Tool for Exposure and Risk Assessment
DEVELOPED BY THE U.S. ENVIRONMENTAL PROTECTION AGENCY
OFFICE OF RESEARCH AND DEVELOPMENT
NATIONAL EXPOSURE RESEARCH LABORATORY
RESEARCH TRIANGLE PARK, NC 27711
Office of Research and Development
National Exposure Research Laboratory
-------
-------
EPA/600/R-12/039 | January 2012 | www.epa.gov/ord
Biomonitoring An Exposure Science
Tool for Exposure and Risk Assessment
DEVELOPED BY THE U.S. ENVIRONMENTAL PROTECTION AGENCY
OFFICE OF RESEARCH AND DEVELOPMENT
NATIONAL EXPOSURE RESEARCH LABORATORY
RESEARCH TRIANGLE PARK, NC 27711
I Office of Research and Development
I National Exposure Research Laboratory
-------
CONTRIBUTORS
Cecilia Tan, Curtis Dary, Daniel Chang, Elin Ulrich, Jeanette Van Emon.
Jianping Xue, Joachim Pleil, John Kenneke, Jon Sobus, Linda Sheldon, Marsha Morgan.
Rocky Goldsmith, Rogelio Tornero-Velez, Ross Highsmith, Roy Fortmann.
Tim Collette, Valerie Zartarian (EPA/ORD/NERL)
PRIMARY CONTACT
Cecilia Tan
National Exposure Research Laboratory.
109 T.W. Alexander Dr., Research Triangle Park, NC 27711
DISCLAIMER
The United States Environmental Protection Agency, through its Office of Research
and Development, funded and managed the research described herein. It has been
subjected to the Agency's administrative review and approved for publication.
-------
Table of Contents
Abstract v
1.0 Introduction 1
2.0 A Source-To-Outcome Continuum for Human Health Research 3
3.0 Biomonitoring Tiers for Exposure and Health Research 7
3.1 Tier 1: Biomonitoring for exposure surveillance 7
3.2 Tier 2: Biomonitoring to support exposure assessment 8
3.3 Tier 3: Biomonitoring to support risk assessment 8
3.4 Special consideration for timed events, sampling strategies, and repeated measures 11
3.5 Tier 4: Biomonitoring for exposure and risk assessments 11
3.6 Tier 5: Biomonitoring to advance exposure and risk assessments 12
4.0 Filling in Information Gaps 15
4.1 When data informing exposure estimates are missing 15
4.2 When kinetic information is missing 15
4.3 When toxicity information is missing 16
4.4 When biologically relevant biomarkers are unidentified 16
5.0 Additional Considerations for Interpreting Biomonitoring Data 19
5.1 Categories and uses of biomarkers 19
5.2 Different categories of chemicals based on their biological half life in relation to
exposure patterns 20
5.3 Exposure reconstruction 20
5.4 Non detect data 21
5.5 Non specific biomarkers 22
5.6 Stereochemistry of biomarkers 23
6.0 Future Directions for Research on Biomarkers of Exposure 25
-------
List of Figures
Figure 1. A source-to-outcome continuum for human health research 4
Figure 2. Requirements and examples of tier 1 analyses of biomarker data 7
Figure 3. Requirements and an example of a tier 2 analysis of biomarker data 9
Figure 4. Requirements and an example of a tier 3 analysis of biomarker data 9
Figure 5. Requirements and examples of tier 4 analyses of biomarker data 10
Figure 6. Requirements and examples of tier 4 analyses of biomarker data 12
Figure 7. Requirements and examples of tier 5 analyses of biomarker data 13
Figure 8. Biomarker measurements to estimate dose levels, exposure levels and environmental stressor
levels, for comparison to reference values 21
Figure 9. The type of exposure information which a biomarker measurement can infer depends on the
availability of kinetic and exposure data 22
List of Tables
Table 1. The uses and requirements of the five biomonitoring tiers 14
-------
Abstract
Biomonitoring studies of environmental stressors are useful
for confirming exposures, estimating dose levels, and
evaluating human health risks. However, the complexities
of exposure-biomarker and biomarker-response relation-
ships have limited the use of biomarkers in exposure science
studies. In this document, an updated source-to-outcome
continuum is presented to better define biomarkers as tools
for human health research; specific attention is given to
biomarker applications in exposure research. This continuum
links exposure sources and health outcomes using a compi-
lation of measurements, mathematical models, and model
estimates. A tiered approach to biomonitoring analyses is
presented, based on this continuum, to categorize the uses
of biomonitoring data given various research objectives and
the availability of specific measurements and models. Tools
that can be used to infer critical model parameters and model
estimates (when they are unavailable) also are discussed to
improve biomarker utilization for exposure and risk assess-
ments. Finally, frequently encountered complications in
biomonitoring studies are discussed, along with suggestions
to address these challenges.
-------
-------
1.0 Introduction
Human exposure assessment for environmental chemicals
traditionally has focused on identifying sources, determin-
ing chemical fate and transport, and quantifying the resulting
microenvironmental concentrations in media with which
humans come in contact. Exposure assessment provides
information for use in risk assessment on the magnitude.
frequency, and duration of the intersection between a stressor
and a receptor. During the assessment process, uncertain-
ties can arise from numerous sources, including extrapola-
tions from external concentration to internal dose. Some of
these uncertainties may be reduced by directly measuring
chemicals and/or their metabolites in biological samples
through biomonitoring studies. Historically, biomonitoring
studies have been used to confirm environmental exposures.
When carefully planned, these studies have the potential to
provide estimates of internal dose or evaluate possible health
risks. The full potential of biomonitoring, however, is yet to
be realized because obtaining the maximum possible value
from biomonitoring requires information, which is often
lacking, about exposure, toxicology, pharmacokinetics, and
epidemiology.1
In 2006, the National Research Council (NRC) of the
National Academy of Sciences (NAS) conducted an indepen-
dent study to review the current practices of interpretation
and uses of conventional biomonitoring data (i.e., chemicals.
their metabolites in human tissues/specimens). In its report.
the NRC identified data gaps when considering biomarkers
for specific applications, such as risk assessment. Specifically.
the NRC recommended the need to "develop biomonitoring-
based epidemiologic, toxicologic, and exposure-assessment
investigations and public-health surveillance to interpret
the risks posed by low-level exposure to environmental
chemicals."
In response to the NRC's recommendations, this document
outlines the research strategies proposed by researchers in
the U.S. Environmental Protection Agency (EPA) National
Exposure Research Laboratory (NERL) to generate data and
develop/refine tools for improving the use and interpretation
of biomonitoring data in human exposure and risk assess-
ment. These research strategies will address the following
key science questions.
1. What are the key elements of a source-to-outcome
continuum that includes biomarkers as a critical link for
exposure and health effects research?
2. How do we interpret biomonitoring data to improve
exposure and risk assessments using the methods.
measurements, and models developed or in use by the
research community?
3. How do we develop and incorporate new biomarkers
and apply emerging technologies to better assess human
exposure and resulting health effects?
The enhanced science and other products from this research
program will lead to the following expected outcomes.
1. A framework that provides guidance for assessing expo-
sures and/or health risks using biomonitoring data.
2. Innovative application of emerging/evolving technologies
for determining and analyzing biomarkers of exposure for
small molecules.
3. Incorporation of biomonitoring data and relevant expo-
sure, pharmacokinetic, and lexicological data/tools for
informing the Agency's risk assessment and management
decisions for human and wildlife health.
NRC (National Research Council) (2006). Human Biomonitoring for
Environmental Chemicals. National Research Council Committee on
Human Biomonitoring for Environmental Toxicants. National Academies
Press, Washington, DC.
-------
-------
2.0 A Source-To-Outcome Continuum for Human
Health Research 2
In this document, an updated source-to-outcome continuum
is presented to better define biomarkers as tools for human
health research. The components and linkages of this
updated continuum (Figure 1) highlight specific research
needs (i.e., measurements, models, estimated values) for
observational studies of human populations. The left side
highlights traditional components of exposure research
whereas the right side highlights contemporary components
of health effects research. Biomarker measurements are
central to the continuum and, therefore, link the exposure
and health effects components. The following sections define
individual components of the continuum and describe how
they can be used to answer exposure- and risk-based
questions for human health research.
The source-to-outcome continuum for human health research
(shown in Figure 1) contains the following eleven compo-
nents (not including source and outcome).
1. Environmental measurements are observed stressors
in environmental media that reflect (either directly or
indirectly) an exposure source. Although stressors can
be biological (e.g., bacteria), physical (e.g., radiation), or
even psychosocial (e.g., stress), chemical stressors in the
environment are the focus of this discussion. Examples
include chemical concentrations in foods (chemical mass
per unit food mass), in drinking water (chemical mass per
unit water volume), in consumer products (chemical mass
per item), and in outdoor and indoor air (chemical mass
per unit air volume).
2. Exposure models generate exposure estimates by math-
ematically combining environmental measurements with
human activity observations. Example human activities
in exposure models include eating food, drinking water.
applying consumer products, and spending time indoors
versus outdoors.
3. An exposure estimate is the predicted amount of chemical
(total mass) that comes into contact with a human. These
estimates can be route-specific (e.g., inhalation exposure.
ingestion exposure, dermal exposure) or summed across
all routes (i.e., aggregate exposure). An example expo-
sure estimate from the diet is food concentration (ug/
mg food) x food consumption (mg food/meal) = dietary
exposure (ug/meal). Exposure levels generally are fixed
in animal studies (e.g., mg/kg/day, referred to as "external
dose") but are inferred in observational exposure studies
using environmental measurements and human activity
observations.
2 Context in this section has been summarized in a published article: Sobus
et al. (2011). A biomonitoring framework to support exposure and risk
assessments. Science of the Total Environment, 409(22):4875-4884.
4. Dose models generate dose estimates by mathemati-
cally combining exposure estimates with parameters
that describe chemical movement into the body from
the site(s) of contact. For inhalation, ingestion, and
dermal exposure, chemical movement into the body
(i.e., absorption) often occurs through the lungs, gut, and
skin, respectively. Although exposure and dose models
are shown separately in Figure 1, parameters describing
chemical exposure and absorption often are included in
combined exposure-dose models.
5. A dose estimate is the amount of a chemical (total mass)
that enters the body. In health effects studies, dose levels
are unambiguous because conditions are generally
deliberate and well controlled. Furthermore, dose levels
in animal studies are adjusted by body weight (e.g., ug/
kg) to allow inter- and intraspecies comparisons and are
a basis for toxicity reference values (e.g., reference dose
[RfD]). Alternatively, dose estimates from observational
exposure studies rely on environmental measurements.
human activity observations, and absorption predictions
and are, therefore, subject to uncertainty.
6. Kinetic models mathematically describe the movement
of a chemical through the body; that is, the chemical's
distribution to various tissues, metabolism by vari-
ous processes, and ultimate elimination from the body.
(Absorption parameters from dose models frequently
are included in kinetic models to simultaneously address
absorption, distribution, metabolism, and elimination).
Parameters for these models generally are estimated using
in vitro metabolism experiments, in vitro kinetic stud-
ies with animals, and controlled human exposure stud-
ies. Parameterized models from these experiments can
be applied in observational exposure studies to predict
tissue/fluid levels of chemicals and metabolites following
exposure events.
7. Biomarker measurements are observations of chemicals.
chemical metabolites, and target molecules in media, such
as blood, urine, breath, fingernails, hair, milk, and feces.
These observations can reflect exposure events (biomark-
ers of exposure), health status (biomarkers of effect).
and systemic functions (biomarkers of susceptibility).
However, discussions herein pertain strictly to biomark-
ers of exposure. As such, example biomarkers include
native (unmetabolized) chemicals, phase-I metabolites
(e.g., oxidized, reduced, or hydrolyzed chemicals), and
phase-II metabolites (e.g., glutathione-, glucuronic acid-.
and sulfate-conjugated chemicals).
-------
Components of human health research
Symbol
A
n
Estimated
value
Measured
value
Empirical
model
Mechanistic
model
Parameter
1) Exposure estimate
2) Dose estimate
3) BR dose estimate
1) Environmental
measurement
2) Biomarker measurement
3) BR biomarker
measurement
1) Statistical model (blue)
2) Exposure model (red)
3) Dose model (red)
4) Kinetic model (red)
5) Dynamic model (red)
Definition
1) Estimated level of human contact with an analyte
2) Estimated level of analyte that enters a human
3) Estimated level of analyte that reaches a target within a human
1) Measured level of analyte in environmental media that reflects an
exposure source
2) Measured level of analyte in biological media that reflects a dose
3) Measured level of analyte in biological media that reflects a BR
dose
1) Model that evaluates observed variables for hypothesis testing
2) Model that estimates exposure using environmental
measurements and human activities
3) Model that estimates how much analyte enters a human
4) Models that describe how an analyte moves through and is
removed from a human
5) Model that describes the effect of an analyte on the human body
Figure 1. A source-to-outcome continuum for human health research.
-------
8. A biologically relevant (BR) dose estimate is the amount
of the dose that reaches a target and is available to
contribute to health impacts. For example, the BR dose
of a neurotoxic chemical may be the amount of chemi-
cal or metabolite that reaches the brain. For a genotoxic
chemical, the BR dose may be the amount of chemical
or metabolite that interacts with genetic material. These
values can be determined directly in some animal studies.
generally via biopsies of target tissues. However, BR dose
levels in observational exposure studies are estimated
using kinetic models and rely on preceding exposure and
dose estimates.
9. Dynamic models mathematically describe the impacts
of the BR dose on biological systems. For example.
dynamic models for neurotoxic chemicals may describe
the rates of enzyme inhibition in the brain, whereas those
for genotoxic chemicals may describe the rates of DNA
damage and repair. In general, these models are param-
eterized using data from in vitro experiments and in vitro
studies with animals. Combined kinetic and dynamic
parameter estimates then can be applied in observational
exposure studies to predict systemic changes as functions
of the estimated BR dose.
10. BR biomarker measurements are observations of analytes
in biological media that reflect (directly or indirectly)
the BR dose of a chemical or group of chemicals. BR
biomarkers can be chemical metabolites, chemically
altered molecules (e.g., adducts of reactive electrophiles).
or non specific markers of systemic processes (e.g., levels
of hormones, antibodies, or gene expression). They differ
from "biomarkers of effect" in that they are not strictly
markers of impaired function or disease endpoints. That
is, they may be, but are not required to be, associated
with key events in a disease process. For example, BR
biomarkers for genotoxic electrophiles can be markers of
genetic damage (direct markers) or products of reactions
with blood nucleophiles (indirect markers).
11. Statistical models are empirical models that compare
observed random variables for hypothesis testing. For
example, statistical models can evaluate associations
between environmental and biomarker measurements
of the same chemical and between biomarker measure-
ments and BR biomarker measurements. Statistical
models also can evaluate the effects on these relationships
of confounding variables, such as age, gender, human
activities, health status, and time (e.g., when measure-
ments are made). Therefore, statistical models are used to
attribute measurement variation to explanatory factors in
observational human health studies.
12. Figure 1 shows that components of the source-to-
outcome continuum align along two planes: (1) a plane
of measured values (i.e., environmental, biomarker, and
BR biomarker measurements) that are shown with blue
boxes; and (2) a plane of estimated values (i.e., exposure.
dose, and BR dose estimates) that are shown with red
triangles. Exposure, dose, and BR dose can be linked
to health outcome in controlled studies, yet all of these
values are estimated in observational studies. As such.
these values rely on measurements, activity observations.
and model parameter estimates and are, therefore, subject
to uncertainty. Biomarker measurements, which are at
the center of the continuum, can reduce uncertainties
by answering specific exposure- and risk-based ques-
tions. The following section demonstrates these uses of
biomonitoring via the following five research tiers.
Tier 1: Biomonitoring for exposure surveillance
Tier 2: Biomonitoring to support exposure
assessment
Tier 3: Biomonitoring to support risk assessment
Tier 4: Biomonitoring for exposure and risk
assessment
Tier 5: Biomonitoring to advance exposure and risk
assessments
Tier 1 considers only biomarker measurements, and subse-
quent tiers consider additional measurements, models.
and estimated values. Example biomarkers for each tier
are assumed to be measurable using reliable sampling and
analytical methods and to reflect exposure to environmen-
tal chemicals (a discussion of these assumptions is given
in section 5). Simple theoretical examples are given for the
biomonitoring tiers to demonstrate how biomarker data can
be used to answer important exposure- and risk-based ques-
tions. Theoretical examples are given, rather than results
from published studies, to enable continuity from one tier to
the next and to simplify the interpretation and discussion.
-------
-------
3.0 Biomonitoring Tiers for Exposure and
Health Research3
3.1 Tier 1: Biomonitoring for exposure
surveillance
Tier 1 analyses of biomarker data aim to answer one or more
of the following questions for exposure surveillance.
1. Who is exposed?
2. What are the exposure trends?
3. Which chemicals should be prioritized for higher tier
analyses?
Figure 2 is an adaptation of the source-to-outcome continuum
that shows biomarker measurements as the only requirement
for a tier 1 analysis. Specifically, cross-sectional biomarker
measurements are used in tier 1 analyses for evaluating
exposures across populations, and longitudinal biomarker
measurements are used for evaluating exposure trends within
3 Context in this section has been summarized in a published article: Sobus
et al. (2011). A biomonitoring framework to support exposure and risk
assessments. Science of the Total Environment, 409(22):4875-4884.
a population. To demonstrate these uses, two theoretical
examples are given in Figure 2. Example 1 displays cumula-
tive distributions of biomarker levels for two groups (a cross-
sectional analysis), and example 2 shows average biomarker
levels for one group as a function of time (a longitudinal
analysis).
In example 1, the two distributions represent biomarker
measurements that have been separated into groups for
hypothesis testing. Example groups include those separated
by gender (i.e., male versus female), geographic location
(i.e., location 1 versus location 2), age (e.g., < 18 years
old versus > 18 years old), source impact (e.g., product
users vs. nonusers), and health status (e.g., healthy versus
heath impaired). Observed differences between grouped
measurements indicate an effect of the grouping variable on
biomarker levels, and suggest exposure differences between
the groups. This type of tier 1 cross-sectional analysis is used
for identifying populations with elevated exposure levels
Kinetic
model
/\
I
Biomarker
measurements
Example 1: Cross-sectional analysis
/\
Example 2: Longitudinal analysis
*_
a °>
II
o
2
25 50 75
cumulative percentile
time
Figure 2. Requirements and examples of tier 1 analyses of biomarker data. (Gray objects are
unavailable in a tier 1 analysis.)
-------
(question 1 above) and increased risk of health impacts. In
particular, these comparisons are used for evaluating expo-
sures among vulnerable and susceptible subpopulations.
In example 2, longitudinal biomarker measurements for a
population are shown decreasing over time, suggesting a
similar decrease in exposure levels. Trends in longitudinal
biomonitoring studies (either increasing or decreasing) can
indicate a change in the source impacts on exposure (e.g..
deregistration of a consumer product) or a change in human
activities through which contact occurs (e.g., product use
patterns). However, higher tier analyses of the biomarker
data generally are needed to pinpoint the cause of a trend.
As such, tier 1 longitudinal analyses of biomarker data are
used to answer questions 2 and 3 above; that is, to identify
chemicals with changing health risks from exposure and to
prioritize chemicals for higher tier analyses.
3.2 Tier 2: Biomonitoring to support exposure
assessment
Tier 2 analyses of biomarker data can answer the following
questions to support exposure assessments.
1. What are the likely exposure sources?
2. What are the likely exposure routes?
As shown in Figure 3, tier 2 analyses consider environmental
and biomarker measurements (paired at the subject level)
and focus on statistical comparisons of these data. A graph in
Figure 3 shows a regression of spot biomarker measurements
on corresponding environmental measurements. A positive
linear trend is shown in this example with a R2 value of 0.3.
This indicates that biomarker levels increased with increas-
ing environmental levels, and that 30% of the biomarker
measurement variance was explained by corresponding
environmental measurements.
If, for example, the environmental measurements in this
example were concentrations of a chemical in food, and the
biomarker measurements were corresponding blood levels of
the same chemical, then the results of the regression analysis
would point to dietary ingestion as a likely exposure route.
Furthermore, the results would point to food or, perhaps, a
specific food item as an exposure source.
Considerable unexplained variance in the biomarker data
(i.e., 70%), however, would suggest additional exposure
routes and/or considerable covariate effects (e.g., timing of
sampling events) on biomarker levels. Therefore, additional
data would be necessary to better explain the observed
biomarker variance and to further support the exposure
assessment. These data could be part of a more complex
tier 2 analysis (e.g., environmental measurements collected
of different media to identify additional exposure routes) or
of a highertier analysis as described in the next sections.
3.3 Tier 3: Biomonitoring to support risk
assessment
Human health risk assessments for environmental chemicals
traditionally are based on environmental measurements.
observations of human activities, and information on chemi-
cal toxicity. Tier 3 analyses of biomarker data can be used to
support risk assessments because they can answer the follow-
ing questions:
1. What are the likely exposure levels?
2. What are the likely dose levels?
The requirements of a tier 3 analysis of biomarker data
are shown in Figure 4 and build on the tier 2 parameters
by adding exposure and dose models. Exposure and dose
estimates are not linked directly to biomarker measurements
in tier 3 analyses (see Figure 4) but can be indirectly linked
via statistical models (e.g., multiple regression models) that
collectively consider environmental measurements, human
activities, and other covariate effects. Given these linkages.
and that exposure and dose estimates can be compared to
risk-based reference values (e.g., RfDs), tier 3 analyses can
evaluate biomarker measurements within a risk context.
It is shown in Figure 4 that exposure and dose are estimated
with environmental measurements, human activity data
(included in exposure models), and uptake predictions but
generally not with biomarker measurements. Biomarkers are.
however, useful for evaluating exposure and dose estimates.
If, for example, air levels of a chemical were not associated
with corresponding biomarker levels, then exposure and dose
estimates based on inhalation likely would be incorrect and.
therefore, not comparable with reference values. On the other
hand, strong associations between food and biomarker levels
would support exposure and dose estimates based on dietary
ingestion.
In our tier 2 regression example (Figure 3), we showed how
measurements of a chemical in food explained 30% of the
observed biomarker variance. This result suggests that expo-
sure and dose estimates based on dietary ingestion would be
reasonable, and therefore comparable to reference values for
risk evaluation. However, given the added information in a
tier 3 analysis, it would be possible to explain more bio-
marker variance, thus increasing confidence in the exposure
and dose estimates.
For example, a graph in Figure 4 shows a regression of
biomarker levels on covariate-adjusted environmental levels
(as an analogue for dose). Here, the adjusted environmental
levels reflect for each individual the combined effects of
food concentration, food consumption, and factors affecting
dietary uptake (e.g., food allergies). A regression R2 value
of 0.6 in this example suggests that the combined effects of
food concentration and covariates could explain 30% more
biomarker variance than food concentration alone (shown
in Figure 3). Therefore, the additional information in tier 3
analyses can highlight other determinants of exposure e.g..
activities and support exposure and dose estimates for risk
assessment.
-------
Exposure
1
Environmental
Sftrtfetfcrf
modcfe
Blomirtwr
m^asuromonts
Simple regression analysts
R2 = 0.3
»-"***«5
environmental level
Figure 3. Requirements and an example of a tier 2 analysis of biomarker data. (Gray objects are
unavailable in a tier 2 analysis.)
RfD
/\
t
*"p
Environmental
measurement
mwftto
B4ofnartur
Complex regression analysis
R2 = 0.6
covariatc-adj.
environmental level
Figure 4. Requirements and an example of a tier 3 analysis of biomarker data. (Gray objects are
unavailable in a tier 3 analysis.)
-------
A
Example 1: Regression wing randomly
selected spot measurem&nls
slope-0.23
R2 = 0.22
10 successive biomarker
measurements of 50 subjects
6:GO am 12:00 prn 6:00 am 12-00 a
Example 2: Regression using end-of-day
measurements (12:00 am)
\
slope = 0.40
Rz-0.49
Exampte 3: Regression of 3 pooled
(averaged) measurements
slop* - 0.82
ff - 0.7S
biomarher
level
B
Example 1: Regression using randomly
selected spot measurements
12 14 1C
10 successive biomarker
measurements of 50 subjects
1
\
Example 2: Regression using end-of-day
s (12:00 am)
Example 3: Regression of 3 pooled
(averaged) measurements
slope «1.0
R2 = 0.99
3 4 <
S 10 13 14 16 IB
biomarker
level
! 4 6 B 10 12 14 16 18
Figure 5. Requirements and examples of tier 4 analyses of biomarker data. (Gray objects are
unavailable in a tier 4 analysis.)
-------
3.4 Special consideration for timed events,
sampling strategies, and repeated
measures
Timed events (e.g., frequency and duration of exposure, time
of biomarker sampling) for tier 2 and tier 3 analyses can
impact the interpretation of biomarker measurements with
respect to environmental measurements, exposure estimates.
and dose estimates. The magnitude of these impacts largely
depends on the between- and within-person components of
biomarker variance.
Figures 5 A and B show repeated biomarker measurements of
individuals from two theoretical groups. Both figures show
10 consecutive measurements from 50 subjects, with the first
measurements made at 6:00 a.m. and the final measurements
at midnight (12:00 a.m.) on the same day. The biomarker
levels in these figures vary between- and within-individuals
according to dose levels and kinetic processes. In Figure 5 A.
biomarker measurements are highly varied over time for each
individual and overlap considerably across individuals. These
observations suggest that individuals have similar dose levels
(based on daily average biomarker levels), and that chemical
uptake and elimination occurs rapidly throughout the day.
In Figure 5B, biomarker measurements are less varied with
time, and are more easily distinguished between individuals.
These observations suggest that individuals have different
dose levels (long-term averages) and that kinetic processes
occur more slowly.
Three example regressions of dose estimates on biomarker
levels are given in both Figures 5A and 5B. Dose is approxi-
mated for each individual as their average biomarker level
across all 10 measurements. In example 1 in both figures.
dose is regressed on randomly selected spot biomarker levels;
this simulates studies where one random biomarker measure-
ment is made for each subject. Example 2 in both figures
shows a regression of dose on end-of-day biomarker levels;
this simulates studies where one biomarker measurement is
made for each subject at a specific time point. In
example 3 in both figures, dose is regressed on the average of
three randomly selected measurements; this simulates studies
where repeated measurements are made for each subject, and
the measurements (or the biological samples themselves) are
pooled (averaged) prior to analysis.
The regression results from the three examples in Fig 5B
show very similar slopes (ranging from 0.92 to 1.0) and R2
values (ranging from 0.93 to 0.99). These results indicate that
timed events have little impact on biomarker interpretation
with respect to dose when the between-person component
variance is large. Specifically, these results suggest that the
biomarker measurements from each of these examples could
be used to accurately and precisely estimate dose levels.
In contrast, dissimilar regression results are shown from the
three examples in Figure 5A, indicating increased impacts of
timed events on biomarker interpretation when the within-
person component of variance is large. The best linear
association is shown in example 3 using the average of three
random biomarker measurements (slope = 0.82, R2 = 0.76).
This suggests that repeated biomarker measurements are
preferred over spot measurements for improving the accuracy
and precision of dose estimates. Furthermore, the regression
slopes in Figure 5 A show that spot biomarker measurements
(collected randomly or at a fixed time) can severely under-
estimate dose levels when the within-person component of
variance is large (attenuation bias).
3.5 Tier 4: Biomonitoring for exposure and risk
assessments
Figure 6 shows that tier 4 analyses of biomarker data include
the components for tier 3 analyses, as well as kinetic models
to link dose estimates and biomarker measurements, and
to predict BR dose levels. As such, tier 4 analyses can
answer the following the questions for exposure and risk
assessments.
1. What are the predominant exposure routes?
2. What are the best estimates of exposure and dose?
3. What is the estimated BR dose?
In the previous example of a tier 3 analysis (Figure 4), dose
estimates and biomarker measurements were not directly
linked. Rather, results from statistical comparisons were used
as support for dose estimates. Risk-based decisions can be
supported by statistical associations but can be further refined
with an understanding of mass transfer from exposure to dose
to biomarker levels; kinetic models are used to describe these
mass transfer processes. More specifically, they are used to
predict biological levels of chemicals and their metabolites
following exposure events.
Example 1 in Figure 6 shows a theoretical comparison of
observed and predicted biomarker levels over time. Here.
the predicted values are estimated blood levels of a chemi-
cal following three dietary exposure events (e.g., breakfast.
lunch, dinner). Assuming a well-parameterized and calibrated
model, good agreement between predicted and observed
values support the diet as the primary exposure route and
help validate exposure and dose estimates. Alternatively.
overestimation of the observed values would suggest incor-
rect exposure and dose estimates, whereas underestimation
could suggest additional exposure routes or endogenous
sources of the biomarker. In these situations, exposure and
dose estimates could be reconstructed to be consistent with
model predictions (discussion of exposure reconstruction is
given in section 4.1.2).
Given the appropriate model parameters, the same kinetic
models used to predict biomarker levels may be used to
predict the BR dose. Example 2 in Figure 6 shows target
levels over time of the same chemical from example 1. In
this theoretical example, the parent chemical is neurotoxic.
and the predicted values are brain tissue levels. As samples of
brain tissue are generally unavailable in observational human
studies, these predicted values are comparable only to obser-
vations from animals. Specifically, the area under the target-
level curve (AUCtarget, which is the time-integrated BR
dose) or the maximum level at the target, could be interpreted
-------
Example 2: Predicted target levels
t t
Exposure model Kinetic model
*
Environmental
measurement
I
Statistical
models
Biomarker
measurement
\Jyii3tr..
time
Example 1: Predicted vs. observed
biomarker levels
o
.0
predicted
observed
time
Figure 6. Requirements and examples of tier 4 analyses of biomarker data. (Gray objects are unavailable in
a tier 4 analysis.)
based on existing dose-response relationships. These types
of comparisons can inform exposure impacts on health risks
but may be refutable because of missing biomarker measure-
ments at the target (i.e., BR biomarker measurements).
3.6 Tier 5: Biomonitoring to advance exposure
and risk assessments
Tier 5 analyses of biomarker data include all components
of the source-to-outcome continuum, as shown in Figure 7.
That is, tier 5 analyses predict biomarker and BR biomarker
levels for comparison to measured values. These comparisons
enable tier 5 analyses to answer the following research ques-
tions to advance exposure and risk assessments.
1. What are the best estimates of BR dose?
2. What are the likely impacts of exposure on health
outcome?
3. What other factors may affect health outcome?
The previous section demonstrated how, in tier 4 analyses,
BR dose can be estimated and interpreted using existing
dose-response relationships. Because the BR dose estimates
in tier 4 analyses are not confirmed with measured values,
there is uncertainty in model predictions. Tier 5 analyses
can reduce this uncertainty via comparison of predicted and
observed BR biomarker levels.
Example 1 in Figure 7 shows predicted versus observed
levels of a BR biomarker. This is an extension of the
examples in Figure 6 where the brain was a target tissue, and
the stressor was a chemical neurotoxin found in food. In this
example, enzyme inhibition (e.g., cholinesterase) in the brain
was the desired outcome, but, because brain tissue is general-
ly inaccessible, blood enzyme levels were used as surrogates.
Kinetic and dynamic models were used to predict blood
enzyme levels following three theoretical dietary exposure
events. Predicted and observed levels were then compared to
evaluate the BR dose estimate.
In example 1, good agreement between measured and
predicted values indicates an accurate estimation of BR dose
and a good understanding of dynamic processes. Therefore,
the biomarker measurements in this example could be quan-
titatively linked backward to exposure (dietary intake) and
forward to potential health outcome (effects caused by inhib-
ited enzyme levels). However, poor agreement between BR
biomarker levels (observed versus predicted), combined with
accurate dose estimates from tier 4 analyses, would suggest
-------
t t *
Brposuremorfef Wnsffcmods/ dynamic motel
*
Environmental
measurement
I
Statistical
models
Biornarker
measurement
+
Statistical
models
BR biomarker
measurements
Example 1: Predicted vs. observed
BR biomarker levels
predicted
observed
03
73
_ cr
< I
IP 3
--
Example 2: Complex statistical model
time
ra _
E
CD
covariate-adj.
biomarker level
Figure 7. Requirements and examples of tier 5 analyses of biomarker data. (Gray objects are unavailable in
a tier 5 analysis.)
an incomplete understanding of dynamic processes in vitro
(possibly caused by uncertainties in interspecies extrapola-
tion). For example, overestimation of BR biomarker levels
could suggest the omission of important recovery processes,
whereas underestimation could suggest additional exogenous
or endogenous sources. In these instances, clarification would
be necessary before utilizing biomarker measurements from
observational studies for health effects research.
In addition to kinetic and dynamic models, statistical compar-
isons of biomarker and BR biomarker measurements are used
in tier 5 analyses to elucidate exposure impacts on health
outcome. These comparisons are generally more complex
than those in lower tier analyses, given repeated observations
of individuals. For example, in vitro dose-response associa-
tions can be informed using regressions of BR biomarker
levels (representing response) on biomarker levels (repre-
senting dose). Nondose related effects also can be observed
by including in the models covariates such as age, gender,
family health history, and genetic information.
Example 2 in Figure 7 shows a regression of BR biomarker
levels on covariate-adjusted biomarker levels. Continuing
from the previous example, this plot suggests that blood
enzyme levels decreased with increasing adjusted biomarker
levels. In other words, normal biological activities were
suppressed given elevated dose levels. The effects of indi-
vidual predictor variables on BR biomarker levels could be
examined to further explain this observation. Moreover, the
model results could help identify important in vivo processes
that could improve dynamic models. Overall, the combined
results of the kinetic, dynamic, and statistical models could
inform exposure and susceptibility effects on health outcome.
This section presented a simple biomonitoring framework
aimed at improving biomarker use and interpretation in
exposure and health research. Throughout the section,
simple definitions and examples were given to articulate
uses of biomonitoring data. These examples were not meant
to represent all biomonitoring research options but, rather,
to serve as a road map for establishing new studies and for
interpreting existing biomarker data. A summary of the uses
and requirements of the five biomonitoring tiers is given in
Table 1. Table 1 lists specific questions that can be answered
in a tiered analysis, as well as the measurements, models, and
model estimates that are required to complete an analysis.
-------
Table 1. The uses and requirements of the five biomonitoring tiers.
'
Exposure surveillance:
Who is exposed?
What are the exposure trends?
Which chemicals should be prioritized
for higher tier analyses?
Supporting exposure assessment:
What are the likely exposure sources?
What are the likely exposure routes?
Supporting risk assessment:
What are the likely exposure levels?
What are the likely dose levels?
Exposure and risk assessment:
What are the predominant exposure
routes?
What are the best estimates of exposure
and dose?
What are the likely BR dose levels?
Advancing exposure and risk
assessment:
What are the best estimates ofBR
dose?
What are the likely impacts of exposure
on health outcome?
What other factors may affect health
outcome?
1) Environmental
2) Biomarker
1) Environmental
2) Biomarker
1) Environmental
2) Biomarker
1) Statistical
1) Statistical
2) Exposure
3) Dose
1) Statistical
2) Exposure
3) Dose
4) Kinetic
1) Environmental
2) Biomarker
3) BR biomarker
1) Statistical
2) Exposure
3) Dose
4) Kinetic
5) Dynamic
None
1) Exposure
2) Dose
1) Exposure
2) Dose
3) BR dose
1) Exposure
2) Dose
3) BR dose
1 Single or repeated measurements
' Sepeated measurements only
-------
4.0 Filling in Information Gaps
As described in the previous section, the ability to utilize
biomonitoring data to assess human exposures and to evalu-
ate human health risks depends on the availability of various
types of information. Besides collecting additional exposure.
kinetic, or toxicity data, there are other approaches that can
be used to inform missing elements in the source-to-outcome
continuum, thus enabling biomonitoring data to be more
fully utilized in exposure and risk assessments. Approaches
presented in this section include exposure and dose modeling.
in silico modeling, and chemical surrogates, as well as the
exploration of 'omics biomarkers.
4.1 When data informing exposure estimates
are missing
Problem: When data that are used to inform exposure esti-
mates are insufficient (e.g., missing environmental measure-
ments), it is difficult to identify exposure pathways and routes
based on biomarker measurements. In this case, biomarker
measurements may only be useful as a surveillance tool
(i.e., tier 1 approach) or for developing hypotheses for future
research.
Potential solutions: In quantitative exposure assessment.
human exposures are estimated by combining environ-
mental measurements with information regarding human
activities, demographic and activity attributes of the intake/
uptake rates, and other exposure factors. Given correspond-
ing measurement and activity data, simple formulas (e.g..
concentrationxcontact timexexposure factors) with point
estimate inputs can be used to estimate exposure. In the
absence of corresponding exposure information, advanced
probabilistic simulations and mathematical algorithms can be
used to generate exposure and intake dose estimates. NERL's
Stochastic Human Exposure and Dose Simulation (SHEDS)4
is an example of such an advanced model. The output of
these probabilistic exposure models can be used as input
terms into pharmacokinetic models to further describe the
exposure- and dose-biomarker relationship.
Limitations: Besides the technical quality of the model
development process, the predictive ability of an exposure
model depends largely on the representativeness, relevancy.
and quality of the input data. When possible, input data
should be obtained through carefully designed observa-
tional exposure or survey studies (e.g., the Food and Drug
Administration's Total Diet Survey) that are representative of
a given population of interest. An exposure model that is built
based on these input data, however, is generally unsuitable
for predicting biomarker levels at the subject level because
of using population and non subject-specific inputs. Instead.
it is more appropriate to use probabilistic models to gener-
ate a distribution of estimated biomarker concentrations to
compare with an observed distribution of biomarker concen-
trations. These distributional comparisons may be helpful for
identifying potential exposure sources, pathways, and routes.
4.2 When kinetic information is missing
Problem: Kinetic data inform the specific metabolism/
biotransformation pathways of xenobiotic chemicals within
a biological system. Without this kinetic information, param-
eterization and performance of a pharmacokinetic model.
which is a predictive tool for describing the time course
of the exposure-biomarker relationship, become highly
uncertain.
Potential solutions: In the absence of kinetic data, chemoin-
formatics-based techniques, such as quantitative structure-
activity relationship (QSAR), can provide pragmatic estima-
tions of chemical-specific parameters for a provisional phar-
macokinetic model. Currently, many software packages (e.g..
MOE5 or QikProP6) exist whereby one can develop, augment.
and utilize new or existing QSAR. In addition, there are also
"trainable" QSAR models, such as ACD/Labs' "Suite of
predictors for PhysChen, ADME, and Tox," that may be used
to adapt and filter a model's predictive capability based on
chemical similarity indices.
Limitations: The effective use of QSAR or any molecular
model relies on the understanding of a model's domain of
applicability. When developing a QSAR model, a specific set
of chemicals (training set) is used to parameterize the model.
Bounded by the molecular properties of the training set, a
QSAR model is limited to a specific chemical space. In other
words, a QSAR model is best suited for interpolating data
within the model specification, but ill-suited for extrapolat-
ing outside the chemical space of the training set. To evaluate
chemicals outside of this chemical space, one will need to
reparameterize an existing model or to create a new one. In
such cases, the best approach is to generate relevant in vitro
chemical data to inform QSAR modeling efforts.
4 http://www.epa.gov/heasd/products/sheds multimedia/sheds mm.html
5 MOE (Molecular Operating Environment) is a software package
developed by Chemical Computing Group, Inc. It contains Structure-
Based Design, Pharmacophore Discovery, Protein & Antibody Modeling;
Molecular Modeling & Simulations, Cheminformatics & HTS QSAR, and
Medicinal Chemistry Applications.
6 https://www.schrodinger.com/products/14/17/
-------
4.3 When toxicity information is missing
Problem: Toxicity information provides an anchor for a
chemical in a risk context. Although having toxicity data
would not necessarily be sufficient to evaluate risk, in lieu
of data on apical toxicity, no clear risk assessment or risk
management activity can be performed. Thus, alternative
methods for estimating non-existent toxicity information are
discussed here.
Potential solutions: In addition to being used to parameter-
ize pharmacokinetic models, chemical surrogates and in
silica models may be used to infer toxicity when no such data
exist. Chemical surrogates provide a starting point for inter-
polating toxicity between chemicals. For instance, nonane
data may be an adequate surrogate for decane data if there
is an established trend of toxicity or other relevant data with
alkane chain length (i.e., in the case of nonane with C6, 7, 8.
and 10 alkanes). When chemical surrogates are unavailable
or deemed unreliable (e.g., low chemical similarity), the rela-
tionship of chemical structure to toxicity may be estimated
using in silica predictions. Several in silica techniques exist
by which predictions are made using chemical functional
group approaches. In this approach, the potential for health
risk is determined by a chemical substructure search against
a precomputed set of structural fragments that give rise to
toxicity alerts (e.g., the Osiris Property Explorer, http://www.
organic-chemistry.org/prog/peo/. and ADME/Tox Boxes.
http://pharma-algorithms.com/webboxesA. Aside from esti-
mating potential for health risks, QSARs can also be gener-
ated in the same manner to quantify toxicity on the basis of
chemical descriptors - historically, this has been reported for
biological outcomes such as Lethal Dose, 50% (LD50) in rats
and mice.
Limitations: The use of chemical surrogates relies heavily
on the following two assumptions: (1) the mode of action is
preserved across a class of chemicals, and (2) chemical simi-
larity among chemical class constituents is highly conserved.
Experimental validation of actual toxicity still should be
carried out when possible, and care should be exercised when
choosing index chemicals in the context of chemical similar-
ity. For both chemical surrogates and in silica approaches.
false positives and false negatives may reside in the predicted
outcomes because of inherent model structure assumptions
and/or training set limitations. In addition, to develop a reli-
able in silica approach, relevant toxicity data and relation-
ships must exist.
4.4 When biologically relevant biomarkers are
unidentified
Problem: To date, few biomonitoring studies have incor-
porated both biomarkers and biologically relevant biomark-
ers to expand the traditional biomonitoring study through
the further mapping of biologically relevant biomarkers to
toxicity starting points. This paucity in the combined use of
biomarkers and biologically relevant biomarkers stems, in
part, from insufficient routine analytical methods with appro-
priate quality assurance documentation for known biomark-
ers of interest.
Potential solutions: Researchers have incorporated new
cost-effective, high-sample-capacity analytical methods into
biomonitoring studies to help quantify chemical levels in
environmental and biological samples and to determine new
'omics-based biomarkers and BR biomarkers with which to
link exposure sources to health outcomes. A more detailed
discussion of such 'omics-based biomarker research is
provided below.
Proteomics is a bioanalytical tool that identifies proteins that
are altered through interactions with environmental chemi-
cals. The concept of protein expression signatures is based
on the measureable protein responses to chemicals in animal
studies. Sensitive, precise, and fast multianalyte methods for
measuring proteins in parallel can lead to protein fingerprint-
ing to aid in identifying new biomarkers and BR biomarkers.
In addition, multiplexed immunoassays (e.g., microarrays)
are becoming robust and reliable tools for high throughput
proteomic analyses to study the structure and functional inter-
actions between proteins and how these interactions control
complex processes in biological systems.
Another emerging technology that shows promise in biomon-
itoring research is metabolomics. Metabolomics includes
the study of toxicant-induced perturbations in endogenous
metabolites that result from exposures. Metabolomic stud-
ies conducted on various species have shown that systematic
patterns of change occur in the metabolome following expo-
sures to pesticides and other environmental chemicals. The
integration of metabolomics with proteomics and genomics
has great potential for providing the systems biology infor-
mation that will elucidate the complex relationships between
measurements of biomarkers/BR biomarkers and health
effects.
-------
It is important to note that, in the context of the 'omics tech-
nologies, a biomarker is a unique pattern of a large number
and variety of endogenous gene/protein/metaborite/biochemi-
cal changes. It has been proposed that these signatures (or
fingerprints) may be more informative and more chemical-
specific or exposure-pathway-specific than one or a few
conventional biomarkers (chemical concentrations in blood.
tissue, or excreta). Also, the flux (or dynamic time course) of
changes in 'omics pattern may yield a better estimate of the
time of occurrence of a single-event exposure. Furthermore.
'omics pattern changes may persist well after the chemical
stressor has cleared the body, which will be important when
interpreting biomonitoring results for nonpersistent chemi-
cals. Finally, because 'omic pattern changes often can be
linked to a specific mechanism of action, they may enable
identification of the harmful component following exposure
to a chemical mixture. Although these concepts have not yet
been fully tested and proven, preliminary work suggests that
studies with conventional biomarkers could be improved
significantly by augmenting them with information from
'omic techniques.
Preliminary work in NERL suggests that Nuclear Magnetic
Resonance (NMR)-based metabolomics may be particularly
appealing in biomonitoring studies, in large part because the
technique is well suited for relatively high-throughput analy-
sis. For example, the per-sample cost is low, little sample
preparation is required, and the instrument can be configured
for automated analysis. This is important when designing
experiments to establish endogenous biomarkers as indicators
of exposure because these investigations require analysis of a
great many samples (e.g., as a function of contaminant iden-
tity, magnitude, duration and timing of exposure). In addition.
the technique is readily amenable to blood as well as bioflu-
ids that can be taken noninvasively from humans (e.g., urine.
saliva, breath condensate).
Limitations: All of these highly multiplexed approaches
require well-characterized methods that are cost-effective.
Reagents must be standardized and screened in a
multiplexed environment before use. Methods development
must encompass quality assurance measures, standard operat-
ing procedures, data handling, interpretation, and reporting.
Current informatic systems are pressed to keep pace with the
increasing multiplexing capability of microarrays and flow
cytometry methods as the interpretation of data is extremely
challenging. Finally, linking 'omics-based biomarkers' or
biologically relevant biomarkers to exposures or health
outcomes requires significant resources.
-------
-------
5.0 Additional Considerations for Interpreting
Biomonitoring Data
Even when sufficient measurement data and predictive
models exist to fill in each component of the source-to-
outcome continuum (Figure 1), there are still many uncertain-
ties and data gaps that can complicate the interpretation of
biomonitoring data. In this section, some of these compli-
cations are discussed and approaches to minimize their
impact on the interpretability of biomonitoring data are
recommended.
5.1 Categories and uses of biomarkers7
The study design and interpretive options for a biomonitoring
study depend largely on the category to which a biomarker
belongs, as well as on the previously available information.
As such, there are sometimes surprises lurking in otherwise
simple biomonitoring strategies when a researcher misclas-
sifies a biomarker. Thus, it is important to be aware of the
category to which a biomarker belongs, as well as the poten-
tial inferences and uses in each category.
Biomarkers can be partitioned into four basic categories
based on their origin.
Group 1. Exogenous (native) chemicals are comprised
of exogenous biomarkers that are anthropogenic in origin and
are not formed via human oxygen/hydrocarbon metabolism.
Examples of exogenous chemicals include dioxins, poly cyclic
aromatic hydrocarbons (PAH), benzene/toluene/ethylbenzene/
xylenes (BTEX), polychlorinated biphenyls (PCB), and
organophosphate (OP) pesticides. Because these chemicals
have no endogenous sources, they can usually be attributed to
their respective environmental sources. It may be noted that
there may be multiple sources and routes of exposures and.
therefore, deducing the specific pathway requires additional
metadata, such as personal activity, geographic location.
occupation, and environmental measurements.
Group 2. Endogenous metabolites are comprised of a
great variety of "life process" chemicals widely studied in the
medical community, and they may be mapped quantitatively
to a lexicological initiating event. Examples of endogenous
biomarkers include cholesterol, liver enzymes, fatty acid
7 The four categories and uses of biomarkers are published in Tan et al.
(2012). Reconstructing Human Exposures Using Biomarkers and Other
"Clues", Journal of Toxicology and Environmental Health, Part B, 15(1).
22-38.
esters, urea, triglycerides, cytokines, and creatinine. These
biomarkers generally have no major environmental sources.
They, however, can be used to provide a baseline or a
statistical range for "normality" within a non- or minimally
exposed subpopulation; as such, outlier individuals can be
spotted quickly and further evaluated to determine whether
there is an environmental chemical that triggers those
biological responses.
Group 3. Ubiquitous organic compounds contain a
wide variety of chemicals that are known, trace-level, human
metabolites that are also present in similar concentrations in
the environment. Examples of ubiquitous organic compounds
include ketones, aldehydes, alcohols, phenols, amines, and
organic acids. Mostly, these compounds pose a dilemma
in microenvironments where, for example, an endogenous
exhaled chemical from subject A becomes an exogenous
inhalation exposure for subject B. As long as confounding
sources are properly identified and monitored, this group of
chemicals may aid in predicting health outcome or assessing
current health status when case-control or pattern recognition
strategies are used.
Group 4. Phase-1 andPhase-2 metabolites are
exogenous biomarkers but differ from Group 1 in that the
biomarkers are metabolites formed by biological processes
known as phase-1 and phase-2 metabolism. Group 4
biomarkers are the most frequently used biomarkers in
exposure assessment. Phase-1 and -2 metabolism generally
results in formation of more polar species that can be
eliminated readily eliminated in urine or feces. Phase-1
metabolism involves oxidation, reduction, or hydrolysis of
the parent chemical. Examples of phase-1 metabolites include
3,5,6-trichloro-2-pyridinol (a metabolite of chlorpyrifos) and
mono-ethyl phthalate (a metabolite of phthalate). Phase-2
metabolites are the result of further conjugation reactions
between phase-1 metabolites and larger biomolecules such
as glutathione, glucuronic acid, sulfates, and other peptides.
Some Phase-1 metabolites or their conjugated forms are
electrophilic and form stable adducts with nucleophilic sites
on proteins and DNA. Phase-2 metabolites are measured
primarily in blood and urine, although analytical methods
also exist for other media, such as exhaled breath condensate
and bronchoalveolar lavage fluid.
-------
5.2 Different categories of chemicals based on
their biological half life in relation to
exposure patterns
Information regarding the half-life of a chemical in the
sampled tissue can be used determine the exposure period
reflected by a biomarker measurement. If a chemical has long
half-life, the biomonitoring data are likely to reflect long-
term exposures. If a chemical has short half-life, the biomoni-
toring data often reflect the daily variation in exposure
patterns. Based on a priori knowledge regarding a chemical's
biological persistence and exposure patterns, NRC (2006)8
categorizes chemicals into one of four groups: (1) lipid-
soluble, bioaccumulative chemicals at steady state exposure;
(2) lipid-soluble, bioaccumulative chemicals not at steady
state exposure; (3) shorter half-life chemicals at pseudosteady
state exposure; and (4) short half-life chemicals that do not
approach steady state. For each category, NRC has provided
an example to demonstrate how human pharmacokinetic
models can be used to convert biomonitoring data to expo-
sure dose when kinetic and exposure data are available.
Unfortunately, kinetic or exposure data do not exist for
a majority of chemicals. When persistence data are lack-
ing for a chemical, QSAR can be used to infer persistence
on the basis of structural similarity to chemicals of known
persistence. In the absence of exposure data, NERL has the
expertise to collect data on exposure pathways and environ-
mental concentrations. Given that exposure and persistence
information is obtainable, different modeling approaches can
be applied to estimate exposure/dose based on biomarker
measurements.
For persistent chemicals that have half-lives in the
order of months to years, their long half-lives tend
to smooth out daily variations in exposure. In other
words, a biomarker measurement is more likely to
reflect chronic exposures or a single historical expo-
sure. For these chemicals, a simple pharmacokinetic
model with first-order clearance can be used to estimate
exposure dose for a given biomarker measurement. For
lipophilic chemicals, life-stage simulation will need to
be considered.
For semi-persistent chemicals that have half-lives in the
order of days to weeks, their biomarker concentrations
reflect exposures over a period on the order of a few
half-lives prior to sampling. For these chemicals, it is
critical to identify the time of biomarker collection with
respect to exposure events (e.g., the last meal eaten).
If this information is obtainable from biomonitoring
studies, exposure-pharmacokinetic modeling can be
conducted to identify key exposure events reflected by
a biomarker measurement. Otherwise, exposure
pharmacokinetic modeling can be used only to identify
the set of exposure scenarios plausibly associated with
the biomarker measurements.
For nonpersistent chemicals that have half-lives in the
order of minutes to hours, their biomarker concentra-
tion is heavily influenced by recent exposure events.
Uncertainties in time and duration of transient expo-
sures make interpreting these biomarker results chal-
lenging. One approach for dealing with nonpersistent
chemicals for which limited exposure information
exists is to use a probabilistic approach that estimates
a distribution of exposures that is consistent with the
observed distribution of biomarker measurements.
5.3 Exposure reconstruction9
Although most biomonitoring surveys only have data for
a tier 1 analysis, the call for exposure reconstruction from
biomarker data has challenged the use of all available
biomonitoring data to reconstruct exposures. This call is
motivated by the traditional risk assessment paradigm, and
the resulting estimates of safe "exposures" are based on
measures of administered dose (e.g., RfD, NOAEL, LOAEL)
or environmental concentrations (e.g., reference concentra-
tion [RfC], maximum contaminant level [MCL], National
Ambient Air Quality Standards [NAAQS]). In cases where
exposure guidance values exist, dose levels or environmental
concentrations converted by biomarker measurements can
then be compared with exposure guidance values for assess-
ing human health risks (Figure 8).
NRC (2006) recommended two main computational
approaches for such conversion.
a. Use human pharmacokinetic models to convert
biomarker measurements to dose levels that are
comparable to an RfD or other dose-based toxicity
value (reverse dosimetry)
b. Use animal pharmacokinetic models to convert
the administered dose-response relationship from
toxicology studies to a target site dose-response
relationship that can be used to evaluate human
biomonitoring results (forward dosimetry)
Both approaches begin with either a well-vetted kinetic
model (or physiologically based pharmacokinetic (PBPK)
model, often originating from controlled animal studies) or
a provisional model that is structurally representative and
complete as possible based on available kinetic data. Ideally.
if a human model exists or can be constructed using time
course data from controlled human studies, and the biomark-
er of interest is one of the model outputs, then the availability
of exposure information will determine whether a biomarker
measurement can be used to reconstruct exposure concentra-
tions. In cases where reliable exposure information exists.
a human pharmacokinetic model can be utilized to recon-
struct exposure concentration as the only unknown exposure
parameter (Figure 9, Steps 1, 2, and 3). The caveat is that
some variability and uncertainty in pharmacokinetics still
exist around the estimated exposure concentrations. When
NRC (National Research Council) (2006). Human Biomonitoring for
Environmental Chemicals. National Research Council Committee on
Human Biomonitoring for Environmental Toxicants. National Academies
Press, Washington, DC.
Part of this subsection is published in Tan et al. (2012). Reconstructing
Human Exposures Using Biomarkers and Other "Clues", Journal of
Toxicology and Environmental Health, PartB, 15(1), 22-38.
-------
RfDs
NOAELs
LOAELs
1 4
Exposur. model Kinetic model
Environmental
measurement
I
fTfCJ
Biomarker
measurement
+
moi
BR biomarker
AMAQSs
/WCLs
RfCs
Figure 8. Biomarker measurements to estimate dose levels, exposure levels and environmental stressor levels for
comparison to reference values. The blue box shows measured values, the red triangles and box show estimated values,
and the green arrows show the reconstruction pathways.
Human FK/Ftf K
model exists or can
be constructed
from human data
Animal HUM
model exists or can
he constructed
tiomarker is one of
the model outputs
All exposure-related
Exposure-related
parameteis tan be
reasonably estimated
parameteis except
concentration/dose are
characterized
Q Construct a
provisional human
model to support
exposure/risk
assessment
GConduct animal
kinetic studies to
inform the ADME
processes and the
dose-biomarker
relationship
GTest hypotheses
and identify data
gaptregaicting
exposures
Q Reconstruct
concentration/dose
Q Estimate a distribution
of dose/concentration
Figure 9. The type of exposure information that a biomarker measurement can infer depends on the availability of
kinetic and exposure data.
-------
large variability and uncertainty are expected for other expo-
sure parameters (e.g., duration, frequency), a probabilistic
approach should be conducted for estimating a distribution of
potential exposure concentrations (Figure 9, Steps 1,2,4, and
5). In the case where exposure pathways are not well charac-
terized or understood, a pharmacokinetic model only can be
used to test hypotheses of potential exposure scenarios and
identify data gaps regarding exposure information (Figure 9.
Steps 1,2, 4, and 6).
If most exposure information can be obtained, different
modeling approaches can be applied in exposure reconstruc-
tion based on the biological half-life of the chemical in body
(see Section 5.2). When only an animal pharmacokinetic
model is available, a provisional human model may be
constructed to identify exposures of greater concern or to
complement other exposure measurements to assist in expo-
sure and risk assessment (Figure 9, Steps 7 and 8).
A provisional model is expected to produce a "first approxi-
mation" of chemical disposition that requires additional
human data to support. Sometimes, when the biomarker of
interest is not one of the model outputs (Figure 9, Steps 1
& 9), or neither a pharmacokinetic model nor kinetic data
exist (Figure 9, Steps 1, 7, and 10), animal kinetics studies
will need to be conducted to inform the ADME process and
the dose-biomarker relationship. These in vivo data can be
used to develop and parameterize an animal pharmacokinetic
model, which can subsequently be "scaled up" to a provision-
al human model. Alternatively, in silica molecular models
may be utilized to parameterize a provisional human model.
The provisional model then can be used to test hypothesis or
to identify data gaps.
5.4 Nondetect data
Analytical chemistry methods have limits of detection (LOD)
and limits of quantification (LOQ) that rarely (if ever) include
zero. Improvements in analytical capability would be expect-
ed to lower the LOD and possibly the LOQ toward zero, but
never at zero. In addition, it is problematic to treat nondetects
statistically.10 Assigning zero to nondetects is a problem
for transformation (angular, square root, or logarithmic) to
approximate the normal distribution. A large number of zero
values is equally challenging for normal score transformation
(too many ties) and the use of distribution-free procedures.
One common solution to this problem is to truncate the
distribution of environmental/biomarker measurements at
the LOD and to consider only the positive values for statisti-
cal analysis. This approach, however, would suggest that the
nondetect values are uninformative or missing values rather
than true values indicative of low or nonexposure. Censoring
zero values when they are true is like imposing a type 1 error.
rejecting a hypothesis that is actually true, before the start of
the test. In the case of biomonitoring in which comparisons
are sought between populations, subpopulation, and cohorts
by time and location, left-truncation leaves the Poisson or
log-normal distributions biased in favor of positive (overesti-
mated) exposure.
An alternative to left-truncation of a continuous variable
involves imputation of replacement values for non-detects
(allowing some to be actual zero) at the relevant LOD (e.g..
2/3X, X/2, X/A/2). However, this imputation assumes that the
nondetect values are members of that same population (i.e..
each trial has only a single possible outcome). Thus, a test of
normality is expected to follow. Rejection of the hypothesis
of normality would render the need for the imputations moot.
and consideration must be given to alternative binomial or
multinomial treatments (i.e., each outcome can have two or
more possible outcomes).
5.5 Nonspecific biomarkers
The presence of nonspecific biomarkers is relatively
common. Nonspecific biomarkers can arise when multiple
parent chemicals can yield the same metabolite that is the
biomarker or when both the parent chemical and its degradate
(which is also the metabolite used as the biomarker) co-occur
in the environment. Concurrent exposures to multiple chemi-
cals that yield the same biomarker can lead to false positive
interpretations related to an exposure. Successful attribu-
tion of nonspecific biomarkers to actual exposure lies with
studious understanding of metabolism and pharmacokinetics.
combined with knowledge of the spatial and temporal condi-
tions of exposures. Additional information of environmental
conditions, such as prior determination of ratios of chemical
of interest/other chemicals can ameliorate false-positive inter-
pretations and make dose-biomarker relationships predict-
able. When exposure to the chemical of interest overshadows
concurrent exposures to other chemicals, one may assume
that the biomarker is originated from one chemical. When
exposures to other chemicals are also significant, one may
use the predetermined environmental concentration ratio as
input for a pharmacokinetic model to estimate the relative
contribution of each parent chemical to the output biomarker
concentration by tracking the time course of the metabolite
from each parent chemical. However, if the ratio of multiple
chemicals varies spatially and temporally because of differ-
ent rates of production/use and environmental degradation.
it will be necessary to collect more environmental measure-
ments and/or use a fate and transport model in conjunction
with an exposure model to estimate the ratio of different
parent compounds' presence in the environment as a function
of time and space. Once the fate and transport and exposure
pathways are characterized, the uncertainty in the environ-
mental concentration of the parent chemicals is reduced, and
such information then can be used as input for a pharmaco-
kinetic model to estimate the dose-biomarker relationship.
On the other hand, if the metabolite, rather than the parent
compound, is the toxic moiety, a nonspecific biomarker may
be advantageous in that it integrates exposure to all parent
chemicals within a single measure. Thus, the necessary
approach for establishing exposure-biomarker relationship is
dictated largely by the nature of exposure conditions, as well
as the toxic mechanism of action.
0 Taylor et al. (2001). A Mixture Model for Occupational Exposure Mean
Testing with a Limit of Detection. Biometrics, 57(3), 681-688.
-------
5.6 Stereochemistry of biomarkers
Many environmental chemicals and biologically relevant
molecules are "optically active" or chiral by nature.
Anthropogenic (e.g., synthetic pesticides) chiral chemicals
often are synthesized as racemic mixtures, containing 2n
isomeric components (where n = number of asymmetric or
stereogenic centers, atoms with four different substituents).
Enantiomers are stereoisomers that are nonsuperimposable
mirror images of one another and have the same inher-
ent physical-chemical properties (i.e., boiling point, vapor
pressure, and molecular weight), whereas diastereomers and
nonsuperimposable nonmirror images have similar but
nonequivalent physicochemical properties, making them
separable. Most chiral natural products (e.g., sugars, amino
acids, lipids, and nucleic acids) exist as an enriched single
isomeric form (i.e., homochiral). Many anthropogenic chemi-
cals (e.g., pesticides and therapeutic drugs) may be designed
or formulated to favor one biologically active racemate over
the other to improve efficacy. Known enrichment/enhance-
ment factors of stereoisomeric mixtures (natural vensus
anthropogenic) can provide a simple basis for differences in
exposure when considering multimedia environmental factors
(e.g., biological degradation pathways). However, molecular
interactions involving enantiomers with endogenous mole-
cules usually have different binding kinetics (e.g., affinities)
with enzymes involved in metabolism or interactions with
target ligands. Therefore, chirality may be expected to impact
exposure through ADME and, in turn, stereospecific differ-
ences in risk. Although consideration of chirality may seem
to complicate exposure and risk assessments, consideration
of the impact of chirality (or stereochemistry) on exposure
and risk assessment is unavoidable. Because enantiomers
can undergo reactions and metabolize differently in the body.
chiral biomarkers may be used as unique markers that can be
related back to exposure. By characterizing stereoselectivity
in exposure for parent compounds and metabolites, additional
information can be gleaned about exposure pathways and
internal chemical disposition that cannot be obtained from
remedial assessments based on achirality or two dimen-
sional planar assumptions. Ultimately, boundary conditions
between tissue distribution, metabolism and toxicity require
knowledge of individual isomeric fate in the same way that
mixtures of multiple chemicals with uniquely defined kinetics
cannot be reduced by simple additive action as additional
mass of each single chemical under the rubric of aggregate
exposure and cumulative risk assessment.
-------
-------
6.0 Future Directions for Research on Biomarkers
of Exposure
The future directions for research on biomarkers conducted
in NERL will support several new research programs in the
Office of Research and Development (ORD): Air, Climate.
and Energy (ACE), Sustainable and Health Communities
(SHC), and Chemical Safety for Sustainability (CSS). More
specifically, one of the eight topics in CSS is "Biomarkers."
Currently, the linkage and translation of exposure and hazard
data into human or ecological risk are conducted indepen-
dently, which can often lead to data gaps and scientific
uncertainties. One of the promising tools for linking different
elements along the source-to-outcome continuum to under-
stand the public health implications of exposure to environ-
mental chemicals is biomarkers. Thus, the overall goals of the
biomarkers research are (1) to develop the scientific knowl-
edge and tools that will improve the use of biomonitoring
data in both single and multiple chemical risk assessment and
risk management decisions and (2) to improve our under-
standing of the fundamental processes and linkages along the
exposure-dose-effects continuum that lead to risk.
Research in the biomarkers research topic will be led
by scientists in NERL and the National Health and
Environmental Effects Research Laboratory (NHEERL) and
is organized around two projects outlined below.
1. Project 1 will identify biomarkers/bioindicators11 and
approaches of interpretation in the context of establishing
exposure to dose to outcome linkages. With biomarkers/
bioindicators linkages defined, this project will contrib-
ute to Project 2 that uses systems models for predicting
adverse health and environmental effects of human and
wildlife exposures.
2. Project 2 will evaluate the predictive models and
develop robust tools for monitoring exposure and effects
in clinical, epidemiological, and ecological field stud-
ies. These efforts will be coordinated with other CSS.
ACE, and SHC topics to identify and understand the most
important exposure sources, routes, and pathways for
high-priority chemicals for human and wildlife species
and how exposure is related to adverse outcome.
11 Bioindicators are defined as measurements of biochemical or physiological
changes within an organism that reflect biological responses.
For NERL, the research direction will focus in the following
three areas.
1. Data collection and analysis
a. Develop a knowledge base of novel and existing
biomarkers of exposure for high-priority and high-
interest emerging chemicals
b. Develop or improve measurement and analytical
methods for environmental and biological samples
c. Perform observational studies to collect data on
environmental and biomarker concentrations, human
time/location activities, and product/chemical use
patterns
d. Develop and identify new biomarkers of exposure
that are better indicators of exposure
2. Predictive modeling tools
a. Develop and apply in silico models to estimate inher-
ent and derived chemical properties based on chemi-
cal structures for informing the selection of proper
biomarkers
b. Evaluate the correlation among empirical environ-
mental and biomarker data using statistical models
c. Describe the exposure-biomarker relationship using
pharmacokinetic models
d. Use in vitro data or in silico models to estimate phar-
macokinetic data for provisional or screening-level
pharmacokinetic models
3. Integrated research
a. Integrate environmental media measurements, human
activity observations, and other exposure factors with
pharmacokinetic models for linking exposures to
biomarkers
b. Identify normal ranges of "probative" biomarkers and
quantify resiliency and shifts in homeostasis using
biomarkers
c. Evaluate variability and susceptibility using bio-
marker measurements
-------
------- |