United States
Environmental Protection
1=1 m m Agency
EPA/690/R-09/023F
Final
9-10-2009
Provisional Peer-Reviewed Toxicity Values for
Ethylbenzene (CASRN 100-41-4)
Derivation of a Subchronic Oral Provisional-RfD
and a Subchronic Inhalation Provisional-RfC
Superfund Health Risk Technical Support Center
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH 45268

-------
Commonly Used Abbreviations
BMD
Benchmark Dose
IRIS
Integrated Risk Information System
IUR
inhalation unit risk
LOAEL
lowest-observed-adverse-effect level
LOAELadj
LOAEL adjusted to continuous exposure duration
LOAELhec
LOAEL adjusted for dosimetric differences across species to a human
NOAEL
no-ob served-adverse-effect level
NOAELadj
NOAEL adjusted to continuous exposure duration
NOAELhec
NOAEL adjusted for dosimetric differences across species to a human
NOEL
no-ob served-effect level
OSF
oral slope factor
p-IUR
provisional inhalation unit risk
p-OSF
provisional oral slope factor
p-RfC
provisional inhalation reference concentration
p-RfD
provisional oral reference dose
RfC
inhalation reference concentration
RfD
oral reference dose
UF
uncertainty factor
UFa
animal to human uncertainty factor
UFC
composite uncertainty factor
UFd
incomplete to complete database uncertainty factor
UFh
interhuman uncertainty factor
UFl
LOAEL to NOAEL uncertainty factor
UFS
subchronic to chronic uncertainty factor
1

-------
FINAL
9-10-2009
PROVISIONAL PEER-REVIEWED TOXICITY VALUES
FOR ETHYLBENZENE (CASRN 100-41-4)
DERIVATION OF A SUBCHRONIC ORAL PROVISIONAL-RFD
AND A SUBCHRONIC INHALATION PROVISIONAL-RFC
Background
On December 5, 2003, the U.S. Environmental Protection Agency's (U.S. EPA) Office of
Superfund Remediation and Technology Innovation (OSRTI) revised its hierarchy of human
health toxicity values for Superfund risk assessments, establishing the following three tiers as the
new hierarchy:
1)	U.S. EPA's Integrated Risk Information System (IRIS).
2)	Provisional Peer-Reviewed Toxicity Values (PPRTVs) used in U.S. EPA's Superfund
Program.
3)	Other (peer-reviewed) toxicity values, including
~	Minimal Risk Levels produced by the Agency for Toxic Substances and Disease
Registry (ATSDR),
~	California Environmental Protection Agency (CalEPA) values, and
~	EPA Health Effects Assessment Summary Table (HEAST) values.
A PPRTV is defined as a toxicity value derived for use in the Superfund Program when
such a value is not available in U.S. EPA's IRIS. PPRTVs are developed according to a Standard
Operating Procedure (SOP) and are derived after a review of the relevant scientific literature
using the same methods, sources of data, and Agency guidance for value derivation generally
used by the U.S. EPA IRIS Program. All provisional toxicity values receive internal review by
two U.S. EPA scientists and external peer review by three independently selected scientific
experts. PPRTVs differ from IRIS values in that PPRTVs do not receive the multiprogram
consensus review provided for IRIS values. This is because IRIS values are generally intended
to be used in all U.S. EPA programs, while PPRTVs are developed specifically for the Superfund
Program.
Because new information becomes available and scientific methods improve over time,
PPRTVs are reviewed on a 5-year basis and updated into the active database. Once an IRIS
value for a specific chemical becomes available for Agency review, the analogous PPRTV for
that same chemical is retired. It should also be noted that some PPRTV documents conclude that
a PPRTV cannot be derived based on inadequate data.
Disclaimers
Users of this document should first check to see if any IRIS values exist for the chemical
of concern before proceeding to use a PPRTV. If no IRIS value is available, staff in the regional
Superfund and Resource Conservation and Recovery Act (RCRA) program offices are advised to
carefully review the information provided in this document to ensure that the PPRTVs used are
appropriate for the types of exposures and circumstances at the Superfund site or RCRA facility
in question. PPRTVs are periodically updated; therefore, users should ensure that the values
contained in the PPRTV are current at the time of use.
1

-------
FINAL
9-10-2009
It is important to remember that a provisional value alone tells very little about the
adverse effects of a chemical or the quality of evidence on which the value is based. Therefore,
users are strongly encouraged to read the entire PPRTV document and understand the strengths
and limitations of the derived provisional values. PPRTVs are developed by the U.S. EPA
Office of Research and Development's National Center for Environmental Assessment,
Superfund Health Risk Technical Support Center for OSRTI. Other U.S. EPA programs or
external parties who may choose of their own initiative to use these PPRTVs are advised that
Superfund resources will not generally be used to respond to challenges of PPRTVs used in a
context outside of the Superfund Program.
Questions Regarding PPRTVs
Questions regarding the contents of the PPRTVs and their appropriate use (e.g., on
chemicals not covered, or whether chemicals have pending IRIS toxicity values) may be directed
to the U.S. EPA Office of Research and Development's National Center for Environmental
Assessment, Superfund Health Risk Technical Support Center (513-569-7300), or OSRTI.
INTRODUCTION
Ethylbenzene has a chronic RfD, a chronic RfC, and a cancer descriptor of "Not
classifiable as to human carcinogenicity" on IRIS (U.S. EPA, 1991a). Thus, only subchronic
toxicity values are presented in this toxicity assessment. There is an Agency for Toxic
Substances and Disease Registry (ATSDR, 1999) assessment of ethylbenzene. ATSDR has
since posted an updated version of Toxicological Profile for Ethylbenzene (ATSDR, 2007) in
September 2007, but it is only a draft for public comment—not the official citable final report.
Therefore, any potential changes or updates in toxicity values (critical effects, principal study,
etc) are not described here in the PPRTV document. In order to determine whether newer data
might be available to support subchronic noncancer toxicity values for ethylbenzene, a targeted
literature search was conducted to identify human or in vivo animal studies of appropriate
duration and quality to serve this purpose. Literature searches were limited to studies published
between 1999 and August 2007 in the following databases: MEDLINE, TOXLINE, BIOSIS,
TSCATS, DART/ETIC, GENETOX, HSDB, and Current Contents. The searches included terms
to identify human exposure studies (epidemiologic, occupational) and animal studies for
noncancer endpoints and less-than-chronic durations. The searches included health effects and
toxicity information available from the U.S. EPA (IRIS), ATSDR, and other relevant federal,
state or international governmental or quasi-governmental agencies, including, but not limited to,
ACGIH, NIOSH, OSHA, NTP, IARC, WHO, and CalEPA. In addition, electronic databases,
including CURRENT CONTENTS, MEDLINE, TOXLINE, BIOSIS/TOXCENTER,
TSCATS/TSCATS2, CCRIS, DART/ETIC, GENETOX, HSDB, and RTECS, were searched.
Studies having the potential ability to inform the derivation of subchronic noncancer toxicity
values were retrieved and a critical study was selected.
The derivation of subchronic toxicity values for ethylbenzene is discussed below. A brief
rationale is provided for the selection of the critical study and endpoint, a summary of the critical
study is presented, and the subchronic toxicity value derivations are described. Further
information on the toxicology and toxicokinetics of ethylbenzene can be found in the
2

-------
FINAL
9-10-2009
ATSDR (1999) Toxicological Profile for Ethylbenzene or on IRIS (www.epa.gov/iris). The
health effects associated with ethylbenzene exposure are currently being reassessed by the IRIS
Program (see IRIS Track at http://cfpub.epa.gov/ncea/iristrac/index.cfm).
REVIEW OF PERTINENT DATA AND DERIVATION OF PROVISIONAL
SUBCHRONIC TOXICITY VALUES FOR ETHYLBENZENE
Subchronic p-RfD
The chronic RfD for ethylbenzene on IRIS (0.1 mg/kg-day) was verified in May 1985
based on liver and kidney toxicity in a "subchronic-to-chronic" rat study (Wolf et al., 1956). It
includes a UF of 10 for subchronic-to-chronic extrapolation. There is no intermediate duration
oral MRL for ethylbenzene; ATSDR (1999) considered the Wolf et al. (1956) study to be of
inadequate quality for the purpose of MRL derivation.
Only one oral study potentially useful for subchronic p-RfD derivation was identified in
the update literature searches: a 13-week rat gavage study (Mellert et al., 2007). This study was
conducted in compliance with GLP and OECD guidelines, used both male and female rats,
evaluated a wide variety of endpoints, and reported both data and results of statistical analysis on
all relevant findings. In contrast, the study by Wolf et al. (1956) utilized for the IRIS RfD used
only female rats, evaluated only a subset of endpoints, and reported results qualitatively.
Mellert et al. (2007) identifies LOAEL and NOAEL values (250 and 75 mg/kg-day,
respectively), which are very near the values reported by Wolf et al. (1956) (291 and
97 mg/kg-day), and they identified the same target organs (liver and kidney). The
Mellert et al. (2007) study is considered a more suitable study for determining a POD than
Wolf et al. (1956) and, therefore, is used to derive the subchronic p-RfD for ethylbenzene.
Mellert et al. (2007) treated groups of Wistar rats (10/sex/dose) with ethylbenzene
(99.7% pure) by gavage at doses of 0, 75, 250, or 750 mg/kg-day 7 days/week1 for 13 weeks.
Animals were examined daily for mortality and signs of toxicity, while a detailed clinical
examination was performed weekly. Weekly measurements were made of food and water intake
and body weights. Urine was collected for analysis (color, turbidity, volume, specific gravity,
pH, protein, glucose, ketones, urobilinogen, bilirubin, blood, microscopic examination of
sediment) and blood samples for hematology and clinical chemistry (details of each not given)
were collected at study termination. Ophthalmology, functional observational battery (FOB) for
neurobehavioral effects and motor activity were evaluated during the final week of treatment.
All animals were necropsied, and major organs (adrenal glands, brain, epididymes, heart,
kidneys, liver, ovaries, spleen, testes, thymus, thyroid, and uterus) were weighed. Microscopic
examination of a comprehensive list of tissues (>45 tissues) was performed in the control and
high-dose animals, while the liver, kidney, and pancreas were examined in all groups. Male
kidneys were also examined using Mallory-Heidenhain staining for hyaline droplets.
Clinical signs in treated animals included postdosing salivation (all mid- and high-dose
animals, as well as one low-dose male) and discolored urine observed in the bedding (but not on
urinalysis) in high-dose animals of both sexes (Mellert et al., 2007). Body weights were
1 Daily gavage administration confirmed by personal communication, Dr. Bennard van Ravenzwaay.
3

-------
FINAL
9-10-2009
significantly lower than controls in high-dose males beginning in Week 5; terminal body weights
in this group were 14% lower than controls (p < 0.01). Water consumption was significantly
increased in mid- and high-dose males and in high-dose females (p < 0.01), and food
consumption was increased in high-dose males (p < 0.05). The FOB revealed a significant
decrease in landing foot-splay in high-dose males, which the authors attributed to decreased body
weight. Motor activity was significantly increased in high-dose females (p < 0.01), but the
pattern of changes was considered inconsistent with treatment-related effects; the authors
reported that treatment-related effects are usually observed at the beginning or end of
measurement, whereas effects in the high-dose females were observed intermittently.
Hematology analysis indicated a statistically significant (p < 0.01) increase in mean corpuscular
volume in high-dose males (5% higher than controls) and mid- and high-dose females (2-4%), as
well as a significant reduction (p < 0.01) in platelet count in high-dose females (15%); there were
no other hematology changes. Absolute and relative thymus weights were decreased in mid- and
high-dose females, but no pathology was observed upon microscopic examination of this organ.
Table 1 shows relevant changes in clinical chemistry, urinalysis, organ weights, and
histopathology. Evidence for liver toxicity at the mid- and high-doses included clinical
chemistry effects (e.g., increases in ALT, GGT, and bilirubin), dose-related increases in absolute
and relative liver weight, and dose-related increases in the incidence of centrilobular hepatocyte
hypertrophy. Kidney effects in mid- and high-dose animals included clinical chemistry changes
in both sexes (e.g., increased serum urea, potassium, and calcium in males; increased serum
potassium in females), urinalysis findings (increased incidences of transitional epithelial cells
and granular and epithelial cell casts in males), dose-related increases in relative (both sexes) and
absolute (males only) kidney weights, and increased severity of hyaline droplet nephropathy in
male rats. The only treatment-related finding at the low dose was increased relative—but not
absolute—liver weight in male rats (4% higher than controls, p < 0.01).
The study authors identified the low dose (75 mg/kg-day) as a NOAEL and the mid-dose
(250 mg/kg-day) as a LOAEL for centrilobular hepatocyte hypertrophy with clinical chemistry
changes indicative of liver and kidney effects. Findings at 250 mg/kg-day that support the
identification of a LOAEL based on liver toxicity include histopathology (centrilobular
hepatocyte hypertrophy) and increased absolute and relative liver weight, in conjunction with
changes in several clinical chemistry measures (increased ALT, GGT, bilirubin, and cholesterol
in males; increased cholesterol and decreased prothrombin time in females).
Mellert et al. (2007) observed liver and kidney effects at lower doses than other
endpoints. Evidence of mild kidney impairment in males exposed to the LOAEL included
urinalysis changes (transitional epithelial cells and granular and epithelial cell casts in urine),
clinical chemistry findings (increased potassium and calcium), increased relative kidney weights,
and an increase in the severity of hyaline droplet nephropathy. Hyaline droplet nephropathy is
related to the accumulation of a2U-globulin, an effect that is specific to the male rat and not
relevant to humans (U.S. EPA, 1991b). Evidence for the role of a2U-globulin includes the
increased incidence and severity of hyaline droplet formation and granular cell casts in the urine.
The only kidney effect observed in female rats exposed to the LOAEL was a slight—but
statistically significant (p < 0.01)—increase in relative kidney weight (7% above controls). In
females at the high-dose, there were slight increases in sodium, potassium, and magnesium
concentrations along with increased relative kidney weight (13%) that indicate a potential effect
4

-------
FINAL
9-10-2009
Table 1. Significant Effects on Liver and Kidney in Rats Treated with Ethylbenzene
via Gavage for 13 Weeksa

Control 75 mg/kg-day
250 mg/kg-day
750 mg/kg-day
Males
Clinical Chemistry
ALT (ukat/L)
0.62 ±0.12"
0.70 ±0.12
0.89 ± 0.26c
1.11 ± 0.23°
GGT (nkat/L)
2 ± 3
6 ± 6
10 ± 6C
10 ± 6C
Total bilirubin (nmol/L)
2.20 ±0.35
2.41 ±0.53
3.05 ± 0.80c
3.40 ± 0.83c
Albumin (g/L)
36.8 ± 1.0
37.0 ±0.6
37.6 ±0.6
38.1 ± 1.0d
Cholesterol (mmol/L)
1.76 ±0.26
1.68 ± 0.11
2.23 ± 0.24c
2.21 ±0.26c
Creatinine (mmol/L)
54.7 ±3.8
53.6 ±3.2
52.1 ±4.2
49.2 ± 3.2C
Serum urea (mmol/L
4.13 ±0.46
4.01 ±0.44
4.07 ±0.59
4.87 ± 0.78d
Potassium (mmol/L)
4.59 ±0.26
4.71 ±0.27
4.89 ± 0.32d
4.98 ± 0.36d
Calcium (mmol/L)
2.75 ±0.04
2.78 ±0.07
2.84 ± 0.05c
2.82 ± 0.10d
Magnesium (mmol/L)
0.92 ±0.04
0.94 ±0.03
0.94 ±0.06
1.00 ± 0.04c
Urinalysis'
Transitional epithelial cells, grade >2
1/10
4/10
7/10°
8/10°
Granular and epithelial cell casts
3/10
7/10
9/10°
8/10d
Body Weight on Day 91 (% change
from control)
372 g
-0.9%
-2.7%
-13.8%c
Organ Weights
Absolute liver weight (g)
8.02 ±0.55
8.26 ±0.81
10.25 ±0.98c
9.88 ± 0.98c
Liver/body weight (%)
2.26 ±0.08
2.36 ± 0.08c
3.01 ± 0.14°
3.31 ± 0.13°
Absolute kidney weight (g)
2.08 ±0.13
2.14 ± 0.15
2.37 ± 0.15c
2.37 ± 0.31c
Kidney/body weight (%)
0.59 ±0.04
0.61 ±0.03
0.70 ± 0.05c
0.79 ± 0.06c
Histopathology
Centrilobular hepatocyte hypertrophy
1/10
1/10
6/10d
8/10°
Hyaline droplet nephropathy
(severity)
8/10(1.5)
9/10(1.7)
10/10(3.1)
10/10(3.2)
Females
Clinical Chemistry
ALT (ukat/L)
0.58 ± 0.18
0.55 ±0.08
0.60 ±0.12
0.73 ± 0.19d
Total bilirubin (nmol/L)
2.75 ±0.39
2.56 ±0.42
2.94 ±0.37
3.65 ± 0.64c
Total protein (g/L)
71.8 ± 2.6
71.3 ±3.0
72.4 ±2.6
75.6 ± 3.ld
Albumin (g/L)
40.2 ± 1.5
39.9 ±1.5
40.9 ± 1.3
41.8 ± 1.5d
Globulins (g/L)
31.6 ±1.3
31.4 ±1.8
31.6 ± 2.0
33.8 ± 1.9°
Cholesterol (mmol/L)
1.35 ±0.27
1.41 ±0.31
1.81 ± 0.31°
2.16 ± 0.20c
Prothrombin time (s)
28.4 ±0.8
28.3 ±2.2
27.0 ± 1.1°
26.3 ± 1.8d
Sodium (mmol/L)
141.4 ± 1.4
140.6 ± 1.3
141.5 ±0.8
139.1 ± 1.0C
Potassium (mmol/L)
4.28 ±0.21
4.38 ±0.29
4.28 ± 0.27
4.62 ± 0.21c
Magnesium (mmol/L)
0.97 ±0.06
0.99 ±0.06
0.97 ±0.04
1.04 ± 0.03c
Body Weight on Day 91(% change
from control)
222 g
+3.1%
-1.6%
-1.4%
Organ Weights
Absolute liver weight (g)
5.40 ±0.30
5.72 ±0.53
6.11 ± 0.36°
7.15 ± 0.50c
Liver/body weight (%)
2.63 ±0.13
2.70 ±0.16
3.03 ± 0.12c
3.52 ± 0.18°
Kidney/body weight (%)
0.67 ±0.03
0.68 ±0.04
0.72 ± 0.03c
0.76 ± 0.03c
Histopathology
Centrilobular hepatocyte hypertrophy
(incidence)
0/10
0/10
5/10d
10/10°
aMellert et al., 2007
bMean ± standard deviation
><0.01
Sp < 0.05
eIncidence of effect; statistical analysis conducted for this review using Fisher's exact test
5

-------
FINAL
9-10-2009
on the kidney unrelated to hyaline droplet nephropathy. However, given that liver effects were
observed in the same dose range as the kidney effects and the possible role of a2u-globulin
accumulation in the kidney effects observed in male rats at the LOAEL and the limited effects
observed in female rats even at the high-dose of ethylbenzene, kidney effects were not
considered as the basis for the subchronic p-RfD.
Several measures of liver toxicity were significantly affected (p < 0.01) at the LOAEL:
incidences of centrilobular hepatocyte hypertrophy (males and females); absolute and relative
liver and kidney weights (males and females); and serum alanine aminotransferase [ALT]
(males), gamma glutamyl transferase [GGT] (males), bilirubin (males); and cholesterol (males
and females). Examination of these clinical chemistry findings and organ weight changes
suggests that male rats may be slightly more sensitive to the liver effects of ethylbenzene than
females, as there were more significant findings at the LOAEL in males than in females.
Furthermore, a 4-fold increase in GGT was observed in male rats exposed at the LOAEL, while
no change in GGT was observed in female rats at this dose. GGT is a sensitive indicator of liver
toxicity (U.S. EPA, 2002). Based on these observations, the data on liver changes in male rats
were considered for BMD modeling.
Endpoints to which benchmark dose modeling was applied include the following: GGT,
bilirubin, cholesterol, absolute and relative liver weight, and incidence of centrilobular
hepatocyte hypertrophy in male rats (see Table 1 for data). Biologically relevant benchmark
response (BMR) values for the continuous endpoints (serum chemistry changes and liver weight)
were not located; thus, the default BMR of 1 standard deviation (SD) from the control mean
(U.S. EPA, 2000) was used for these endpoints. The BMR used for modeling incidence of
centrilobular hepatocyte hypertrophy was the default value of 10% increase over the control
incidence. Serum ALT in males was not modeled because the observed increases, while
statistically significant (p < 0.01), were less than 2-fold increase, and were of unknown
biological significance. Because body weights were significantly reduced (14%; p < 0.01) in the
high-dose males, this dose group was not used in modeling of absolute and relative liver weight,
as the liver weights were confounded by body weight changes. However, even without the
high-dose group, efforts to apply benchmark dose modeling to the data on relative liver weight
were not successful (the model failed to converge and no results were produced). In addition,
modeling of cholesterol changes did not result in any model fit (see Appendix A).
Details of the benchmark dose modeling and results, as well as graphs of the best-fitting
model for each endpoint, are provided in Appendix A. Benchmark dose modeling of serum
GGT in male rats resulted in model fit using the linear model with modeled variance. The
BMD isd (benchmark dose associated with 1SD from the control mean response) and BIVIDLisd
(lower confidence limit on this benchmark dose) calculated from these data were 96 and
53 mg/kg-day, respectively. Model fit was achieved using the linear model with modeled
variance for the data on total serum bilirubin. The BMDisd and BMDLisd calculated from these
data were 105 and 62 mg/kg-day, respectively. Modeling of absolute liver weight gave
reasonable fit using the linear model with homogenous variance. The BMDisd and BMDLisd
predicted by the linear model were 84 and 63 mg/kg-day, respectively. BMD modeling of
centrilobular hepatocyte hypertrophy in male rats resulted in model fit for several quantal
models. The log-probit model provided the best fit and the BMDio and BMDLio predicted by the
6

-------
FINAL
9-10-2009
log-probit model for the data on centrilobular hepatocyte hypertrophy in male rats were 79 and
48 mg/kg-day (respectively). Table 2 shows the BMDs and BMDLs calculated from each of the
liver toxicity endpoints.
Table 2. Comparison of BMDs and BMDLs Predicted by Modeling of Liver Effect
Endpoints in Male Rats
Endpoint Modeled
Best-fitting Model
BMD (mg/kg-
day)
BMDL
(mg/kg-day)
Serum GGT
Linear (modeled variance)
95.95
53.06
Total Serum Bilirubin
Linear (modeled variance)
105.43
62.04
Absolute Liver Weight
Linear (constant variance)
83.80
63.30
Incidence of Centrilobular
Hepatocyte Hypertrophy
Log-probit
78.95
48.26
The lowest BMDL (48 mg/kg-day), derived from modeling centrilobular hepatocyte
hypertrophy in male rats, was used as the point of departure (POD) for subchronic p-RfD
derivation. Using this BMDL as the POD is expected to provide protection against potential
kidney effects, since there was no evidence of kidney effects at the NOAEL. The subchronic
p-RfD for ethylbenzene is derived as follows:
Subchronic p-RfD = BMDLio UF
= 48 mg/kg-day ^ 1,000
= 0.05 mg/kg-day or 5 x 10"2 mg/kg-day
A composite Uncertainty Factor (UF) of 1,000 was applied to the BMDLioto calculate
the subchronic p-RfD for ethylbenzene. The composite UF included a factor of 10 for
interspecies extrapolation, a factor of 10 for human variability, and a factor of 10 for database
uncertainties, as follows:
•	A full UFa of 10 was applied for interspecies extrapolation to account for potential
pharmacokinetic and pharmacodynamic differences between rats and humans. There are
no data to determine whether humans are more or less sensitive than rats to the liver
and/or kidney toxicity of ethylbenzene.
•	A full UFa of 10 was applied for intraspecies differences to account for potentially
susceptible individuals in the absence of information on the variability of response in
humans.
•	A full UFd of 10 was applied to account for database uncertainty. There are only two
subchronic studies of oral exposure to ethylbenzene (i.e., Mellert et al., 2007;
Wolf et al., 1956) and no oral studies of developmental or reproductive toxicity. Further,
studies of inhaled ethylbenzene have identified ototoxicity as the most sensitive endpoint
(see subchronic p-RfC derivation below). A short-term (2-week) study of the ototoxicity
of orally administered ethylbenzene (Gagnaire and Langlais, 2005) reported
histopathological evidence of ototoxicity at the only dose tested, 8.47 mmol/kg-day
(900 mg/kg-day), which indicates that this endpoint may be relevant to oral exposure but
cannot be evaluated with current information.
7

-------
FINAL
9-10-2009
Confidence in the principal study (Mellert et al., 2007) is high because the study tested
10 rats per sex at 4 dose levels (including controls), and a broad array of endpoints was
evaluated. Confidence in the database is low reflecting the limited oral toxicity data (only two
subchronic studies), the lack of multigeneration reproductive and developmental toxicity studies,
and the lack of information on potential ototoxicity from oral exposure. Reflecting high
confidence in the principal study and low confidence in the database, confidence in the
provisional subchronic p-RfD is medium.
Comparison of the subchronic p-RfD (0.05 mg/kg-day) with the chronic RfD for
ethylbenzene (0.1 mg/kg-day) on IRIS indicates that the subchronic p-RfD is lower than
(one-half of) the existing chronic RfD. The chronic RfD for ethylbenzene that is currently
posted on IRIS was derived in 1985 using U.S. EPA guidance and methods that have since been
updated and revised. The subchronic p-RfD for ethylbenzene is derived using a new study and
current U.S. EPA guidance and methods, resulting in a lower subchronic value. The chronic
RfD for ethylbenzene on IRIS is currently being reassessed (see IRIS Track at
www.epa.gov/iris); when the reassessment is complete, the chronic RfD will also reflect the new
data and the use of current EPA guidance and methods.
Subchronic p-RfC
The chronic RfC for ethylbenzene (1 mg/m3) on IRIS was verified in December 1990
based on developmental toxicity studies in rats and rabbits exposed during gestation or for
3 weeks prior to gestation and during gestation (Andrew et al., 1981; Hardin et al., 1981), and it
is supported by subchronic and chronic studies in several species (NTP, 1989, 1990;
Cragg et al., 1989; Elovaara et al., 1985; Clark, 1983; Wolf et al., 1956). No UF for exposure
duration was used. The intermediate duration inhalation MRL (1 ppm, or 4.3 mg/m3) was
derived in 1999 and was also based on Andrew et al. (1981).
A number of inhalation studies in animals were identified in the update literature
searches: a multigeneration reproductive toxicity study in rats (Faber et al., 2006, 2007), three
developmental toxicity studies in rats (Saillenfait et al., 2003, 2006, 2007), and four studies of
the ototoxic effects of ethylbenzene in rats (Cappaert et al., 1999, 2000, 2001, 2002;
Gagnaire et al., 2007). Of the ototoxicity studies, only Gagnaire et al. (2007) employed a
subchronic exposure duration (13 weeks). The other ototoxicity studies (Cappaert et al., 1999,
2000, 2002) were of short-term duration (5 days) and, thus, were not considered pertinent to the
derivation of a subchronic RfC.
Table 3 shows a comparison of the recent studies identified through the literature search
and the developmental toxicity study that was used as the basis of both the chronic RfC and the
intermediate duration inhalation MRL for ethylbenzene. As the table shows, the recent
reproductive and developmental toxicity studies (Saillenfait et al., 2003, 2006, 2007;
Faber et al., 2006, 2007) support the LOAEL (1,000 ppm or 4,340 mg/m3) identified in the study
by Andrew et al. (1981). In contrast, the ototoxic effects were observed at a lower concentration
than developmental toxicity; Gagnaire et al. (2007) identified a LOAEL of 200 ppm (868 mg/m3)
for persistent ototoxic effects. Ototoxicity studies of shorter duration (Cappaert et al., 1999,
2000, 2001, 2002) identify LOAELs in the range of 300-400 ppm, providing support for the
sensitivity of this endpoint when compared with developmental toxicity. Thus, the study by
Gagnaire et al. (2007) was selected as the critical study for derivation of the subchronic p-RfC.
8

-------
FINAL
9-10-2009
Table 3. Comparison of Recent (1999-2007) Inhalation Studies with Critical Study Used for Chronic RfC and
Intermediate MRL Derivation for Ethylbenzene
Species
Sex
Exposure
Concentration
(ppm)
Exposure
NOAEL
(ppm)
LOAEL
(ppm)
Responses
Comments
Reference
Rat
Subchronic
Ototoxicity
Study
M
0, 200, 400, 600,
and 800
6 hr/d, 6 d/wk for
13 wks
NA
200
Minimal LOAEL; loss of 3rd
row outer hair cells from organ
of Corti
Increased audiometric thresholds
were observed at >400 ppm
Gagnaire et al., 2007
Rats
Developmental
Toxicity Study
F
0, 100, 500,
1,000, and 2,000
6 hr/d during
GD 6-20.
500 (maternal
and fetal)
1,000 (maternal
and fetal)
Reduced weight gain (maternal)
Reduced fetal body weight
(fetal)

Saillenfait et al., 2003
Rats
Developmental
Toxicity Study
F
0, 250, and
1,000
6 hr/d during
GD 6-20.
250 (maternal
and fetal)
1,000 (maternal
and fetal)
Reduced weight gain (maternal)
Reduced fetal body weight
(fetal)
Data collected as part of a study
on interaction with methyl ethyl
ketone
Saillenfait et al., 2006
Rats
Developmental
Toxicity Study
F
0, 250, and
1,000
6 hr/d during
GD 6-20.
250 (maternal
and fetal)
1,000 (maternal
and fetal)
Reduced weight gain (maternal)
Reduced fetal body weight
(fetal)
Data collected as part of a study
on interaction with butyl acetate
Saillenfait et al., 2007
Rats
2-generation
Reproductive
Toxicity
M/F
0, 25, 100, and
500
6 hr/d for at least
70 days
premating,
through mating
and gestation for
2 generations
500
NA

Reduced estrous cycle length
observed in F0, but there was no
effect on fertility or time to
mating and Fi females were not
affected at this concentration
Faber et al., 2006,
2007
Rat and rabbit
Developmental
Toxicity
M/F
0, 100, and
1,000
7 hr/d, 5 d/wk for
3 weeks
premating,
through mating
pregnancy daily
through GD 19.
100 (maternal
and fetal)
1,000 (maternal
and fetal)
Increased incidence
supernumerary ribs in rats;
slightly reduced litter size in
rabbits
This study was used for the IRIS
chronic RfC and ATSDR
intermediate MRL. A weight of
evidence approach was used by
IRIS to identify the LOAEL
based on a cluster of mild effects
in rats and rabbits
Andrew et al., 1981;
Hardin etal., 1981
9

-------
FINAL
9-10-2009
Gagnaire et al. (2007) exposed groups of 14 male Sprague-Dawley rats to ethylbenzene
(99% pure) vapors (whole body exposure) at concentrations of 0, 200, 400, 600, or 800 ppm
6 hours/day, 6 days/week for 13 weeks followed by 8 untreated weeks. The rats were about
14 weeks of age at the time testing commenced. Mortality was monitored and body weights
were recorded weekly. Auditory thresholds at different sound frequencies (2, 4, 8, and 16 kHz)
were measured by brainstem auditory-evoked responses (using surgically implanted electrodes
and a computerized recording device) assessed at the end of the 4th, 8th, and 13th weeks of
exposure and at the end of the 8-week recovery period. After the 8 untreated weeks,
8 rats/exposure-concentration were sacrificed for microscopic examination of the organ of Corti.
The microscopic examination was used to quantify loss of outer hair cells in the organ of Corti;
these results were presented as histocochleograms (graphs of cell loss of inner hair cells and the
three rows of outer hair cells).
A single rat in the 800-ppm group died of unknown causes and a second was sacrificed
after developing a large tumor on the neck (Gagnaire et al., 2007). A third rat lost its head plug
and could not undergo audiometric threshold testing. Ethylbenzene treatment did not affect
body-weight gain in any group. Audiometric thresholds at all four frequencies were statistically
significantly (p < 0.05) increased over controls in groups exposed to 400 ppm and higher
beginning in the 4th week of exposure. The magnitudes of the threshold shifts, depending on
frequency, exhibited some dose-dependency, ranging from 23 to 27 decibels (dB) in the
400-ppm group and from 44 to 49 dB in the 600- and 800-ppm groups. The threshold increases
observed at 4 weeks did not change with additional exposure and persisted through the 8-week
untreated period, with no evidence of recovery. There was no change in audiometric threshold in
the rats exposed to 200-ppm ethylbenzene. Microscopic examination of the organs of Corti
revealed significant dose-related and, in some cases, marked losses of both inner and outer hair
cells. There was no evidence of biological significant hair cell loss in the controls. At the
highest concentrations (600 and 800 ppm), there was nearly complete loss of all three rows of
outer hair cells, as well as less marked inner hair cell loss (14% and 32% in the 600 and 800 ppm
groups, respectively). At 400 ppm, there was limited loss of the inner hair cells, but still marked
loss of outer hair cells, especially in the third row. At 200 ppm, significant outer hair cell loss
(up to 30%) in the mid-frequency region) was observed in the third row in 4/8 rats examined.
This study identified a minimal LOAEL of 200 ppm for histopathological evidence of
ototoxicity without functional changes in audiometric threshold; no NOAEL can be determined
from these data. Because a NOAEL was not identified, Gagnaire et al. (2007) calculated
theoretical lowest adverse effect levels2 (TLAELs) based on the upper confidence limits of the
average hair cell losses observed in the controls. TLAELs were calculated to be 114, 120, and
130 ppm for the 95, 99, and 99.9% upper confidence limits.
The LOAEL identified from the data reported by Gagnaire et al. (2007) was associated
with histopathological evidence of ototoxicity (loss of outer hair cells). The data for this
endpoint were reported graphically and were not amenable to BMD modeling. Thus, the
2To calculate the TLAELs, the concentration-response relationship (mean cell loss in the third row of the OHC
versus exposure concentration) was fitted using a logistic regression model. The regression analysis was used to
estimate the concentrations associated with the upper confidence limits (95%, 99%, and 99.9%) on the control mean
response; these concentrations were termed the TLAELs. The study authors did not report the parameters of the
regression model.
10

-------
FINAL
9-10-2009
LOAEL of 200 ppm (868 mg/m3) was used as the POD for derivation of the subchronic p-RfC.
No adjustment for continuous exposure was made because available data indicate that inhaled
ethylbenzene is rapidly absorbed, metabolized, and excreted through the urine (ATSDR, 1999).
As a result, effects of inhaled exposure are considered to be more correlated with concentration
than with duration of exposure. Studies of 5-day exposures (Cappaert et al., 1999, 2000, 2001,
2002) identify ototoxicity LOAELs only slightly higher (300-400 ppm) than the subchronic
study (200 ppm; Gagnaire et al., 2007), providing support for a minimal effect of exposure
duration on otoxicity.
The LOAEL was converted to a human equivalent concentration (LOAELHec) based on
the guidance provided in U.S. EPA (1994). It is not clear from the available information whether
exposure to the inner ear of the rats occurred primarily via direct contact or via absorption in the
lungs and transport via the bloodstream. However, because ototoxicity is an extrarespiratory
effect, ethylbenzene was treated as a Category 3 gas, and the ratio of blood:gas partition
coefficients was used to make the dosimetric adjustment, as shown below:
LOAELhec = LOAEL x [(Hb/g)A/(Hb/g)A]
Where: (Hb/g)A = blood/gas partition coefficient in rats
(Hb/g)H = blood/gas partition coefficient in humans
Abraham et al. (2005) reported human and rat blood:gas partition coefficients of 28 and
30, respectively, for ethylbenzene. Because (Hb/g)A > (Hb/g)H, a default value of 1 was used for
the animal-to-human blood:gas ratio in accordance with U.S. EPA (1994) guidance. Thus, the
LOAELhec is equal to 868 mg/m3, calculated as follows:
LOAELhec = LOAEL x [(Hb/g)A/(Hb/g)A]
= 868 mg/m3 x 1
= 868 mg/m3
This value would be the same if no dosimetric adjustment was made under the assumption that
exposure to the inner ear occurred via direct contact. The subchronic p-RfC for ethylbenzene is
derived as follows:
Subchronic p-RfC = LOAEL UF
= 868 mg/m3 100
= 9 or 9 x 10° mg/m3
The composite UF of 100 includes the following:
•	A partial UFA of 3 (10°5) was applied for interspecies extrapolation to account for
potential toxicodynamic differences between rats and humans when a dosimetric
adjustment is used.
•	A full UFh of 10 was used to account for intraspecies differences for potentially
susceptible individuals in the absence of information on the variability of
response in humans.
11

-------
FINAL
9-10-2009
•	A partial UFL of 3 (10°5) was applied for use of a minimal LOAEL. The effects
observed at the LOAEL consisted of histopathological evidence of limited outer
hair cell loss in the third row only and in four/eight rats, without functional
changes in auditory threshold. Further, Gagnaire et al. (2007) estimated TLAELs
in the range of 114-130 ppm based on the statistical upper confidence limits on
outer hair cell loss in the control group. These values are in the range of one-half
the LOAEL, providing further support for a partial UF for LOAEL-to-NOAEL
extrapolation.
•	A UFd of 1 was applied for database uncertainty. No database UF was required,
because the toxicological database for inhaled ethylbenzene includes high-quality
subchronic bioassays, as well as developmental toxicity and multi-generation
reproduction studies and a number of studies of ototoxicity.
Confidence in the principal study is medium. Gagnaire et al. (2007) is an adequate
subchronic oral toxicity study using a sufficient number of animals, an appropriate range of
exposure levels and measuring sensitive endpoints, but the study used only one gender (males)
and did not identify a NOAEL. Confidence in the database is high. The animal database
contains high quality studies on a variety of endpoints and in multiple species. Further, there are
some limited data suggesting that the critical effect (ototoxicity) is relevant to humans; hearing
loss has been reported in solvent abusers and in workers exposed to both solvents and sound,
which may interact synergistically (Gagnaire et al., 2007). Confidence in the subchronic p-RfC
is, therefore, medium.
Table 4 summarizes the subchronic noncancer assessments for ethylbenzene.
Table 4. Summary of Subchronic Noncancer Reference Values for Ethylbenzene

POD Type
POD
UF
Reference
Value
Critical Effect
Species/
Sex
Principal Study
p-sRfD
BMDLio
48
mg/kg-day
1,000
5 x 10"2
mg/kg-day
Centrilobular
hepatocyte
hypertrophy
Rat/M
Mellert et al., 2007
p-sRfC
LOAELhec
868
mg/m3
100
9 x 10°
mg/m3
Ototoxicity
Rat/M
Gagnaire et al., 2007
12

-------
FINAL
9-10-2009
REFERENCES
Abraham, M.H., A. Ibrahim and W.E. Acree, Jr. 2005. Air to blood distribution of volatile
organic compounds: A linear free energy analysis. Chem. Res. Toxicol. 18(5):904—911.
Andrew, F.D., R.L. Buschbom, W.C. Cannon et al. 1981. Teratologic assessment of
ethylbenzene and 2-ethoxyethanol. Battelle Pacific Northwest Laboratory, Richland, WA.
PB83- 208074.
ATSDR (Agency for Toxic Substances and Disease Registry). 1999. Toxicological Profile for
Ethylbenzene. Agency for Toxic Substances and Disease Registry, Public Health Service, U.S.
Department of Health and Human Services. PB/99/166647. Online, http://www.ntis.gov/
search/product. aspx?ABBR=PB99166647.
ATSDR (Agency for Toxic Substances and Disease Registry). 2007. Toxicological profile for
Ethylbenzene (Draft for Public Comment). Atlanta, GA: U.S. Department of Health and Human
Services, Public Health Service. Online. http://www.atsdr.cdc.gov/toxprofiles/tpllO.html.
Cappaert, N.L.M., S.F.L. Klis, H. Muijser et al. 1999. The ototoxic effects of ethyl benzene in
rats. Hear. Res. 137:91-102.
Cappaert, N.L.M., S.F.L. Klis, A.B. Baretta et al. 2000. Ethyl benzene-induced ototoxicity in
rats: A dose-dependent mid-frequency hearing loss. J. Assoc. Res. Otolaryngol. l(4):292-299.
Cappaert, N.L.M., S.F.L. Klis, H. Muijser et al. 2001. Simultaneous exposure to ethyl benzene
and noise: Synergistic effects on outer hair cells. Hear. Res. 162(1-2):67-79.
Cappaert, N.L.M., S.F.L. Klis, H. Muijser et al. 2002. Differential susceptibility of rats and
guinea pigs to the ototoxic effects of ethyl benzene. Neurotoxicol. Teratol. 24(4):503-510.
Clark, D.G. 1983. Ethylbenzene hydroperoxide (EBHP) and ethylbenzene (EB): 12 week
inhalation study in rats. (Group research report with attachments and cover sheet.) EPA OTS
Public Files. Shell Oil Co. Document No. 86870001629. Fiche Number 0516206 (2). (As cited
in U.S. EPA, 1991a IRIS Record for Ethylbenzene.
Cragg, S.T., E.A. Clarke, I.W. Daly et al. 1989. Subchronic inhalation toxicity of ethylbenzene
in mice, rats, and rabbits. Fund. Appl. Toxicol. 13(3):399-408. (As cited in U.S. EPA, 1991a
IRIS Record for Ethylbenzene.)
Elovaara, E., K. Engstrom, J. Nickels et al. 1985. Biochemical and morphological effects of
long-term inhalation exposure of rats to ethylbenzene. Xenobiotica. 15(4):299-308. (As cited
in U.S. EPA, 1991a IRIS Record for Ethylbenzene.)
Faber, W.D., L.S.G. Roberts, D.G. Stump et al. 2006. Two generation reproduction study of
ethylbenzene by inhalation in Crl-CD rats. Birth Def. Res. B Dev. Reprod. Toxicol. 77(1):
10-21.
13

-------
FINAL
9-10-2009
Faber, W.D., L.S.G. Roberts, D.G. Stump et al. 2007. Inhalation developmental neurotoxicity
study of ethylbenzene in Crl-CD rats. Birth Def. Res. B Dev. Reprod. Toxicol. 80:34-48.
Gagnaire, F. and C. Langlais. 2005. Relative ototoxicity of 21 aromatic solvents. Arch.
Toxicol. 79(6):346-354.
Gagnaire, F., C. Langlais, S. Grossmann et al. 2007. Ototoxicity in rats exposed to ethylbenzene
and to two technical xylene vapours for 13 weeks. Arch. Toxicol. 81(2): 127-143.
Hardin, B.D., G.P. Bond, M.R. Sikov et al. 1981. Testing of selected workplace chemicals for
teratogenic potential. Scand. J. Work Environ. Health. 7(Suppl 4):66-75.
MADEP (Massachusetts Department of Environmental Protection). 2003. Updated Petroleum
Hydrocarbon Fraction Toxicity Values for the VPH/EPH/APH Methodology. Prepared by the
Office of Research and Standards, MADEP, Boston MA. November.
Mellert, W., K. Deckardt, W. Kaufmann et al. 2007. Ethylbenzene: 4- and 13-week rat oral
toxicity. Arch. Toxicol. 81:361-370.
NTP (National Toxicology Program). 1989. Chairperson's report. Pathology Working Group
(PWG) review of subchronic toxicity testing on ethylbenzene administered by inhalation in F344
rats and B6C3F1 mice. (As cited in U.S. EPA, 1991a IRIS Record for Ethylbenzene.)
NTP (National Toxicology Program). 1990. Draft NTP technical report on the toxicity studies
of ethylbenzene in F344 rats and B6C3F1 mice (inhalation studies). (As cited in U.S. EPA,
1991a IRIS Record for Ethylbenzene.)
Saillenfait, A.M., F. Gallissot, G. Morel et al. 2003. Developmental toxicities of ethylbenzene,
ortho-, meta-, para-xylene and technical xylenes in rats following inhalation exposure. Food
Chem. Toxicol. 41:415-429.
Saillenfait, A.M., F. Gallissot, J.P. Sabate et al. 2006. Developmental toxicity of combined
ethylbenzene and methylethylketone administered by inhalation to rats. Food Chem. Toxicol.
44(8): 1287-1298.
Saillenfait, A.M., F. Gallissot, J.P. Sabate et al. 2007. Developmental toxic effects of
ethylbenzene to toluene alone and in combination with butyl acetate in rats after inhalation
exposure. J. Appl. Toxicol. 27(l):32-42.
U.S. EPA (Environmental Protection Agency). 1991a. Integrated Risk Information System
(IRIS). IRIS Summary of Ethylbenzene (CASRN 100-41-4). Office of Research and
Development, National Center for Environmental Assessment, Washington, DC. Available
online at http://www.epa.gov/iris/.
U.S. EPA. 1991b. Alpha 2u-globulin: Association with Chemically Induced Renal Toxicity
and Neoplasia in the Male Rat. Risk Assessment Forum, Washington, DC. EPA/625/3-91/019F.
14

-------
FINAL
9-10-2009
U.S. EPA. 1994. Methods for Derivation of Inhalation Reference Concentrations and
Application of Inhalation Dosimetry. U.S. Environmental Protection Agency, Office of Research
and Development, Office of Health and Environmental Assessment, Washington, DC,
EPA/600/8-90/066F.
U.S. EPA. 2000. Benchmark Dose Technical Guidance Document [external review draft],
EPA/630/R-00/001. Online, http://www.epa.gov/iris/backgr-d.htm.
U.S. EPA. 2002. Hepatocellular Hypertrophy. HED Guidance Document #G2002.01. U.S.
Environmental Protection Agency, Office of Pesticide Programs, Health Effects Division,
Washington, DC.
Wolf, M.A., V.K. Rowe, D.D. McCollister et al. 1956. Toxicological studies of certain
alkylated benzenes and benzene. Arch. Ind. Health. 14:387-398.
15

-------
FINAL
9-10-2009
APPENDIX A. DETAILS OF BENCHMARK DOSE MODELING FOR
SUBCHRONIC ORAL RfD
Modeling Procedure Continuous Data Modeling
The model fitting procedure for continuous data is as follows. When a
biologically-defined BMR is not available, the default BMR of 1 standard deviation from the
control mean response is used (U.S. EPA, 2000). The simplest model (linear) is first applied to
the data while assuming constant variance. If the data are consistent with the assumption of
constant variance (p> 0.1), then the fit of the linear model to the means is evaluated. If the
linear model adequately fits the means (p> 0.1), then it is selected as the model for BMD
derivation. If the linear model does not adequately fit the means, then the more complex models
are fit to the data while assuming constant variance. Among the models providing adequate fit to
the means (p> 0.1), the one with the lowest AIC for the fitted model is selected for BMD
derivation. If the test for constant variance is negative, the linear model is run again while
applying the power model integrated into the BMDS to account for nonhomogenous variance. If
the nonhomogenous variance model provides an adequate fit (p > 0.1) to the variance data, then
the fit of the linear model to the means is evaluated. If the linear model does not provide
adequate fit to the means while the nonhomogenous variance model is applied, then the
polynomial, power and Hill models are fit to the data and evaluated while the variance model is
applied. Among those providing adequate fit to the means (p> 0.1), the one with the lowest AIC
for the fitted model is selected for BMD derivation. If the test for constant variance is negative
and the nonhomogenous variance model does not provide an adequate fit to the variance data,
then the data set is considered unsuitable for modeling.
Modeling of Data on Serum GGT in Male Rats
Following the above procedure, continuous-variable models in the EPA BMDS
(version 1.3.2) were fit to the data shown in Table 2 for increased serum GGT in male rats
(Mellert et al., 2007) using a default BMR of 1 standard deviation from the control mean. Using
these data, the constant variance model provided adequate fit to the variance data. However, the
linear model with constant variance did not provide an adequate fit to the means, as shown in
Table A-l. Further, none of the remaining models provided adequate fit to the means (there
were not enough dose groups to apply the Hill model). In order to achieve model fit, the
high-dose group was dropped from the analysis. With the reduced data set, the homogenous
variance model did not fit the variance data adequately. With the modeled variance, the linear
model provided adequate fit to the means (Figure A-l). The BMDs and the 95% lower
confidence limits (BMDLs) associated with a change of 1 standard deviation (SD) from the
control were calculated using the linear model with modeled variance.
16

-------
FINAL
9-10-2009
Table A-l. Model Predictions for Serum GGT in Male Rats Exposed Orally to

Ethylbenzene for 13 Weeks3



Variance
Means
bmd1sd
BMDL1sd
Model
/7-Valueb
/7-Valueb
(mg/kg-day)
(mg/kg-day)
All dose groups
Linear (constant variance)0
0.1352
0.03645
651.15
404.85
Polynomial (constant variance)0'd
0.1352
0.01006
651.15
404.85
Power (constant variance)0
0.1352
0.01006
651.15
404.85
Hill (constant variance)0
NAf
NA
NA
NA
Without high-dose group
Linear (constant variance)0
0.07308
0.4167
164.06
108.12
Linear (modeled variance)0
0.4848
0.1416
95.95
53.06
aMellert et al., 2007
bValues <0.10 fail to meet conventional goodness-of-fit criteria
Coefficients restricted to be positive
d2-degree polynomial selected
"Power restricted to >1
fNA = not applicable (insufficient dose groups available to apply this model)
17

-------
FINAL
9-10-2009
Linear Model with 0.95 Confidence Level
Dose
10:29 08/10 2007
BMDs and BMDLs indicated are associated with a change of 1 SD from the control and are in units of mg/kg-day.
Figure A-l. Fit of Linear Model (Modeled Variance) to Data on Serum GGT in Male Rats
(Mellert et al., 2007)
18

-------
FINAL
9-10-2009
Modeling of Data on Total Serum Bilirubin in Male Rats
Following the above procedure, continuous-variable models in the EPA BMDS
(version 1.3.2) were fit to the data shown in Table 2 for increased total serum bilirubin in male
rats (Mellert et al., 2007) using a default BMR of 1 standard deviation from the control mean.
Using these data, the constant variance model did not provide adequate fit to the variance data.
When the variance was modeled using the power model in the BMDS, the linear model did not
provide an adequate fit to the means, as shown in Table A-2. Further, none of the remaining
models provided adequate fit to the means (there were not enough dose groups to apply the Hill
model). In order to achieve model fit, the high-dose group was dropped from the analysis. With
the reduced data set, the homogenous variance model did not fit the variance data adequately.
With the modeled variance, the linear model provided adequate fit to the means (Figure A-2).
The BMDs and the 95% lower confidence limits (BMDLs) associated with a change of
1 standard deviation (SD) from the control were calculated using the linear model with modeled
variance.
Table A-2. Model Predictions for Total Serum Bilirubin in Male Rats Exposed Orally

Ethylbenzene for 13 Weeks3



Variance
Means
bmd1sd
BMDL1sd
Model
p-V alueb
p-V alueb
(mg/kg-day)
(mg/kg-day)
All dose groups
Linear (constant variance)0
0.03826
0.1883
427.04
300.05
Linear (modeled variance)0
0.7171
0.04065
252.03
141.00
Polynomial (modeled variance)0'd
0.7171
0.01138
252.03
141.00
Power (constant variance)0
0.717
0.01138
252.03
141.00
Hill (constant variance)0
NAf
NA
NA
NA
Without high-dose group
Linear (constant variance)0
0.0393
0.8397
162.49
107.38
Linear (modeled variance)0
0.5809
0.9404
105.43
62.04
aMellert et al., 2007
bValues <0.10 fail to meet conventional goodness-of-fit criteria
Coefficients restricted to be positive
d2-degree polynomial selected
"Power restricted to >1
fNA = not applicable (insufficient dose groups available to fit this model)
19

-------
FINAL
9-10-2009
Linear Model with 0.95 Confidence Level
Linear
3.5
o
Q.
3
2.5
2
BMDL
BMD
0
50
100
150
200
250
Dose
10:21 08/10 2007
BMDs and BMDLs indicated are associated with a change of 1 SD from the control and are in units of mg/kg-day.
Figure A-2. Fit of Linear Model (Modeled Variance) to Data on Total Serum Bilirubin in
Male Rats (Mellert et al., 2007)
Modeling of Data on Serum Cholesterol in Male Rats
Following the above procedure, continuous-variable models in the EPA BMDS
(version 1.3.2) were fit to the data shown in Table 2 for increased serum cholesterol in male rats
(Mellert et al., 2007) using a default BMR of 1 standard deviation from the control mean. Using
these data, the constant variance model did not provide adequate fit to the variance data. Further,
the variance model included in the BMDS did not provide an adequate fit to the variance, as
shown in Table A-3. In an attempt to achieve model fit, the high-dose group was dropped from
the analysis. However, the results were the same as with the full dataset; neither the
homogenous nor modeled variance options resulted in adequate fit to the variance data. Thus,
this data set was not considered suitable for BMD analysis.
20

-------
FINAL
9-10-2009
Table A-3. Model Predictions for Serum Cholesterol in Male Rats Exposed Orally to

Ethylbenzene for 13 Weeks3


Variance
Means
bmd1sd
BMDLisd
Model
p-\alueb
p-\ alueb
(mg/kg-day)
(mg/kg-day)
All dose groups
Linear (constant variance)0
0.05241
0.0001022
413.53
292.77
Linear (modeled variance)0
0.05377
<0.001
372.14
210.33
Without high-dose group
Linear (constant variance)0
0.02868
0.009652
107.86
78.30
Linear (modeled variance)0
0.01563
0.005013
109.90
77.92
aMellert et al., 2007
bValues <0.10 fail to meet conventional goodness-of-fit criteria
Coefficients restricted to be positive
Modeling of Data on Absolute Liver Weight in Male Rats
Following the above procedure, continuous-variable models in the EPA BMDS
(version 1.3.2) were fit to the data shown in Table 2 for increased absolute liver weight in male
rats (Mellert et al., 2007) using a default BMR of 1 standard deviation from the control mean.
As noted in the text, the high-dose group was excluded from the analysis a priori due to the
confounding effect of reduced body weight on liver-weight changes. Using this reduced data set,
the linear model with constant variance model provided adequate fit to both the variance and
means data (Table A-4 and Figure A-3). The BMDs and the 95% lower confidence limits
(BMDLs) associated with a change of 1 standard deviation (SD) from the control were calculated
using the linear model with constant variance.
21

-------
FINAL
9-10-2009
Table A-4. Model Predictions for Absolute Liver Weight in Male Rats Exposed Orally

to Ethylbenzene for 13 Weeks3


Variance
Means
bmd1sd
BMDL1sd
Model
p-V alueb
p-V alueb
(mg/kg-day)
(mg/kg-day)
Without high-dose group
Linear (constant variance)0
0.2048
0.1617
83.80
63.30
aMellert et al., 2007
bValues <0.10 fail to meet conventional goodness-of-fit criteria
Coefficients restricted to be positive
Linear Model with 0.95 Confidence Level
Dose
09:44 08/10 2007
BMDs and BMDLs indicated are associated with a change of 1 SD from the control and are in units of mg/kg-day.
Figure A-3. Fit of Linear Model (Constant Variance) to Data on Absolute Liver Weight in
Male Rats (Mellert et al., 2007)
22

-------
FINAL
9-10-2009
Modeling Procedure for Dichotomous Data
The benchmark dose (BMD) modeling for dichotomous data was conducted with the
EPA's BMD software (BMDS version 2.1). For all the dichotomous data, the original data were
modeled with all the dichotomous models (i.e., Gamma, Multistage, Logistic, Log-logistic,
Probit, Log-Probit, Weibull, and Quantal linear models) available within the software with a
default benchmark response (BMR) of 10% extra risk. An adequate fit was judged based on the
goodness of fit /;-value (p > 0.1), scaled residual at the range of benchmark response (BMR), and
visual inspection of the model fit. Among all the models provided adequate data fit, the lowest
BMDL will be selected if the BMDLs estimated from different models if the range is considered
sufficiently large; otherwise, the BMDL from the model with the lowest Akaike's Information
Criterion (AIC) would be considered appropriate for the data set.
Modeling of Data on Centrilobular Hepatocyte Hypertrophy in Male Rats
Table 2 shows the dose-response data for incidence of centrilobular hepatocyte
hypertrophy in male rats (Mellert et al., 2007). These data were modeled according to the
procedure outlined above. As assessed by the % goodness-of-fit test, all models in the software
provided adequate fits to the data for the incidence of centrilobular hepatocyte hypertrophy in
male rats (% p > 0.1) (Table A-5). The Log-probit model provided the best fit, as assessed by
AIC. The fit of the log-probit model to the data is shown in Figure A-4.
23

-------
FINAL
9-10-2009
Table A-5. Model Predictions for Incidence of Centrilobular Hepatocyte
Hypertrophy in the Male Rats Exposed Orally to Ethylbenzene for 13 Weeks3
Model
Degrees
of
Freedom
x2
X2 Goodness
of Fit
/7-Value
AIC
BMC10
(mg/m3)
BMCL10
(mg/m3)
Log Logistic
1
1.04
0.3068
43.55
73.08
18.24
Gamma
1
1.60
0.2059
44.15
61.59
29.13
Multistage (degree of polynomial = l)b
2
1.62
0.4457
42.24
46.8785
28.92
Multistage (degree of polynomial = 2)b
2
1.62
0.4457
42.24
46.8785
28.92
Multistage (degree of polynomial = 3)b
2
1.62
0.4457
42.24
46.8785
28.92
Weibull
1
1.62
0.2026
44.21
54.54
29.00
Quantal Linear
2
1.62
0.4457
42.24
46.88
28.92
Log Probit
2
0.98
0.6119
41.48
78.95
48.26
Probit
2
3.44
0.1792
43.91
112.02
76.32
Logistic
2
3.46
0.1772
43.95
114.34
73.53
aMellert etal.,2007
bDegree of polynomial initially set to (n-1) where n = number of dose groups including control. Betas
restricted to >0.
24

-------
FINAL
9-10-2009
LogProbit Model with 0.95 Confidence Level
Dose
14:37 07/06 2009
BMDs and BMDLs indicated are associated with an extra risk of 10% and are in units of mg/kg-day.
Figure A-4. Fit of Log-Probit Model to Incidence of Centrilobular Hepatocyte
Hypertrophy in the Male Rat (Mellert et al., 2007)
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21Beta\Temp\4tmpllOE.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21Beta\Temp\4tmpllOE.plt
Mon Jul 06 14:44:29 2009
BMDS Model Run
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope*Log(Dose)),
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = Incidence
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 4
25

-------
FINAL
9-10-2009
Total number of records with missing values = 0
Maximum number of iterations = 25 0
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial	(and Specified) Parameter Values
background =	0.1
intercept =	-6.15205
slope =	1.07357
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by
the user,
and do not appear in the correlation matrix )
background intercept
background	1	-0.37
intercept	-0.37	1
Interval
Variable
Limit
background
0.223726
intercept
5.02734
slope
Estimate
0.0810598
-5.65033
1
Parameter Estimates
Std. Err.
0.07279
0.317859
NA
95.0% Wald Confidence
Lower Conf. Limit Upper Conf.
-0.061606
-6.27332
NA - Indicates that this parameter has hit a bound
implied by some ineguality constraint and thus
has no standard error.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
AIC:
Log(likelihood)
-18.2358
-18.741
-26.9205
41.482
# Param's	Deviance	Test d.f.	P-value
4
2	1.01037	2	0.6034
1	17.3693	3	0.0005933
Goodness of Fit
Dose	Est._Prob. Expected Observed	Size
0.0000	0.0811	0.811	1.000	10
Scaled
Residual
0.219
26

-------
Benchmark Dose Computation
Specified effect =
Risk Type	=
Confidence level =
BMD =
BMDL =
0.1
Extra risk
0.95
78.9472
48.2564
FINAL
9-10-2009
75.0000
250.0000
750.0000
Chi^2
0.1650
0.4934
0.8474
1. 650
4.934
8.474
1. 000
6. 000
8.000
0.98
d.f. = 2
P-value
0.6119
10
10
10
-0.553
0.674
-0.417
27

-------