat
Case
A
D
B
E
C
F
NTP
NTP
Control animals (combined
All NTP
All NTP
inhalation
inhalation
Concurrent
Concurrent
with concurrent controls)
historical3
historical3
historical3
historical3
only3
only3
Cell replication dose
response
J-shaped
Hockey stick
J-shaped
Hockey stick
J-shaped
Hockey stick
Log-likelihood
-1692.65
-1693.68
-1,493.21
-1,493.35
-1,474.29
-1,474.29
I^Nbasal
1.87 x l(T6
2.12 x 10~6
7.32 x 10~7
9.32 x 10~7
0.0
0.0
KMU
1.12 x l(T7
0.0
6.84 x 10~7
6.18 x 10~7
1.20 x 10~6
1.20 x 10"6
KMU/ |iNbasal
0.06
0.0
0.94
0.66
co
co
(0.0, 0.40)
(0.0, 0.25)
(0.26, 6.20)
(0.2, 5.20)
(0.42, oo)
(0.41, oo)
C^max
0.045
0.045
0.045
0.045
0.045
0.045
(0.029, 0.045)
(0.029, 0.045)
(0.026, 0.045)
(0.027, 0.045)
(0.027, 0.045)
(0.027, 0.045)
Crq o
§ 4
O
>!
aValues in parentheses denote lower and upper 90% confidence bounds.
Source: Adapted from Subramaniam et al. (2007).
ffl
to
o
H
L
O
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
It may be noted, however, that absence of tumors in the limited number of concurrent
animals does not imply that the calculation will necessarily predict a zero background
probability of tumor (i.e., a parameter estimate of (j,Nbasai = 0). We observed such a
counterexample estimate for [j,Nbasai in simulations involving the alternate dose-response curves
for ocn and ai that are discussed later. Nonetheless, when [j,Nbasai = 0, an upper bound for [j,Nbasai
using the concurrent controls can be inferred. Accordingly, the 90% statistical lower confidence
bound on the ratio KMU/|iNbasai is also reported in Table E-2. Such a value would of course
provide a lower bound on risk by using this model and would therefore not be conservative.
Conolly et al. (2003) estimated KMU to be zero for both the hockey-stick and J-shape
cell replication models. However, the estimate for the coefficient KMU (obtained using the
solution of Crump et al. [2005]) is zero only for the case of the model with the hockey-stick
curve for cell replication and with control data as used by Conolly et al. (2003). It is positive in
all other cases and statistically significantly so in all cases in which either inhalation control data
or concurrent controls were used. With concurrent controls only and the J-shape cell replication
model, the MLE estimate for KMU (1.2 x 10 6) is larger than the statistical upper bound
obtained by Conolly et al. (2003) (8.2 x 10 7), It should also be kept in mind that the estimate
would be about 4.2 times larger still had the Conolly et al. (2003) DPX model been used.
E.3.1.3. Influence on Dose-Response Curve
Subramaniam et al. (2007) showed that inclusion of historical controls had a strong
impact on the tumor probability curve below the range of exposures over which tumors were
observed in the formaldehyde bioassays. As shown there, the MLE probabilities for occurrence
of a fatal tumor at exposure concentrations below 6 ppm were roughly an order of magnitude
higher when all the NTP historical controls were used, compared with MLE probabilities
predicted when historical controls were drawn only from inhalation bioassays, and many orders
of magnitude higher than MLE probabilities predicted when only concurrent controls were used
in the analysis. (Note that this comparison should not be inferred to apply to upper bound risk
estimates since there were many fewer concurrent than historical controls, so error bounds could
be much larger in the case where concurrent controls were used.)
However, as shown by these authors, model fits to the tumor data in the 6-15 ppm
exposure concentration range were qualitatively indifferent to which of these control data sets
was used. This observation emphasizes the statistical aspect of the CUT modeling—that
significant interplay among the various adjustable parameters allows the model to achieve a
good fit to the tumor incidence data independent of the control data used. On the other hand, the
results in Subramaniam et al. (2007) show that changes in the control data affect parameter
This document is a draft for review purposes only and does not constitute Agency policy.
E-13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
KMU, resulting in significantly different tumor predictions at lower exposure concentrations.
Therefore, the strong influence of using all the NTP historical controls on the low-dose region of
the time-to-tumor curves presented in Subramaniam et al. (2007) suggests that large
uncertainties may arise in extrapolating to both human and rat (in the low-dose region) from
such considerations alone.
E.3.1.4. Problem Including 1976 Study for Inhalation Historical Control
A crucial point needs to be noted with regard to the use of inhalation NTP historical
controls (i.e., cases B and E) in the two-stage clonal growth modeling. The single relevant tumor
in the NTP inhalation studies came from the very first NTP inhalation study, dated 1976, and the
animals in this study were from Hazelton Laboratories, whereas the concurrent animals were all
from Charles River Laboratories. Similar problems arise with inclusion of several other NTP
inhalation studies. As mentioned before, genetic and other time-related variation can lead to
different tumor and survival rates, and in general it is recommended that use of historical
controls be restricted to the same kind of bioassays and to studies within a 5-7 year span of the
concurrent animals (Peddada et al., 2007). Thus, it is problematic to assume that the tumor in
the 1976 NTP study is representative of the risk of SCCs in the formaldehyde bioassays. Even if
it were appropriate to consider the 1976 study, this leads to the unstable situation in which,
despite all of the "upstream" mechanistic information used to construct the BBDR model, the
only piece of data that might keep the model predictions of human risk bounded is a single tumor
found among several thousand rats from NTP bioassays (Crump et al., 2008). In summary,
although it can be argued that the rate of SCCs among the controls in the rat bioassay is probably
not zero, it is also problematic to assume that this rate can be adequately represented by the
background rate in NTP historical controls or even in NTP inhalation historical controls.
E.3.1.5. Inference on MOA
Subramaniam et al. (2007) also examined the contribution of the DPX component (which
represents the directly mutagenic potential of formaldehyde in the model) to the calculated tumor
probability, choosing for their case study the optimized models that use the NTP inhalation
control data. In the range of exposures where tumors were observed (6.0-15.0 ppm), the DPX
term was found to be responsible for 58-74% of the added tumor probability. Below 6.0 ppm
the estimated DPX contribution was extremely sensitive to the shape of apha n and varied
between 2 and 80%.
The CUT BBDR cancer modeling has contributed to the weight-of-evidence process in
various formaldehyde risk assessment efforts and papers by lending weight to the argument that
This document is a draft for review purposes only and does not constitute Agency policy.
E-14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
the direct mutations induced by formaldehyde are relatively irrelevant compared to the
importance of cytotoxicity-induced cell proliferation in explaining the observed tumorigenicity
in rodent bioassays and in projecting those observations to human exposures (Conolly et al.,
2004, 2003; Slikker et al., 2004; Bogdanffy et al., 2001, 1999; Conolly, 1995). The reanalyses in
Subramaniam et al. (2007) (in particular, the results in the above paragraph) indicate that, if the
CUT mathematical modeling were utilized to inform this debate, it would in fact indicate the
contrary—that a large contribution from formaldehyde's mutagenic potential may be needed to
explain formaldehyde carcinogenicity. This discussion is resumed in the context of uncertainties
in model specification for initiated cells.
E.3.2. Characterization of Uncertainty-Variability in cell Replication Rates
E.3.2.1. Dose-Response for an as Used in the CUT Clonal Growth Modeling
Monticello et al. (1996, 1991) used ULLI to quantify cell replication within the respiratory
epithelium. ULLI is a ratio between a count of labeled cells and the corresponding length (in
millimeters) of basal membrane examined, whereas the per-cell LI is the ratio of labeled cells to
all epithelial cells, in this case, along some length of basal membrane and its associated layer of
epithelial cells. Monticello et al. (1996, 1991) published ULLI values averaged over replicate
animals for each combination of exposure concentration, exposure time, and nasal site. These
values are plotted in Figure E-l. Conolly et al. (2003) adopted the following procedure in using
these values (Subramaniam et al., 2008):
1. The injection labeled ULLI data were first normalized by the ratio of the average
minipump ULLI for controls to the average injection labeled ULLI for controls.
2. Next, these ULLI average values were weighted by the exposure times in Monticello et
al. (1996, 1991) and averaged over the nasal sites. Thus, the data were combined into
one TWA for each exposure concentration.
3. LI was linearly related to the measured ULLI by using data from a different experiment
(Monticello et al., 1990) where both quantities had been measured for two sites in the
nose.
4. Cell replication rates of normal cells (oin) were then calculated as oin = (-0.5/t)log(l - LI)
(Moolgavkar and Luebeck, 1992), where LI is the labeling index and t is the period of
labeling.
5. This was repeated for each exposure concentration of formaldehyde, resulting in one
value of oin for each exposure concentration.
6. Correspondingly, for a given exposure concentration, the steady-state formaldehyde flux
into tissue, computed by CFD modeling, was averaged over all nasal sites. Thus, the
afflux) constructed by Conolly et al. (2003) consisted of a single oin and a single
average flux for each of six exposures.
This document is a draft for review purposes only and does not constitute Agency policy.
E-l5 DRAFT—DO NOT CITE OR QUOTE
-------
Exposure
< 12 wk (pulse
labeling)
Exposure > 12 wk (continuous labeling)
O'
O'
o1
o ¦
LO
LO
o
. •'.y'
o'
LO '
LO '
LO
LO '
o'
PMS
o'
PMS
0 5 10 15 0 5 10 15
Formaldehyde exposure concentration, ppm
1
2 Figure E-l. ULLI data for pulse and continuous labeling studies.
3 Note: Data are from pulse labeling study, left-hand side, at 1-42 days of exposure
4 and from the continuous-labeling study, right-hand side, at 13-78 weeks of
5 exposure for five nasal sites ALM, AMS, MMT, PLM, and posterior mid septum
6 [PMS]). Within each graph, lines with more breaks correspond to shorter
7 exposure times. Data source: Monticello et al. (1996, 1991).
8 This document is a draft for review purposes only and does not constitute Agency policy.
E-16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
This yielded a J-shaped dose-response curve for cell replication (when viewed on a non-
transformed scale for ocn), as shown in Figure D-l (Appendix D) for the full range of flux values
used in their modeling. The authors also considered a hockey-stick threshold representation of
their J-shaped curve for aN in order to make a health-protective choice, and the differences
between the two can be seen from the insets in Figure D-l. In these curves, the cell replication
rate is less than or the same as the baseline cell replication rate at low formaldehyde flux values.
The shape of the dose-response curve for cell replication as characterized in Conolly et al.
(2003) is seen as representing regenerative cell proliferation secondary to the cytotoxicity of
formaldehyde (Conolly, 2002). Considerable uncertainty and variability, both quantitative and
qualitative, exist in the use and interpretation of these labeling data for characterizing a dose
response for cell replication rates. The primary issues are discussed here. Unlike the preceding
sections, these have largely not been published elsewhere, so more details are provided.
E.3.2.2. Uncertainty in the Use of Pulse Labeling Data, and Short-Time Exposure Effects on
Cell Replication
The formula used for obtaining aN from LI in Conolly et al. (2003) was due to
Moolgavkar and Luebeck (1992) who derived this formula for continuous LI, cautioning that it
is not applicable for pulse labeled data. However, Conolly et al. (2003) applied this formula to
the injection (pulse) labeled data also. Such an application is problematic because 2-hour pulse
labeled data represent the pool of cells in S-phase rather than the rate at which cells are recruited
to the pool and because the baseline values of aN obtained in this manner from both data sets
differ considerably. As such, we are not aware of any reasonable manner to derive cell
replication rates from these pulse data without acquisition of data at additional time points.
Therefore, the quantitative analysis of cell replication rates is restricted in this document to the
continuous labeled data (Monticello et al., 1996), which does not include measurements made
before 13 weeks of exposure.
It is unfortunate that the continuous labeled data do not include any early measurements
because, as indicated by Figure E-l, the temporal variation in the unit-length LI (ULLI, the raw
data) is quite different between the "early time" (left panel) and "later time" (right panel) and
these early-time effects may be quite important to the cancer modeling. At the earliest times in
the left panel, the data show an increased trend in labeling at 2 ppm for the sites anterior lateral
meatus (ALM), anterior medial septum (AMS), posterior lateral meatus (PLM), and medial
maxilloturbinate (MMT) relative to control. (Also see the dose-response plotted as a function of
flux in Figure E-4 for the 13-week exposure time, where such an increase is generally indicated
for low flux values.)
This document is a draft for review purposes only and does not constitute Agency policy.
E-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
The early times would be important if, say, repeated episodic exposures were considered,
where adequate time has not elapsed for adaptive effects to take place. Such an exposure
scenario may be the norm in the human context. However, the contribution of the early-time
labeling data is minimized in the CUT cancer modeling since the LI was weighted by exposure
time. Because of the problems described above in incorporating the pulse-labeled data, the
sensitivity analysis will be restricted to only the continuous labeling data.
E.3.2.3. Site and Time Variability
In the remainder of this section, the factors that are considered in order to represent the
uncertainty and variability in the cell replication data when developing alternate dose-response
curves for afflux) will be elaborated. Figure E-2 (from Subramaniam et al., 2008) shows the
variability due to replicated animals, exposure times, and nasal sites in the continuous labeled
data obtained by Monticello et al. (1996). The ULLI data for individual animals were provided
by CUT. In this figure, log ocn versus site-specific flux are plotted for six sites and four exposure
times for four to six replicate animals in each case. (The mean ULLI over these replicates were
shown in Figure E-l for each site and time as a function of exposure concentration.) It needs to
be noted that these nasal sites differ considerably in the number of cells estimated at these
locations as shown in Table E-3. Each point in Figure E-2 represents data from a single site for a
single animal at a given time. For comparison, the afflux) in Conolly et al. (2003) is also
plotted in this figure at their averaged flux values (filled circles). For flux >9,340 pmol/mm2-
hour, Conolly et al. (2003) extrapolated this empirically derived aN(flux) by using a scheme
discussed in Appendix D (section D.5) on the upward extrapolation of cell replication rate. The
curves shown connecting the filled circles in the figure represent their linear interpolation (long
dashes) between the six points. Their linear extrapolation for flux value >9,340 pmol/mm2-hour
is also shown (short dashes). Note that the linear interpolation/extrapolation is shown
transformed to a logarithmic scale.
In Figures E-3, fitted dose-response curves are plotted for logio(a#) versus flux with
simultaneous confidence limits separately for each time point for two of the largest sites in Table
E-3 (ALM and PLM). Note that flux levels are different at each site. Simple polynomial models
in flux (as a continuous predictor), with time included as a factor (i.e., a class or indicator
variable, x, representing the effect of the ith time) were used as follows:
log(aw) = a + bxflux + cxflux2 + dxflux3 + Ti (9)
This document is a draft for review purposes only and does not constitute Agency policy.
E-l8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
2000 4000 6000 8000 10000 12000
Figure E-2. Logarithm of normal cell replication rate ocn versus
formaldehyde flux (in units of pmol/mm -hour) for the F344 rat nasal
epithelium.
Note: Values were derived from continuous unit length labeled data obtained by
Monticello et al. (1996) for four to six individual animals at all six nasal sites
(legend, sites as denoted in original paper) and four exposure durations (13, 26,
52, 78 weeks). Each point represents a measurement for one rat, at one nasal site,
and at a given exposure time. Filled red circles: aN(flux) used in Conolly et al.
(2003) plotted at their averaged flux values (see text for details). Long dashed
lines: their linear interpolation between points. Short dashed line: their linear
extrapolation for flux value >9,340 pmol/mm2-hour (see Figure D-l for full range
of extrapolation). Linear interpolation/extrapolation is shown with y-axis
transformed to logarithmic scale.
Source: Subramaniam et al. (2008).
This document is a draft for review purposes only and does not constitute Agency policy.
E-19 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table E-3. Variation in number of cells across nasal sites in the F344 rat
2
Nasal site
No. of cells
Anterior lateral meatus
976,000
Posterior lateral meatus
508,000
Anterior mid septum
184,000
Posterior mid septum
190,000
Anterior dorsal septum
128,000
Anterior medial maxilloturbinate
104,000
3
4 Note: Mean number of cells in each side of the nose of control animals.
5
6 Source: Monticello et al. (1996).
7
This document is a draft for review purposes only and does not constitute Agency policy.
E-20 DRAFT—DO NOT CITE OR QUOTE
-------
1
13 weeks
26 weeks
0 4000 8000 12000
Flux
a
o
CM .
¦
CO .
1
^ o
>~ *
**- -''6
LO .
1
0 4000 8000 12000
Flux
52 weeks
78 weeks
0 4000 8000 12000
Flux
0 4000 8000 12000
Flux
2
3
4 Figure E-3A. Logarithm of normal cell replication rate versus formaldehyde
5 flux with simultaneous confidence limits for the ALM.
6
7 Source: Subramaniam et al. (2008).
This document is a draft for review purposes only and does not constitute Agency policy.
E-21 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
13 weeks
0 2000 6000
Flux
52 weeks
0 2000 6000
Flux
26 weeks
0 2000 6000
Flux
78 weeks
0 2000 6000
Flux
Figure E-3B. Logarithm of normal cell replication rate versus formaldehyde
flux with simultaneous confidence limits for the PLM.
Source: Subramaniam et al. (2008).
The variability considered is that among animals and any measurement error as well as
any other design-related components of error. Simultaneous 95% confidence limits for log(aN)
were produced using Scheffe's method (Snedecor and Cochran, 1980). These 95% confidence
limits span a range of 0.96 in loglO(aN), or nearly a 10-fold range in median o,n. There is
additional dispersion in these data that does not appear in Figures E-2 and E-3; due to variation
in the number of cells per mm basement membrane, the ratio of ULLI/LI had a spread of
approximately ±25% (0.45 to 0.71, mean 0.60) among the eight observations considered in
Monticello et al. (1990). Thus:
This document is a draft for review purposes only and does not constitute Agency policy.
E-22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
1. As suggested by Table E-3, and Figures E-2 and E-3, the shape of afflux) in Conolly et
al. (2003) is therefore likely to be very sensitive to how oin is weighted and averaged over
site and time.
2. Averaging of sites could significantly affect model calibration because of substantial
nonlinearity in model dependence on ocn at the 10 and 15 ppm doses associated with high
cancer incidence.
3. Monticello et al. (1996) found a high correlation between tumor rate and the ULLI
weighted by the number of cells at a site. Therefore, considering these factors while
regressing aN against tissue dose would be important in the context of site differences in
tumor response.
4. A further complexity arises because of histologic changes and thickening that occurs in
the nasal epithelium over time in the higher dose groups (Morgan, 1997), factors that are
likely to affect estimates of local formaldehyde flux, uptake, and replication rates
(Subramaniam et al., 2008).
Figure E-l indicates that the time dependence in ULLI is significant. It would also be
useful to examine if the time dependence affects the results of the time-to-tumor modeling and if
early temporal changes in replication rate are important to consider because of the generally
cumulative nature of cancer risk. The time window over which formaldehyde-induced cancer
risk is most influenced is not known, but the time weighting used by Conolly et al. (2003)
assigns a relatively low weight to labeling observed at early times compared with those observed
at later time points. Finally, initiated cells are likely to be replicating at higher rates than normal
cells as evidenced in several studies on premalignant lesions (Coste et al., 1996; Dragan et al.,
1995; Rotstein et al., 1986). Therefore, LI data as an estimator of normal cell replication rate
would be most reliable at early times when the mix of cells sampled include fewer preneoplastic
or neoplastic cells.
The more relevant question, therefore, is whether afflux) derived by a TWA over all
sites (as carried out by Conolly et al. [2003]) has an effect on low-dose risk estimates. Given the
above uncertainties and variability not characterized in CUT (1999) or in Conolly et al. (2003), it
is important to examine whether additional dose-response curves that fit the cell replication data
reasonably well have an impact on estimated risk. Such sensitivity analyses are carried out in
the sections that follow. Clearly, a large number of alternative afflux) can be developed. In
conjunction with the other uncertainties, mainly the use of control data and alternative model
structures for initiated cell kinetics, the number of plausible clonal growth models to be
exercised soon require a prohibitively large investment of time. Therefore, detailed analyses
This document is a draft for review purposes only and does not constitute Agency policy.
E-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
were restricted to a select set of biologically plausible choices of curves for afflux), which
would allow the identification of a range of plausible risk estimates (MLEs and statistical
bounds).
E.3.2.4. Alternate dose-response curves for cell replication
Six alternative equations for o,n were developed by regression analysis of the Monticello
et al. (1996) ULLI data. The replicate data corresponding to the summary data presented in this
paper were kindly provided to EPA by CUT for further analyses. In each of these equations, oin
is expressed as a function of formaldehyde flux to nasal tissue (pmol/mm2-hour) and, in one
equation (eq 15) that explored time-dependence, the duration of exposure to formaldehyde in
weeks. All the graphs use flux/10,000 for the x-axis, and the y-axis expresses logio ocn.
One source of uncertainty in the cell proliferation dose response in Conolly et al. (2003)
is the large value of amax in the upward extrapolation (the cell replication rate corresponding to
the upper end of the flux range at 15 ppm exposure). The optimal value of amax was found by
Conolly et al. (2003) to be 0.0435 hour-1. As noted by the authors, an argument in support of
this value is that it corresponds to the inverse of the fastest cell cycle times found in the
literature. Since the model treats the induced replication rates as being time invariant, this means
that cells in the high-flux region(s) divide at the highest cell turnover rate ever observed
throughout most of an animal's life. This does not seem to be biologically plausible
(Subramaniam et al., 2008).
In the analysis, it was found that a 20% increase or decrease in the estimated value for
amax degraded the fit to the tumor incidence data considerably. Because of the interplay between
the parameters estimated by optimization, this sensitivity of the model to amax indicates that it is
necessary to examine if other plausible values of amax are also indicated by the data and to what
extent low dose estimates of risk are influenced by the uncertainty in its value. The need for
such an analysis is also indicated by Figure E-2. The value of amax (logioamax = -1.37) in the
modeling of Conolly et al. (2003) is roughly an order of magnitude greater than the values of
aN(flux) at the highest flux levels in this figure. If the data pooled over all sites and times are to
be used for aN(flux), then, based solely on the trend in aN(flux) in Figure E-2, it appears unlikely
that aN(flux) could increase up to this value of amax. Visually, these empirically derived data
collectively suggest that aN versus flux could be leveling off rather than increasing 10-fold.
Therefore, as an alternative to the approach taken in Conolly et al. (2003) of estimating amax via
likelihood optimization against the tumor data, regressions of the empirical cell replication data
were used to extrapolate afflux) outside the range of observation (recognizing the uncertainty
This document is a draft for review purposes only and does not constitute Agency policy.
E-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
and model dependence that still results from extrapolating well outside the range of observed
data).
In fitting dose-response curves to the cell replication data, a functional form was used
that was flexible to allow a variety of monotonic and non-monotonic shapes, with a parameter
that determined the asymptotic behavior of the dose-response function. This allowed the
extrapolation by only relying on the empirical cell replication data without using an adjustable
parameter estimated by fitting to the tumor data. However, the plausible asymptotes obtained
spanned a large range. In one case below, the asymptote suggested by the fit was judged to be
abnormally high. In this case, the versus flux curve was followed until the biological
maximum of amax (as given in Conolly et al. [2003]) was reached.
In three of the regression models below, the data were restricted to the earliest exposure
time (13 weeks) in Monticello et al. (1996) for which the cell proliferation rate (aN) could be
calculated. The interest in using only the 13-week exposure time arises from observations
(Monticello et al., 1996, 1991) that at later times there were more frequent and severe histologic
changes, which may have altered formaldehyde uptake and cell proliferation response.
Consequently, given that the data in Monticello et al. (1991) for times earlier than 13 weeks
could not be utilized as explained earlier, the 13-week responses might better represent
proliferation rates for use in a two-stage model of the cancer process than the rest of the
Monticello et al.(1996) data.
Second, the LI data showed considerable variation among nasal sites, which may be
related to the variation in tumor response among sites. Since the cell replication dose-response
curves used in the cancer model represent all of the sites, it was attempted to include this
variation by weighting the regression by the relative cell populations at risk at each of the sites.
This was carried out for some of the models as stated below. The following models (denoted
N1-N6), shown in Figure E-4, have been included in addition to using the hockey stick- and J-
shaped curves in Conolly et al. (2003). Applicable equations are as follows:
Nl: Quadratic; monotone increasing in flux, derived from fit to all of the Monticello et al. (1996)
ULLI data.
aN = Exp{-2.015 - 6.513 x Exp[— (6.735* 10 4 x flux)2]} (10)
N2: Linear-quadratic; decreasing in flux for small values of flux, derived from fit to all of the
Monticello et al. (1996) ULLI data.
This document is a draft for review purposes only and does not constitute Agency policy.
E-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
CO
co
0.0
0.2
0.4
0.6
0.8
1.0
x
Figure E-4, Nl. Various dose-response modeling of normal cell replication
rate.
Note: See text for definitions of N1-N6. Nl: Quadratic; monotone increasing in
flux, derived from fit to all of the Monticello et al. (1996) ULLI data.
Iog10(alpha) = -2.565 -0.987 * exp{+2.188*X-(2.162*X)A2 }
O
CO
LO
CO
o
0.0
0.2
0.4
0.6
0.8
1.0
weighted mean flux/10,000
Figure E-4, N2: Various dose-response modeling of normal cell replication
rate.
Note: See text for definitions of N1-N6. N2: Linear-quadratic; decreasing in flux
for small values of flux, derived from fit to all of the Monticello et al. (1996)
ULLI data.
This document is a draft for review purposes only and does not constitute Agency policy.
E-26 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Time = 13 weeks
CD
CO
_Q_
CO
c
CO
a;
E
-o
a;
D)
1
0.0
0.2
0.4
0.6
0.8
1.0
weighted mean flux/10,000
Figure E-4, N3. Various dose-response modeling of normal cell replication
rate.
Note: See text for definitions of N1-N6. N3: Linear-quadratic; decreasing in flux
for small values of flux, derived from fit to the 13-week Monticello et al. (1996)
ULLI data, using average flux over all sites for a given ppm exposure and
weighting regression by estimates of the numbers of cells at each of five sites.
Time = 13 weeks
CD
co
o
0.0
0.2
0.4
0.6
0.8
1.0
1.2
flux/10,000
Figure E-4, N4. Various dose-response modeling of normal cell replication
rate.
Note: See text for definitions of N1-N6. N4: Quadratic; monotone increasing in
flux, derived from unweighted fit to 13-week Monticello et al. (1996) ULLI data.
This document is a draft for review purposes only and does not constitute Agency policy.
E-27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
Time = 13 weeks
CD
co
0.0
0.2
0.4
0.6
0.8
1.0
weighted mean flux/10,000
Figure E-4, N5. Various dose-response modeling of normal cell replication
rate.
Note: See text for definitions of N1-N6. N5: Linear-quadratic-cubic; initially
increasing slightly with increasing flux, then decreasing slightly, and finally
increasing, derived from fit to 13-week Monticello et al. (1996) ULLI data, using
average flux over all sites for a given ppm exposure and weighting regression by
estimates of the numbers of cells at each of five sites.
This document is a draft for review purposes only and does not constitute Agency policy.
E-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
All Sites, - Time + 2nd order in Flux
Time = 52
Time = 78
0.0 0.2 0.4 0.6 0.8 1.0 1.2
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Figure E-4, N6. Various dose-response modeling of normal cell replication
rate.
Note: See text for definitions of N1-N6. N6: Linear-quadratic-cubic; initially
increasing slightly with increasing flux, then decreasing slightly, and finally
increasing, derived from fit to all Monticello et al. (1996) ULLI data, using weeks
of exposure as a covariate. In this model, time was a regression (continuous)
predictor, not a class variable, and its coefficient represents the decrease in logio
ocn per week of exposure time.
aN = Exp{-5.906-2.272 x Exp [2.188* 10'4 x flux - (2.162* 1CT4 x flux)2]} (11)
N3: Linear-quadratic; decreasing in flux for small values of flux, derived from fit to the 13-week
Monticello et al. (1996) ULLI data, using average flux over all sites for a given ppm exposure
and weighting regression by estimates of the numbers of cells at each of five sites.
aN = Exp{-5.274 - 2.792 x Exp/1.407* 10 4 x flux - (1.986x 1(T4 x flux)2]} (12)
This document is a draft for review purposes only and does not constitute Agency policy.
E-29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
N4: Quadratic; monotone increasing in flux, derived from unweighted fit to 13-week Monticello
et al. (1996) ULLI data.
aN = Exp{-3.858 - 4.809 * Exp[- (9.293* 1(T5 * flux)2]} (13)
N5: Linear-quadratic-cubic; initially increasing slightly with increasing flux, then decreasing
slightly, and finally increasing, derived from fit to 13-week Monticello et al. (1996) ULLI data,
using average flux over all sites for a given ppm exposure and weighting regression by estimates
of the numbers of cells at each of five sites.
aN = Exp{-5.488 - 2.755 x /-jc/;/ 7.808 * 10 5 x flux + (2.349 *10 4 x flux)2
-(2.166x1 ft* x flux)3]} (14)
N6: Linear-quadratic-cubic; initially increasing slightly with increasing flux, then decreasing
slightly, and finally increasing, derived from fit to all Monticello et al. (1996) ULLI data, using
weeks of exposure as a covariate. In this model, time was a regression (continuous) predictor,
not a class variable, and its coefficient represents the decrease in logio aN per week of exposure
time.
aN = Exp{7.785* 10 3 * (weeks) - 5.722 - 2.501 * Exp[1.103*10 4 * flux
- (7.223* 10'5 * flux)2 (1.575*10 4 * flux)3]} (15)
Further details on the above regressions are provided in the appendix. These regressions
of the cell replication data as well as the hockey-stick and J-shaped curves used by Conolly et al.
(2003) (shown in Figure D-l, Appendix D) are used next as inputs to the clonal growth model
for cancer.
E.3.3. Uncertainty in Model Specification of Initiated Cell Replication and Death
E.3.3.1. Biological Inferences of Assumptions in Conolly et al. (2003)
The results of a two-stage MVK model are extremely sensitive to the values for initiated
cell division (ai) and death (J3i) rates, particularly in the case of a sharply rising dose-response
curve as in the case of formaldehyde. The pool of cells used for obtaining the available LI data
(Monticello et al., 1996, 1991) consists of largely normal cells with perhaps increasing numbers
of initiated cells at higher exposure concentrations. As such there is no way of inferring the
This document is a draft for review purposes only and does not constitute Agency policy.
E-30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
division rates of initiated cells in the nasal epithelium, either spontaneous (baseline) or induced
by exposure to formaldehyde, from the available empirical data. Conolly et al. (2003)
considered ai(flux) as a function of afflux) as given by eq 2 in Appendix D. As shown in
Figure D-l (Appendix D), ai is estimated in Conolly et al. (2003) to be very similar to ocn. That
is, with eq 2 assumed to relate ai(flux) to afflux), a J- or hockey-shaped dose-response curve
for aN(flux) results in a J or hockey shape for ai(flux).
The J shape for the TWA afflux) in Conolly et al. (2003) could plausibly be explained,
as suggested by the examples in Conolly and Lutz (2004), by a mathematical superposition of
dose-response curves describing the effects of the inhibition of cell replication by the formation
of DPXs (Heck and Casanova, 1999) and cytotoxicity-induced regenerative replication (Conolly,
2002). However, as explained earlier, there is considerable uncertainty and variability, both
qualitative and quantitative, in the interpretation of the LI data and in the derivation of normal
cell replication rates from the ULLI data. While the TWA values of ULLI indicate a J-shaped
dose response for some sites, as also concluded by Gaylor et al. (2004), this is not consistently
the case for all exposure times and sites as discussed earlier. Notwithstanding this uncertainty
variability, and in the absence of data, the following essential questions have a significant impact
on risk predictions and need resolution if the model structure in eq 2 is to be used in a
biologically based (or motivated) sense to predict risk outside observable data:
• Should mechanisms that might explain a J-shaped dose response for normal cell
replication or a cytotoxicity-driven threshold in dose response (as indicated by a hockey-
stick-shaped curve) be expected to prevail also for initiated cells?
• Would the formaldehyde flux at which the cell replication dose-response curve rises
above its baseline be similar in value for both normal and initiated cells as inferred by the
CUT model in Figure D-l?
The next critical assumption was that made for J3i (the death rate of initiated cells),
namely, Pi(flux) = afflux) (eq 3). In Subramaniam et al. (2008), the rationale for this
assumption in Conolly et al. (2003) is explained by assuming formaldehyde to be equally
cytotoxic to initiated and normal cells (since the mechanism is presumed to be via its general
chemical reactivity). In essence, this assumption brings the cytotoxic action of formaldehyde to
bear strongly on the parameterization of the CUT model.
There are no data to evaluate the strength of these assumptions, so Subramaniam et al.
(2008) studied the plausibility of various inferences that arise as a result of these assumptions.
These inferences are only briefly listed here (see the paper for further discussion).
This document is a draft for review purposes only and does not constitute Agency policy.
E-31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
• For flux <27,975 pmol/mm2-hour, ai > aN (Figures D-l & D-2 of Appendix D).
Qualitatively, this is in line with data on epithelial and other tissue types with or without
exposure to specific chemicals.
• For higher flux levels, however, the model indicates ai < aN (Figure D-2). There are no
data to shed further light on this inference.
• At these higher flux levels, initiated cells in the model die at a faster rate than they
divide, indicating the extinction of initiated cell clones in regions subject to these flux
levels. There are no data indicating formaldehyde to have this effect.
In evaluating these inferences, Subramaniam et al. (2008) point to various data that
indicate that initiated cells represent distinctly different cell populations (from that of normal
cells) with regard to proliferation response (Ceder et al., 2007; Bull, 2000; Schulte-Hermann et
al., 1997; Coste et al., 1996; Dragan et al., 1995), have excess capacity to clear formaldehyde
and, in general, are considerably more resistant to cytotoxicity (such a resistance is manifested
variably as decreased ability of the toxicant to induce cell death or to inhibit cell proliferation
compared to corresponding effects in normal cells), and may already have altered cell cycle
control; thus, the influence of formaldehyde on apoptosis likely differs between normal and
initiated cells.
As concluded in Subramaniam et al. (2008), taken together, there is much data to suggest
that inferring ai < aN at cytotoxic formaldehyde flux levels is problematic and that death rates of
initiated cells are likely to be very different from those of normal cells. In the absence of data to
indicate that eq 2 and eq 3 (in Appendix D) are biologically reasonable approaches to link the
kinetics of initiated cells with those of normal cells, alternate model structures other than those
represented by these relationships considered by Conolly et al. (2003) need to be explored, given
that the two-stage model is extremely sensitive to ai and J3i. Such an evaluation needs to
primarily explore if the assumptions in eq 2 and eq 3 significantly impact the intended use of the
model, namely extrapolation to low-dose human cancer risk and the calculation of an upper
bound on human risk. Any such alternate model structure needs to provide a good fit to the
time-to-tumor data.
This document is a draft for review purposes only and does not constitute Agency policy.
E-32 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.3.2. Plausible Alternative Assumptions for ai and /?/
Therefore, in the additional sensitivity analysis presented here, initiated cell kinetics are
considered to be independent of normal cells, and initiated cell proliferation cannot take a J
shape (motivated by the consideration that lower-than-baseline turnover rate represents an
increased amount of DNA repair taking place, which may not be consistent with impaired DNA
repair in initiated cells).
Thus, two alternatives were considered to eq 2 for «/(flux):
II: a7 = j! x [1 + exp(y2/y3)]/{1 + exp[-(flux- y2) / y3]} (16)
12: aI = max[aI(eq.Il),aNBasai] (17)
Here yi, }'¦>, and ys are parameters estimated by fitting the cancer model to the rat bioassay
data. In eq 16, a/ increases monotonically with flux from a background level of yi asymptotically
up to a maximum value of y} x [1 + Exp(y2 / ys)]. The choice of this functional form in eq 16
and eq 17 was considered in order to be parsimonious while at the same time allowing for a
flexible shape to the dose-response curve. The sigmoidal curve allows for the possibility of a
slow rise in the curve at low dose and an asymptote.
Equation 17 is a modification of equation 16 that restricts the rate of division of initiated
cells to be at least as large as the spontaneous division rate of unexposed normal cells. There is
evidence to suggest (e.g., in the case of liver foci) that initiated cells have a growth advantage
over normal cells, with or without exposure to specific chemicals (Ceder et al., 2007; Grasl-
Kraupp et al., 2000; Schulte-Hermann et al., 1999; Coste et al., 1996; Dragan et al., 1995).
In addition, in most runs, an upper bound {ahigh) is selected for both aN and a/. This value
is assumed to represent the largest biologically plausible rate of cell division. Following Conolly
et al. (2003), in most cases augh is set equal to 0.045 hours-1. If a value of a/ or aN computed
using one of the above formulas exceeded ahigh, the value of ahigh was used in the computation
rather than the value obtained by using the formula.
As noted above, Conolly et al. (2003) set the rate of death for intermediate cells, /?/, equal
to the division rate of normal cells, fi/ = aN. On the other hand, apoptotic rates and cell
proliferation rates are thought to be coupled (Schulte-Hermann, 1999; Moolgavkar, 1994), so
that death rates of initiated cells would rise concomitantly with an increase in their division rates
(Grasl-Kraupp et al., 2000; Schulte-Hermann et al., 1999). Therefore, as an alternative to the
Conolly et al. (2003) formulation, it is assumed that the death rate of intermediate cells is
proportional to the division rate of intermediate cells.
This document is a draft for review purposes only and does not constitute Agency policy.
E-33 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Pl — Kp x al
(18)
where the constant of proportionality, Kp, is an additional parameter to be estimated by
optimization against the tumor incidence data. Such an assumption has also been made by other
authors (Luebeck et al., 2000, 1995; Moolgavkar et al., 1993).
Since most of the SCCs in the rat bioassays occurred in rats exposed to the highest
formaldehyde concentration (15 ppm), the data from this exposure level have a big impact on the
estimated model parameters. In most runs that incorporated the 15 ppm data, the model
appeared, based on inspection of the KM plots, to fit the 15 ppm data quite well but to fit the
lower exposure data less well. Because of the high level of necrosis occurring at 15 ppm, it is
possible that the data at this exposure may not be particularly relevant to modeling the sharp
upward rise in the dose response at 6 ppm. Furthermore, the principal interest is in the
predictions of the model at lower levels to which human populations may be exposed.
Consequently, in order to improve the fit of the model at lower exposures, some of the
alternative models were constructed with the 15 ppm data omitted.
E.3.4. Results of Sensitivity Analyses on cun, oci, and Pi
E.3.4.1. Further Constraints
The number of models that might be constructed if all the possibilities listed above for
ccn, cci, and Pi are to be tried in a systematic manner clearly become exponential and daunting.
(Optimally, it would have been desirable to elucidate the role of a specific modification while
keeping others unchanged to determine risk.) Therefore, in order to carry out a viable sensitivity
analysis while at the same time examining the plausible range of risks resulting from variations
in parameters and model structures, various uncertainties were combined in any given
simulation. By using the constraints described above (eq 10-17 and associated text) for ai, J3i,
and ocn, 19 models were obtained that provided similarly good fits to the time-to-tumor data
(which in some cases contained only five dose groups).
However, for many of these models, the optimal ai(flux) displayed a threshold in flux
even when the model utilized for aN(flux) was a monotonic increasing curve without a threshold
(i.e., model N4 for aN in Figure E-4). Indeed, if a thresholded dose-response curve was
plausible for ai based on arguments of cytotoxicity, then a threshold is all the more plausible for
aN, and such models are removed from consideration.
Secondly, the basal value of ai was required to be at least as large as the basal value of
aN. Another constraint was placed on the baseline initiated cell replication rate. In the absence
This document is a draft for review purposes only and does not constitute Agency policy.
E-34 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
of formaldehyde exposure, ai was not allowed to be greater than two or four times ocn, even if
such models described the tumor data, including the control data, very well. There are some data
that suggest that baseline initiated cells have a small growth advantage over normal cells, so a
huge advantage was thought to be biologically less plausible.
E.3.4.2. Sensitivity of Risk Estimates for the F344 Rat
Figure E-5 contains plots of the MLE of additional risk computed for the F344 rat at
formaldehyde exposures of 0.001, 0.01, 0.1, and 1 ppm for eight models. Two log-log plots are
provided. For those models for which the estimates of additional risk are all positive, the
additional risks are plotted (panel A), and, for those for which estimates of additional risk are
negative, the negatives of additional risks are plotted (panel B). Only five dose groups were
considered (i.e., 15 ppm data omitted) for models 8, 5, 15, and 16. Figure E-6 shows the dose-
response curves for ocn and ai for these eight cases (panels A and B corresponding to those in
Figure E-5). The primary results are as follows:
1. Among the models considered, negative values for additional risk can arise only in
models in which the dose response for normal cells is J shaped. Thus, all of the models
with negative dose responses for risk have J-shaped dose responses for normal cells.
However, the converse is not necessarily true as may be noted from model 8. This model
has both a positive dose response for risk and a J-shaped dose response for normal cells.
In this case, the strong positive increase in response of initiated cells at low dose was
sufficient to counteract the negative response of normal cells.
2. The risk estimates predicted by the different models span a very large range for doses
below which no tumors were observed. This result points to large uncertainties in model
specification (how to relate the kinetics of normal and initiated cells) as well as in
parameter values. As mentioned above, the analysis does not attempt to separate the
influence of the different sources of uncertainty, so this range also incorporates the
uncertainty arising from the use of different control data and that due to amax.
3. At the 10 ppb (0.01 ppm) concentration, MLE risks range from -4.0 xlO-6 to +1.3 xlO~7
At this dose, models that gave only positive risks resulted in a five orders of magnitude
risk range from 1.2 x 10~12 to 1.3 x 10 7, while narrowing to a four orders of magnitude
risk range from 1.2 x 10~10 to 1.3 x 10 6 at the 0.1 ppm level. This narrowing continues as
exposure concentration increases, and the curves coalesce to substantially similar values
at 6 ppm and above (not shown).
This document is a draft for review purposes only and does not constitute Agency policy.
E-3 5 DRAFT—DO NOT CITE OR QUOTE
-------
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
JX.
(/)
"C 1.E-09
T3
d)
^ 1.E-10
<
1.E-11
1.E-12
1.E-13
1.E-14X
1.E-15
0.001
1
2
3
4
5
~
O
A
~
O
A
~
O
A
X
,\fide *¦=>
X
O model 8
~ model 15
A model 16
XmodeM7j)
X
As in Conoiiy et ai. (2003), Hockey
stick aN and a, but using concurrent
0.01 0.1
formaldehyde exposure cone (ppm)
Figure E-5A. BBDR models for the rat—models with positive added risk.
Note: All four models provide "similar" fits to tumor data (see text).
This document is a draft for review purposes only and does not constitute Agency policy.
E-36 DRAFT—DO NOT CITE OR QUOTE
-------
0.001
-1.E-08 H—
-1.E-07
in
¦c
"g-I.E-06
<
LU
>
<-1E-05
0
LU
-1.E-04
-1.E-03
Exposure cone (ppm)
0.01 0.1
w,
X
'a,..,
"®h.
* fn
'Of _
x
O model 3
~ model 4
A model 5
model 13"^
X
As in Conolly et al. (2003), J-shape aN
and a, but using Inhalation NTP controls
added to concurrent
1
2
3
4
5
6
Figure E-5B. BBDR rat models resulting in negative added risk.
Note: All four models provide "similar" fits to tumor data (see text).
This document is a draft for review purposes only and does not constitute Agency policy.
E-37 DRAFT—DO NOT CITE OR QUOTE
-------
1
«N
-""I
0.01 -
1E-3-
Model 8
1E-4
0
10000
20000
30000
40000
«N
S
c
o
m o.oi ^
>
2-
d
1E-3-
Model 15
1E-4
0
10000
20000
30000
40000
Rux (pmole/mm2/hr) Rux (pmole/mm2/hr)
AM 8 models in A & B provide
similar fits to tumor data
m o.oi ^
>
d
1E-3-
"On
Model 16
0 10000 20000 30000 40000
Rux (pmole/mm2/hr)
1 Hockey stick aN and a, with concurrent
^ controls
0.1 - Ct|
Model 17
-i—i—i—i—i—i—i—i—i—i—i—i—i—i
10000 20000 30000 40000
Flux (pmole/mm2/hr)
2
3 Figure E-6A. Models resulting in positive added rat risk: Dose-response for
4 normal and initiated cell replication
5
6
This document is a draft for review purposes only and does not constitute Agency policy.
E-3 8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
55 0.01 -=
w 0.01 -
>
S
1E-3-
Model 3
0 10000 20000 30000 40000
Flux (pmole/mm2/hr)
«N
---CC|
Model 4
10000 20000 30000 40000
Flux (pmole/mm2/hr)
All 8 models in A & B provide
similar fits to tumor data
05 0.01 -
J-shape aN, a, in Conolly (2003) with
Inhalation NTP + concurrent controls
Model 5
10000 20000 30000 40000
Flux (pmole/mm2/hr)
c/j 0.01 -
Mode 13
10000 20000 30000 40000
Flux (pmole/mm2/hr)
Figure E-6B. Models resulting in negative added rat risk: Dose-response for
normal and initiated cell replication
4. There does not seem to be any systematic effect on additional risk that depends on
whether the 15 ppm data are included in the analysis.
5. For all of the models except models 13 and 17 in Figures E-5 & E-6, the additional risk
varies substantially linearly with exposure at low exposures between 0.001 and 1.0 ppm
(departing only to a small extent from linearity between 0.1 and 1.0 ppm). Models 13
and 17 (the models in Conolly et al. [2003] except for different control data being used)
show a quadratic dependence.
The various model choices presented in Figure E-5 all provided equally good fits to the
time-to-tumor data although within the context of a significant qualification. It was not possible
This document is a draft for review purposes only and does not constitute Agency policy.
E-39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
to simply use the log-likelihood values as a means of comparing the goodness-of-fit to the tumor
incidence data across these model choices. This is because many of the model choices differed
in the number of doses or in the number of control animals that were used, so the fits were
compared across such models only visually. Within model choices where such a comparison did
not pose a problem, the log-likelihood values did not differ statistically significantly.
Wherever results from the BBDR modeling are discussed, values of added risk, as
opposed to extra risk, are reported. This is purely for convenience in interpretation. Because of
the low background incidence, these values are only negligibly different from the corresponding
extra risk estimate. The final risk (or unit risk) estimates provided in this document are based on
extra risk estimates.
E.3.4.3. MOA Inferences Revisited
The ratio K.MU///\basai represents the added fractional probability of mutation per cell
generation (ju^- //Nbasai)//
-------
1 tumors were observed), and 6 ppm (the lowest dose where tumors were observed) with the MLE
2 risk estimates at these doses. In both cases, these intervals were quite narrow compared with the
3 differences in risk predicted by different models in Figure E-5. This suggests that model
4 uncertainty is of more consequence in the formaldehyde animal model than is statistical
5 uncertainty. We also estimated confidence bounds using the bootstrap method for select models,
6 and determined that these estimates were in agreement with the bounds calculated using the
7 profile likelihood method. These results are not presented here. We return to the calculation of
8 confidence limits when determining points of departure (PODs).
9
10 Table E-4. Comparison of statistical confidence bounds on added risk for
11 two models
12
Dose (ppm)
Model
Lower
bound
MLE
Upper
bound
0.001
Model 15
4.4 x l(T9
1.3 x l(T8
1.6 x l(T8
Model 17
1.2 x l(T14
1.2 x l(T
14
1.3 x l(T14
0.1
Model 15
4.5 x l(T7
1.3 x 10"6
1.7 x 10"6
Model 17
1.2 x l(T10
1.2 x i(T
10
1.3 x l(T10
6
Model 15
1.8 x l(T2
2.1 x 1(T2
2.3 x l(T2
Model 17
1.3 x l(T2
1.8 x l(T2
3.0 x l(T2
13
14
15 In conclusion, it is demonstrated that the different formaldehyde clonal growth models
16 can fit the data about equally well and still produce considerable variation in additional risk and
17 biological inferences at low exposures. However, even with these large variations, the highest
18 MLE added risk for the F344 rat is only of the order of 10 6 at 0.1 ppm. Thus, with regard to
19 calculating a reasonable upper bound that includes model and statistical uncertainty, the relevant
20 question is whether the range arising out of uncertainties in the rat model amplifies when
21 extrapolated to the human. Thus, in Appendix F, the human model in Conolly et al. (2004) will
22 be examined.
This document is a draft for review purposes only and does not constitute Agency policy.
E-41 DRAFT—DO NOT CITE OR QUOTE
-------
This page intentionally left blank.
This document is a draft for review purposes only and does not constitute Agency policy.
E-42 DRAFT—DO NOT CITE OR QUOTE
-------
Appendix F
i
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
APPENDIX F
SENSITIVITY ANALYSIS OF BBDR MODEL FOR FORMALDEHYDE INDUCED
RESPIRATORY CANCER IN HUMANS
F.l. MAJOR UNCERTAINTIES IN THE FORMALDEHYDE HUMAN BBDR MODEL
Subsequent to the BBDR model for modeling rat cancer, Conolly et al. (2004) developed
a corresponding model for humans for the purpose of extrapolating the risk to humans estimated
by the rat model. Also, rather than considering only nasal tumors, it is used to predict the risk of
all human respiratory tumors. The human model for formaldehyde carcinogenicity (Conolly et
al., 2004) is conceptually very similar to the rat model and follows the schematic in Figure 5-11
in Chapter 5. The model structure, notations, and calibration are described in Appendix D.
Unlike the sensitivity analysis of the rat modeling where a number of issues were examined, a
much more restricted analysis will be presented here for the sake of brevity. A more extensive
analysis was carried out initially that carried forward several of the rat models to the human, and
the lessons learned from those exercises are in agreement with the more restricted presentation
that follows. Table F-l lists the major uncertainties and assumptions in the human extrapolation
model in Conolly et al. (2004).
This document is a draft for review purposes only and does not constitute Agency policy.
F-2 DRAFT—DO NOT CITE OR QUOTE
-------
Table F-l. Summary of evaluation of major uncertainties in CUT human BBDR model
Assumptions, approach, and
characterization of input data"
Rationale in Conolly et al. (2003) or
CUT (1999)
EPA uncertainty evaluation
Further elaboration
Cell division rates derived from rat
labeling data assumed applicable for
human (except for assuming
different fraction of cells with
replicative potential).
No equivalent LI data for human or
guidance in extrapolating cell division
rate across species.
Enzymatic metabolism plays a role in mitosis.
Therefore, we expect interspecies difference in cell
division rate. Basal cell division rates in humans
expected to be much more variable than in laboratory
animals.
Subramaniam et al.
(2008)
Development of PBPK model for
DPX concentration in human
respiratory lining.
See text (Chap 3)
See text (Chap 3)
Chap 3; Conolly et
al. (2000);
Subramaniam et al.
(2008); Klein et al.
(2009)
Anatomically realistic representation
of nasal passages.
Reduces uncertainty (over default
calculation carried out by averaging dose
over entire nasal surface).
Computer representation pertains to that of one
individual (Caucasian male adult). Considerable
interindividual variability in nasal anatomy.
Susceptible individuals even more variable.
Kimbell et al. (2001a,
b); Subramaniam et
al. (2008, 1998)
KMU/|i\|:i(IS(l is species invariant
(used to estimate human).
Human cells are more difficult to
transform than rodent, both
spontaneously and by exposure to
formaldehyde.
|i\i;i(is:i is 0 when concurrent controls or inhalation NTP
controls in time frame of concurrent bioassays are
used. Leads to infinitely large KMU for human.
Subramaniam et al.
(2007); Crump et al.
(2009, 2008).
Conservative assumptions were
made. Results are conservative in
the face of model uncertainties.
1) Hockey-stick dose-response for aN was
included even though TWA indicated
J-shape.
2) Overall respiratory tract cancer
incidence data for human baseline rates
were used.
3) Risk was evaluated at statistical upper
bound of the proportionality parameter
relating DPXs to the probability of
mutation.
CUT result cannot be characterized as conservative in
the face of model uncertainties and as a plausible upper
bound on human risk. Human model is unstable.
Conolly et al. (2004);
Subramaniam et al.
(2007); Crump et al.
(2009, 2008).
aAssumptions in this table are in addition to those listed for the BBDR model for the F344 rat.
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.2. SENSITIVITY ANALYSIS OF HUMAN BBDR MODELING
Crump et al. (2008) carried out a limited sensitivity analysis of the Conolly et al. (2004)
human model. This analysis was limited to evaluating the effect on the human model of the
following. These evaluations have been the subject of some debate in the literature and in
various conferences (Conolly, 2009; Conolly et al., 2009, 2008; Crump et al. 2009).
1. The use of the alternative sets of control data for the rat bioassay data that were
considered in the sensitivity analysis of the rat model (Subramaniam et al., 2007).
2. Minor perturbations in model assumptions regarding the effect of formaldehyde on the
division and death rates of initiated cells (ai, J3i). Now, recall from the description of the
structure of the human model that one (of the two) adjustable parameter in the expression
for the human ai was determined from the model fit to the rat tumor incidence data while
the second parameter was determined from background rates of cancer incidence in the
human. Therefore, variations considered in ai were constrained to only those that (a) did
not meaningfully degrade the fit of the model to the rat tumor incidence data and (b) were
in concordance with background rates in the human. Crump et al. (2008) also evaluated
these variations with respect to their biological plausibility. The sensitivity analysis on
assumed initiated cell kinetics was thought to be particularly important since there were
no data to even crudely inform the kinetics of initiated cells for use in the models, even in
rats, and the two-stage clonal expansion model is very sensitive to initiated cell kinetics
(Gaylor and Zheng, 1996; Crump, 1994a, b).
Crump et al. (2008) note that, since the purpose of their analysis was to carry out a
sensitivity analysis, in order to illustrate certain points, only risks to the general U.S. population
from constant lifetime exposure to various levels of formaldehyde under the Conolly et al.
(2004) environmental scenario (8 hours/day sleeping, 8 hours/day sitting, and 8 hours/day
engaged in light activity) are considered. Fits based on the hockey-stick and J-shape models
were identical, and, of the three estimated parameters (|ibasal, multb, and D), only the estimate
of |ibasal differed between the two models.
F.2.1. Effect of background Rates of Nasal Tumors in Rats on Human Risk Estimates
Crump et al. (2008) quantitatively evaluated the impact of different control groups on
estimates of additional human risk as follows:
This document is a draft for review purposes only and does not constitute Agency policy.
F-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
1. Concurrent controls plus all NTP controls:, the same as used by Conolly et al. (2004);
2. Concurrent controls plus controls from NTP inhalation studies;
3. Only concurrent controls;
4. Each set of control data was applied with both the J shape and hockey-stick models in
Conolly et al. (2004) for afflux) and ai(flux) for a total of six analyses;.
5. Uncertainties associated with oin or ai are not addressed. Parameters amax, multfc, and
KMU were estimated in exactly the same manner as in Conolly et al. (2004).
Crump et al. (2008) present the following dose-response predictions of additional risk in
humans from constant lifetime exposure to various levels of formaldehyde arising from
exercising the above six cases. Their plots are reproduced in Figure F-l, where the
corresponding curves based on Conolly et al. (2004) are also shown for comparison.
The lowest dotted curve in Figure F-l represents the highest estimates of human risk
developed by Conolly et al. (2004). This resulted from use of the hockey-stick model for cell
division rates in conjunction with the statistical upper bound for the parameter KMU. As
indicated by the downward block arrows in the figure, their corresponding estimates based on
the J-shape model were all negative for exposures below 1 ppm.
Consider next the solid curves in the figure, which show predicted MLE added risks that
were positive and less than 0.5. Crump et al. (2008) next examined the added risk obtained
when the MLE estimate of (KMU/fibasai) in these cases is replaced by the 95% upper bound of
this parameter ratio. The upper bound risk estimates in Conolly et al. (2004) were calculated in a
similar manner (but using all NTP historical controls). Except for minor differences, risk
estimates corresponding to such an upper bound and using all NTP controls were very similar in
the two efforts (Crump et al., 2008; Conolly et al., 2004).
Figure F-l shows that the choice of controls to include in the rat model can make an
enormous difference in estimates of additional human risk. For the J-shaped model for cell
replication rate both estimates based on the MLE and those based on the 95% upper bound on
KK4UIfibasai are negative for formaldehyde exposures below 1 ppm. However, when only
concurrent controls are used in the model in Crump et al. (2008), the MLE from the J-shape
model is positive and is more than three orders of magnitude higher than the highest estimates
obtained by Conolly et al. (2004). Using only concurrent controls, estimates based on the 95%
upper bound on KMUI/ibasai are unboundedly large (block arrows at the top of the figure). For
the hockey-stick shaped model for cell replication rate, when all NTP controls are used, the
estimates based on the MLEs are zero for exposures less than about 0.5 ppm. If only inhalation
controls are added, the MLEs are about seven times larger than the Conolly et al. (2004) upper
This document is a draft for review purposes only and does not constitute Agency policy.
F-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
bound estimates, and the estimates based on the 95% upper bound on KMUI/ibasai are about 50
times larger than the Conolly et al. (2004) estimates. If only concurrent controls are used, both
the MLE estimates and those based on the 95% upper bound on KMU/ubasai are unboundedly
large.
Figure F-l. Effect of choice of NTP bioassays for historical controls on
human risk.
Note: Estimates of additional human risk of respiratory cancer by age 80 from
lifetime exposure to formaldehyde are obtained by using different control groups
of rats.
Source: Crump et al. (2008).
F.2.2. Alternative Assumptions Regarding the Rate of Replication of Initiated Cells
For the human model, Conolly et al. (2004) made the same assumptions for relating
ai(flux) and Pi(flux) to afflux) as in their rat model (Conolly et al., 2003). That is, these
quantities were related by using eq 2 and eq 3. As discussed in the context of the rat modeling,
by extending the shape of these curves to humans, the authors' model brings the cytotoxic action
of formaldehyde to bear strongly on the parameterization of the human model as well.
Hockey, Concurrent Controls, MLE and 95% UB;
0.1 J-Shape, Concurrent Controls, 95% UB
Hockey, Inh. NTP
Controls, 95% UB
0.01 -
: J-Shape, Concurrent
1E-3 .Controls, MLE _
Hockey, All NTP
Controls, 95% UB,
1 E-4
•Hockey, Inh. NTP
IE_5 -.Controls, MLE , -
Conolly et al. (20C4),
Hockey UB /
J-Shape, All NTP Controls, MLE and 95% UB;
J-Shape, Inh. NTP Controls, MLE and 95% IpB;
Conolly et al. (2004) J-Shape UB
Hockey,
All NTP
Controls;
MLE
Formaldehyde Exposure (ppm)
This document is a draft for review purposes only and does not constitute Agency policy.
F-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
In the sensitivity analyses of the rat modeling, it was concluded that other biologically
plausible assumptions for ai and J3i resulted in several orders of magnitude variations in the low
dose risk relative to those obtained by models based on the assumptions in Conolly et al. (2003)
but that the highest risks were nonetheless of the order of 10 6 at the 10 ppb level. This section
examines how these uncertainties in the rat model propagate to the human model.
Crump et al. (2008) made minor modifications to the assumed division rates of initiated
cells in Conolly et al. (2004), while all other aspects of the model and input data were kept
unchanged. Two alternatives were considered for each of the J-shape and hockey-stick models.
Figure F-2 shows the hockey-stick model for initiated cells in rats. In the first modification to
the hockey-stick model (hockey-stick Mod 1), rather than having a threshold at a flux of
1,240 pmol/m2-hour, the division rate increases linearly with increasing flux until the graph
intersects the original curve at 4,500 pmol/m2-hour, where it then assumes the same value as in
the original curve for larger values of flux. The second modification (hockey-stick Mod 2) is
similar, except the modified curve intersects the original curve at a flux of 3,000 pmol/m2-hour.
Mod 1
0.05 n
.0010
Mod 2
0.04
.0005
0.03
.0000
1000 2000 3000 4000 5000
03
sz
Q.
CO
0.02
0.01
0.00
0
10000
20000
30000
40000
Formaldehyde Flux [pmol/(m2-h)]
Figure F-2. Conolly et al. (2003) hockey-stick model for division rates of
initiated cells in rats and two modified models.
Source: Crump et al. (2008).
Figure F-3 shows the rat J-shape model for initiated cells. In the first modification to this
dose response (J-shape Mod 1), rather than having a J shape, the division rate of initiated cells
remains constant at the basal value until the original curve rises above the basal value and has
This document is a draft for review purposes only and does not constitute Agency policy.
F-7 DRAFT—DO NOT CITE OR QUOTE
-------
1 the same value as the original curve for larger values of flux. In the second modification
2 (J-shape Mod 2), the J shape is retained but somewhat mitigated. In this modification, the
3 division rate initially decreases in a linear manner similar to that of the original model but with a
4 less negative slope until it intersects the original curve at a flux of 1,240 |im/m2-hour, where it
5 then follows the original curve for higher values of flux.
6
0.05
Mod 1
Mod 2
0.04
0.03
0.02
0.01
0.00
0
10000
20000
30000
40000
Formaldehyde Flux [pmol/(m2-h)]
7
8
9 Figure F-3. Conolly et al. (2003) J-shape model for division rates of initiated
10 cells in rats and two modified models.
11
12 Source: Crump et al. (2008).
13
14
15 Since the first constraint on the variation in ai was in concordance with the rat time-to-
16 tumor incidence data, Crump et al. (2008) applied each of the modified models in Figures F-2
17 and F-3 to the version of the formaldehyde models in Subramaniam et al. (2007) that employed
18 all NTP controls and the hockey-stick curve for ocn. These authors restricted their analysis to
19 this case since their stated purpose was only a sensitivity analysis as opposed to developing
20 alternate credible risk estimates. Figure F-4 reproduces (from Crump et al. [2008]) curves of the
21 cumulative probability of a rat dying from a nasal SCC by a given age for bioassay exposure
22 groups of 6, 10, and 15 ppm. For comparison purposes, the corresponding KM (nonparametric)
23 estimates of the probability of death from a nasal tumor are also shown. Three sets of
24 probabilities are graphed: the original unmodified one and the ones obtained by using hockey-
25 stick Mod 1 and Mod 2. Crump et al. (2008) state that the changes in the tumor probability
26 resulting from these modifications are so slight that the three models cannot be readily
This document is a draft for review purposes only and does not constitute Agency policy.
F-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
distinguished in this graph.4 Thus, the modifications considered to the models for the division
rates of initiated cells caused an inconsequential change in the fit of the model-predicted tumor
incidence to the animal tumor data.
0.9 n
0.8-
| °-7"i
8- 06-
(/> .
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
The above modifications did not affect the basal rate of cell division in the model and
likewise had no effect on the fit to the human background data (Crump et al., 2008).
Crump et al. (2008) noted that, although the threshold model for initiated cells in Conolly
et al. (2003) was replaced with a model that had a small positive slope at the origin, the resulting
curves, hockey-stick Mod 1 and hockey-stick Mod 2, could have been shifted slightly to the right
along the flux axis in order to introduce a threshold for a/without materially affecting the risk
estimates resulting from these modified curves. Thus, "the assumption of a linear no-threshold
response is not an essential feature of the modifications to the hockey-stick model; clearly
threshold models exist that would produce essentially the same effect" (Crump et al. 2008).
F.2.3. Biological Plausibility of Alternate Assumptions
These very small variations made to the ai in Conolly et al. (2003) are seen to be
consistent with the tumor-incidence data (just demonstrated above); small compared with the
variability and uncertainty in the cell replication rates characterized from the available empirical
data (at the formaldehyde flux where ai was varied); supported (qualitatively) by limited data,
suggesting increased cell proliferation at doses below cytotoxic; perturbations that one should
expect on any dose response derived from laboratory animal data because of human population
variability in cell replication; and biologically plausible because cell cycle control in initiated
cells is likely to be disrupted.
The averaged cell replication rate constants as tabulated in Table 1 of Conolly et al.
(2003) and shown by the red curve in Figure E-2 of Appendix E (for various exposure
concentrations and corresponding average formaldehyde flux values in the F344 rat nose)
demonstrate an increase over baseline values only at exposure concentrations of 6 ppm and
higher. Increased cell proliferation at these concentrations of formaldehyde, whether transient or
sustained, have been associated in the literature with epithelial response to the cytotoxic
properties of formaldehyde (Conolly, 2002; Monticello and Morgan, 1997; Monticello et al.,
1996, 1991). The labeling data are considered to show a lack of cytotoxicity and regenerative
cell proliferation in the F344 rat at exposures of 2 ppm and below (Conolly, 2002). In the
Conolly et al. (2003) modeling, it is further assumed that the formaldehyde flux levels at which
cell replication exceeds baseline rates remain essentially unchanged when extrapolated to the
human and for initiated cells for the rat as well as the human. These assumptions need to be first
viewed in the context of the uncertainty and variability in the data on normal cells discussed
earlier.
Arguments for a hockey-stick or J shape over the background have been made in the
literature for sustained and chronic cell replication rates; the analyses of the cell replication data
This document is a draft for review purposes only and does not constitute Agency policy.
F-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
show that the data are not consistently (over each site and time) indicative of a hockey-stick or
J shape as the best representation of the data. This uncertainty is particularly prominent when
examining the cell replication data at the 13-week exposure time and the pooled data from the
PLM nasal site from Monticello et al. (1996) (Figures E-l [dotted curve], E-3B, E-4 of Appendix
E). The earliest exposure time in this experiment was at 13 weeks, and the 13-week cell
replication data appear to be more representative of a monotonic increasing dose response
without a threshold. It is possible that early times are of more relevance to the carcinogenesis as
well as for considering typical (short duration) human exposures.
For initiated cells, there are no data on which to evaluate the modifications made to these
rates. However, some perspective can be gained by comparing them to the variability in the
division rates obtained from the data on normal cells used to construct the formaldehyde model.
As shown in Figure E-2 and discussed further in Subramaniam et al. (2008), these data show
roughly an order of magnitude variation in the cell replication rate at a given flux. As part of a
statistical evaluation of these data, a standard deviation of 0.32 was calculated for the log-
transforms of individual measurements of division rates of normal cells. By comparison, the
maximum change in the log-transform division rate of initiated cells resulting from hockey-stick
Mod 2 was only 0.20, and the average change would be considerably smaller. Thus, although
there are no data for initiated cells, it can be said that the modifications introduced in Crump et
al. (2008) for initiated cells are extremely small in comparison to the dispersion in the data for
normal cells.
Subramaniam et al. (2008) also point to some additional, albeit limited, data, suggesting
that exposure to formaldehyde could result in increased cell replication at doses far below those
that are considered to be cytotoxic. Tyihak et al. (2001) treated different human cell lines in
culture to various doses (0.1-10 mM) of formaldehyde and found that the mitotic index
increased at the lowest dose of 0.1 mM. These findings considered along with human population
variability and susceptibility (for example, polymorphisms in ADH3 [Hedberg et al., 2001])
indicate that it is necessary to consider the possibility of small increases in the human a7 over
baseline levels at exposures well below those at which cytotoxicity-driven proliferative response
is thought to occur.
Heck and Casanova (1999) have provided arguments to explain that the formation of
DPXs by formaldehyde leads to inhibition of cell replication (i.e., if this effect alone is
considered, normal cell replication rate of the exposed cells would be less than the baseline rate).
However, this hypothesis was posed for normal cells. Subramaniam et al. (2008) argue that if an
initiated cell is created by a specific mutation that impairs cell cycle control, the effect would be
to mitigate the DPX-induced inhibition in cell replication, either partially or fully, depending on
This document is a draft for review purposes only and does not constitute Agency policy.
F-11 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
the extent to which the cell cycle control has been disrupted. In the absence of data on initiated
cells, the above argument provided biological motivation to the modification applied to the
J-shape model for cell division (Crump et al. 2008).
Thus, the previous paragraphs suggest that the changes made in the analysis in Crump et
al. (2008) to the assumption by Conolly et al. (2003) regarding the dose response for the division
rate of initiated cells are not implausible.
F.2.4. Effect of Alternate Assumptions for Initiated Cell Kinetics on Human Risk Estimates
Figure F-5 contains graphs of the additional human risks estimated (in Crump et al.
[2008]) by applying these modified models for a/ and using all NTP controls, compared with
those obtained by using the original Conolly et al. (2004) model. Each of the four modified
models presents a very different picture from that of Conolly et al. (2004). At low exposures,
these risks are three to four orders of magnitude larger than the largest estimates obtained by
Conolly et al. (2004).
These results have been criticized by Conolly et al. (2009) as being unrealistically large
and above the realm of any epidemiologic estimate for formaldehyde SCC. Thus, they argue that
the parameter adjustments made in Crump et al. (2008) are inappropriate. Crump et al. (2009)
rebutted these points by arguing that the purpose of their work was not to provide a more reliable
or plausible model but to carry out a sensitivity analysis. They argued that the changes made to
the model (in their analyses) were reasonable since they did not violate any biological
constraints or the available data. Further, they pointed out that "by appropriately mitigating the
small modifications [they] made to the division rates of initiated cells, the model [would]
provide any desired risk ranging from that estimated by the original model up to risks 1,000-fold
larger than the conservative estimate in Conolly et al. (2004)."
Crump et al. (2008) also evaluated the assumption in eq 3 of the CUT modeling
pertaining to initiated cell death rates (fti) by making small changes to /?/. They report that they
obtained similarly large values for estimates of additional human risk at low exposures.
Obtaining reliable data on cell death rates in the nasal epithelium appears to be an unusually
difficult proposition (Hester et al., 2003; Monticello and Morgan, 1997), and, even if data are
obtained, they are likely to be extremely variable.
This document is a draft for review purposes only and does not constitute Agency policy.
F-12 DRAFT—DO NOT CITE OR QUOTE
-------
J-Shape Mod 2
0.01
-Shape Mod 1
Hockey Mod 1
E-3
Hockey Mod 2
CD
c
o
"4=
T3
"D
<
E-4
Conolly et al. (2004)
"Risk Conservative"
Hockey Stick Model
E-5
E-6
Unmodified
Hockey Stick
Model
E-7
E-8
1E-3
0.01
0.1
1
Formaldehyde Exposure (ppm)
1
2
3 Figure F-5. Graphs of the additional human risks estimated by applying
4 these modified models for «/, using all NTP controls, compared to those
5 obtained using the original Conolly et al. (2004) model.
6
7 Source: Crump et al. (2008).
This document is a draft for review purposes only and does not constitute Agency policy.
F-13 DRAFT—DO NOT CITE OR QUOTE
-------
This page intentionally left blank.
This document is a draft for review purposes only and does not constitute Agency policy.
F-14 DRAFT—DO NOT CITE OR QUOTE
-------
Appendix G
i
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
APPENDIX G
EVALUATION OF THE CANCER DOSE-RESPONSE MODELING
OF GENOMIC DATA FOR FORMALDEHYDE RISK ASSESSMENT
G.l. MAJOR CONCLUSIONS IN ANDERSEN ET AL. (2008)
In Chapter 4, the gene microarray data from animal studies on formaldehyde (Andersen
et al., 2008; Thomas et al., 2007) were described. The analysis of these animal high throughput
data and the conclusions reached in these two groundbreaking papers were closely examined for
use in this assessment. Studies on high throughput animal data provide a wealth of information
that helps further understanding of the relevant mechanisms. However, such studies have
generally not made quantitative bottom-line inferences that inform low dose human risk. The
above-mentioned studies are a notable exception due to the breadth of their conclusions on low
dose MO As, their pioneering application of the benchmark dose (BMD) methodology to
genomic data, their use of BMD-response analysis that identified dose estimates at which
specific cellular processes were significantly altered, the fact that they were accompanied by
recommendation in the literature urging use of these results in setting exposure standards for
formaldehyde (Daston, 2008).
We focus here on the conclusions in these papers with regard to modeling the cancer
dose-response for formaldehyde. In addition to supporting our disposition of these analyses for
this assessment, this write-up serves the purpose of exemplifying critical issues that need to be
considered for the future.
The overall BMD determined in Andersen et al. (2008) for all genes with significant
dose-response averaged 6.4 ppm. These analyses indicated a general progression with the lowest
BMD values (i.e., the most sensitive epithelial responses) for extracellular and cell membrane
components and higher BMD values for intracellular processes. Overall, these authors
concluded that
¦ Genomic changes, including those suggestive of mutagenic effects, did not temporally
precede or occur at lower doses than phenotypic changes in the tissue
¦ Genomic changes were no more sensitive than tissue responses
¦ Formaldehyde, being an endogenous chemical, is well handled until some threshold is
achieved. Above these doses, toxicity rapidly ensues with concomitant genomic and
histologic changes.
¦ Linear extrapolations, or extrapolations that specify similar MO As at high and low doses
would be inappropriate.
This document is a draft for review purposes only and does not constitute Agency policy.
G-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
These findings were judged to have significant implications on the debated MOA for
formaldehyde carcinogenicity, confirming results from earlier bioassays and dose-response
modeling that the mutagenicity of formaldehyde was too weak to be of relevance to its
carcinogenicity. Daston (2008) judged the method in these efforts to be extremely sensitive and
therefore suited to examining whether responses at the molecular level take place at doses below
which frank adverse effects occur. Daston (2008) argued that"... if there are pleiotropic effects
at lower exposure levels that would elicit a different profile of gene expression, those genes
would not go unnoticed" and thus concluded that "the gene expression data confirm that the
responses are not linear at low doses."
In the analyses that follow, we point to some significant quantitative factors that impact
on these conclusions.
G.2. USE OF MULTIPLE FILTERS ON THE DATA
The analyses in these papers involved the following sequence of data filters.
1. Gene probe sets that differed in expression in response to treatment were identified by
one-way analysis of variance. Probability values were adjusted for multiple comparisons
by using a false discovery rate of 5%.
2. Next, in addition to the above statistical filter, the output was further screened by
selecting only those genes that exhibited a change from the control group that was greater
than or equal to 1.5-fold (logarithmic).
3. The gene probe sets that demonstrated significant dose-response behavior were then
matched to their corresponding biological process and molecular function gene ontology
(GO) categories (considering only those involving more than three genes) and grouped
into process categories such as cell division, DNA repair, cellular proliferation,
apoptosis, and related molecular function categories.
A large number of genes are expressed in these studies; therefore, clearly some
appropriate filter needs to be used for meaningful interpretation of the vast database. Tissue
pathology served as a phenotypic anchor for the interpretation of microarray results, and the
genomic study confirmed (and improved on) the qualitative and quantitative understanding
derived from the histopathology and observation of frank effects. It is possible that the
combination of filters used by these authors is adequate for an inquiry into some mechanisms
associated with the specific phenotypic effects. However, the studies reached bottom-line
conclusions with regard to the low-dose MOA and approach to be considered for quantitative
This document is a draft for review purposes only and does not constitute Agency policy.
G-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
extrapolation. These conclusions necessarily involve questions as to whether there were gene
expression changes at low dose and at early exposure times that may be relevant to initiating
carcinogenesis and finally as to whether there is a threshold in dose associated with
formaldehyde carcinogenesis. However, collectively, the three filters employed in these studies
likely constitute overly stringent criteria, taking away the resolution needed to observe critical
gene changes needed to delineate low dose effects. An indication that this may indeed be the
case can be seen by examining the correlations in their findings with the observed trend in the
data on DPXs formed by formaldehyde. This is detailed in the following section.
G.3. DATA FOR LOW-DOSE CANCER RESPONSE
A significant finding in Thomas et al. (2007) is that BMD estimates for the GO
categories applicable to cell proliferation and DNA damage were similar to values obtained for
cell labeling indices and DPXs in earlier studies and to BMD estimates obtained for the onset of
nasal tumors. The mean BMD for the GO category of "positive regulation of cell proliferation"
was 5.7 ppm; in comparison, Schlosser et al. (2003) obtained a 10% BMD of 4.9 ppm for the cell
labeling index. The GO category associated with "response to DNA damage stimulus," seen as a
genomic correlate to a mutagenic effect, had a mean BMD of 6.31 ppm. Thomas et al. (2007)
compare this finding with significant increase at 6 ppm of DPXs following a 3-hour exposure in
the study by Casanova et al. (1994). The formation and repair of DPXs have been considered to
be one of the potential mechanisms associated with the genotoxic action of formaldehyde
(Conolly et al., 2003, 2000). Based on earlier work in the same laboratory (Conolly et al., 2004,
2003; Conolly, 2002), Slikker et al. (2004) concluded that there is a dose threshold (at about
6 ppm) to formaldehyde carcinogenicity and that the putative mutagenic action of formaldehyde
is not relevant to its carcinogenicity. Therefore, the finding that a significant genomic response
(e.g., induction of DNA repair genes) is not observed at doses lower than those that induce
tumors in rodent bioassays is seen by these authors (Andersen et al., 2008; Daston, 2008;
Thomas et al., 2007) to further buttress the above conclusions related to the mode of action for
formaldehyde-induced respiratory cancer.
However, phenotypic anchoring to the DPX data drawn only from Casanova et al. (1994)
misses critical low-dose data that informs mode of action. In an earlier study, Casanova et al.
(1989) observed statistically significantly elevated (over controls) levels of DPXs at 2 ppm and a
trend towards elevated DPXs at 0.7 ppm. In analysis of low-dose data, the trend in the dose-
response is critically important because data inherently lack the power to establish statistical
significance. Furthermore, the two studies by Casanova and coworkers are different in some
respects. The earlier study was a 6-hour exposure, while the later study was a 3-hour study; thus,
This document is a draft for review purposes only and does not constitute Agency policy.
G-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
on this account alone, it appears more relevant to compare with the older study. Exposures in
the earlier study were additionally at 0.3 and 10 ppm, thus affording a lower exposure
concentration. In the earlier study, tissue from the whole nose was analyzed, whereas in the later
study tissue from two specific regions was obtained from the "high" tumor (Level II) and "low"
tumor regions. Together, these data suggest that DPXs occur at exposure concentrations
considerably lower than those that elicited transcriptional changes. One possible explanation is
that the increase in DPXs was not sufficient to induce DNA repair genes. Alternatively, these
discrepancies may be due to the stringent filters and the low statistical power of the Andersen et
al. (2008) study. These disparities between the gene array study and the DPXs question the
ability of the studies in Andersen et al. (2008) and Thomas et al. (2007) to inform the presence or
absence of a mutational MOA for formaldehyde, and in essence, to inform the low-dose response
curve for formaldehyde-induced cancer.
In another instance, Andersen et al. (2008) clearly stated that no genes were significantly
altered by exposure to 0.7 ppm, yet they state that there was "a trend toward altered expression at
0.7 ppm" in some genes with U and inverted U shape dose-responses (Figures 4 and 5 of their
paper). While these changes may not be statistically significant, they could be biologically
significant.
G.4. DIFFICULTIES IN INTERPRETING THE BENCHMARK MODELING
The benchmark analyses are summarized in Thomas et al. (2007) as average BMD
estimates for genes in a given GO that were statistically significantly dose related. The
benchmark modeling was then used by the authors to identify that the dose below individual
cellular processes was judged to be "not altered."
The BMD definition used by these authors is quite stringent: it defines an effect so that
only 0.005 of controls will be considered affected and sets the BMR corresponding to this dose
at 0.105. The net effect is that the BMD is the air level, such that the increase in the mean
response is 1.349 x standard deviation. This is essentially an arbitrary definition. For
comparison, if 0.05 of controls are considered affected and the BMR is set at 0.1 (common
values that are applied to whole animal data), the BMD is the air level such that the increase in
the mean response is 0.608 x standard deviation. Thus, if this definition had been used (as is
traditionally the case), the BMD estimates would all be 2.2 times smaller than those obtained by
Schlosser et al. (2003). Furthermore, the analysis assumes equal variance in all dose groups.
Thus, further consideration of these issues with regard to interpretation of the BMR obtained
from these studies is needed before it can be used in regulatory exposure setting. Secondly,
lower confidence limits on the BMDs need to be derived for the data in Andersen et al. (2008).
This document is a draft for review purposes only and does not constitute Agency policy.
G-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
G.5. STATISTICAL SENSITIVITY OF THE DATA FOR DOSE-RESPONSE
Another cautionary note pertains to the qualification of gene array studies as being
extremely sensitive. Such a qualification should actually refer to the fact that only tiny amounts
of mRNA are needed, that is, the sensitivity of the assay per se for measuring gene expression.
However, this should not be confused with the sensitivity needed to identify the very small dose-
related changes at low dose. Andersen et al. (2008) reports on results of studies that involve
small numbers of animals in each dose group (five or eight). Despite the limited power in such
studies, the paper equates the absence of a statistically significant effect with no effect. This
limitation is generally true of studies of the dose responses of changes in gene expression
conducted to date; they have generally relied on very few animals (<10 per dose group). Since
there will likely always be background amounts of gene expression, quantifying the dose
response requires statistically significant changes in gene expression as a function of dose. If the
genomic data involve even fewer animals per group than the histopathological data, they have
even less power to delineate the dose response; in particular, whether there is a threshold at low
exposures. This is illustrated by the example in Figure G-l of the dose responses for epithelial
hyperplasia (Andersen et al. 2008, lesion 2). These appear equally consistent with both a
threshold at around 1 ppm and a linear response down to zero.
G.6. LENGTH OF THE STUDY AND STOCHASTIC EVENTS
Another significant consideration with regard to MOA conclusions that are pertinent to
the disease process is the length of the study, 15 days. If formaldehyde-induced tumor formation
is a stochastic process (e.g., genotoxicity), then exposure of a small number of animals to low
concentrations for 15 days may not be long enough to detect changes that might occur under
long-term exposure scenarios.
Relatedly, it has been suggested that gene (and protein) expression is a stochastic process
whereby steady state gene expression obeys Poisson statistics (i.e. distribution of rare events),
and that events of interest may occur in a single cell or small number of cells in which larger
tissue samples can average out such stochastic events and prevent the detection of non-average
behavior (Quakenbush, 2007). Given the implied difficulty in such an analysis, duration of
exposure may be one of the most tenable ways of addressing whether a chemical increases the
probability of an adverse response.
This document is a draft for review purposes only and does not constitute Agency policy.
G-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
0 ~-
0
Day 1
2 3 4 5
Day 1 Recovery
Day 5
Day 6 Recovery
1 2 3 4 5
Day 15
<
' ,
Figure G-l: Graphs of epithelial hyperplasia (Lesion 2) versus formaldehyde
concentration (ppm) with 95% confidence intervals (with linear fit by eye)
G.7. OVERALL CONCLUSION
We believe our analyses of the presentations in Andersen et al. (2008) and Daston (2008)
are generally useful with regard to future developments in quantitative analyses of genomic data
if they are to be of relevance to risk assessment. For risk assessment, rather than focusing on
what responses are statistically significant, an analysis should focus on 1) what range of values
of critical parameters (e.g., gene expression) are consistent with the data, and 2) what these
values imply for whole animal risk. This is of course, an extremely difficult proposition because
This document is a draft for review purposes only and does not constitute Agency policy.
G-l DRAFT—DO NOT CITE OR QUOTE
-------
1 we do not know nearly enough about how changes in genes quantitatively affect whole animal
2 risk, or even which genes are important.
This document is a draft for review purposes only and does not constitute Agency policy.
G-8 DRAFT—DO NOT CITE OR QUOTE
-------
Appendix H
i
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
APPENDIX H
EXPERT PANEL CONSULTATION ON QUANTITATIVE EVALUATION OF ANIMAL
TOXICOLOGY DATA FOR ANALYZING CANCER RISK DUE TO INHALED
FORMALDEHYDE
The National Center for Environmental Assessment convened an expert panel of
scientists for advice on evaluating available approaches for incorporating biological information
in analyzing animal tumor data for assessing cancer risk due to inhaled formaldehyde. This
Appendix pertains to the major deliberations and results of that meeting and is divided into three
sections.
A. Scope and Agenda of Meeting on Quantitative Evaluation of Animal Toxicology Data for
Analyzing Cancer Risk due to Inhaled Formaldehyde. October 28 & 29, 2004.
B. Summary of Consultative Meeting on CUT Formaldehyde Model. October 28 & 29,
2004.
C. Meeting Report from Dr. Rory B. Conolly
This document is a draft for review purposes only and does not constitute Agency policy.
H-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
A. Scope and Agenda of Meeting on Quantitative Evaluation of Animal Toxicology
Data for Analyzing Cancer Risk due to Inhaled Formaldehyde
October 28 & 29, 2004. Washington, DC.
This meeting is to assist EPA in evaluating available approaches for incorporating biological
information in analyzing animal tumor data for assessing cancer risk due to inhaled
formaldehyde. The CUT Centers for Health Research (CUT) has published a novel risk
assessment that links site-specific predictions of flux using computational fluid dynamics (CFD)
modeling with a two-stage clonal growth model of cancer to analyze nasal tumor incidence in
two rodent bioassays. The rodent models are used with corresponding human models for low-
dose extrapolation of cancer risk to people.
Key predictions of the CUT effort are a zero maximum likelihood estimate of the probability of
formaldehyde-induced mutation per cell generation in the rat and a de minimus additional
lifetime risk in non-smokers due to continuous environmental exposure below 0.2 ppm. The
National Center for Environmental Assessment is carrying out sensitivity analyses and
examining variations of the CUT model in order to understand the implications of the model
structure and parameters on model predictions. In this meeting, we wish to focus on the
strengths and key uncertainties of this model, the extent to which assumptions in the CUT model
are supported by biological data, and examine the impact of uncertainty and variability on the
overall quantitative risk characterization.
Broadly, the discussions will focus on the following areas:
• Impact of uncertainties in dosimetry on human risk estimates
• Uncertainties in the use of experimental data on labeling index
• The model structure related to initiated cells and DNA protein cross-links
• Considerations of time-to-tumor in the clonal growth modeling
• Inferences and information on the role of mutation and cytotoxicity in estimating human risk
• Relative merits of benchmark dose modeling vs. the 2-stage clonal growth model
Discussions on Mode of Action are expected to be an integral part of several of the sessions.
Therefore a specific time-slot is not set aside for this purpose.
The meeting will have a panel discussion format. There will be no formal presentations unless
necessary to elucidate an issue. Various attachments referred to in the Agenda below, as well as
the relevant manuscripts will be sent separately.
Specifically, we suggest the following issues upon which to focus the discussion in the above
areas, and approximate time frames and discussion leads, although discussants should feel free to
bring up other critical issues.
This document is a draft for review purposes only and does not constitute Agency policy.
H-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
I. Introduction and purpose of discussion
Peter Preuss.
9:00 AM, Oct 28
II. Impact of uncertainties in dosimetry on risk estimates
Lead discussant: Linda Hanna
9:15 - 11 AM Oct 28
Boundary conditions
The CFD modeling specified a mass transfer coefficient as a boundary condition on the
nasal lining, adjusting the value of this coefficient on the "absorbing" portion of the
lining so as to match simulated overall uptake in the rat nose to the experimentally
determined average overall uptake. This value was then used for the corresponding
human nasal lining. Are these boundary conditions appropriate surrogates for the
underlying pharmacokinetics, including saturation in metabolism and mucociliary
clearance, particularly with reference to humans?
Turbulence
Turbulent flow has been seen to occur in experimental models of the human nose at some
of the higher flow rates at which the CFD models were used in CIIT=s assessment. It is
not likely that the CUT CFD model can reliably identify signatures of transition to
turbulent behavior. Turbulent flow can significantly alter regional uptake patterns.
Additionally, significant mass balance errors were seen at the higher flow rates in the
human flow models. Discuss if these are likely to impact significantly on risk estimates.
Interindividual variability
The CUT assessment has focused on the nasal anatomy of a single individual. Discuss
the implications of interindividual variations in nasal anatomy on the population
distribution in risk.
III. Uncertainties in the use of experimental data on labeling index
Lead discussant: George Lucier
11AM- 11:45 AM, 1:00 - 3:15 PM Oct 28
Cell-replication rate and its relationship to flux is a critical determinant of risk. Therefore
uncertainties and variability in measurement of the unit length labeling index and its use in the
CUT clonal growth modeling need to be characterized.
1. Discuss the strengths, uncertainties and limitations associated with estimating cell
replication rates from the unit length labeling index (ULLI).
a. For example, a constant ratio of the measured ULLI to the labeling index (LI) that
is used in the model is assumed. Is it valid to assume this ratio to be constant
across nasal sites, dose and exposure time.
b. How uncertain is this ratio?
This document is a draft for review purposes only and does not constitute Agency policy.
H-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
2. Considering the large patterns of variability in the ULLI data, discuss the validity of
using ULLI averaged over site and exposure times.
a. The averaging loses information on the sequential effect of change with
time, and on significant differences among sites.
b. How sensitive is the clonal growth modeling result to these variations in the dose-
response function for cell replication rates vs. flux to the tissue? A discussion of
this question in this session is intended to serve as input to later deliberations on
the issue.
3. Discuss the validity of combining data collected in different experiments using different
labeling methods, and the validity of estimating cell replication rates from LI or ULLI
measured in a single pulse labeling experiment.
See attachment C: "ULLI Dose-Response Modeling and Statistical Analysis "for a
discussion of these issues, and Moolgavkar and Luebeck (1992).
IV. Model Structure: Birth and death rates for Initiated cells, Role of DPX
Lead discussant: Kenny Crump
3:30 - 6:00 PM Oct 28.
Parameters for initiated cells
1. The CUT analysis of ULLI data allows for a virtual threshold in dose in the replication
rate of normal cells. Discuss the validity of ascribing such a behavior to initiated cells
considering the sensitivity of 2-stage model results to the initiated cell replication rates.
2. Discuss the treatment of death rate for initiated cells in the model (set equal to birth rate
of normal cells in Conolly et al., 2003) and implications for confidence in model
predictions.
Also see Attachment A (memo from Rory Conolly) and Attachment D (EPA discussion of
CUT clonal growth modeling and some sensitivity analyses. . .)
Treatment of DNA protein cross-links (DPX) in clonal expansion model
3. FORMALDEHDYE-INDUCED MUTATION IS MODELED AS TAKING
PLACE ONLY WHILE DPX ARE IN PLACE WITH DPX UNDERGOING
RAPID REPAIR. DISCUSS THE POSSIBILITY OF PERSISTENT GENETIC
DAMAGE THAT EXTENDS BEYOND THE DPX HALF-LIFE AND
ENHANCES MUTATION. HOW MIGHT THIS ISSUE BE INCLUDED IN
THE MODEL STRUCTURE?
This document is a draft for review purposes only and does not constitute Agency policy.
H-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
V. Considerations of time-to-tumor in the CUT clonal growth modeling
Lead discussant: Christopher Portier
8:30- 11:00 AM, Oct 29.
1. A number of issues affect likelihood values and the model fit to the time-to-tumor data.
Discuss assumptions in the treatment of time-to-tumor in the CUT clonal expansion
model, and their impact on parameter estimates. For example,
a. Results in Conolly et al. (2003, 2004) are derived considering all tumors to be
fatal. Note in this context that serially sacrificed animals have been combined
with those experiencing mortalityBthe effect of this is visible as irregularities in
the time-to-tumor curve.
b. How is the time variability in ULLI likely to impact on the time-to-
tumor predictions?
2. Long delay times are predicted by the model for observation of detectable tumor. Is this
compatible with the assumption of rapidly fatal tumors?
3. Discuss the weight to be given to differences in likelihood when comparing with
variations on the Conolly et al (2003) model structure such as in Attachment A or D.
VI. Inferences on the role of formaldehyde-induced mutation and cell proliferation
Lead discussant: Dale Hattis
11:15 - 12:00 PM, 1:00 - 4:00 PM, Oct 29.
1. The model structure in Conolly et al. (2003) predicts a zero maximum likelihood estimate
for the constant of proportionality (KMU) linking DPX to the probability of
formaldehyde-induced mutation per cell generation. Examine the strength of this
conclusion, and the extent to which an insignificant probability of formaldehyde-induced
mutation per cell generation is supported by data.
2. Discuss the biological relevance and validity of model-estimated parameters, particularly
in the context of low-dose predictions.
a. Discuss possible avenues to validate CUT cancer model predictions.
3. Discuss the validity of using cell replication rates determined for the rat to predict human
risk in a population.
4. In the face of uncertainties, are the results in Conolly et al. (2003, 2004) conservative in
the sense of overpredicting risk?
a. Discuss the extent to which sensitivity analyses have addressed this issue and the
extent to which sensitivity analyses can speak to the strength of the model. [See
Attachments A: Memo from Conolly, andD: EPA discussion of CUT clonal
growth modeling and some sensitivity analyses . . .].
This document is a draft for review purposes only and does not constitute Agency policy.
H-6 DRAFT—DO NOT CITE OR QUOTE
-------
1 VII. Benchmark Dose Modeling
2 Lead discussant: Kenny Crump
3 4:15-5:30 PM, Oct 29.
4
5 Discuss the relative merits of using a benchmark dose approach that incorporates
6 biological modeling (such as estimating flux to tissue or DPX levels) as compared with
7 the CUT 2-stage model for cancer. (See attachment E and Schlosser et al., 2003.)
8
9
This document is a draft for review purposes only and does not constitute Agency policy.
H-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
B. Summary of Consultative Meeting on CUT Formaldehyde Model
October 28 & 29 2004, NCEA, Washington, DC
Date: November 10, 2004
Ravi P. Subramaniam, Ph.D.
Quantitative Risk Methods Group
National Center for Environmental Assessment, ORD, US EPA
This is a broad summary of the most important issues at the formaldehyde meeting.
It was generally felt by consultants that the broad framework of the approach adopted by
CUT, namely the use of a two-stage model for cancer, the linking of localized flux to cell
replication rates and DPX concentration, and the expression of formaldehyde-induced mutation
as a linear function of DPX, was reasonable.
Potential errors in the dosimetry modeling were seen not to have a significant effect on
risk estimates. The boundary conditions used were discussed to be a reasonable representation
of the pharmacokinetics for both rats and humans. The discussion on the impact of
interindividual variability of nasal anatomy was not particularly conclusive. It was determined
that there was likely to be much less variability in reactive gas uptake than that seen in
particulates.
Crucial errors were however identified on several fronts in the manner in which the
clonal growth model had been implemented in the CUT effort. Dr. Portier felt that the
calculation of probability was seriously flawed on account of lumping serially-sacrificed animals
and animals that died of tumor together, while at the same time assuming rapid fatality of all
tumors. This was seen to significantly alter the calculation of tumor probability (the shape of the
dose-response curve), and his insight was that a correction was likely to allow for a substantially
higher value for the probability of formaldehyde-induced mutation at low-dose. The best
estimate for this probability is now zero in the model. Drs. Crump, Portier and Hattis argued that
replacing this estimate by an upper confidence bound on KMU (the coefficient determining the
role of DPX in the probability of mutation per cell generation), keeping other structural problems
in the model unexplored, or other parameters fixed, would not be enough. There was a
discussion on the need to provide confidence bounds on risk determined by allowing all the
parameters to vary. Drs. Crump and Hattis (and Portier?) felt such an estimate would be very
different from that calculated based on individual parameters.
Drs. Crump, Hattis and Portier urged us not to be constrained by the optimal likelihood
values of a single plausible model, and underscored the need to explore a variety of biologically
reasonable model structures as a requisite for utilizing such a model in risk assessment.
This document is a draft for review purposes only and does not constitute Agency policy.
H-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Likelihood was seen to be an inadequate expression of what is to be considered an optimal
model (okay only for comparing models that were nested, etc.). These models should allow the
expression of variability and uncertainty in the data, as well as in underlying assumptions in
model specification. Dr. Crump (and Hattis also?) felt that alternate model structures, if
explored, could potentially lead to risk estimates, for the range below the observed data, that
were higher by several thousands.
Dr. Crump cautioned that extrapolating to human using the hockey or J-shaped cell
replication curve used in the rodent carried with it a large uncertainty that had not been
characterized in the Conolly modeling.
Dr. Portier expressed concern over the manner in which historical and concurrent
controls were lumped together. The thrust of Portier's comments was that such a combination of
controls was generally not done. The large number of historical controls was likely to
significantly bias the impact of the bioassay data in determining the time-to-tumor fits.
There were various discussions about the pros and cons of constructing a joint likelihood
of the cell replication data and the tumor data, and the weights to be assigned to the separate
likelihoods. This was considered to be problematic by Dr. Portier.
Dr. Crump's opinion was that the Conolly model, and those explored by EPA, fit the
tumor data poorly, and that an improved description of the tumor data was needed before the
model could be used for low-dose and inter-species extrapolation.
Drs. Lucier and Hattis placed emphasis on including the early-time cell replication data
instead of constructing a time-weighted average. It was felt that the two Monticello experiments
could not be combined together as in Conolly et al. Dr. Lucier felt that the early-time data would
have a greater impact in the progression of carcinogenesis. In general, the effect of "time" was
considered to have significant effects on the time-to-tumor modeling, and they urged us to
incorporate time-dependent terms in the modeling. CUT expressed willingness to provide the
original cell replication data to us for further analysis. (Further discussion on this matter did not
take place in the open forum.)
Preliminary indications are, particularly based on Dr. Portier's insight, that the currently-
held "de-minimus" picture of low-dose risk, as expressed in Conolly et al. (2004), is not likely to
be the case if these various suggestions are incorporated in the modeling.
This document is a draft for review purposes only and does not constitute Agency policy.
H-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
C. Meeting Report from Dr. Rory B. Conolly
Rory B. Conolly, Sc.D., D.A.B.T.
106 Michael's Way
Chapel Hill, NC 27516
Voice: 919.929.2258
July 24, 2005
Dr. Bobette Norse
ORAU Procurement - MS-04
P.O. Box 117
Oak Ridge, TN 37831-0117
Phone: 865-576-3051
Fax: 865-576-9385
Dear Dr. Nourse,
The following is my final written report on the formaldehyde review meeting held at the
U.S. EPA in Washington, D.C. on 28-29 October, 2004.
EPA provided no guiding philosophical statement about the criteria being used to
evaluate the CUT assessment. The new Guidelines for Carcinogen Assessment state that the
preferred default approach is to use a biologically based model. Since the key components of the
CUT assessment have been published in the peer-reviewed literature and have undergone several
peer reviews other than the current NCEA effort, one has to wonder just how high the bar is set
for acceptance of biologically based assessments. Given the time and resources expended on the
CUT assessment and the richness of the supporting data base, I find it difficult to imagine what
an acceptable biologically-based assessment might look like if in the end the CUT assessment is
deemed not acceptable by NCEA. If this is in fact the outcome it will have major implications
for the likelihood that anyone will be willing to commit the significant resources needed to
develop of these kinds of risk assessment models.
The documents provided in advance of the October 2004 review meeting were
collectively a discussion of uncertainty about the CUT work. With respect to the clonal growth
model, however, no new risk predictions were provided, so there was no way to judge how the
uncertainties that NCEA identified might impact predicted risk. Evaluation of the significance
of "uncertainties" when the impact of the uncertainties on the predicted risk is not known is itself
an uncertain process.
A related concern is that there did not seem to be any consideration of the historical
context of the CUT assessment. EPA developed formaldehyde assessments in 1987 and 1991.
The 1987 assessment used ppm as the input and the LMS model for the dose-response
prediction. The 1991 assessment used DPX as a dosimeter and the LMS model. BMD
assessments have since become available from other sources such as Paul Schlosser's work. The
risk predictions of the BMD models are similar to the 1991 LMS assessment. Both the DPX-
LMS and BMD assessments predicted somewhat less risk than the 1987 assessment, establishing
the trend of less risk with increased incorporation of relevant data. I have always argued
This document is a draft for review purposes only and does not constitute Agency policy.
H-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
(probably initially at the 1998 Ottawa review) that the historical context is the appropriate
context for evaluating the CUT clonal growth model. For a "level playing field" the
uncertainties of the 1987 and 1991 assessments, and of the more recent BMD models, should be
analyzed to the same degree as the clonal growth model. Does NCEA think that, because the
LMS and BMD approaches used structurally simpler dose-response models and much more
limited data inputs, they are less uncertain? The NCEA analysis seemed to be implying that use
of more data and of a biologically more realistic model structure actually makes the CUT
approach more uncertain than the LMS and BMD approaches. I encourage NCEA to consider
how uncertainties that can be evaluated explicitly in the structurally rich CUT model compare to
hidden uncertainties in the simpler models, where the hidden uncertainties encompass, for
example:
1. Missing or incomplete descriptions of the regional dosimetry of formaldehyde.
2. Lack of simultaneous incorporation of the directly mutagenic and
cytolethal/regenerative proliferation modes of action.
3. Lack of explicit consideration of the multistage nature of cancer.
4. Lack of consideration of the growth kinetics of initiated cell populations
5. Lack of evaluation of the measured J-shaped dose response for regenerative cellular
proliferation.
A careful, balanced comparison of the CUT assessment with the previous assessments along
these lines would be informative with respect to the suitability of the CUT assessment as the
basis for a new IRIS listing for formaldehyde.
A further concern involves the peer-review of the CUT formaldehyde assessment held in
Ottawa in 1998. This review was sponsored by the U.S. EPA and Health Canada and involved
what was arguably a world-class review panel. The CUT assessment was not in its final form at
that time, though we did provide a detailed description of the overall approach and the specific
methods we were using to generate dose-response predictions. The 1999 CUT document and the
subsequent peer-reviewed publications are responsive to the comments and suggestions raised by
the reviewers. My concern is that no information was provided on the role that Ottawa review
plays in the ongoing review of the CUT formaldehyde assessment by NCEA. Should the
October 2004 review be viewed as standing on the shoulders of the 1998 review or as being in
parallel to it? It was not at all clear to me that the October 2004 review in any way utilized the
judgments of the 1998 review. It seems that the 2004 review was more of a parallel effort and
that the 1998 review was ignored and was effectively a waste of time and money. I would like to
have some clear understanding of how the 2004 review effort should be viewed relative to that of
1998.
In closing, let me reiterate that while the detailed examination of the CUT formaldehyde
assessment is laudable, this examination should be conduced with an eye to the historical context
of formaldehyde risk assessment on the one had and, on the other hand, to a concern for
encouraging, and not discouraging, development of biologically based risk assessment models.
Sincerely yours,
Rory B. Conolly, Sc.D., D.A.B.T.
This document is a draft for review purposes only and does not constitute Agency policy.
H-11 DRAFT—DO NOT CITE OR QUOTE
-------
- End of Volume IV -
This document is a draft for review purposes only and does not constitute Agency policy.
H-12 DRAFT—DO NOT CITE OR QUOTE
------- |